alistair23-linux/arch/powerpc/sysdev/xics/icp-native.c

294 lines
6.4 KiB
C
Raw Normal View History

/*
* Copyright 2011 IBM Corporation.
*
* 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.
*
*/
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/irq.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/cpu.h>
#include <linux/of.h>
#include <linux/spinlock.h>
#include <asm/prom.h>
#include <asm/io.h>
#include <asm/smp.h>
#include <asm/irq.h>
#include <asm/errno.h>
#include <asm/xics.h>
struct icp_ipl {
union {
u32 word;
u8 bytes[4];
} xirr_poll;
union {
u32 word;
u8 bytes[4];
} xirr;
u32 dummy;
union {
u32 word;
u8 bytes[4];
} qirr;
u32 link_a;
u32 link_b;
u32 link_c;
};
static struct icp_ipl __iomem *icp_native_regs[NR_CPUS];
static inline unsigned int icp_native_get_xirr(void)
{
int cpu = smp_processor_id();
return in_be32(&icp_native_regs[cpu]->xirr.word);
}
static inline void icp_native_set_xirr(unsigned int value)
{
int cpu = smp_processor_id();
out_be32(&icp_native_regs[cpu]->xirr.word, value);
}
static inline void icp_native_set_cppr(u8 value)
{
int cpu = smp_processor_id();
out_8(&icp_native_regs[cpu]->xirr.bytes[0], value);
}
static inline void icp_native_set_qirr(int n_cpu, u8 value)
{
out_8(&icp_native_regs[n_cpu]->qirr.bytes[0], value);
}
static void icp_native_set_cpu_priority(unsigned char cppr)
{
xics_set_base_cppr(cppr);
icp_native_set_cppr(cppr);
iosync();
}
static void icp_native_eoi(struct irq_data *d)
{
unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
iosync();
icp_native_set_xirr((xics_pop_cppr() << 24) | hw_irq);
}
static void icp_native_teardown_cpu(void)
{
int cpu = smp_processor_id();
/* Clear any pending IPI */
icp_native_set_qirr(cpu, 0xff);
}
static void icp_native_flush_ipi(void)
{
/* We take the ipi irq but and never return so we
* need to EOI the IPI, but want to leave our priority 0
*
* should we check all the other interrupts too?
* should we be flagging idle loop instead?
* or creating some task to be scheduled?
*/
icp_native_set_xirr((0x00 << 24) | XICS_IPI);
}
static unsigned int icp_native_get_irq(void)
{
unsigned int xirr = icp_native_get_xirr();
unsigned int vec = xirr & 0x00ffffff;
unsigned int irq;
if (vec == XICS_IRQ_SPURIOUS)
return NO_IRQ;
irq = irq_radix_revmap_lookup(xics_host, vec);
if (likely(irq != NO_IRQ)) {
xics_push_cppr(vec);
return irq;
}
/* We don't have a linux mapping, so have rtas mask it. */
xics_mask_unknown_vec(vec);
/* We might learn about it later, so EOI it */
icp_native_set_xirr(xirr);
return NO_IRQ;
}
#ifdef CONFIG_SMP
powerpc: Consolidate ipi message mux and demux Consolidate the mux and demux of ipi messages into smp.c and call a new smp_ops callback to actually trigger the ipi. The powerpc architecture code is optimised for having 4 distinct ipi triggers, which are mapped to 4 distinct messages (ipi many, ipi single, scheduler ipi, and enter debugger). However, several interrupt controllers only provide a single software triggered interrupt that can be delivered to each cpu. To resolve this limitation, each smp_ops implementation created a per-cpu variable that is manipulated with atomic bitops. Since these lines will be contended they are optimialy marked as shared_aligned and take a full cache line for each cpu. Distro kernels may have 2 or 3 of these in their config, each taking per-cpu space even though at most one will be in use. This consolidation removes smp_message_recv and replaces the single call actions cases with direct calls from the common message recognition loop. The complicated debugger ipi case with its muxed crash handling code is moved to debug_ipi_action which is now called from the demux code (instead of the multi-message action calling smp_message_recv). I put a call to reschedule_action to increase the likelyhood of correctly merging the anticipated scheduler_ipi() hook coming from the scheduler tree; that single required call can be inlined later. The actual message decode is a copy of the old pseries xics code with its memory barriers and cache line spacing, augmented with a per-cpu unsigned long based on the book-e doorbell code. The optional data is set via a callback from the implementation and is passed to the new cause-ipi hook along with the logical cpu number. While currently only the doorbell implemntation uses this data it should be almost zero cost to retrieve and pass it -- it adds a single register load for the argument from the same cache line to which we just completed a store and the register is dead on return from the call. I extended the data element from unsigned int to unsigned long in case some other code wanted to associate a pointer. The doorbell check_self is replaced by a call to smp_muxed_ipi_resend, conditioned on the CPU_DBELL feature. The ifdef guard could be relaxed to CONFIG_SMP but I left it with BOOKE for now. Also, the doorbell interrupt vector for book-e was not calling irq_enter and irq_exit, which throws off cpu accounting and causes code to not realize it is running in interrupt context. Add the missing calls. Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-10 13:29:39 -06:00
static void icp_native_cause_ipi(int cpu, unsigned long data)
{
icp_native_set_qirr(cpu, IPI_PRIORITY);
}
static irqreturn_t icp_native_ipi_action(int irq, void *dev_id)
{
int cpu = smp_processor_id();
icp_native_set_qirr(cpu, 0xff);
powerpc: Consolidate ipi message mux and demux Consolidate the mux and demux of ipi messages into smp.c and call a new smp_ops callback to actually trigger the ipi. The powerpc architecture code is optimised for having 4 distinct ipi triggers, which are mapped to 4 distinct messages (ipi many, ipi single, scheduler ipi, and enter debugger). However, several interrupt controllers only provide a single software triggered interrupt that can be delivered to each cpu. To resolve this limitation, each smp_ops implementation created a per-cpu variable that is manipulated with atomic bitops. Since these lines will be contended they are optimialy marked as shared_aligned and take a full cache line for each cpu. Distro kernels may have 2 or 3 of these in their config, each taking per-cpu space even though at most one will be in use. This consolidation removes smp_message_recv and replaces the single call actions cases with direct calls from the common message recognition loop. The complicated debugger ipi case with its muxed crash handling code is moved to debug_ipi_action which is now called from the demux code (instead of the multi-message action calling smp_message_recv). I put a call to reschedule_action to increase the likelyhood of correctly merging the anticipated scheduler_ipi() hook coming from the scheduler tree; that single required call can be inlined later. The actual message decode is a copy of the old pseries xics code with its memory barriers and cache line spacing, augmented with a per-cpu unsigned long based on the book-e doorbell code. The optional data is set via a callback from the implementation and is passed to the new cause-ipi hook along with the logical cpu number. While currently only the doorbell implemntation uses this data it should be almost zero cost to retrieve and pass it -- it adds a single register load for the argument from the same cache line to which we just completed a store and the register is dead on return from the call. I extended the data element from unsigned int to unsigned long in case some other code wanted to associate a pointer. The doorbell check_self is replaced by a call to smp_muxed_ipi_resend, conditioned on the CPU_DBELL feature. The ifdef guard could be relaxed to CONFIG_SMP but I left it with BOOKE for now. Also, the doorbell interrupt vector for book-e was not calling irq_enter and irq_exit, which throws off cpu accounting and causes code to not realize it is running in interrupt context. Add the missing calls. Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-10 13:29:39 -06:00
return smp_ipi_demux();
}
#endif /* CONFIG_SMP */
static int __init icp_native_map_one_cpu(int hw_id, unsigned long addr,
unsigned long size)
{
char *rname;
int i, cpu = -1;
/* This may look gross but it's good enough for now, we don't quite
* have a hard -> linux processor id matching.
*/
for_each_possible_cpu(i) {
if (!cpu_present(i))
continue;
if (hw_id == get_hard_smp_processor_id(i)) {
cpu = i;
break;
}
}
/* Fail, skip that CPU. Don't print, it's normal, some XICS come up
* with way more entries in there than you have CPUs
*/
if (cpu == -1)
return 0;
rname = kasprintf(GFP_KERNEL, "CPU %d [0x%x] Interrupt Presentation",
cpu, hw_id);
if (!request_mem_region(addr, size, rname)) {
pr_warning("icp_native: Could not reserve ICP MMIO"
" for CPU %d, interrupt server #0x%x\n",
cpu, hw_id);
return -EBUSY;
}
icp_native_regs[cpu] = ioremap(addr, size);
if (!icp_native_regs[cpu]) {
pr_warning("icp_native: Failed ioremap for CPU %d, "
"interrupt server #0x%x, addr %#lx\n",
cpu, hw_id, addr);
release_mem_region(addr, size);
return -ENOMEM;
}
return 0;
}
static int __init icp_native_init_one_node(struct device_node *np,
unsigned int *indx)
{
unsigned int ilen;
const u32 *ireg;
int i;
int reg_tuple_size;
int num_servers = 0;
/* This code does the theorically broken assumption that the interrupt
* server numbers are the same as the hard CPU numbers.
* This happens to be the case so far but we are playing with fire...
* should be fixed one of these days. -BenH.
*/
ireg = of_get_property(np, "ibm,interrupt-server-ranges", &ilen);
/* Do that ever happen ? we'll know soon enough... but even good'old
* f80 does have that property ..
*/
WARN_ON((ireg == NULL) || (ilen != 2*sizeof(u32)));
if (ireg) {
*indx = of_read_number(ireg, 1);
if (ilen >= 2*sizeof(u32))
num_servers = of_read_number(ireg + 1, 1);
}
ireg = of_get_property(np, "reg", &ilen);
if (!ireg) {
pr_err("icp_native: Can't find interrupt reg property");
return -1;
}
reg_tuple_size = (of_n_addr_cells(np) + of_n_size_cells(np)) * 4;
if (((ilen % reg_tuple_size) != 0)
|| (num_servers && (num_servers != (ilen / reg_tuple_size)))) {
pr_err("icp_native: ICP reg len (%d) != num servers (%d)",
ilen / reg_tuple_size, num_servers);
return -1;
}
for (i = 0; i < (ilen / reg_tuple_size); i++) {
struct resource r;
int err;
err = of_address_to_resource(np, i, &r);
if (err) {
pr_err("icp_native: Could not translate ICP MMIO"
" for interrupt server 0x%x (%d)\n", *indx, err);
return -1;
}
if (icp_native_map_one_cpu(*indx, r.start, r.end - r.start))
return -1;
(*indx)++;
}
return 0;
}
static const struct icp_ops icp_native_ops = {
.get_irq = icp_native_get_irq,
.eoi = icp_native_eoi,
.set_priority = icp_native_set_cpu_priority,
.teardown_cpu = icp_native_teardown_cpu,
.flush_ipi = icp_native_flush_ipi,
#ifdef CONFIG_SMP
.ipi_action = icp_native_ipi_action,
powerpc: Consolidate ipi message mux and demux Consolidate the mux and demux of ipi messages into smp.c and call a new smp_ops callback to actually trigger the ipi. The powerpc architecture code is optimised for having 4 distinct ipi triggers, which are mapped to 4 distinct messages (ipi many, ipi single, scheduler ipi, and enter debugger). However, several interrupt controllers only provide a single software triggered interrupt that can be delivered to each cpu. To resolve this limitation, each smp_ops implementation created a per-cpu variable that is manipulated with atomic bitops. Since these lines will be contended they are optimialy marked as shared_aligned and take a full cache line for each cpu. Distro kernels may have 2 or 3 of these in their config, each taking per-cpu space even though at most one will be in use. This consolidation removes smp_message_recv and replaces the single call actions cases with direct calls from the common message recognition loop. The complicated debugger ipi case with its muxed crash handling code is moved to debug_ipi_action which is now called from the demux code (instead of the multi-message action calling smp_message_recv). I put a call to reschedule_action to increase the likelyhood of correctly merging the anticipated scheduler_ipi() hook coming from the scheduler tree; that single required call can be inlined later. The actual message decode is a copy of the old pseries xics code with its memory barriers and cache line spacing, augmented with a per-cpu unsigned long based on the book-e doorbell code. The optional data is set via a callback from the implementation and is passed to the new cause-ipi hook along with the logical cpu number. While currently only the doorbell implemntation uses this data it should be almost zero cost to retrieve and pass it -- it adds a single register load for the argument from the same cache line to which we just completed a store and the register is dead on return from the call. I extended the data element from unsigned int to unsigned long in case some other code wanted to associate a pointer. The doorbell check_self is replaced by a call to smp_muxed_ipi_resend, conditioned on the CPU_DBELL feature. The ifdef guard could be relaxed to CONFIG_SMP but I left it with BOOKE for now. Also, the doorbell interrupt vector for book-e was not calling irq_enter and irq_exit, which throws off cpu accounting and causes code to not realize it is running in interrupt context. Add the missing calls. Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2011-05-10 13:29:39 -06:00
.cause_ipi = icp_native_cause_ipi,
#endif
};
int icp_native_init(void)
{
struct device_node *np;
u32 indx = 0;
int found = 0;
for_each_compatible_node(np, NULL, "ibm,ppc-xicp")
if (icp_native_init_one_node(np, &indx) == 0)
found = 1;
if (!found) {
for_each_node_by_type(np,
"PowerPC-External-Interrupt-Presentation") {
if (icp_native_init_one_node(np, &indx) == 0)
found = 1;
}
}
if (found == 0)
return -ENODEV;
icp_ops = &icp_native_ops;
return 0;
}