Re: Regression 2.6.34->2.6.35-rc4: radeaon KMS an RS690 broken

From: Alex Deucher
Date: Wed Jul 14 2010 - 18:17:14 EST

Next message: Nicolas Kaiser: "[PATCH] drm/i830: fixed brace and spacing coding style issues"
Previous message: David Rientjes: "Re: [S+Q2 06/19] slub: Check kasprintf results inkmem_cache_init()"
In reply to: Jerome Glisse: "Re: Regression 2.6.34->2.6.35-rc4: radeaon KMS an RS690 broken"
Next in thread: Torsten Kaiser: "Re: Regression 2.6.34->2.6.35-rc4: radeaon KMS an RS690 broken"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Wed, Jul 14, 2010 at 4:05 PM, Torsten Kaiser
<just.for.lkml@xxxxxxxxxxxxxx> wrote:
> On Wed, Jul 14, 2010 at 9:30 PM, Jerome Glisse <glisse@xxxxxxxxxxxxxxx> wrote:
>> On 07/14/2010 02:51 PM, Torsten Kaiser wrote:
>>>
>>> On Tue, Jul 13, 2010 at 9:10 PM, Alex Deucher<alexdeucher@xxxxxxxxx>
>>> wrote:
>>>>
>>>> On Tue, Jul 13, 2010 at 2:29 PM, Torsten Kaiser
>>>> <just.for.lkml@xxxxxxxxxxxxxx> wrote:
>>>>>
>>>>> But the CP is still broken:
>>>>
>>>> Is this a regression? If so, can you bisect it?
>>>>
>>>> Alex
>>>
>>> I bisected it to this commit:
>>>
>>> d594e46ace22afa1621254f6f669e65430048153 is the first bad commit
>>> commit d594e46ace22afa1621254f6f669e65430048153
>>> Author: Jerome Glisse<jglisse@xxxxxxxxxx>
>>> Date: Wed Feb 17 21:54:29 2010 +0000
>>>
>>> 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@xxxxxxxxxx>
>>> Signed-off-by: Dave Airlie<airlied@xxxxxxxxxx>
>>>
>>> :040000 040000 05c1e456fcf6565aa8711e4933807956d0055cca
>>> 792c6be2bd161a52500c5e8d685ee651cd5af07e M drivers
>>>
>>> HTH, Torsten
>>>
>>>>> [ 0.426931] Linux agpgart interface v0.103
>>>>> [ 0.427092] [drm] Initialized drm 1.1.0 20060810
>>>>> [ 0.427196] [drm] radeon defaulting to kernel modesetting.
>>>>> [ 0.427255] [drm] radeon kernel modesetting enabled.
>>>>> [ 0.427372] radeon 0000:01:05.0: PCI INT A -> GSI 18 (level, low) ->
>>>>> IRQ 18
>>>>> [ 0.429659] [drm] initializing kernel modesetting (RS690
>>>>> 0x1002:0x791E).
>>>>> [ 0.429817] [drm] register mmio base: 0xFE9F0000
>>>>> [ 0.429876] [drm] register mmio size: 65536
>>>>> [ 0.430457] ATOM BIOS: ATI
>>>>> [ 0.430532] radeon 0000:01:05.0: VRAM: 32M 0xDE000000 - 0xDFFFFFFF
>>>>> (32M used)
>>>>> [ 0.430592] radeon 0000:01:05.0: GTT: 512M 0xBE000000 - 0xDDFFFFFF
>>>>> [ 0.430675] [drm] radeon: irq initialized.
>>>>> [ 0.430737] mtrr: type mismatch for fc000000,2000000 old:
>>>>> write-back new: write-comb
>>>>> ining
>>>>> [ 0.430811] [drm] Detected VRAM RAM=32M, BAR=32M
>>>>> [ 0.430868] [drm] RAM width 128bits DDR
>>>>> [ 0.431011] [TTM] Zone kernel: Available graphics memory: 2010234
>>>>> kiB.
>>>>> [ 0.431070] [TTM] Initializing pool allocator.
>>>>> [ 0.431147] [drm] radeon: 32M of VRAM memory ready
>>>>> [ 0.431205] [drm] radeon: 512M of GTT memory ready.
>>>>> [ 0.431266] [drm] GART: num cpu pages 131072, num gpu pages 131072
>>>>> [ 0.434654] [drm] radeon: 1 quad pipes, 1 z pipes initialized.
>>>>> [ 0.441719] [drm] Loading RS690/RS740 Microcode
>>>>> [ 0.441926] [drm] radeon: ring at 0x00000000BE000000
>>>>> [ 0.577118] [drm:r100_ring_test] *ERROR* radeon: ring test failed
>>>>> (sracth(0x15E4)=0x
>>>>> CAFEDEAD)
>>>>> [ 0.577192] [drm:r100_cp_init] *ERROR* radeon: cp isn't working
>>>>> (-22).
>>>>> [ 0.577252] radeon 0000:01:05.0: failled initializing CP (-22).
>>>>> [ 0.577310] radeon 0000:01:05.0: Disabling GPU acceleration
>>>>> [ 0.577440] [drm] radeon: cp finalized
>>>>> [ 0.578078] [drm] Default TV standard: NTSC
>>>>> [ 0.578314] [drm] Default TV standard: NTSC
>>>>> [ 0.578590] [drm] Radeon Display Connectors
>>>>> [ 0.578648] [drm] Connector 0:
>>>>> [ 0.578706] [drm] VGA
>>>>> [ 0.578764] [drm] DDC: 0x7e50 0x7e40 0x7e54 0x7e44 0x7e58 0x7e48
>>>>> 0x7e5c 0x7e4c
>>>>> [ 0.578837] [drm] Encoders:
>>>>> [ 0.578894] [drm] CRT1: INTERNAL_KLDSCP_DAC1
>>>>> [ 0.578952] [drm] Connector 1:
>>>>> [ 0.579010] [drm] S-video
>>>>> [ 0.579067] [drm] Encoders:
>>>>> [ 0.579124] [drm] TV1: INTERNAL_KLDSCP_DAC1
>>>>> [ 0.579182] [drm] Connector 2:
>>>>> [ 0.579239] [drm] HDMI-A
>>>>> [ 0.579297] [drm] DDC: 0x7e40 0x7e50 0x7e44 0x7e54 0x7e48 0x7e58
>>>>> 0x7e4c 0x7e5c
>>>>> [ 0.579369] [drm] Encoders:
>>>>> [ 0.579427] [drm] DFP3: INTERNAL_LVTM1
>>>>> [ 0.773375] [drm] fb mappable at 0xFC040000
>>>>> [ 0.773434] [drm] vram apper at 0xFC000000
>>>>> [ 0.773491] [drm] size 786432
>>>>> [ 0.773549] [drm] fb depth is 8
>>>>> [ 0.773606] [drm] pitch is 1024
>>>>> [ 0.773737] fbcon: radeondrmfb (fb0) is primary device
>>>>> [ 0.793240] Console: switching to colour frame buffer device 128x48
>>>>> [ 0.794833] fb0: radeondrmfb frame buffer device
>>>>> [ 0.794852] drm: registered panic notifier
>>>>> [ 0.794871] Slow work thread pool: Starting up
>>>>> [ 0.794932] Slow work thread pool: Ready
>>>>> [ 0.794953] [drm] Initialized radeon 2.5.0 20080528 for
>>>>> 0000:01:05.0 on minor 0
>>>>>
>>>>>
>>>>> Torsten
>>>>>
>>>>
>>
>> Does the attached patch works ? (try to change the if 0 to if 1 too
>
> The patch doesn't compile, but after changing mc->foo to rdev->mc.foo it built.
>

I discussed this with Jerome and I think we found the root cause.
Does this patch help?

Alex

> Result from the original patch ("if 0"):
> [ 0.429603] [drm] initializing kernel modesetting (RS690 0x1002:0x791E).
> [ 0.429751] [drm] register mmio base: 0xFE9F0000
> [ 0.429809] [drm] register mmio size: 65536
> [ 0.430385] ATOM BIOS: ATI
> [ 0.430460] radeon 0000:01:05.0: VRAM: 32M 0xDE000000 - 0xDFFFFFFF (32M used)
> [ 0.430520] radeon 0000:01:05.0: GTT: 512M 0xB0000000 - 0xCFFFFFFF
> [ 0.430603] [drm] radeon: irq initialized.
> [ 0.430666] mtrr: type mismatch for fc000000,2000000 old:
> write-back new: write-combining
> [ 0.430739] [drm] Detected VRAM RAM=32M, BAR=32M
> [ 0.430797] [drm] RAM width 128bits DDR
> [ 0.430940] [TTM] Zone kernel: Available graphics memory: 2010234 kiB.
> [ 0.430999] [TTM] Initializing pool allocator.
> [ 0.431075] [drm] radeon: 32M of VRAM memory ready
> [ 0.431133] [drm] radeon: 512M of GTT memory ready.
> [ 0.431194] [drm] GART: num cpu pages 131072, num gpu pages 131072
> [ 0.434577] [drm] radeon: 1 quad pipes, 1 z pipes initialized.
> [ 0.441645] [drm] Loading RS690/RS740 Microcode
> [ 0.441853] [drm] radeon: ring at 0x00000000B0000000
> [ 0.576773] [drm:r100_ring_test] *ERROR* radeon: ring test failed
> (sracth(0x15E4)=0xCAFEDEAD)
> [ 0.576847] [drm:r100_cp_init] *ERROR* radeon: cp isn't working (-22).
> [ 0.576907] radeon 0000:01:05.0: failled initializing CP (-22).
> [ 0.576965] radeon 0000:01:05.0: Disabling GPU acceleration
>
> Result from patch after changing it to "if 1":
> [ 0.400348] [drm] initializing kernel modesetting (RS690 0x1002:0x791E).
> [ 0.400497] [drm] register mmio base: 0xFE9F0000
> [ 0.400556] [drm] register mmio size: 65536
> [ 0.401097] ATOM BIOS: ATI
> [ 0.401171] radeon 0000:01:05.0: VRAM: 32M 0xDE000000 - 0xDFFFFFFF (32M used)
> [ 0.401231] radeon 0000:01:05.0: GTT: 512M 0x00000000 - 0x1FFFFFFF
> [ 0.401314] [drm] radeon: irq initialized.
> [ 0.401377] mtrr: type mismatch for fc000000,2000000 old:
> write-back new: write-combining
> [ 0.401451] [drm] Detected VRAM RAM=32M, BAR=32M
> [ 0.401509] [drm] RAM width 128bits DDR
> [ 0.401597] [TTM] Zone kernel: Available graphics memory: 2010234 kiB.
> [ 0.401656] [TTM] Initializing pool allocator.
> [ 0.401732] [drm] radeon: 32M of VRAM memory ready
> [ 0.401791] [drm] radeon: 512M of GTT memory ready.
> [ 0.401852] [drm] GART: num cpu pages 131072, num gpu pages 131072
> [ 0.405242] [drm] radeon: 1 quad pipes, 1 z pipes initialized.
> [ 0.412298] [drm] Loading RS690/RS740 Microcode
> [ 0.412507] [drm] radeon: ring at 0x0000000000000000
> [ 0.412582] [drm] ring test succeeded in 1 usecs
> [ 0.412726] [drm] radeon: ib pool ready.
> [ 0.412792] [drm] ib test succeeded in 0 usecs
>
> Torsten
>
From d9a401e87e770949ebebf1e741cd74acc9fbabaf Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexdeucher@xxxxxxxxx>
Date: Wed, 14 Jul 2010 17:58:15 -0400
Subject: [PATCH] drm/radeon/kms: fix gtt MC base alignment on rs4xx/rs690/rs740 asics

The asics in question have the following requirements with regard to
their gart setups:

1. The GART aperture size has to be in the form of 2^X bytes, where X is from 25 to 31
2. The GART aperture MC base has to be aligned to a boundary equal to the size of the
aperture.
3. The GART page table has to be aligned to the boundary equal to the size of the table.
4. The GART page table size is: table_entry_size * (aperture_size / page_size)
5. The GART page table has to be allocated in non-paged, non-cached, contiguous system
memory.

This patch takes care 2. The rest should already be handled properly.

This should fix a regression noticed by: Torsten Kaiser <just.for.lkml@xxxxxxxxxxxxxx>

Signed-off-by: Alex Deucher <alexdeucher@xxxxxxxxx>
---
drivers/gpu/drm/radeon/r100.c | 1 +
drivers/gpu/drm/radeon/r300.c | 1 +
drivers/gpu/drm/radeon/r520.c | 1 +
drivers/gpu/drm/radeon/r600.c | 1 +
drivers/gpu/drm/radeon/radeon.h | 1 +
drivers/gpu/drm/radeon/radeon_device.c | 8 ++++----
drivers/gpu/drm/radeon/rs400.c | 5 ++++-
drivers/gpu/drm/radeon/rs600.c | 1 +
drivers/gpu/drm/radeon/rs690.c | 1 +
drivers/gpu/drm/radeon/rv515.c | 1 +
10 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 366bcd8..a7726f1 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -2354,6 +2354,7 @@ void r100_mc_init(struct radeon_device *rdev)
if (rdev->flags & RADEON_IS_IGP)
base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = 0;
if (!(rdev->flags & RADEON_IS_AGP))
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
diff --git a/drivers/gpu/drm/radeon/r300.c b/drivers/gpu/drm/radeon/r300.c
index 7e81db5..0a1638c 100644
--- a/drivers/gpu/drm/radeon/r300.c
+++ b/drivers/gpu/drm/radeon/r300.c
@@ -481,6 +481,7 @@ void r300_mc_init(struct radeon_device *rdev)
if (rdev->flags & RADEON_IS_IGP)
base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = 0;
if (!(rdev->flags & RADEON_IS_AGP))
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
diff --git a/drivers/gpu/drm/radeon/r520.c b/drivers/gpu/drm/radeon/r520.c
index 34330df..694af7c 100644
--- a/drivers/gpu/drm/radeon/r520.c
+++ b/drivers/gpu/drm/radeon/r520.c
@@ -125,6 +125,7 @@ void r520_mc_init(struct radeon_device *rdev)
r520_vram_get_type(rdev);
r100_vram_init_sizes(rdev);
radeon_vram_location(rdev, &rdev->mc, 0);
+ rdev->mc.gtt_base_align = 0;
if (!(rdev->flags & RADEON_IS_AGP))
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index f2f509c..ad38ca8 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -1194,6 +1194,7 @@ void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
if (rdev->flags & RADEON_IS_IGP)
base = (RREG32(MC_VM_FB_LOCATION) & 0xFFFF) << 24;
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = 0;
radeon_gtt_location(rdev, mc);
}
}
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5f4a2c0..e7b6e89 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -354,6 +354,7 @@ struct radeon_mc {
int vram_mtrr;
bool vram_is_ddr;
bool igp_sideport_enabled;
+ u64 gtt_base_align;
};

bool radeon_combios_sideport_present(struct radeon_device *rdev);
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index 9b092b6..3fca920 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -226,20 +226,20 @@ void radeon_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
{
u64 size_af, size_bf;

- size_af = 0xFFFFFFFF - mc->vram_end;
- size_bf = mc->vram_start;
+ size_af = ((0xFFFFFFFF - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
+ size_bf = mc->vram_start & ~mc->gtt_base_align;
if (size_bf > size_af) {
if (mc->gtt_size > size_bf) {
dev_warn(rdev->dev, "limiting GTT\n");
mc->gtt_size = size_bf;
}
- mc->gtt_start = mc->vram_start - mc->gtt_size;
+ mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size;
} else {
if (mc->gtt_size > size_af) {
dev_warn(rdev->dev, "limiting GTT\n");
mc->gtt_size = size_af;
}
- mc->gtt_start = mc->vram_end + 1;
+ mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
}
mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
dev_info(rdev->dev, "GTT: %lluM 0x%08llX - 0x%08llX\n",
diff --git a/drivers/gpu/drm/radeon/rs400.c b/drivers/gpu/drm/radeon/rs400.c
index 9e4240b..f454c9a 100644
--- a/drivers/gpu/drm/radeon/rs400.c
+++ b/drivers/gpu/drm/radeon/rs400.c
@@ -57,7 +57,9 @@ void rs400_gart_adjust_size(struct radeon_device *rdev)
}
if (rdev->family == CHIP_RS400 || rdev->family == CHIP_RS480) {
/* FIXME: RS400 & RS480 seems to have issue with GART size
- * if 4G of system memory (needs more testing) */
+ * if 4G of system memory (needs more testing)
+ */
+ /* XXX is this still an issue with proper alignment? */
rdev->mc.gtt_size = 32 * 1024 * 1024;
DRM_ERROR("Forcing to 32M GART size (because of ASIC bug ?)\n");
}
@@ -263,6 +265,7 @@ void rs400_mc_init(struct radeon_device *rdev)
r100_vram_init_sizes(rdev);
base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = rdev->mc.gtt_size - 1;
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
}
diff --git a/drivers/gpu/drm/radeon/rs600.c b/drivers/gpu/drm/radeon/rs600.c
index 7bb4c3e..6dc15ea 100644
--- a/drivers/gpu/drm/radeon/rs600.c
+++ b/drivers/gpu/drm/radeon/rs600.c
@@ -698,6 +698,7 @@ void rs600_mc_init(struct radeon_device *rdev)
base = G_000004_MC_FB_START(base) << 16;
rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = 0;
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
}
diff --git a/drivers/gpu/drm/radeon/rs690.c b/drivers/gpu/drm/radeon/rs690.c
index f4f0a61..ce4ecbe 100644
--- a/drivers/gpu/drm/radeon/rs690.c
+++ b/drivers/gpu/drm/radeon/rs690.c
@@ -162,6 +162,7 @@ void rs690_mc_init(struct radeon_device *rdev)
rs690_pm_info(rdev);
rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
radeon_vram_location(rdev, &rdev->mc, base);
+ rdev->mc.gtt_base_align = rdev->mc.gtt_size - 1;
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
}
diff --git a/drivers/gpu/drm/radeon/rv515.c b/drivers/gpu/drm/radeon/rv515.c
index 7d9a7b0..0c9c169 100644
--- a/drivers/gpu/drm/radeon/rv515.c
+++ b/drivers/gpu/drm/radeon/rv515.c
@@ -195,6 +195,7 @@ void rv515_mc_init(struct radeon_device *rdev)
rv515_vram_get_type(rdev);
r100_vram_init_sizes(rdev);
radeon_vram_location(rdev, &rdev->mc, 0);
+ rdev->mc.gtt_base_align = 0;
if (!(rdev->flags & RADEON_IS_AGP))
radeon_gtt_location(rdev, &rdev->mc);
radeon_update_bandwidth_info(rdev);
--
1.7.1.1

Next message: Nicolas Kaiser: "[PATCH] drm/i830: fixed brace and spacing coding style issues"
Previous message: David Rientjes: "Re: [S+Q2 06/19] slub: Check kasprintf results inkmem_cache_init()"
In reply to: Jerome Glisse: "Re: Regression 2.6.34->2.6.35-rc4: radeaon KMS an RS690 broken"
Next in thread: Torsten Kaiser: "Re: Regression 2.6.34->2.6.35-rc4: radeaon KMS an RS690 broken"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]