2009-06-05 06:42:42 -06:00
|
|
|
/*
|
|
|
|
* Copyright 2008 Advanced Micro Devices, Inc.
|
|
|
|
* Copyright 2008 Red Hat Inc.
|
|
|
|
* Copyright 2009 Jerome Glisse.
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
|
|
* copy of this software and associated documentation files (the "Software"),
|
|
|
|
* to deal in the Software without restriction, including without limitation
|
|
|
|
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
|
|
|
|
* and/or sell copies of the Software, and to permit persons to whom the
|
|
|
|
* Software is furnished to do so, subject to the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice shall be included in
|
|
|
|
* all copies or substantial portions of the Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
|
|
* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
|
|
|
|
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
|
|
|
|
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
|
|
|
|
* OTHER DEALINGS IN THE SOFTWARE.
|
|
|
|
*
|
|
|
|
* Authors: Dave Airlie
|
|
|
|
* Alex Deucher
|
|
|
|
* Jerome Glisse
|
|
|
|
*/
|
2012-10-02 11:01:07 -06:00
|
|
|
#include <drm/drmP.h>
|
2009-06-05 06:42:42 -06:00
|
|
|
#include "radeon.h"
|
2010-03-11 14:19:17 -07:00
|
|
|
#include "radeon_asic.h"
|
2014-12-22 15:26:51 -07:00
|
|
|
#include "radeon_audio.h"
|
2009-07-13 13:04:08 -06:00
|
|
|
#include "atom.h"
|
2009-10-01 01:39:24 -06:00
|
|
|
#include "rs690d.h"
|
2009-06-05 06:42:42 -06:00
|
|
|
|
2012-02-23 15:53:38 -07:00
|
|
|
int rs690_mc_wait_for_idle(struct radeon_device *rdev)
|
2009-06-05 06:42:42 -06:00
|
|
|
{
|
|
|
|
unsigned i;
|
|
|
|
uint32_t tmp;
|
|
|
|
|
|
|
|
for (i = 0; i < rdev->usec_timeout; i++) {
|
|
|
|
/* read MC_STATUS */
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS);
|
|
|
|
if (G_000090_MC_SYSTEM_IDLE(tmp))
|
2009-06-05 06:42:42 -06:00
|
|
|
return 0;
|
2009-10-01 01:39:24 -06:00
|
|
|
udelay(1);
|
2009-06-05 06:42:42 -06:00
|
|
|
}
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
static void rs690_gpu_init(struct radeon_device *rdev)
|
2009-06-05 06:42:42 -06:00
|
|
|
{
|
|
|
|
/* FIXME: is this correct ? */
|
|
|
|
r420_pipes_init(rdev);
|
|
|
|
if (rs690_mc_wait_for_idle(rdev)) {
|
2017-02-28 05:55:52 -07:00
|
|
|
pr_warn("Failed to wait MC idle while programming pipes. Bad things might happen.\n");
|
2009-06-05 06:42:42 -06:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-03-17 23:04:01 -06:00
|
|
|
union igp_info {
|
|
|
|
struct _ATOM_INTEGRATED_SYSTEM_INFO info;
|
|
|
|
struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 info_v2;
|
|
|
|
};
|
|
|
|
|
2009-07-13 13:04:08 -06:00
|
|
|
void rs690_pm_info(struct radeon_device *rdev)
|
|
|
|
{
|
|
|
|
int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo);
|
2010-03-17 23:04:01 -06:00
|
|
|
union igp_info *info;
|
2009-07-13 13:04:08 -06:00
|
|
|
uint16_t data_offset;
|
|
|
|
uint8_t frev, crev;
|
|
|
|
fixed20_12 tmp;
|
|
|
|
|
2010-03-17 23:04:01 -06:00
|
|
|
if (atom_parse_data_header(rdev->mode_info.atom_context, index, NULL,
|
|
|
|
&frev, &crev, &data_offset)) {
|
|
|
|
info = (union igp_info *)(rdev->mode_info.atom_context->bios + data_offset);
|
|
|
|
|
|
|
|
/* Get various system informations from bios */
|
|
|
|
switch (crev) {
|
|
|
|
case 1:
|
2010-04-27 19:46:42 -06:00
|
|
|
tmp.full = dfixed_const(100);
|
2011-02-14 14:16:22 -07:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_const(le32_to_cpu(info->info.ulBootUpMemoryClock));
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
|
2011-02-14 14:16:22 -07:00
|
|
|
if (le16_to_cpu(info->info.usK8MemoryClock))
|
2010-06-30 10:02:03 -06:00
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(le16_to_cpu(info->info.usK8MemoryClock));
|
|
|
|
else if (rdev->clock.default_mclk) {
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(rdev->clock.default_mclk);
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
|
|
|
|
} else
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(400);
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_ht_link_clk.full = dfixed_const(le16_to_cpu(info->info.usFSBClock));
|
|
|
|
rdev->pm.igp_ht_link_width.full = dfixed_const(info->info.ucHTLinkWidth);
|
2010-03-17 23:04:01 -06:00
|
|
|
break;
|
|
|
|
case 2:
|
2010-04-27 19:46:42 -06:00
|
|
|
tmp.full = dfixed_const(100);
|
2011-02-14 14:16:22 -07:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_const(le32_to_cpu(info->info_v2.ulBootUpSidePortClock));
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
|
2011-02-14 14:16:22 -07:00
|
|
|
if (le32_to_cpu(info->info_v2.ulBootUpUMAClock))
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(le32_to_cpu(info->info_v2.ulBootUpUMAClock));
|
2010-06-30 10:02:03 -06:00
|
|
|
else if (rdev->clock.default_mclk)
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(rdev->clock.default_mclk);
|
|
|
|
else
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(66700);
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
|
2011-02-14 14:16:22 -07:00
|
|
|
rdev->pm.igp_ht_link_clk.full = dfixed_const(le32_to_cpu(info->info_v2.ulHTLinkFreq));
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_ht_link_clk.full = dfixed_div(rdev->pm.igp_ht_link_clk, tmp);
|
|
|
|
rdev->pm.igp_ht_link_width.full = dfixed_const(le16_to_cpu(info->info_v2.usMinHTLinkWidth));
|
2010-03-17 23:04:01 -06:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
/* We assume the slower possible clock ie worst case */
|
2010-06-30 10:02:03 -06:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_const(200);
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(200);
|
|
|
|
rdev->pm.igp_ht_link_clk.full = dfixed_const(1000);
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_ht_link_width.full = dfixed_const(8);
|
2010-03-17 23:04:01 -06:00
|
|
|
DRM_ERROR("No integrated system info for your GPU, using safe default\n");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
} else {
|
2009-07-13 13:04:08 -06:00
|
|
|
/* We assume the slower possible clock ie worst case */
|
2010-06-30 10:02:03 -06:00
|
|
|
rdev->pm.igp_sideport_mclk.full = dfixed_const(200);
|
|
|
|
rdev->pm.igp_system_mclk.full = dfixed_const(200);
|
|
|
|
rdev->pm.igp_ht_link_clk.full = dfixed_const(1000);
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.igp_ht_link_width.full = dfixed_const(8);
|
2009-07-13 13:04:08 -06:00
|
|
|
DRM_ERROR("No integrated system info for your GPU, using safe default\n");
|
|
|
|
}
|
|
|
|
/* Compute various bandwidth */
|
|
|
|
/* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4 */
|
2010-04-27 19:46:42 -06:00
|
|
|
tmp.full = dfixed_const(4);
|
|
|
|
rdev->pm.k8_bandwidth.full = dfixed_mul(rdev->pm.igp_system_mclk, tmp);
|
2009-07-13 13:04:08 -06:00
|
|
|
/* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8
|
|
|
|
* = ht_clk * ht_width / 5
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
tmp.full = dfixed_const(5);
|
|
|
|
rdev->pm.ht_bandwidth.full = dfixed_mul(rdev->pm.igp_ht_link_clk,
|
2009-07-13 13:04:08 -06:00
|
|
|
rdev->pm.igp_ht_link_width);
|
2010-04-27 19:46:42 -06:00
|
|
|
rdev->pm.ht_bandwidth.full = dfixed_div(rdev->pm.ht_bandwidth, tmp);
|
2009-07-13 13:04:08 -06:00
|
|
|
if (tmp.full < rdev->pm.max_bandwidth.full) {
|
|
|
|
/* HT link is a limiting factor */
|
|
|
|
rdev->pm.max_bandwidth.full = tmp.full;
|
|
|
|
}
|
|
|
|
/* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7
|
|
|
|
* = (sideport_clk * 14) / 10
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
tmp.full = dfixed_const(14);
|
|
|
|
rdev->pm.sideport_bandwidth.full = dfixed_mul(rdev->pm.igp_sideport_mclk, tmp);
|
|
|
|
tmp.full = dfixed_const(10);
|
|
|
|
rdev->pm.sideport_bandwidth.full = dfixed_div(rdev->pm.sideport_bandwidth, tmp);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
|
|
|
|
2012-08-31 11:43:50 -06:00
|
|
|
static void rs690_mc_init(struct radeon_device *rdev)
|
2009-06-05 06:42:42 -06:00
|
|
|
{
|
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-17 14:54:29 -07:00
|
|
|
u64 base;
|
2013-04-08 15:25:47 -06:00
|
|
|
uint32_t h_addr, l_addr;
|
|
|
|
unsigned long long k8_addr;
|
2009-06-05 06:42:42 -06:00
|
|
|
|
|
|
|
rs400_gart_adjust_size(rdev);
|
|
|
|
rdev->mc.vram_is_ddr = true;
|
2009-12-03 14:18:19 -07:00
|
|
|
rdev->mc.vram_width = 128;
|
2009-07-21 04:39:30 -06:00
|
|
|
rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
|
|
|
|
rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
|
2010-05-27 13:40:24 -06:00
|
|
|
rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
|
|
|
|
rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
|
2010-02-19 07:33:54 -07:00
|
|
|
rdev->mc.visible_vram_size = rdev->mc.aper_size;
|
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-17 14:54:29 -07:00
|
|
|
base = RREG32_MC(R_000100_MCCFG_FB_LOCATION);
|
|
|
|
base = G_000100_MC_FB_START(base) << 16;
|
2010-01-05 09:27:29 -07:00
|
|
|
rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
|
2013-12-02 16:15:51 -07:00
|
|
|
/* Some boards seem to be configured for 128MB of sideport memory,
|
|
|
|
* but really only have 64MB. Just skip the sideport and use
|
|
|
|
* UMA memory.
|
|
|
|
*/
|
|
|
|
if (rdev->mc.igp_sideport_enabled &&
|
|
|
|
(rdev->mc.real_vram_size == (384 * 1024 * 1024))) {
|
|
|
|
base += 128 * 1024 * 1024;
|
|
|
|
rdev->mc.real_vram_size -= 128 * 1024 * 1024;
|
|
|
|
rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
|
|
|
|
}
|
2013-04-08 15:25:47 -06:00
|
|
|
|
|
|
|
/* Use K8 direct mapping for fast fb access. */
|
|
|
|
rdev->fastfb_working = false;
|
|
|
|
h_addr = G_00005F_K8_ADDR_EXT(RREG32_MC(R_00005F_MC_MISC_UMA_CNTL));
|
|
|
|
l_addr = RREG32_MC(R_00001E_K8_FB_LOCATION);
|
|
|
|
k8_addr = ((unsigned long long)h_addr) << 32 | l_addr;
|
|
|
|
#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
|
|
|
|
if (k8_addr + rdev->mc.visible_vram_size < 0x100000000ULL)
|
|
|
|
#endif
|
|
|
|
{
|
|
|
|
/* FastFB shall be used with UMA memory. Here it is simply disabled when sideport
|
|
|
|
* memory is present.
|
|
|
|
*/
|
|
|
|
if (rdev->mc.igp_sideport_enabled == false && radeon_fastfb == 1) {
|
|
|
|
DRM_INFO("Direct mapping: aper base at 0x%llx, replaced by direct mapping base 0x%llx.\n",
|
|
|
|
(unsigned long long)rdev->mc.aper_base, k8_addr);
|
|
|
|
rdev->mc.aper_base = (resource_size_t)k8_addr;
|
|
|
|
rdev->fastfb_working = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-08-02 17:39:15 -06:00
|
|
|
rs690_pm_info(rdev);
|
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-17 14:54:29 -07:00
|
|
|
radeon_vram_location(rdev, &rdev->mc, base);
|
2010-07-15 08:51:10 -06:00
|
|
|
rdev->mc.gtt_base_align = rdev->mc.gtt_size - 1;
|
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-17 14:54:29 -07:00
|
|
|
radeon_gtt_location(rdev, &rdev->mc);
|
2010-03-16 18:54:38 -06:00
|
|
|
radeon_update_bandwidth_info(rdev);
|
2009-12-06 17:45:17 -07:00
|
|
|
}
|
|
|
|
|
2009-07-13 13:04:08 -06:00
|
|
|
void rs690_line_buffer_adjust(struct radeon_device *rdev,
|
|
|
|
struct drm_display_mode *mode1,
|
|
|
|
struct drm_display_mode *mode2)
|
|
|
|
{
|
|
|
|
u32 tmp;
|
|
|
|
|
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2)
commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many
vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core
more fragile to drivers which don't update hw vblank counters and
vblank timestamps in sync with firing of the vblank irq and
essentially at leading edge of vblank.
This exposed a problem with radeon-kms/amdgpu-kms which do not
satisfy above requirements:
The vblank irq fires a few scanlines before start of vblank, but
programmed pageflips complete at start of vblank and
vblank timestamps update at start of vblank, whereas the
hw vblank counter increments only later, at start of vsync.
This leads to problems like off by one errors for vblank counter
updates, vblank counters apparently going backwards or vblank
timestamps apparently having time going backwards. The net result
is stuttering of graphics in games, or little hangs, as well as
total failure of timing sensitive applications.
See bug #93147 for an example of the regression on Linux 4.4-rc:
https://bugs.freedesktop.org/show_bug.cgi?id=93147
This patch tries to align all above events better from the
viewpoint of the drm core / of external callers to fix the problem:
1. The apparent start of vblank is shifted a few scanlines earlier,
so the vblank irq now always happens after start of this extended
vblank interval and thereby drm_update_vblank_count() always samples
the updated vblank count and timestamp of the new vblank interval.
To achieve this, the reporting of scanout positions by
radeon_get_crtc_scanoutpos() now operates as if the vblank starts
radeon_crtc->lb_vblank_lead_lines before the real start of the hw
vblank interval. This means that the vblank timestamps which are based
on these scanout positions will now update at this earlier start of
vblank.
2. The driver->get_vblank_counter() function will bump the returned
vblank count as read from the hw by +1 if the query happens after
the shifted earlier start of the vblank, but before the real hw increment
at start of vsync, so the counter appears to increment at start of vblank
in sync with the timestamp update.
3. Calls from vblank irq-context and regular non-irq calls are now
treated identical, always simulating the shifted vblank start, to
avoid inconsistent results for queries happening from vblank irq vs.
happening from drm_vblank_enable() or vblank_disable_fn().
4. The radeon_flip_work_func will delay mmio programming a pageflip until
the start of the real vblank iff it happens to execute inside the shifted
earlier start of the vblank, so pageflips now also appear to execute at
start of the shifted vblank, in sync with vblank counter and timestamp
updates. This to avoid some races between updates of vblank count and
timestamps that are used for swap scheduling and pageflip execution which
could cause pageflips to execute before the scheduled target vblank.
The lb_vblank_lead_lines "fudge" value is calculated as the size of
the display controllers line buffer in scanlines for the given video
mode: Vblank irq's are triggered by the line buffer logic when the line
buffer refill for a video frame ends, ie. when the line buffer source read
position enters the hw vblank. This means that a vblank irq could fire at
most as many scanlines before the current reported scanout position of the
crtc timing generator as the number of scanlines the line buffer can
maximally hold for a given video mode.
This patch has been successfully tested on a RV730 card with DCE-3 display
engine and on a evergreen card with DCE-4 display engine, in single-display
and dual-display configuration, with different video modes.
A similar patch is needed for amdgpu-kms to fix the same problem.
Limitations:
- Line buffer sizes in pixels are hard-coded on < DCE-4 to a value
i just guessed to be high enough to work ok, lacking info on the true
sizes atm.
Fixes: fdo#93147
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Michel Dänzer <michel.daenzer@amd.com>
Cc: Harry Wentland <Harry.Wentland@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
(v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net>
(v2) Refine radeon_flip_work_func() for better efficiency:
In radeon_flip_work_func, replace the busy waiting udelay(5)
with event lock held by a more performance and energy efficient
usleep_range() until at least predicted true start of hw vblank,
with some slack for scheduler happiness. Release the event lock
during waits to not delay other outputs in doing their stuff, as
the waiting can last up to 200 usecs in some cases.
Retested on DCE-3 and DCE-4 to verify it still works nicely.
(v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-25 12:14:31 -07:00
|
|
|
/* Guess line buffer size to be 8192 pixels */
|
|
|
|
u32 lb_size = 8192;
|
|
|
|
|
2009-07-13 13:04:08 -06:00
|
|
|
/*
|
|
|
|
* Line Buffer Setup
|
|
|
|
* There is a single line buffer shared by both display controllers.
|
2009-10-01 01:39:24 -06:00
|
|
|
* R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
|
2009-07-13 13:04:08 -06:00
|
|
|
* the display controllers. The paritioning can either be done
|
|
|
|
* manually or via one of four preset allocations specified in bits 1:0:
|
|
|
|
* 0 - line buffer is divided in half and shared between crtc
|
|
|
|
* 1 - D1 gets 3/4 of the line buffer, D2 gets 1/4
|
|
|
|
* 2 - D1 gets the whole buffer
|
|
|
|
* 3 - D1 gets 1/4 of the line buffer, D2 gets 3/4
|
2009-10-01 01:39:24 -06:00
|
|
|
* Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual
|
2009-07-13 13:04:08 -06:00
|
|
|
* allocation mode. In manual allocation mode, D1 always starts at 0,
|
|
|
|
* D1 end/2 is specified in bits 14:4; D2 allocation follows D1.
|
|
|
|
*/
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT;
|
|
|
|
tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE;
|
2009-07-13 13:04:08 -06:00
|
|
|
/* auto */
|
|
|
|
if (mode1 && mode2) {
|
|
|
|
if (mode1->hdisplay > mode2->hdisplay) {
|
|
|
|
if (mode1->hdisplay > 2560)
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q;
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
|
2009-07-13 13:04:08 -06:00
|
|
|
} else if (mode2->hdisplay > mode1->hdisplay) {
|
|
|
|
if (mode2->hdisplay > 2560)
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
|
2009-07-13 13:04:08 -06:00
|
|
|
} else
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
|
2009-07-13 13:04:08 -06:00
|
|
|
} else if (mode1) {
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY;
|
2009-07-13 13:04:08 -06:00
|
|
|
} else if (mode2) {
|
2009-10-01 01:39:24 -06:00
|
|
|
tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2009-10-01 01:39:24 -06:00
|
|
|
WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp);
|
drm/radeon: Fixup hw vblank counter/ts for new drm_update_vblank_count() (v2)
commit 4dfd6486 "drm: Use vblank timestamps to guesstimate how many
vblanks were missed" introduced in Linux 4.4-rc1 makes the drm core
more fragile to drivers which don't update hw vblank counters and
vblank timestamps in sync with firing of the vblank irq and
essentially at leading edge of vblank.
This exposed a problem with radeon-kms/amdgpu-kms which do not
satisfy above requirements:
The vblank irq fires a few scanlines before start of vblank, but
programmed pageflips complete at start of vblank and
vblank timestamps update at start of vblank, whereas the
hw vblank counter increments only later, at start of vsync.
This leads to problems like off by one errors for vblank counter
updates, vblank counters apparently going backwards or vblank
timestamps apparently having time going backwards. The net result
is stuttering of graphics in games, or little hangs, as well as
total failure of timing sensitive applications.
See bug #93147 for an example of the regression on Linux 4.4-rc:
https://bugs.freedesktop.org/show_bug.cgi?id=93147
This patch tries to align all above events better from the
viewpoint of the drm core / of external callers to fix the problem:
1. The apparent start of vblank is shifted a few scanlines earlier,
so the vblank irq now always happens after start of this extended
vblank interval and thereby drm_update_vblank_count() always samples
the updated vblank count and timestamp of the new vblank interval.
To achieve this, the reporting of scanout positions by
radeon_get_crtc_scanoutpos() now operates as if the vblank starts
radeon_crtc->lb_vblank_lead_lines before the real start of the hw
vblank interval. This means that the vblank timestamps which are based
on these scanout positions will now update at this earlier start of
vblank.
2. The driver->get_vblank_counter() function will bump the returned
vblank count as read from the hw by +1 if the query happens after
the shifted earlier start of the vblank, but before the real hw increment
at start of vsync, so the counter appears to increment at start of vblank
in sync with the timestamp update.
3. Calls from vblank irq-context and regular non-irq calls are now
treated identical, always simulating the shifted vblank start, to
avoid inconsistent results for queries happening from vblank irq vs.
happening from drm_vblank_enable() or vblank_disable_fn().
4. The radeon_flip_work_func will delay mmio programming a pageflip until
the start of the real vblank iff it happens to execute inside the shifted
earlier start of the vblank, so pageflips now also appear to execute at
start of the shifted vblank, in sync with vblank counter and timestamp
updates. This to avoid some races between updates of vblank count and
timestamps that are used for swap scheduling and pageflip execution which
could cause pageflips to execute before the scheduled target vblank.
The lb_vblank_lead_lines "fudge" value is calculated as the size of
the display controllers line buffer in scanlines for the given video
mode: Vblank irq's are triggered by the line buffer logic when the line
buffer refill for a video frame ends, ie. when the line buffer source read
position enters the hw vblank. This means that a vblank irq could fire at
most as many scanlines before the current reported scanout position of the
crtc timing generator as the number of scanlines the line buffer can
maximally hold for a given video mode.
This patch has been successfully tested on a RV730 card with DCE-3 display
engine and on a evergreen card with DCE-4 display engine, in single-display
and dual-display configuration, with different video modes.
A similar patch is needed for amdgpu-kms to fix the same problem.
Limitations:
- Line buffer sizes in pixels are hard-coded on < DCE-4 to a value
i just guessed to be high enough to work ok, lacking info on the true
sizes atm.
Fixes: fdo#93147
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Michel Dänzer <michel.daenzer@amd.com>
Cc: Harry Wentland <Harry.Wentland@amd.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
(v1) Tested-by: Dave Witbrodt <dawitbro@sbcglobal.net>
(v2) Refine radeon_flip_work_func() for better efficiency:
In radeon_flip_work_func, replace the busy waiting udelay(5)
with event lock held by a more performance and energy efficient
usleep_range() until at least predicted true start of hw vblank,
with some slack for scheduler happiness. Release the event lock
during waits to not delay other outputs in doing their stuff, as
the waiting can last up to 200 usecs in some cases.
Retested on DCE-3 and DCE-4 to verify it still works nicely.
(v2) Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2015-11-25 12:14:31 -07:00
|
|
|
|
|
|
|
/* Save number of lines the linebuffer leads before the scanout */
|
|
|
|
if (mode1)
|
|
|
|
rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
|
|
|
|
|
|
|
|
if (mode2)
|
|
|
|
rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
|
2009-06-05 06:42:42 -06:00
|
|
|
}
|
|
|
|
|
2009-07-13 13:04:08 -06:00
|
|
|
struct rs690_watermark {
|
|
|
|
u32 lb_request_fifo_depth;
|
|
|
|
fixed20_12 num_line_pair;
|
|
|
|
fixed20_12 estimated_width;
|
|
|
|
fixed20_12 worst_case_latency;
|
|
|
|
fixed20_12 consumption_rate;
|
|
|
|
fixed20_12 active_time;
|
|
|
|
fixed20_12 dbpp;
|
|
|
|
fixed20_12 priority_mark_max;
|
|
|
|
fixed20_12 priority_mark;
|
|
|
|
fixed20_12 sclk;
|
|
|
|
};
|
|
|
|
|
2012-08-31 11:43:50 -06:00
|
|
|
static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
|
2012-10-25 14:58:55 -06:00
|
|
|
struct radeon_crtc *crtc,
|
|
|
|
struct rs690_watermark *wm,
|
|
|
|
bool low)
|
2009-07-13 13:04:08 -06:00
|
|
|
{
|
|
|
|
struct drm_display_mode *mode = &crtc->base.mode;
|
|
|
|
fixed20_12 a, b, c;
|
|
|
|
fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width;
|
|
|
|
fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency;
|
2012-10-25 14:58:55 -06:00
|
|
|
fixed20_12 sclk, core_bandwidth, max_bandwidth;
|
|
|
|
u32 selected_sclk;
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
if (!crtc->base.enabled) {
|
|
|
|
/* FIXME: wouldn't it better to set priority mark to maximum */
|
|
|
|
wm->lb_request_fifo_depth = 4;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-10-25 14:58:55 -06:00
|
|
|
if (((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880)) &&
|
|
|
|
(rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled)
|
|
|
|
selected_sclk = radeon_dpm_get_sclk(rdev, low);
|
|
|
|
else
|
|
|
|
selected_sclk = rdev->pm.current_sclk;
|
|
|
|
|
|
|
|
/* sclk in Mhz */
|
|
|
|
a.full = dfixed_const(100);
|
|
|
|
sclk.full = dfixed_const(selected_sclk);
|
|
|
|
sclk.full = dfixed_div(sclk, a);
|
|
|
|
|
|
|
|
/* core_bandwidth = sclk(Mhz) * 16 */
|
|
|
|
a.full = dfixed_const(16);
|
|
|
|
core_bandwidth.full = dfixed_div(rdev->pm.sclk, a);
|
|
|
|
|
2010-04-27 19:46:42 -06:00
|
|
|
if (crtc->vsc.full > dfixed_const(2))
|
|
|
|
wm->num_line_pair.full = dfixed_const(2);
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2010-04-27 19:46:42 -06:00
|
|
|
wm->num_line_pair.full = dfixed_const(1);
|
|
|
|
|
|
|
|
b.full = dfixed_const(mode->crtc_hdisplay);
|
|
|
|
c.full = dfixed_const(256);
|
|
|
|
a.full = dfixed_div(b, c);
|
|
|
|
request_fifo_depth.full = dfixed_mul(a, wm->num_line_pair);
|
|
|
|
request_fifo_depth.full = dfixed_ceil(request_fifo_depth);
|
|
|
|
if (a.full < dfixed_const(4)) {
|
2009-07-13 13:04:08 -06:00
|
|
|
wm->lb_request_fifo_depth = 4;
|
|
|
|
} else {
|
2010-04-27 19:46:42 -06:00
|
|
|
wm->lb_request_fifo_depth = dfixed_trunc(request_fifo_depth);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Determine consumption rate
|
|
|
|
* pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000)
|
|
|
|
* vtaps = number of vertical taps,
|
|
|
|
* vsc = vertical scaling ratio, defined as source/destination
|
|
|
|
* hsc = horizontal scaling ration, defined as source/destination
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(mode->clock);
|
|
|
|
b.full = dfixed_const(1000);
|
|
|
|
a.full = dfixed_div(a, b);
|
|
|
|
pclk.full = dfixed_div(b, a);
|
2009-07-13 13:04:08 -06:00
|
|
|
if (crtc->rmx_type != RMX_OFF) {
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(2);
|
2009-07-13 13:04:08 -06:00
|
|
|
if (crtc->vsc.full > b.full)
|
|
|
|
b.full = crtc->vsc.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_mul(b, crtc->hsc);
|
|
|
|
c.full = dfixed_const(2);
|
|
|
|
b.full = dfixed_div(b, c);
|
|
|
|
consumption_time.full = dfixed_div(pclk, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
|
|
|
consumption_time.full = pclk.full;
|
|
|
|
}
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(1);
|
|
|
|
wm->consumption_rate.full = dfixed_div(a, consumption_time);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
|
|
|
|
/* Determine line time
|
|
|
|
* LineTime = total time for one line of displayhtotal
|
|
|
|
* LineTime = total number of horizontal pixels
|
|
|
|
* pclk = pixel clock period(ns)
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
|
|
|
|
line_time.full = dfixed_mul(a, pclk);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
/* Determine active time
|
|
|
|
* ActiveTime = time of active region of display within one line,
|
|
|
|
* hactive = total number of horizontal active pixels
|
|
|
|
* htotal = total number of horizontal pixels
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(crtc->base.mode.crtc_htotal);
|
|
|
|
b.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
|
|
|
|
wm->active_time.full = dfixed_mul(line_time, b);
|
|
|
|
wm->active_time.full = dfixed_div(wm->active_time, a);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
/* Maximun bandwidth is the minimun bandwidth of all component */
|
2012-10-25 14:58:55 -06:00
|
|
|
max_bandwidth = core_bandwidth;
|
2010-06-12 09:50:13 -06:00
|
|
|
if (rdev->mc.igp_sideport_enabled) {
|
2012-10-25 14:58:55 -06:00
|
|
|
if (max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
|
2009-07-13 13:04:08 -06:00
|
|
|
rdev->pm.sideport_bandwidth.full)
|
2012-10-25 14:58:55 -06:00
|
|
|
max_bandwidth = rdev->pm.sideport_bandwidth;
|
2013-11-07 16:16:23 -07:00
|
|
|
read_delay_latency.full = dfixed_const(370 * 800);
|
|
|
|
a.full = dfixed_const(1000);
|
|
|
|
b.full = dfixed_div(rdev->pm.igp_sideport_mclk, a);
|
|
|
|
read_delay_latency.full = dfixed_div(read_delay_latency, b);
|
|
|
|
read_delay_latency.full = dfixed_mul(read_delay_latency, a);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2012-10-25 14:58:55 -06:00
|
|
|
if (max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
|
2009-07-13 13:04:08 -06:00
|
|
|
rdev->pm.k8_bandwidth.full)
|
2012-10-25 14:58:55 -06:00
|
|
|
max_bandwidth = rdev->pm.k8_bandwidth;
|
|
|
|
if (max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
|
2009-07-13 13:04:08 -06:00
|
|
|
rdev->pm.ht_bandwidth.full)
|
2012-10-25 14:58:55 -06:00
|
|
|
max_bandwidth = rdev->pm.ht_bandwidth;
|
2010-04-27 19:46:42 -06:00
|
|
|
read_delay_latency.full = dfixed_const(5000);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(16);
|
2012-10-25 14:58:55 -06:00
|
|
|
sclk.full = dfixed_mul(max_bandwidth, a);
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(1000);
|
2012-10-25 14:58:55 -06:00
|
|
|
sclk.full = dfixed_div(a, sclk);
|
2009-07-13 13:04:08 -06:00
|
|
|
/* Determine chunk time
|
|
|
|
* ChunkTime = the time it takes the DCP to send one chunk of data
|
|
|
|
* to the LB which consists of pipeline delay and inter chunk gap
|
|
|
|
* sclk = system clock(ns)
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(256 * 13);
|
2012-10-25 14:58:55 -06:00
|
|
|
chunk_time.full = dfixed_mul(sclk, a);
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(10);
|
|
|
|
chunk_time.full = dfixed_div(chunk_time, a);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
/* Determine the worst case latency
|
|
|
|
* NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
|
|
|
|
* WorstCaseLatency = worst case time from urgent to when the MC starts
|
|
|
|
* to return data
|
|
|
|
* READ_DELAY_IDLE_MAX = constant of 1us
|
|
|
|
* ChunkTime = time it takes the DCP to send one chunk of data to the LB
|
|
|
|
* which consists of pipeline delay and inter chunk gap
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
if (dfixed_trunc(wm->num_line_pair) > 1) {
|
|
|
|
a.full = dfixed_const(3);
|
|
|
|
wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
|
2009-07-13 13:04:08 -06:00
|
|
|
wm->worst_case_latency.full += read_delay_latency.full;
|
|
|
|
} else {
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(2);
|
|
|
|
wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
|
2009-07-13 13:04:08 -06:00
|
|
|
wm->worst_case_latency.full += read_delay_latency.full;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Determine the tolerable latency
|
|
|
|
* TolerableLatency = Any given request has only 1 line time
|
|
|
|
* for the data to be returned
|
|
|
|
* LBRequestFifoDepth = Number of chunk requests the LB can
|
|
|
|
* put into the request FIFO for a display
|
|
|
|
* LineTime = total time for one line of display
|
|
|
|
* ChunkTime = the time it takes the DCP to send one chunk
|
|
|
|
* of data to the LB which consists of
|
|
|
|
* pipeline delay and inter chunk gap
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
if ((2+wm->lb_request_fifo_depth) >= dfixed_trunc(request_fifo_depth)) {
|
2009-07-13 13:04:08 -06:00
|
|
|
tolerable_latency.full = line_time.full;
|
|
|
|
} else {
|
2010-04-27 19:46:42 -06:00
|
|
|
tolerable_latency.full = dfixed_const(wm->lb_request_fifo_depth - 2);
|
2009-07-13 13:04:08 -06:00
|
|
|
tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
tolerable_latency.full = dfixed_mul(tolerable_latency, chunk_time);
|
2009-07-13 13:04:08 -06:00
|
|
|
tolerable_latency.full = line_time.full - tolerable_latency.full;
|
|
|
|
}
|
|
|
|
/* We assume worst case 32bits (4 bytes) */
|
2010-04-27 19:46:42 -06:00
|
|
|
wm->dbpp.full = dfixed_const(4 * 8);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
/* Determine the maximum priority mark
|
|
|
|
* width = viewport width in pixels
|
|
|
|
*/
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(16);
|
|
|
|
wm->priority_mark_max.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
|
|
|
|
wm->priority_mark_max.full = dfixed_div(wm->priority_mark_max, a);
|
|
|
|
wm->priority_mark_max.full = dfixed_ceil(wm->priority_mark_max);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
/* Determine estimated width */
|
|
|
|
estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
estimated_width.full = dfixed_div(estimated_width, consumption_time);
|
|
|
|
if (dfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
|
|
|
|
wm->priority_mark.full = dfixed_const(10);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2010-04-27 19:46:42 -06:00
|
|
|
a.full = dfixed_const(16);
|
|
|
|
wm->priority_mark.full = dfixed_div(estimated_width, a);
|
|
|
|
wm->priority_mark.full = dfixed_ceil(wm->priority_mark);
|
2009-07-13 13:04:08 -06:00
|
|
|
wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-10-25 14:58:55 -06:00
|
|
|
static void rs690_compute_mode_priority(struct radeon_device *rdev,
|
|
|
|
struct rs690_watermark *wm0,
|
|
|
|
struct rs690_watermark *wm1,
|
|
|
|
struct drm_display_mode *mode0,
|
|
|
|
struct drm_display_mode *mode1,
|
|
|
|
u32 *d1mode_priority_a_cnt,
|
|
|
|
u32 *d2mode_priority_a_cnt)
|
2009-07-13 13:04:08 -06:00
|
|
|
{
|
|
|
|
fixed20_12 priority_mark02, priority_mark12, fill_rate;
|
|
|
|
fixed20_12 a, b;
|
|
|
|
|
2012-10-25 14:58:55 -06:00
|
|
|
*d1mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
|
|
|
|
*d2mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
|
2009-07-13 13:04:08 -06:00
|
|
|
|
|
|
|
if (mode0 && mode1) {
|
2012-10-25 14:58:55 -06:00
|
|
|
if (dfixed_trunc(wm0->dbpp) > 64)
|
|
|
|
a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = wm0->num_line_pair.full;
|
|
|
|
if (dfixed_trunc(wm1->dbpp) > 64)
|
|
|
|
b.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2012-10-25 14:58:55 -06:00
|
|
|
b.full = wm1->num_line_pair.full;
|
2009-07-13 13:04:08 -06:00
|
|
|
a.full += b.full;
|
2012-10-25 14:58:55 -06:00
|
|
|
fill_rate.full = dfixed_div(wm0->sclk, a);
|
|
|
|
if (wm0->consumption_rate.full > fill_rate.full) {
|
|
|
|
b.full = wm0->consumption_rate.full - fill_rate.full;
|
|
|
|
b.full = dfixed_mul(b, wm0->active_time);
|
|
|
|
a.full = dfixed_mul(wm0->worst_case_latency,
|
|
|
|
wm0->consumption_rate);
|
2009-07-13 13:04:08 -06:00
|
|
|
a.full = a.full + b.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark02.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = dfixed_mul(wm0->worst_case_latency,
|
|
|
|
wm0->consumption_rate);
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark02.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2012-10-25 14:58:55 -06:00
|
|
|
if (wm1->consumption_rate.full > fill_rate.full) {
|
|
|
|
b.full = wm1->consumption_rate.full - fill_rate.full;
|
|
|
|
b.full = dfixed_mul(b, wm1->active_time);
|
|
|
|
a.full = dfixed_mul(wm1->worst_case_latency,
|
|
|
|
wm1->consumption_rate);
|
2009-07-13 13:04:08 -06:00
|
|
|
a.full = a.full + b.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark12.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = dfixed_mul(wm1->worst_case_latency,
|
|
|
|
wm1->consumption_rate);
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark12.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2012-10-25 14:58:55 -06:00
|
|
|
if (wm0->priority_mark.full > priority_mark02.full)
|
|
|
|
priority_mark02.full = wm0->priority_mark.full;
|
|
|
|
if (wm0->priority_mark_max.full > priority_mark02.full)
|
|
|
|
priority_mark02.full = wm0->priority_mark_max.full;
|
|
|
|
if (wm1->priority_mark.full > priority_mark12.full)
|
|
|
|
priority_mark12.full = wm1->priority_mark.full;
|
|
|
|
if (wm1->priority_mark_max.full > priority_mark12.full)
|
|
|
|
priority_mark12.full = wm1->priority_mark_max.full;
|
|
|
|
*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
|
|
|
|
*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
|
2010-03-30 22:33:27 -06:00
|
|
|
if (rdev->disp_priority == 2) {
|
2012-10-25 14:58:55 -06:00
|
|
|
*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
|
|
|
|
*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
|
2010-03-30 22:33:27 -06:00
|
|
|
}
|
2009-07-13 13:04:08 -06:00
|
|
|
} else if (mode0) {
|
2012-10-25 14:58:55 -06:00
|
|
|
if (dfixed_trunc(wm0->dbpp) > 64)
|
|
|
|
a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = wm0->num_line_pair.full;
|
|
|
|
fill_rate.full = dfixed_div(wm0->sclk, a);
|
|
|
|
if (wm0->consumption_rate.full > fill_rate.full) {
|
|
|
|
b.full = wm0->consumption_rate.full - fill_rate.full;
|
|
|
|
b.full = dfixed_mul(b, wm0->active_time);
|
|
|
|
a.full = dfixed_mul(wm0->worst_case_latency,
|
|
|
|
wm0->consumption_rate);
|
2009-07-13 13:04:08 -06:00
|
|
|
a.full = a.full + b.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark02.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = dfixed_mul(wm0->worst_case_latency,
|
|
|
|
wm0->consumption_rate);
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark02.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2012-10-25 14:58:55 -06:00
|
|
|
if (wm0->priority_mark.full > priority_mark02.full)
|
|
|
|
priority_mark02.full = wm0->priority_mark.full;
|
|
|
|
if (wm0->priority_mark_max.full > priority_mark02.full)
|
|
|
|
priority_mark02.full = wm0->priority_mark_max.full;
|
|
|
|
*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
|
2010-03-30 22:33:27 -06:00
|
|
|
if (rdev->disp_priority == 2)
|
2012-10-25 14:58:55 -06:00
|
|
|
*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
|
2010-08-02 10:13:46 -06:00
|
|
|
} else if (mode1) {
|
2012-10-25 14:58:55 -06:00
|
|
|
if (dfixed_trunc(wm1->dbpp) > 64)
|
|
|
|
a.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
|
2009-07-13 13:04:08 -06:00
|
|
|
else
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = wm1->num_line_pair.full;
|
|
|
|
fill_rate.full = dfixed_div(wm1->sclk, a);
|
|
|
|
if (wm1->consumption_rate.full > fill_rate.full) {
|
|
|
|
b.full = wm1->consumption_rate.full - fill_rate.full;
|
|
|
|
b.full = dfixed_mul(b, wm1->active_time);
|
|
|
|
a.full = dfixed_mul(wm1->worst_case_latency,
|
|
|
|
wm1->consumption_rate);
|
2009-07-13 13:04:08 -06:00
|
|
|
a.full = a.full + b.full;
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark12.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
} else {
|
2012-10-25 14:58:55 -06:00
|
|
|
a.full = dfixed_mul(wm1->worst_case_latency,
|
|
|
|
wm1->consumption_rate);
|
2010-04-27 19:46:42 -06:00
|
|
|
b.full = dfixed_const(16 * 1000);
|
|
|
|
priority_mark12.full = dfixed_div(a, b);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2012-10-25 14:58:55 -06:00
|
|
|
if (wm1->priority_mark.full > priority_mark12.full)
|
|
|
|
priority_mark12.full = wm1->priority_mark.full;
|
|
|
|
if (wm1->priority_mark_max.full > priority_mark12.full)
|
|
|
|
priority_mark12.full = wm1->priority_mark_max.full;
|
|
|
|
*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
|
2010-03-30 22:33:27 -06:00
|
|
|
if (rdev->disp_priority == 2)
|
2012-10-25 14:58:55 -06:00
|
|
|
*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2012-10-25 14:58:55 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
void rs690_bandwidth_update(struct radeon_device *rdev)
|
|
|
|
{
|
|
|
|
struct drm_display_mode *mode0 = NULL;
|
|
|
|
struct drm_display_mode *mode1 = NULL;
|
|
|
|
struct rs690_watermark wm0_high, wm0_low;
|
|
|
|
struct rs690_watermark wm1_high, wm1_low;
|
|
|
|
u32 tmp;
|
|
|
|
u32 d1mode_priority_a_cnt, d1mode_priority_b_cnt;
|
|
|
|
u32 d2mode_priority_a_cnt, d2mode_priority_b_cnt;
|
|
|
|
|
2014-11-03 07:57:46 -07:00
|
|
|
if (!rdev->mode_info.mode_config_initialized)
|
|
|
|
return;
|
|
|
|
|
2012-10-25 14:58:55 -06:00
|
|
|
radeon_update_display_priority(rdev);
|
|
|
|
|
|
|
|
if (rdev->mode_info.crtcs[0]->base.enabled)
|
|
|
|
mode0 = &rdev->mode_info.crtcs[0]->base.mode;
|
|
|
|
if (rdev->mode_info.crtcs[1]->base.enabled)
|
|
|
|
mode1 = &rdev->mode_info.crtcs[1]->base.mode;
|
|
|
|
/*
|
|
|
|
* Set display0/1 priority up in the memory controller for
|
|
|
|
* modes if the user specifies HIGH for displaypriority
|
|
|
|
* option.
|
|
|
|
*/
|
|
|
|
if ((rdev->disp_priority == 2) &&
|
|
|
|
((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))) {
|
|
|
|
tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
|
|
|
|
tmp &= C_000104_MC_DISP0R_INIT_LAT;
|
|
|
|
tmp &= C_000104_MC_DISP1R_INIT_LAT;
|
|
|
|
if (mode0)
|
|
|
|
tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
|
|
|
|
if (mode1)
|
|
|
|
tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
|
|
|
|
WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
|
|
|
|
}
|
|
|
|
rs690_line_buffer_adjust(rdev, mode0, mode1);
|
|
|
|
|
|
|
|
if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
|
|
|
|
WREG32(R_006C9C_DCP_CONTROL, 0);
|
|
|
|
if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
|
|
|
|
WREG32(R_006C9C_DCP_CONTROL, 2);
|
|
|
|
|
|
|
|
rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_high, false);
|
|
|
|
rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_high, false);
|
|
|
|
|
|
|
|
rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_low, true);
|
|
|
|
rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_low, true);
|
|
|
|
|
|
|
|
tmp = (wm0_high.lb_request_fifo_depth - 1);
|
|
|
|
tmp |= (wm1_high.lb_request_fifo_depth - 1) << 16;
|
|
|
|
WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
|
|
|
|
|
|
|
|
rs690_compute_mode_priority(rdev,
|
|
|
|
&wm0_high, &wm1_high,
|
|
|
|
mode0, mode1,
|
|
|
|
&d1mode_priority_a_cnt, &d2mode_priority_a_cnt);
|
|
|
|
rs690_compute_mode_priority(rdev,
|
|
|
|
&wm0_low, &wm1_low,
|
|
|
|
mode0, mode1,
|
|
|
|
&d1mode_priority_b_cnt, &d2mode_priority_b_cnt);
|
2010-08-02 10:13:46 -06:00
|
|
|
|
|
|
|
WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
|
2012-10-25 14:58:55 -06:00
|
|
|
WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_b_cnt);
|
2010-08-02 10:13:46 -06:00
|
|
|
WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
|
2012-10-25 14:58:55 -06:00
|
|
|
WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_b_cnt);
|
2009-07-13 13:04:08 -06:00
|
|
|
}
|
2009-06-05 06:42:42 -06:00
|
|
|
|
|
|
|
uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg)
|
|
|
|
{
|
2013-09-03 17:00:09 -06:00
|
|
|
unsigned long flags;
|
2009-06-05 06:42:42 -06:00
|
|
|
uint32_t r;
|
|
|
|
|
2013-09-03 17:00:09 -06:00
|
|
|
spin_lock_irqsave(&rdev->mc_idx_lock, flags);
|
2009-10-01 01:39:24 -06:00
|
|
|
WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg));
|
|
|
|
r = RREG32(R_00007C_MC_DATA);
|
|
|
|
WREG32(R_000078_MC_INDEX, ~C_000078_MC_IND_ADDR);
|
2013-09-03 17:00:09 -06:00
|
|
|
spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
|
2009-06-05 06:42:42 -06:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
void rs690_mc_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
|
|
|
|
{
|
2013-09-03 17:00:09 -06:00
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&rdev->mc_idx_lock, flags);
|
2009-10-01 01:39:24 -06:00
|
|
|
WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg) |
|
|
|
|
S_000078_MC_IND_WR_EN(1));
|
|
|
|
WREG32(R_00007C_MC_DATA, v);
|
|
|
|
WREG32(R_000078_MC_INDEX, 0x7F);
|
2013-09-03 17:00:09 -06:00
|
|
|
spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
|
2009-10-01 01:39:24 -06:00
|
|
|
}
|
|
|
|
|
2012-08-31 11:43:50 -06:00
|
|
|
static void rs690_mc_program(struct radeon_device *rdev)
|
2009-10-01 01:39:24 -06:00
|
|
|
{
|
|
|
|
struct rv515_mc_save save;
|
|
|
|
|
|
|
|
/* Stops all mc clients */
|
|
|
|
rv515_mc_stop(rdev, &save);
|
|
|
|
|
|
|
|
/* Wait for mc idle */
|
|
|
|
if (rs690_mc_wait_for_idle(rdev))
|
|
|
|
dev_warn(rdev->dev, "Wait MC idle timeout before updating MC.\n");
|
|
|
|
/* Program MC, should be a 32bits limited address space */
|
|
|
|
WREG32_MC(R_000100_MCCFG_FB_LOCATION,
|
|
|
|
S_000100_MC_FB_START(rdev->mc.vram_start >> 16) |
|
|
|
|
S_000100_MC_FB_TOP(rdev->mc.vram_end >> 16));
|
|
|
|
WREG32(R_000134_HDP_FB_LOCATION,
|
|
|
|
S_000134_HDP_FB_START(rdev->mc.vram_start >> 16));
|
|
|
|
|
|
|
|
rv515_mc_resume(rdev, &save);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int rs690_startup(struct radeon_device *rdev)
|
|
|
|
{
|
|
|
|
int r;
|
|
|
|
|
|
|
|
rs690_mc_program(rdev);
|
|
|
|
/* Resume clock */
|
|
|
|
rv515_clock_startup(rdev);
|
|
|
|
/* Initialize GPU configuration (# pipes, ...) */
|
|
|
|
rs690_gpu_init(rdev);
|
|
|
|
/* Initialize GART (initialize after TTM so we can allocate
|
|
|
|
* memory through TTM but finalize after TTM) */
|
|
|
|
r = rs400_gart_enable(rdev);
|
|
|
|
if (r)
|
|
|
|
return r;
|
2010-08-27 16:25:25 -06:00
|
|
|
|
|
|
|
/* allocate wb buffer */
|
|
|
|
r = radeon_wb_init(rdev);
|
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
|
2011-11-20 13:45:34 -07:00
|
|
|
r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
|
|
|
|
if (r) {
|
|
|
|
dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
/* Enable IRQ */
|
radeon: Fix system hang issue when using KMS with older cards
The current radeon driver initialization routines, when using KMS, are written
so that the IRQ installation routine is called before initializing the WB buffer
and the CP rings. With some ASICs, though, the IRQ routine tries to access the
GFX_INDEX ring causing a call to RREG32 with the value of -1 in
radeon_fence_read. This, in turn causes the system to completely hang with some
cards, requiring a hard reset.
A call stack that can cause such a hang looks like this (using rv515 ASIC for the
example here):
* rv515_init (rv515.c)
* radeon_irq_kms_init (radeon_irq_kms.c)
* drm_irq_install (drm_irq.c)
* radeon_driver_irq_preinstall_kms (radeon_irq_kms.c)
* rs600_irq_process (rs600.c)
* radeon_fence_process - due to SW interrupt (radeon_fence.c)
* radeon_fence_read (radeon_fence.c)
* hang due to RREG32(-1)
The patch moves the IRQ installation to the card startup routine, after the ring
has been initialized, but before the IRQ has been set. This fixes the issue, but
requires a check to see if the IRQ is already installed, as is the case in the
system resume codepath.
I have tested the patch on three machines using the rv515, the rv770 and the
evergreen ASIC. They worked without issues.
This seems to be a known issue and has been reported on several bug tracking
sites by various distributions (see links below). Most of reports recommend
booting the system with KMS disabled and then enabling KMS by reloading the
radeon module. For some reason, this was indeed a usable workaround, however,
UMS is now deprecated and disabled by default.
Bug reports:
https://bugzilla.redhat.com/show_bug.cgi?id=845745
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/561789
https://bbs.archlinux.org/viewtopic.php?id=156964
Signed-off-by: Adis Hamzić <adis@hamzadis.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
2013-06-02 08:47:54 -06:00
|
|
|
if (!rdev->irq.installed) {
|
|
|
|
r = radeon_irq_kms_init(rdev);
|
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2009-09-30 14:18:43 -06:00
|
|
|
rs600_irq_set(rdev);
|
2010-01-07 04:39:21 -07:00
|
|
|
rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
|
2009-10-01 01:39:24 -06:00
|
|
|
/* 1M ring buffer */
|
|
|
|
r = r100_cp_init(rdev, 1024 * 1024);
|
|
|
|
if (r) {
|
2011-01-28 15:32:04 -07:00
|
|
|
dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
|
2009-10-01 01:39:24 -06:00
|
|
|
return r;
|
|
|
|
}
|
drm/radeon: introduce a sub allocator and convert ib pool to it v4
Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.
v2 Harmonize pool handling and add suspend/resume callback to pin/unpin
sa bo (tested on rv280, rv370, r420, rv515, rv610, rv710, redwood, cayman,
rs480, rs690, rs880)
v3 Simplify allocator
v4 Fix radeon_ib_get error path to properly free fence
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-11-15 09:48:34 -07:00
|
|
|
|
2012-07-05 03:55:34 -06:00
|
|
|
r = radeon_ib_pool_init(rdev);
|
|
|
|
if (r) {
|
|
|
|
dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
|
drm/radeon: introduce a sub allocator and convert ib pool to it v4
Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.
v2 Harmonize pool handling and add suspend/resume callback to pin/unpin
sa bo (tested on rv280, rv370, r420, rv515, rv610, rv710, redwood, cayman,
rs480, rs690, rs880)
v3 Simplify allocator
v4 Fix radeon_ib_get error path to properly free fence
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-11-15 09:48:34 -07:00
|
|
|
return r;
|
2012-07-05 03:55:34 -06:00
|
|
|
}
|
drm/radeon: introduce a sub allocator and convert ib pool to it v4
Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.
v2 Harmonize pool handling and add suspend/resume callback to pin/unpin
sa bo (tested on rv280, rv370, r420, rv515, rv610, rv710, redwood, cayman,
rs480, rs690, rs880)
v3 Simplify allocator
v4 Fix radeon_ib_get error path to properly free fence
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-11-15 09:48:34 -07:00
|
|
|
|
2014-12-22 15:26:51 -07:00
|
|
|
r = radeon_audio_init(rdev);
|
2012-06-04 15:18:51 -06:00
|
|
|
if (r) {
|
|
|
|
dev_err(rdev->dev, "failed initializing audio\n");
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int rs690_resume(struct radeon_device *rdev)
|
|
|
|
{
|
2012-02-20 15:57:20 -07:00
|
|
|
int r;
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
/* Make sur GART are not working */
|
|
|
|
rs400_gart_disable(rdev);
|
|
|
|
/* Resume clock before doing reset */
|
|
|
|
rv515_clock_startup(rdev);
|
|
|
|
/* Reset gpu before posting otherwise ATOM will enter infinite loop */
|
2010-03-09 07:45:11 -07:00
|
|
|
if (radeon_asic_reset(rdev)) {
|
2009-10-01 01:39:24 -06:00
|
|
|
dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
|
|
|
|
RREG32(R_000E40_RBBM_STATUS),
|
|
|
|
RREG32(R_0007C0_CP_STAT));
|
|
|
|
}
|
|
|
|
/* post */
|
|
|
|
atom_asic_init(rdev->mode_info.atom_context);
|
|
|
|
/* Resume clock after posting */
|
|
|
|
rv515_clock_startup(rdev);
|
2009-12-08 21:15:38 -07:00
|
|
|
/* Initialize surface registers */
|
|
|
|
radeon_surface_init(rdev);
|
drm/radeon: introduce a sub allocator and convert ib pool to it v4
Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.
v2 Harmonize pool handling and add suspend/resume callback to pin/unpin
sa bo (tested on rv280, rv370, r420, rv515, rv610, rv710, redwood, cayman,
rs480, rs690, rs880)
v3 Simplify allocator
v4 Fix radeon_ib_get error path to properly free fence
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-11-15 09:48:34 -07:00
|
|
|
|
|
|
|
rdev->accel_working = true;
|
2012-02-20 15:57:20 -07:00
|
|
|
r = rs690_startup(rdev);
|
|
|
|
if (r) {
|
|
|
|
rdev->accel_working = false;
|
|
|
|
}
|
|
|
|
return r;
|
2009-10-01 01:39:24 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
int rs690_suspend(struct radeon_device *rdev)
|
|
|
|
{
|
2013-12-18 12:07:14 -07:00
|
|
|
radeon_pm_suspend(rdev);
|
2014-12-03 15:07:01 -07:00
|
|
|
radeon_audio_fini(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
r100_cp_disable(rdev);
|
2010-08-27 16:25:25 -06:00
|
|
|
radeon_wb_disable(rdev);
|
2009-09-30 14:18:43 -06:00
|
|
|
rs600_irq_disable(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
rs400_gart_disable(rdev);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void rs690_fini(struct radeon_device *rdev)
|
|
|
|
{
|
2013-12-18 12:07:14 -07:00
|
|
|
radeon_pm_fini(rdev);
|
2014-12-03 15:07:01 -07:00
|
|
|
radeon_audio_fini(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
r100_cp_fini(rdev);
|
2010-08-27 16:25:25 -06:00
|
|
|
radeon_wb_fini(rdev);
|
2012-07-05 03:55:34 -06:00
|
|
|
radeon_ib_pool_fini(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
radeon_gem_fini(rdev);
|
|
|
|
rs400_gart_fini(rdev);
|
|
|
|
radeon_irq_kms_fini(rdev);
|
|
|
|
radeon_fence_driver_fini(rdev);
|
2009-11-20 06:29:23 -07:00
|
|
|
radeon_bo_fini(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
radeon_atombios_fini(rdev);
|
|
|
|
kfree(rdev->bios);
|
|
|
|
rdev->bios = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
int rs690_init(struct radeon_device *rdev)
|
|
|
|
{
|
|
|
|
int r;
|
|
|
|
|
|
|
|
/* Disable VGA */
|
|
|
|
rv515_vga_render_disable(rdev);
|
|
|
|
/* Initialize scratch registers */
|
|
|
|
radeon_scratch_init(rdev);
|
|
|
|
/* Initialize surface registers */
|
|
|
|
radeon_surface_init(rdev);
|
2010-07-14 20:13:50 -06:00
|
|
|
/* restore some register to sane defaults */
|
|
|
|
r100_restore_sanity(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
/* TODO: disable VGA need to use VGA request */
|
|
|
|
/* BIOS*/
|
|
|
|
if (!radeon_get_bios(rdev)) {
|
|
|
|
if (ASIC_IS_AVIVO(rdev))
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
if (rdev->is_atom_bios) {
|
|
|
|
r = radeon_atombios_init(rdev);
|
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
} else {
|
|
|
|
dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
/* Reset gpu before posting otherwise ATOM will enter infinite loop */
|
2010-03-09 07:45:11 -07:00
|
|
|
if (radeon_asic_reset(rdev)) {
|
2009-10-01 01:39:24 -06:00
|
|
|
dev_warn(rdev->dev,
|
|
|
|
"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
|
|
|
|
RREG32(R_000E40_RBBM_STATUS),
|
|
|
|
RREG32(R_0007C0_CP_STAT));
|
|
|
|
}
|
|
|
|
/* check if cards are posted or not */
|
2009-11-30 21:06:31 -07:00
|
|
|
if (radeon_boot_test_post_card(rdev) == false)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
/* Initialize clocks */
|
|
|
|
radeon_get_clock_info(rdev->ddev);
|
drm/radeon/kms: simplify memory controller setup V2
Get rid of _location and use _start/_end also simplify the
computation of vram_start|end & gtt_start|end. For R1XX-R2XX
we place VRAM at the same address of PCI aperture, those GPU
shouldn't have much memory and seems to behave better when
setup that way. For R3XX and newer we place VRAM at 0. For
R6XX-R7XX AGP we place VRAM before or after AGP aperture this
might limit to limit the VRAM size but it's very unlikely.
For IGP we don't change the VRAM placement.
Tested on (compiz,quake3,suspend/resume):
PCI/PCIE:RV280,R420,RV515,RV570,RV610,RV710
AGP:RV100,RV280,R420,RV350,RV620(RPB*),RV730
IGP:RS480(RPB*),RS690,RS780(RPB*),RS880
RPB: resume previously broken
V2 correct commit message to reflect more accurately the bug
and move VRAM placement to 0 for most of the GPU to avoid
limiting VRAM.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2010-02-17 14:54:29 -07:00
|
|
|
/* initialize memory controller */
|
|
|
|
rs690_mc_init(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
rv515_debugfs(rdev);
|
|
|
|
/* Fence driver */
|
2011-11-20 13:45:34 -07:00
|
|
|
r = radeon_fence_driver_init(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
/* Memory manager */
|
2009-11-20 06:29:23 -07:00
|
|
|
r = radeon_bo_init(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
r = rs400_gart_init(rdev);
|
|
|
|
if (r)
|
|
|
|
return r;
|
|
|
|
rs600_set_safe_registers(rdev);
|
drm/radeon: introduce a sub allocator and convert ib pool to it v4
Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.
v2 Harmonize pool handling and add suspend/resume callback to pin/unpin
sa bo (tested on rv280, rv370, r420, rv515, rv610, rv710, redwood, cayman,
rs480, rs690, rs880)
v3 Simplify allocator
v4 Fix radeon_ib_get error path to properly free fence
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
2011-11-15 09:48:34 -07:00
|
|
|
|
2013-12-18 12:07:14 -07:00
|
|
|
/* Initialize power management */
|
|
|
|
radeon_pm_init(rdev);
|
|
|
|
|
2009-10-01 01:39:24 -06:00
|
|
|
rdev->accel_working = true;
|
|
|
|
r = rs690_startup(rdev);
|
|
|
|
if (r) {
|
|
|
|
/* Somethings want wront with the accel init stop accel */
|
|
|
|
dev_err(rdev->dev, "Disabling GPU acceleration\n");
|
|
|
|
r100_cp_fini(rdev);
|
2010-08-27 16:25:25 -06:00
|
|
|
radeon_wb_fini(rdev);
|
2012-07-05 03:55:34 -06:00
|
|
|
radeon_ib_pool_fini(rdev);
|
2009-10-01 01:39:24 -06:00
|
|
|
rs400_gart_fini(rdev);
|
|
|
|
radeon_irq_kms_fini(rdev);
|
|
|
|
rdev->accel_working = false;
|
|
|
|
}
|
|
|
|
return 0;
|
2009-06-05 06:42:42 -06:00
|
|
|
}
|