redonkable
/
alistair23-linux
1
0
Fork 0
Code Releases Activity
alistair23-linux/sound/soc/soc-cache.c

1112 lines
27 KiB
C
Raw Normal View History

ASoC: Begin to factor out register cache I/O functions A lot of CODECs share the same register data formats and therefore replicate the code to manage access to and caching of the register map. In order to reduce code duplication centralised versions of this code will be introduced with drivers able to configure the use of the common code by calling the new snd_soc_codec_set_cache_io() API call during startup. As an initial user the 7 bit address/9 bit data format used by many Wolfson devices is supported for write only CODECs and the drivers with straightforward register cache implementations are converted to use it. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2009-07-05 10:24:50 -06:00
/*
* soc-cache.c -- ASoC register cache helpers
*
* Copyright 2009 Wolfson Microelectronics PLC.
*
* Author: Mark Brown <broonie@opensource.wolfsonmicro.com>
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the
* Free Software Foundation; either version 2 of the License, or (at your
* option) any later version.
*/
ASoC: Add I/O control bus information to factored out cache setup While writes tend to be able to use a fairly bus independant format to do the writes reads are all bus specific. To allow us to factor out this code include the bus type as a parameter when setting up the cache. Initially just use this to factor out hw_write_t for I2C. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2009-07-10 15:24:27 -06:00
#include <linux/i2c.h>
ASoC: Factor out 7 bit register 9 bit data SPI write This converts all the Wolfson drivers using this format (the only devices that do) except WM8753 to use it. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2009-07-10 16:28:16 -06:00
#include <linux/spi/spi.h>
ASoC: Begin to factor out register cache I/O functions A lot of CODECs share the same register data formats and therefore replicate the code to manage access to and caching of the register map. In order to reduce code duplication centralised versions of this code will be introduced with drivers able to configure the use of the common code by calling the new snd_soc_codec_set_cache_io() API call during startup. As an initial user the 7 bit address/9 bit data format used by many Wolfson devices is supported for write only CODECs and the drivers with straightforward register cache implementations are converted to use it. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2009-07-05 10:24:50 -06:00
#include <sound/soc.h>
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
#include <linux/lzo.h>
#include <linux/bitmap.h>
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
#include <linux/rbtree.h>
ASoC: Begin to factor out register cache I/O functions A lot of CODECs share the same register data formats and therefore replicate the code to manage access to and caching of the register map. In order to reduce code duplication centralised versions of this code will be introduced with drivers able to configure the use of the common code by calling the new snd_soc_codec_set_cache_io() API call during startup. As an initial user the 7 bit address/9 bit data format used by many Wolfson devices is supported for write only CODECs and the drivers with straightforward register cache implementations are converted to use it. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2009-07-05 10:24:50 -06:00
ASoC: soc-cache: Add trace event for snd_soc_cache_sync() This patch makes it easy to see when the syncing process begins and ends. You can also enable the snd_soc_reg_write tracepoint to see which registers are being synced. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-21 08:29:02 -07:00
#include <trace/events/asoc.h>
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
static bool snd_soc_set_cache_val(void *base, unsigned int idx,
unsigned int val, unsigned int word_size)
{
switch (word_size) {
case 1: {
u8 *cache = base;
if (cache[idx] == val)
return true;
cache[idx] = val;
break;
}
case 2: {
u16 *cache = base;
if (cache[idx] == val)
return true;
cache[idx] = val;
break;
}
default:
BUG();
}
return false;
}
static unsigned int snd_soc_get_cache_val(const void *base, unsigned int idx,
unsigned int word_size)
{
ASoC: Check for NULL register bank in snd_soc_get_cache_val() Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com>
2011-06-06 04:26:15 -06:00
if (!base)
return -1;
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
switch (word_size) {
case 1: {
const u8 *cache = base;
return cache[idx];
}
case 2: {
const u16 *cache = base;
return cache[idx];
}
default:
BUG();
}
/* unreachable */
return -1;
}
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
struct snd_soc_rbtree_node {
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
struct rb_node node; /* the actual rbtree node holding this block */
unsigned int base_reg; /* base register handled by this block */
unsigned int word_size; /* number of bytes needed to represent the register index */
void *block; /* block of adjacent registers */
unsigned int blklen; /* number of registers available in the block */
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
} __attribute__ ((packed));
struct snd_soc_rbtree_ctx {
struct rb_root root;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
struct snd_soc_rbtree_node *cached_rbnode;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
};
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
static inline void snd_soc_rbtree_get_base_top_reg(
struct snd_soc_rbtree_node *rbnode,
unsigned int *base, unsigned int *top)
{
*base = rbnode->base_reg;
*top = rbnode->base_reg + rbnode->blklen - 1;
}
static unsigned int snd_soc_rbtree_get_register(
struct snd_soc_rbtree_node *rbnode, unsigned int idx)
{
unsigned int val;
switch (rbnode->word_size) {
case 1: {
u8 *p = rbnode->block;
val = p[idx];
return val;
}
case 2: {
u16 *p = rbnode->block;
val = p[idx];
return val;
}
default:
BUG();
break;
}
return -1;
}
static void snd_soc_rbtree_set_register(struct snd_soc_rbtree_node *rbnode,
unsigned int idx, unsigned int val)
{
switch (rbnode->word_size) {
case 1: {
u8 *p = rbnode->block;
p[idx] = val;
break;
}
case 2: {
u16 *p = rbnode->block;
p[idx] = val;
break;
}
default:
BUG();
break;
}
}
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
static struct snd_soc_rbtree_node *snd_soc_rbtree_lookup(
struct rb_root *root, unsigned int reg)
{
struct rb_node *node;
struct snd_soc_rbtree_node *rbnode;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int base_reg, top_reg;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
node = root->rb_node;
while (node) {
rbnode = container_of(node, struct snd_soc_rbtree_node, node);
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
snd_soc_rbtree_get_base_top_reg(rbnode, &base_reg, &top_reg);
if (reg >= base_reg && reg <= top_reg)
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
return rbnode;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
else if (reg > top_reg)
node = node->rb_right;
else if (reg < base_reg)
node = node->rb_left;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
return NULL;
}
static int snd_soc_rbtree_insert(struct rb_root *root,
struct snd_soc_rbtree_node *rbnode)
{
struct rb_node **new, *parent;
struct snd_soc_rbtree_node *rbnode_tmp;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int base_reg_tmp, top_reg_tmp;
unsigned int base_reg;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
parent = NULL;
new = &root->rb_node;
while (*new) {
rbnode_tmp = container_of(*new, struct snd_soc_rbtree_node,
node);
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
/* base and top registers of the current rbnode */
snd_soc_rbtree_get_base_top_reg(rbnode_tmp, &base_reg_tmp,
&top_reg_tmp);
/* base register of the rbnode to be added */
base_reg = rbnode->base_reg;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
parent = *new;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
/* if this register has already been inserted, just return */
if (base_reg >= base_reg_tmp &&
base_reg <= top_reg_tmp)
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
return 0;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
else if (base_reg > top_reg_tmp)
new = &((*new)->rb_right);
else if (base_reg < base_reg_tmp)
new = &((*new)->rb_left);
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
/* insert the node into the rbtree */
rb_link_node(&rbnode->node, parent, new);
rb_insert_color(&rbnode->node, root);
return 1;
}
static int snd_soc_rbtree_cache_sync(struct snd_soc_codec *codec)
{
struct snd_soc_rbtree_ctx *rbtree_ctx;
struct rb_node *node;
struct snd_soc_rbtree_node *rbnode;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int regtmp;
ASoC: Suppress restore of default register values for rbtree cache sync Currently the rbtree code will write out the entire register map when doing a cache sync which is wasteful and will slow things down. Check to see if the value we're about to write is the default and don't bother restoring it if it is, either the value will have been retained or the device will have been reset and holds the value already. We should really store the defaults in the nodes but this resolves the immediate issue. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com>
2011-06-03 09:36:30 -06:00
unsigned int val, def;
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
int ret;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
int i;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
rbtree_ctx = codec->reg_cache;
for (node = rb_first(&rbtree_ctx->root); node; node = rb_next(node)) {
rbnode = rb_entry(node, struct snd_soc_rbtree_node, node);
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
for (i = 0; i < rbnode->blklen; ++i) {
regtmp = rbnode->base_reg + i;
val = snd_soc_rbtree_get_register(rbnode, i);
ASoC: Suppress restore of default register values for rbtree cache sync Currently the rbtree code will write out the entire register map when doing a cache sync which is wasteful and will slow things down. Check to see if the value we're about to write is the default and don't bother restoring it if it is, either the value will have been retained or the device will have been reset and holds the value already. We should really store the defaults in the nodes but this resolves the immediate issue. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com>
2011-06-03 09:36:30 -06:00
def = snd_soc_get_cache_val(codec->reg_def_copy, i,
rbnode->word_size);
if (val == def)
continue;
ASoC: Fix register cache sync register_writable WARN_ONs Currently the condition for these WARN_ONs is reversed and they are placed before the actual check whether we are going to write to that register. So if the codec implements the register_writable callback we'll get a warning for each writable register when syncing the register cache. While we are at it change the check to use snd_soc_codec_writable_register instead of open-coding it. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-08-27 10:24:13 -06:00
WARN_ON(!snd_soc_codec_writable_register(codec, regtmp));
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
codec->cache_bypass = 1;
ret = snd_soc_write(codec, regtmp, val);
codec->cache_bypass = 0;
if (ret)
return ret;
dev_dbg(codec->dev, "Synced register %#x, value = %#x\n",
regtmp, val);
}
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
return 0;
}
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
static int snd_soc_rbtree_insert_to_block(struct snd_soc_rbtree_node *rbnode,
unsigned int pos, unsigned int reg,
unsigned int value)
{
u8 *blk;
blk = krealloc(rbnode->block,
(rbnode->blklen + 1) * rbnode->word_size, GFP_KERNEL);
if (!blk)
return -ENOMEM;
/* insert the register value in the correct place in the rbnode block */
memmove(blk + (pos + 1) * rbnode->word_size,
blk + pos * rbnode->word_size,
(rbnode->blklen - pos) * rbnode->word_size);
/* update the rbnode block, its size and the base register */
rbnode->block = blk;
rbnode->blklen++;
if (!pos)
rbnode->base_reg = reg;
snd_soc_rbtree_set_register(rbnode, pos, value);
return 0;
}
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
static int snd_soc_rbtree_cache_write(struct snd_soc_codec *codec,
unsigned int reg, unsigned int value)
{
struct snd_soc_rbtree_ctx *rbtree_ctx;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
struct snd_soc_rbtree_node *rbnode, *rbnode_tmp;
struct rb_node *node;
unsigned int val;
unsigned int reg_tmp;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
unsigned int base_reg, top_reg;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int pos;
int i;
int ret;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
rbtree_ctx = codec->reg_cache;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
/* look up the required register in the cached rbnode */
rbnode = rbtree_ctx->cached_rbnode;
if (rbnode) {
snd_soc_rbtree_get_base_top_reg(rbnode, &base_reg, &top_reg);
if (reg >= base_reg && reg <= top_reg) {
reg_tmp = reg - base_reg;
val = snd_soc_rbtree_get_register(rbnode, reg_tmp);
if (val == value)
return 0;
snd_soc_rbtree_set_register(rbnode, reg_tmp, value);
return 0;
}
}
/* if we can't locate it in the cached rbnode we'll have
* to traverse the rbtree looking for it.
*/
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
rbnode = snd_soc_rbtree_lookup(&rbtree_ctx->root, reg);
if (rbnode) {
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
reg_tmp = reg - rbnode->base_reg;
val = snd_soc_rbtree_get_register(rbnode, reg_tmp);
if (val == value)
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
return 0;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
snd_soc_rbtree_set_register(rbnode, reg_tmp, value);
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
rbtree_ctx->cached_rbnode = rbnode;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
} else {
/* bail out early, no need to create the rbnode yet */
if (!value)
return 0;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
/* look for an adjacent register to the one we are about to add */
for (node = rb_first(&rbtree_ctx->root); node;
node = rb_next(node)) {
rbnode_tmp = rb_entry(node, struct snd_soc_rbtree_node, node);
for (i = 0; i < rbnode_tmp->blklen; ++i) {
reg_tmp = rbnode_tmp->base_reg + i;
if (abs(reg_tmp - reg) != 1)
continue;
/* decide where in the block to place our register */
if (reg_tmp + 1 == reg)
pos = i + 1;
else
pos = i;
ret = snd_soc_rbtree_insert_to_block(rbnode_tmp, pos,
reg, value);
if (ret)
return ret;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
rbtree_ctx->cached_rbnode = rbnode_tmp;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
return 0;
}
}
/* we did not manage to find a place to insert it in an existing
* block so create a new rbnode with a single register in its block.
* This block will get populated further if any other adjacent
* registers get modified in the future.
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
*/
rbnode = kzalloc(sizeof *rbnode, GFP_KERNEL);
if (!rbnode)
return -ENOMEM;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
rbnode->blklen = 1;
rbnode->base_reg = reg;
rbnode->word_size = codec->driver->reg_word_size;
rbnode->block = kmalloc(rbnode->blklen * rbnode->word_size,
GFP_KERNEL);
if (!rbnode->block) {
kfree(rbnode);
return -ENOMEM;
}
snd_soc_rbtree_set_register(rbnode, 0, value);
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
snd_soc_rbtree_insert(&rbtree_ctx->root, rbnode);
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
rbtree_ctx->cached_rbnode = rbnode;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
return 0;
}
static int snd_soc_rbtree_cache_read(struct snd_soc_codec *codec,
unsigned int reg, unsigned int *value)
{
struct snd_soc_rbtree_ctx *rbtree_ctx;
struct snd_soc_rbtree_node *rbnode;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
unsigned int base_reg, top_reg;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int reg_tmp;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
rbtree_ctx = codec->reg_cache;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
/* look up the required register in the cached rbnode */
rbnode = rbtree_ctx->cached_rbnode;
if (rbnode) {
snd_soc_rbtree_get_base_top_reg(rbnode, &base_reg, &top_reg);
if (reg >= base_reg && reg <= top_reg) {
reg_tmp = reg - base_reg;
*value = snd_soc_rbtree_get_register(rbnode, reg_tmp);
return 0;
}
}
/* if we can't locate it in the cached rbnode we'll have
* to traverse the rbtree looking for it.
*/
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
rbnode = snd_soc_rbtree_lookup(&rbtree_ctx->root, reg);
if (rbnode) {
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
reg_tmp = reg - rbnode->base_reg;
*value = snd_soc_rbtree_get_register(rbnode, reg_tmp);
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
rbtree_ctx->cached_rbnode = rbnode;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
} else {
/* uninitialized registers default to 0 */
*value = 0;
}
return 0;
}
static int snd_soc_rbtree_cache_exit(struct snd_soc_codec *codec)
{
struct rb_node *next;
struct snd_soc_rbtree_ctx *rbtree_ctx;
struct snd_soc_rbtree_node *rbtree_node;
/* if we've already been called then just return */
rbtree_ctx = codec->reg_cache;
if (!rbtree_ctx)
return 0;
/* free up the rbtree */
next = rb_first(&rbtree_ctx->root);
while (next) {
rbtree_node = rb_entry(next, struct snd_soc_rbtree_node, node);
next = rb_next(&rbtree_node->node);
rb_erase(&rbtree_node->node, &rbtree_ctx->root);
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
kfree(rbtree_node->block);
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
kfree(rbtree_node);
}
/* release the resources */
kfree(codec->reg_cache);
codec->reg_cache = NULL;
return 0;
}
static int snd_soc_rbtree_cache_init(struct snd_soc_codec *codec)
{
struct snd_soc_rbtree_ctx *rbtree_ctx;
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
unsigned int word_size;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
unsigned int val;
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
int i;
int ret;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
codec->reg_cache = kmalloc(sizeof *rbtree_ctx, GFP_KERNEL);
if (!codec->reg_cache)
return -ENOMEM;
rbtree_ctx = codec->reg_cache;
rbtree_ctx->root = RB_ROOT;
ASoC: soc-cache: Cache a pointer to the last accessed rbnode Whenever we are doing a read or a write through the rbtree code, we'll cache a pointer to the rbnode. To avoid looking up the register everytime we do a read or a write, we first check if it can be found in the cached register block, otherwise we traverse the rbtree and finally cache the rbnode for future use. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:30 -06:00
rbtree_ctx->cached_rbnode = NULL;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (!codec->reg_def_copy)
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
return 0;
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
word_size = codec->driver->reg_word_size;
for (i = 0; i < codec->driver->reg_cache_size; ++i) {
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
val = snd_soc_get_cache_val(codec->reg_def_copy, i,
word_size);
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
if (!val)
continue;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
ret = snd_soc_rbtree_cache_write(codec, i, val);
if (ret)
goto err;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
return 0;
ASoC: soc-cache: Block based rbtree compression This patch prepares the ground for the actual rbtree optimization patch which will save a pointer to the last accessed rbnode that was used in either the read() or write() functions. Each rbnode manages a variable length block of registers. There can be no two nodes with overlapping blocks. Each block has a base register and a currently top register, all the other registers, if any, lie in between these two and in ascending order. The reasoning behind the construction of this rbtree is simple. In the snd_soc_rbtree_cache_init() function, we iterate over the register defaults provided by the driver. For each register value that is non-zero we insert it in the rbtree. In order to determine in which rbnode we need to add the register, we first look if there is another register already added that is adjacent to the one we are about to add. If that is the case we append it in that rbnode block, otherwise we create a new rbnode with a single register in its block and add it to the tree. In the next patch, where a cached rbnode is used by both the write() and the read() functions, we also check if the register we are about to add is in the cached rbnode (the least recently accessed one) and if so we append it in that rbnode block. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-05-19 06:45:29 -06:00
err:
snd_soc_cache_exit(codec);
return ret;
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
}
ASoC: Make LZO cache compression optional Make LZO cache compression optional as it pulls in the kernel wide LZO implementation and rbtree compression is generally more efficient for typical register maps, especially in terms of CPU performance. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:19:56 -07:00
#ifdef CONFIG_SND_SOC_CACHE_LZO
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
struct snd_soc_lzo_ctx {
void *wmem;
void *dst;
const void *src;
size_t src_len;
size_t dst_len;
size_t decompressed_size;
unsigned long *sync_bmp;
int sync_bmp_nbits;
};
#define LZO_BLOCK_NUM 8
static int snd_soc_lzo_block_count(void)
{
return LZO_BLOCK_NUM;
}
static int snd_soc_lzo_prepare(struct snd_soc_lzo_ctx *lzo_ctx)
{
lzo_ctx->wmem = kmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
if (!lzo_ctx->wmem)
return -ENOMEM;
return 0;
}
static int snd_soc_lzo_compress(struct snd_soc_lzo_ctx *lzo_ctx)
{
size_t compress_size;
int ret;
ret = lzo1x_1_compress(lzo_ctx->src, lzo_ctx->src_len,
lzo_ctx->dst, &compress_size, lzo_ctx->wmem);
if (ret != LZO_E_OK || compress_size > lzo_ctx->dst_len)
return -EINVAL;
lzo_ctx->dst_len = compress_size;
return 0;
}
static int snd_soc_lzo_decompress(struct snd_soc_lzo_ctx *lzo_ctx)
{
size_t dst_len;
int ret;
dst_len = lzo_ctx->dst_len;
ret = lzo1x_decompress_safe(lzo_ctx->src, lzo_ctx->src_len,
lzo_ctx->dst, &dst_len);
if (ret != LZO_E_OK || dst_len != lzo_ctx->dst_len)
return -EINVAL;
return 0;
}
static int snd_soc_lzo_compress_cache_block(struct snd_soc_codec *codec,
struct snd_soc_lzo_ctx *lzo_ctx)
{
int ret;
lzo_ctx->dst_len = lzo1x_worst_compress(PAGE_SIZE);
lzo_ctx->dst = kmalloc(lzo_ctx->dst_len, GFP_KERNEL);
if (!lzo_ctx->dst) {
lzo_ctx->dst_len = 0;
return -ENOMEM;
}
ret = snd_soc_lzo_compress(lzo_ctx);
if (ret < 0)
return ret;
return 0;
}
static int snd_soc_lzo_decompress_cache_block(struct snd_soc_codec *codec,
struct snd_soc_lzo_ctx *lzo_ctx)
{
int ret;
lzo_ctx->dst_len = lzo_ctx->decompressed_size;
lzo_ctx->dst = kmalloc(lzo_ctx->dst_len, GFP_KERNEL);
if (!lzo_ctx->dst) {
lzo_ctx->dst_len = 0;
return -ENOMEM;
}
ret = snd_soc_lzo_decompress(lzo_ctx);
if (ret < 0)
return ret;
return 0;
}
static inline int snd_soc_lzo_get_blkindex(struct snd_soc_codec *codec,
unsigned int reg)
{
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
codec_drv = codec->driver;
return (reg * codec_drv->reg_word_size) /
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
DIV_ROUND_UP(codec->reg_size, snd_soc_lzo_block_count());
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
}
static inline int snd_soc_lzo_get_blkpos(struct snd_soc_codec *codec,
unsigned int reg)
{
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
codec_drv = codec->driver;
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
return reg % (DIV_ROUND_UP(codec->reg_size, snd_soc_lzo_block_count()) /
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
codec_drv->reg_word_size);
}
static inline int snd_soc_lzo_get_blksize(struct snd_soc_codec *codec)
{
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
codec_drv = codec->driver;
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
return DIV_ROUND_UP(codec->reg_size, snd_soc_lzo_block_count());
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
}
static int snd_soc_lzo_cache_sync(struct snd_soc_codec *codec)
{
struct snd_soc_lzo_ctx **lzo_blocks;
unsigned int val;
int i;
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
int ret;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
lzo_blocks = codec->reg_cache;
for_each_set_bit(i, lzo_blocks[0]->sync_bmp, lzo_blocks[0]->sync_bmp_nbits) {
ASoC: Fix register cache sync register_writable WARN_ONs Currently the condition for these WARN_ONs is reversed and they are placed before the actual check whether we are going to write to that register. So if the codec implements the register_writable callback we'll get a warning for each writable register when syncing the register cache. While we are at it change the check to use snd_soc_codec_writable_register instead of open-coding it. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-08-27 10:24:13 -06:00
WARN_ON(!snd_soc_codec_writable_register(codec, i));
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
ret = snd_soc_cache_read(codec, i, &val);
if (ret)
return ret;
ASoC: soc-cache: Apply the cache_bypass option Incorporate the use of the cache_bypass functionality in the syncing functions. The snd_soc_flat_cache_sync() need not be hooked as there is no performance benefit from using the cache_bypass option. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-19 07:53:37 -07:00
codec->cache_bypass = 1;
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
ret = snd_soc_write(codec, i, val);
ASoC: soc-cache: Apply the cache_bypass option Incorporate the use of the cache_bypass functionality in the syncing functions. The snd_soc_flat_cache_sync() need not be hooked as there is no performance benefit from using the cache_bypass option. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-19 07:53:37 -07:00
codec->cache_bypass = 0;
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
if (ret)
return ret;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
dev_dbg(codec->dev, "Synced register %#x, value = %#x\n",
i, val);
}
return 0;
}
static int snd_soc_lzo_cache_write(struct snd_soc_codec *codec,
unsigned int reg, unsigned int value)
{
struct snd_soc_lzo_ctx *lzo_block, **lzo_blocks;
int ret, blkindex, blkpos;
size_t blksize, tmp_dst_len;
void *tmp_dst;
/* index of the compressed lzo block */
blkindex = snd_soc_lzo_get_blkindex(codec, reg);
/* register index within the decompressed block */
blkpos = snd_soc_lzo_get_blkpos(codec, reg);
/* size of the compressed block */
blksize = snd_soc_lzo_get_blksize(codec);
lzo_blocks = codec->reg_cache;
lzo_block = lzo_blocks[blkindex];
/* save the pointer and length of the compressed block */
tmp_dst = lzo_block->dst;
tmp_dst_len = lzo_block->dst_len;
/* prepare the source to be the compressed block */
lzo_block->src = lzo_block->dst;
lzo_block->src_len = lzo_block->dst_len;
/* decompress the block */
ret = snd_soc_lzo_decompress_cache_block(codec, lzo_block);
if (ret < 0) {
kfree(lzo_block->dst);
goto out;
}
/* write the new value to the cache */
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
if (snd_soc_set_cache_val(lzo_block->dst, blkpos, value,
codec->driver->reg_word_size)) {
kfree(lzo_block->dst);
goto out;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
}
/* prepare the source to be the decompressed block */
lzo_block->src = lzo_block->dst;
lzo_block->src_len = lzo_block->dst_len;
/* compress the block */
ret = snd_soc_lzo_compress_cache_block(codec, lzo_block);
if (ret < 0) {
kfree(lzo_block->dst);
kfree(lzo_block->src);
goto out;
}
/* set the bit so we know we have to sync this register */
set_bit(reg, lzo_block->sync_bmp);
kfree(tmp_dst);
kfree(lzo_block->src);
return 0;
out:
lzo_block->dst = tmp_dst;
lzo_block->dst_len = tmp_dst_len;
return ret;
}
static int snd_soc_lzo_cache_read(struct snd_soc_codec *codec,
unsigned int reg, unsigned int *value)
{
struct snd_soc_lzo_ctx *lzo_block, **lzo_blocks;
int ret, blkindex, blkpos;
size_t blksize, tmp_dst_len;
void *tmp_dst;
*value = 0;
/* index of the compressed lzo block */
blkindex = snd_soc_lzo_get_blkindex(codec, reg);
/* register index within the decompressed block */
blkpos = snd_soc_lzo_get_blkpos(codec, reg);
/* size of the compressed block */
blksize = snd_soc_lzo_get_blksize(codec);
lzo_blocks = codec->reg_cache;
lzo_block = lzo_blocks[blkindex];
/* save the pointer and length of the compressed block */
tmp_dst = lzo_block->dst;
tmp_dst_len = lzo_block->dst_len;
/* prepare the source to be the compressed block */
lzo_block->src = lzo_block->dst;
lzo_block->src_len = lzo_block->dst_len;
/* decompress the block */
ret = snd_soc_lzo_decompress_cache_block(codec, lzo_block);
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
if (ret >= 0)
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
/* fetch the value from the cache */
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
*value = snd_soc_get_cache_val(lzo_block->dst, blkpos,
codec->driver->reg_word_size);
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
kfree(lzo_block->dst);
/* restore the pointer and length of the compressed block */
lzo_block->dst = tmp_dst;
lzo_block->dst_len = tmp_dst_len;
return 0;
}
static int snd_soc_lzo_cache_exit(struct snd_soc_codec *codec)
{
struct snd_soc_lzo_ctx **lzo_blocks;
int i, blkcount;
lzo_blocks = codec->reg_cache;
if (!lzo_blocks)
return 0;
blkcount = snd_soc_lzo_block_count();
/*
* the pointer to the bitmap used for syncing the cache
* is shared amongst all lzo_blocks. Ensure it is freed
* only once.
*/
if (lzo_blocks[0])
kfree(lzo_blocks[0]->sync_bmp);
for (i = 0; i < blkcount; ++i) {
if (lzo_blocks[i]) {
kfree(lzo_blocks[i]->wmem);
kfree(lzo_blocks[i]->dst);
}
/* each lzo_block is a pointer returned by kmalloc or NULL */
kfree(lzo_blocks[i]);
}
kfree(lzo_blocks);
codec->reg_cache = NULL;
return 0;
}
static int snd_soc_lzo_cache_init(struct snd_soc_codec *codec)
{
struct snd_soc_lzo_ctx **lzo_blocks;
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
size_t bmp_size;
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
int ret, tofree, i, blksize, blkcount;
const char *p, *end;
unsigned long *sync_bmp;
ret = 0;
codec_drv = codec->driver;
/*
* If we have not been given a default register cache
* then allocate a dummy zero-ed out region, compress it
* and remember to free it afterwards.
*/
tofree = 0;
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (!codec->reg_def_copy)
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
tofree = 1;
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (!codec->reg_def_copy) {
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
codec->reg_def_copy = kzalloc(codec->reg_size, GFP_KERNEL);
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (!codec->reg_def_copy)
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
return -ENOMEM;
}
blkcount = snd_soc_lzo_block_count();
codec->reg_cache = kzalloc(blkcount * sizeof *lzo_blocks,
GFP_KERNEL);
if (!codec->reg_cache) {
ret = -ENOMEM;
goto err_tofree;
}
lzo_blocks = codec->reg_cache;
/*
* allocate a bitmap to be used when syncing the cache with
* the hardware. Each time a register is modified, the corresponding
* bit is set in the bitmap, so we know that we have to sync
* that register.
*/
bmp_size = codec_drv->reg_cache_size;
ASoC: soc-cache: A few minor stylistic changes Remove redundant parentheses/spaces in the use of the sizeof operator. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-14 08:15:36 -07:00
sync_bmp = kmalloc(BITS_TO_LONGS(bmp_size) * sizeof(long),
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
GFP_KERNEL);
if (!sync_bmp) {
ret = -ENOMEM;
goto err;
}
ASoC: soc-cache: Fix memory overflow in LZO initialization The bitmap_zero() nbits argument was improperly set to reg_size but the underlying buffer was bmp_size long. This caused the memset to zero past the end of the allocated buffer and into the kernel heap causing strange kernel crashes sometimes by overwriting critical kernel structures. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 04:43:33 -07:00
bitmap_zero(sync_bmp, bmp_size);
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
/* allocate the lzo blocks and initialize them */
for (i = 0; i < blkcount; ++i) {
lzo_blocks[i] = kzalloc(sizeof **lzo_blocks,
GFP_KERNEL);
if (!lzo_blocks[i]) {
kfree(sync_bmp);
ret = -ENOMEM;
goto err;
}
lzo_blocks[i]->sync_bmp = sync_bmp;
ASoC: soc-cache: Fix invalid memory access during snd_soc_lzo_cache_sync() The size of the lzo syncing bitmap was incorrectly set to the size of the cache times the word size, however, the correct size is the size of the cache. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:24:02 -07:00
lzo_blocks[i]->sync_bmp_nbits = bmp_size;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
/* alloc the working space for the compressed block */
ret = snd_soc_lzo_prepare(lzo_blocks[i]);
if (ret < 0)
goto err;
}
blksize = snd_soc_lzo_get_blksize(codec);
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
p = codec->reg_def_copy;
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
end = codec->reg_def_copy + codec->reg_size;
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
/* compress the register map and fill the lzo blocks */
for (i = 0; i < blkcount; ++i, p += blksize) {
lzo_blocks[i]->src = p;
if (p + blksize > end)
lzo_blocks[i]->src_len = end - p;
else
lzo_blocks[i]->src_len = blksize;
ret = snd_soc_lzo_compress_cache_block(codec,
lzo_blocks[i]);
if (ret < 0)
goto err;
lzo_blocks[i]->decompressed_size =
lzo_blocks[i]->src_len;
}
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (tofree) {
kfree(codec->reg_def_copy);
codec->reg_def_copy = NULL;
}
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
return 0;
err:
snd_soc_cache_exit(codec);
err_tofree:
ASoC: soc-cache: Use reg_def_copy instead of reg_cache_default Make sure to use codec->reg_def_copy instead of codec_drv->reg_cache_default wherever necessary. This change is necessary because in the next patch we move the cache initialization code outside snd_soc_register_codec() and by that time any data marked as __devinitconst such as the original reg_cache_default array might have already been freed by the kernel. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 09:11:05 -07:00
if (tofree) {
kfree(codec->reg_def_copy);
codec->reg_def_copy = NULL;
}
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
return ret;
}
ASoC: Make LZO cache compression optional Make LZO cache compression optional as it pulls in the kernel wide LZO implementation and rbtree compression is generally more efficient for typical register maps, especially in terms of CPU performance. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:19:56 -07:00
#endif
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
static int snd_soc_flat_cache_sync(struct snd_soc_codec *codec)
{
int i;
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
int ret;
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
unsigned int val;
codec_drv = codec->driver;
for (i = 0; i < codec_drv->reg_cache_size; ++i) {
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
ret = snd_soc_cache_read(codec, i, &val);
if (ret)
return ret;
ASoC: soc-cache: Ensure flat compression uses a copy of the defaults cache With the addition of the cache fallback functionality, it is necessary to ensure that if the register defaults cache was marked as __devinitconst and the LZO compression is not compiled in the kernel, the fallback to flat compression will still use a copy of the defaults cache. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:22:28 -07:00
if (codec->reg_def_copy)
if (snd_soc_get_cache_val(codec->reg_def_copy,
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
i, codec_drv->reg_word_size) == val)
continue;
ASoC: Fix register cache sync register_writable WARN_ONs Currently the condition for these WARN_ONs is reversed and they are placed before the actual check whether we are going to write to that register. So if the codec implements the register_writable callback we'll get a warning for each writable register when syncing the register cache. While we are at it change the check to use snd_soc_codec_writable_register instead of open-coding it. Signed-off-by: Lars-Peter Clausen <lars@metafoo.de> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-08-27 10:24:13 -06:00
WARN_ON(!snd_soc_codec_writable_register(codec, i));
ASoC: soc-cache: Add error checking in the *_cache_sync functions Ensure that we report any errors encountered during reads/writes in the cache syncing functions. Remove redundant newline in the source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:24:54 -07:00
ret = snd_soc_write(codec, i, val);
if (ret)
return ret;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
dev_dbg(codec->dev, "Synced register %#x, value = %#x\n",
i, val);
}
return 0;
}
static int snd_soc_flat_cache_write(struct snd_soc_codec *codec,
unsigned int reg, unsigned int value)
{
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
snd_soc_set_cache_val(codec->reg_cache, reg, value,
codec->driver->reg_word_size);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
return 0;
}
static int snd_soc_flat_cache_read(struct snd_soc_codec *codec,
unsigned int reg, unsigned int *value)
{
ASoC: soc-cache: Clean up the cache manipulation code Use Takashi's clean up code to make the cache manipulation code more readable. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-11 04:29:49 -07:00
*value = snd_soc_get_cache_val(codec->reg_cache, reg,
codec->driver->reg_word_size);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
return 0;
}
static int snd_soc_flat_cache_exit(struct snd_soc_codec *codec)
{
if (!codec->reg_cache)
return 0;
kfree(codec->reg_cache);
codec->reg_cache = NULL;
return 0;
}
static int snd_soc_flat_cache_init(struct snd_soc_codec *codec)
{
ASoC: Constify struct snd_soc_codec_driver Allow the CODEC driver structure to be marked const by making all the APIs that use it do so. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-02 09:21:08 -07:00
const struct snd_soc_codec_driver *codec_drv;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
codec_drv = codec->driver;
ASoC: soc-cache: Ensure flat compression uses a copy of the defaults cache With the addition of the cache fallback functionality, it is necessary to ensure that if the register defaults cache was marked as __devinitconst and the LZO compression is not compiled in the kernel, the fallback to flat compression will still use a copy of the defaults cache. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:22:28 -07:00
if (codec->reg_def_copy)
codec->reg_cache = kmemdup(codec->reg_def_copy,
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
codec->reg_size, GFP_KERNEL);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
else
ASoC: soc-cache: Add reg_size as a member to snd_soc_codec Simplify the use of reg_size, by calculating it once and storing it in the codec structure for later reference. The value of reg_size is reg_cache_size * reg_word_size. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-12 03:38:58 -07:00
codec->reg_cache = kzalloc(codec->reg_size, GFP_KERNEL);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
if (!codec->reg_cache)
return -ENOMEM;
return 0;
}
/* an array of all supported compression types */
static const struct snd_soc_cache_ops cache_types[] = {
ASoC: If we can't find a cache compression type default to flat This makes it easier to make cache types build time configurable as we don't have a hard dependency on a given cache being built in. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:09:48 -07:00
/* Flat *must* be the first entry for fallback */
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
{
ASoC: soc-cache: Ensure consistent cache naming Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-29 03:54:28 -07:00
.id = SND_SOC_FLAT_COMPRESSION,
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
.name = "flat",
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
.init = snd_soc_flat_cache_init,
.exit = snd_soc_flat_cache_exit,
.read = snd_soc_flat_cache_read,
.write = snd_soc_flat_cache_write,
.sync = snd_soc_flat_cache_sync
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
},
ASoC: Make LZO cache compression optional Make LZO cache compression optional as it pulls in the kernel wide LZO implementation and rbtree compression is generally more efficient for typical register maps, especially in terms of CPU performance. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:19:56 -07:00
#ifdef CONFIG_SND_SOC_CACHE_LZO
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
{
.id = SND_SOC_LZO_COMPRESSION,
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
.name = "LZO",
ASoC: soc-cache: Add support for LZO register caching This patch adds support for LZO compression when storing the register cache. The initial register defaults cache is marked as __devinitconst and the only change required for a driver to use LZO compression is to set the compress_type member in codec->driver to SND_SOC_LZO_COMPRESSION. For a typical device whose register map would normally occupy 25kB or 50kB by using the LZO compression technique, one can get down to ~5-7kB. There might be a performance penalty associated with each individual read/write due to decompressing/compressing the underlying cache, however that should not be noticeable. These memory benefits depend on whether the target architecture can get rid of the memory occupied by the original register defaults cache which is marked as __devinitconst. Nevertheless there will be some memory gain even if the target architecture can't get rid of the original register map, this should be around ~30-32kB instead of 50kB. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:58 -07:00
.init = snd_soc_lzo_cache_init,
.exit = snd_soc_lzo_cache_exit,
.read = snd_soc_lzo_cache_read,
.write = snd_soc_lzo_cache_write,
.sync = snd_soc_lzo_cache_sync
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
},
ASoC: Make LZO cache compression optional Make LZO cache compression optional as it pulls in the kernel wide LZO implementation and rbtree compression is generally more efficient for typical register maps, especially in terms of CPU performance. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:19:56 -07:00
#endif
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
{
.id = SND_SOC_RBTREE_COMPRESSION,
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
.name = "rbtree",
ASoC: soc-cache: Add support for rbtree based register caching This patch adds support for rbtree compression when storing the register cache. It does this by not adding any uninitialized registers (those whose value is 0). If any of those registers is written with a nonzero value they get added into the rbtree. Consider a sample device with a large sparse register map. The register indices are between [0, 0x31ff]. An array of 12800 registers is thus created each of which is 2 bytes. This results in a 25kB region. This array normally lives outside soc-core, normally in the driver itself. The original soc-core code would kmemdup this region resulting in 50kB total memory. When using the rbtree compression technique and __devinitconst on the original array the figures are as follows. For this typical device, you might have 100 initialized registers, that is registers that are nonzero by default. We build an rbtree with 100 nodes, each of which is 24 bytes. This results in ~2kB of memory. Assuming that the target arch can freeup the memory used by the initial __devinitconst array, we end up using about ~2kB bytes of actual memory. The memory footprint will increase as uninitialized registers get written and thus new nodes created in the rbtree. In practice, most of those registers are never changed. If the target arch can't freeup the __devinitconst array, we end up using a total of ~27kB. The difference between the rbtree and the LZO caching techniques, is that if using the LZO technique the size of the cache will increase slower as more uninitialized registers get changed. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:59 -07:00
.init = snd_soc_rbtree_cache_init,
.exit = snd_soc_rbtree_cache_exit,
.read = snd_soc_rbtree_cache_read,
.write = snd_soc_rbtree_cache_write,
.sync = snd_soc_rbtree_cache_sync
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
};
int snd_soc_cache_init(struct snd_soc_codec *codec)
{
int i;
for (i = 0; i < ARRAY_SIZE(cache_types); ++i)
ASoC: Add compress_type as a member to snd_soc_codec We need to keep a copy of the compress_type supplied by the codec driver so that we can override it if necessary with whatever the machine driver has provided us with. The reason for not modifying the codec->driver struct directly is that ideally we'd like to keep it const. Adjust the code in soc-cache and soc-core to make use of the compress_type member in the snd_soc_codec struct. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-02 07:53:01 -07:00
if (cache_types[i].id == codec->compress_type)
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
break;
ASoC: If we can't find a cache compression type default to flat This makes it easier to make cache types build time configurable as we don't have a hard dependency on a given cache being built in. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:09:48 -07:00
/* Fall back to flat compression */
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
if (i == ARRAY_SIZE(cache_types)) {
ASoC: If we can't find a cache compression type default to flat This makes it easier to make cache types build time configurable as we don't have a hard dependency on a given cache being built in. Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
2010-12-21 10:09:48 -07:00
dev_warn(codec->dev, "Could not match compress type: %d\n",
codec->compress_type);
i = 0;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
mutex_init(&codec->cache_rw_mutex);
codec->cache_ops = &cache_types[i];
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
if (codec->cache_ops->init) {
if (codec->cache_ops->name)
dev_dbg(codec->dev, "Initializing %s cache for %s codec\n",
codec->cache_ops->name, codec->name);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
return codec->cache_ops->init(codec);
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
}
ASoC: soc-cache: Return -ENOSYS instead of -EINVAL These functions fail with -EINVAL if the corresponding callbacks are not implemented. Change them to return -ENOSYS as it is more appropriate for unimplemented callbacks. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-22 04:48:49 -06:00
return -ENOSYS;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
/*
* NOTE: keep in mind that this function might be called
* multiple times.
*/
int snd_soc_cache_exit(struct snd_soc_codec *codec)
{
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
if (codec->cache_ops && codec->cache_ops->exit) {
if (codec->cache_ops->name)
dev_dbg(codec->dev, "Destroying %s cache for %s codec\n",
codec->cache_ops->name, codec->name);
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
return codec->cache_ops->exit(codec);
ASoC: soc-cache: Add optional cache name member to snd_soc_cache_ops Added an optional name member to snd_soc_cache_ops to enable more sensible diagnostic messages during cache init, exit and sync. Remove redundant newline in source code. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-12-06 02:51:57 -07:00
}
ASoC: soc-cache: Return -ENOSYS instead of -EINVAL These functions fail with -EINVAL if the corresponding callbacks are not implemented. Change them to return -ENOSYS as it is more appropriate for unimplemented callbacks. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-22 04:48:49 -06:00
return -ENOSYS;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
/**
* snd_soc_cache_read: Fetch the value of a given register from the cache.
*
* @codec: CODEC to configure.
* @reg: The register index.
* @value: The value to be returned.
*/
int snd_soc_cache_read(struct snd_soc_codec *codec,
unsigned int reg, unsigned int *value)
{
int ret;
mutex_lock(&codec->cache_rw_mutex);
if (value && codec->cache_ops && codec->cache_ops->read) {
ret = codec->cache_ops->read(codec, reg, value);
mutex_unlock(&codec->cache_rw_mutex);
return ret;
}
mutex_unlock(&codec->cache_rw_mutex);
ASoC: soc-cache: Return -ENOSYS instead of -EINVAL These functions fail with -EINVAL if the corresponding callbacks are not implemented. Change them to return -ENOSYS as it is more appropriate for unimplemented callbacks. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-22 04:48:49 -06:00
return -ENOSYS;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
EXPORT_SYMBOL_GPL(snd_soc_cache_read);
/**
* snd_soc_cache_write: Set the value of a given register in the cache.
*
* @codec: CODEC to configure.
* @reg: The register index.
* @value: The new register value.
*/
int snd_soc_cache_write(struct snd_soc_codec *codec,
unsigned int reg, unsigned int value)
{
int ret;
mutex_lock(&codec->cache_rw_mutex);
if (codec->cache_ops && codec->cache_ops->write) {
ret = codec->cache_ops->write(codec, reg, value);
mutex_unlock(&codec->cache_rw_mutex);
return ret;
}
mutex_unlock(&codec->cache_rw_mutex);
ASoC: soc-cache: Return -ENOSYS instead of -EINVAL These functions fail with -EINVAL if the corresponding callbacks are not implemented. Change them to return -ENOSYS as it is more appropriate for unimplemented callbacks. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-22 04:48:49 -06:00
return -ENOSYS;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
EXPORT_SYMBOL_GPL(snd_soc_cache_write);
/**
* snd_soc_cache_sync: Sync the register cache with the hardware.
*
* @codec: CODEC to configure.
*
* Any registers that should not be synced should be marked as
* volatile. In general drivers can choose not to use the provided
* syncing functionality if they so require.
*/
int snd_soc_cache_sync(struct snd_soc_codec *codec)
{
int ret;
ASoC: soc-cache: Add trace event for snd_soc_cache_sync() This patch makes it easy to see when the syncing process begins and ends. You can also enable the snd_soc_reg_write tracepoint to see which registers are being synced. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-21 08:29:02 -07:00
const char *name;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
if (!codec->cache_sync) {
return 0;
}
ASoC: soc-cache: dereferencing before checking The patch c358e640a66 "ASoC: soc-cache: Add trace event for snd_soc_cache_sync()" introduced a dereference of "codec->cache_ops" before we had checked it for NULL. I pulled the check forward, and then pulled everything in an indent level. Signed-off-by: Dan Carpenter <error27@gmail.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-02-07 12:01:41 -07:00
if (!codec->cache_ops || !codec->cache_ops->sync)
ASoC: soc-cache: Return -ENOSYS instead of -EINVAL These functions fail with -EINVAL if the corresponding callbacks are not implemented. Change them to return -ENOSYS as it is more appropriate for unimplemented callbacks. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-22 04:48:49 -06:00
return -ENOSYS;
ASoC: soc-cache: dereferencing before checking The patch c358e640a66 "ASoC: soc-cache: Add trace event for snd_soc_cache_sync()" introduced a dereference of "codec->cache_ops" before we had checked it for NULL. I pulled the check forward, and then pulled everything in an indent level. Signed-off-by: Dan Carpenter <error27@gmail.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-02-07 12:01:41 -07:00
ASoC: soc-cache: Add trace event for snd_soc_cache_sync() This patch makes it easy to see when the syncing process begins and ends. You can also enable the snd_soc_reg_write tracepoint to see which registers are being synced. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-21 08:29:02 -07:00
if (codec->cache_ops->name)
name = codec->cache_ops->name;
else
name = "unknown";
ASoC: soc-cache: dereferencing before checking The patch c358e640a66 "ASoC: soc-cache: Add trace event for snd_soc_cache_sync()" introduced a dereference of "codec->cache_ops" before we had checked it for NULL. I pulled the check forward, and then pulled everything in an indent level. Signed-off-by: Dan Carpenter <error27@gmail.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-02-07 12:01:41 -07:00
if (codec->cache_ops->name)
dev_dbg(codec->dev, "Syncing %s cache for %s codec\n",
codec->cache_ops->name, codec->name);
trace_snd_soc_cache_sync(codec, name, "start");
ret = codec->cache_ops->sync(codec);
if (!ret)
codec->cache_sync = 0;
trace_snd_soc_cache_sync(codec, name, "end");
return ret;
ASoC: soc-cache: Add support for flat register caching This patch introduces the new caching API and migrates the old caching interface into the new one. The flat register caching technique does not use compression at all and it is equivalent to the old caching technique. One can still access codec->reg_cache directly but this is not advised as that will not be portable across different caching strategies. None of the existing drivers need to be changed to adapt to this caching technique. There should be no noticeable overhead associated with using the new caching API. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2010-11-11 03:04:57 -07:00
}
EXPORT_SYMBOL_GPL(snd_soc_cache_sync);
ASoC: soc-cache: Add support for default readable()/volatile() functions For common scenarios, device drivers can provide a table of all the registers that are at least either readable/writable/volatile. The idea is that if a register lookup fails, all of its read/write/vol members will be zero and will be treated as default. This also reduces the size of the register access array. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@slimlogic.co.uk> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-01-13 05:20:36 -07:00
static int snd_soc_get_reg_access_index(struct snd_soc_codec *codec,
unsigned int reg)
{
const struct snd_soc_codec_driver *codec_drv;
unsigned int min, max, index;
codec_drv = codec->driver;
min = 0;
max = codec_drv->reg_access_size - 1;
do {
index = (min + max) / 2;
if (codec_drv->reg_access_default[index].reg == reg)
return index;
if (codec_drv->reg_access_default[index].reg < reg)
min = index + 1;
else
max = index;
} while (min <= max);
return -1;
}
int snd_soc_default_volatile_register(struct snd_soc_codec *codec,
unsigned int reg)
{
int index;
if (reg >= codec->driver->reg_cache_size)
return 1;
index = snd_soc_get_reg_access_index(codec, reg);
if (index < 0)
return 0;
return codec->driver->reg_access_default[index].vol;
}
EXPORT_SYMBOL_GPL(snd_soc_default_volatile_register);
int snd_soc_default_readable_register(struct snd_soc_codec *codec,
unsigned int reg)
{
int index;
if (reg >= codec->driver->reg_cache_size)
return 1;
index = snd_soc_get_reg_access_index(codec, reg);
if (index < 0)
return 0;
return codec->driver->reg_access_default[index].read;
}
EXPORT_SYMBOL_GPL(snd_soc_default_readable_register);
ASoC: Add default snd_soc_default_writable_register() callback By using struct snd_soc_reg_access for the read/write/vol attributes of the registers, we provide callbacks that automatically determine whether a given register is readable/writable or volatile. Signed-off-by: Dimitris Papastamos <dp@opensource.wolfsonmicro.com> Acked-by: Liam Girdwood <lrg@ti.com> Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
2011-03-24 07:45:17 -06:00
int snd_soc_default_writable_register(struct snd_soc_codec *codec,
unsigned int reg)
{
int index;
if (reg >= codec->driver->reg_cache_size)
return 1;
index = snd_soc_get_reg_access_index(codec, reg);
if (index < 0)
return 0;
return codec->driver->reg_access_default[index].write;
}
EXPORT_SYMBOL_GPL(snd_soc_default_writable_register);