drm/i915: fix crash in error state readout on non-execlist platforms
Subsystems (2)
PATCH 1/1 drm/i915: fix crash in error state readout on non-execlist platforms
Looks like this was introduced in:
commit d1675198ed1f21aec6e036336e4340c40b726497
Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100
drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/i915_gpu_error.c +13 -6
@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset); +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) {err_printf (m,"[%04x] %08x %08x %08x %08x\n" , offset, - obj->pages[LRC_PPHWSP_PN ][elt], - obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN ][elt+3 ]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3 ]); offset +=16 ; } }
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Jesse,
Will the patch here fix the issue? It should help other cases where LRC_PPHWSP_PN is referenced on non-execlist / guc platforms.
drivers/gpu/drm/i915/intel_lrc.h +1 -1
@@ -71,7 +71,7 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf,/* One extra page is added before LRC for GuC as shared data */ #define LRC_GUCSHR_PN (0) -#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + i915.enable_guc_submission ? 1 : 0)#define LRC_STATE_PN (LRC_PPHWSP_PN + 1)void intel_lr_context_free (struct intel_context *ctx);
Thanks,
Alex
Quoted 45 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
Cc Alex Dai <[email protected]>Signed-off-by Jesse Barnes <[email protected]>--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_gpu_error.c +9
-3@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset); +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) { err_printf(m, "[%04x] %08x %08x %08x %08x\n", offset, - obj->pages[LRC_PPHWSP_PN][elt],
drivers/gpu/drm/i915/i915_gpu_error.c int i915_error_state_to_str(struct drm_i915_error_state_buf *m, · line 480
Quoted 10 lines
drivers/gpu/drm/i915/i915_gpu_error.c +4
-3- obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN][elt+3]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3]); offset +=16 ; } }
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
That looks like it would, but I think it's still confusing to reference LRC state when we haven't initialized execlists at all...
Jesse
Quoted 68 lines · Yu Dai
Jesse,
Will the patch here fix the issue? It should help other cases where
LRC_PPHWSP_PNis referenced on non-execlist / guc platforms.drivers/gpu/drm/i915/intel_lrc.h +1
-1@@ -71,7 +71,7 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf,/* One extra page is added before LRC for GuC as shared data */ #define LRC_GUCSHR_PN (0) -#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + i915.enable_guc_submission ? 1 : 0)#define LRC_STATE_PN (LRC_PPHWSP_PN + 1)void intel_lr_context_free (struct intel_context *ctx);Thanks,
AlexQuoted 46 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
Cc Alex Dai <[email protected]>Signed-off-by Jesse Barnes <[email protected]>--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_gpu_error.c +9
-5@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset); +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) { err_printf(m, "[%04x] %08x %08x %08x %08x\n", offset, - obj->pages[LRC_PPHWSP_PN][elt], - obj->pages[LRC_PPHWSP_PN][elt+1], - obj->pages[LRC_PPHWSP_PN][elt+2],
Quoted 10 lines
drivers/gpu/drm/i915/i915_gpu_error.c int i915_error_state_to_str(struct drm_i915_error_state_buf *m, · line 479
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +4
-1- obj->pages[LRC_PPHWSP_PN][elt+3]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3]); offset +=16 ; } }
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Agree. The LRC prefix is confusing. Thanks for the patch. -Alex
Quoted 84 lines · Jesse Barnes
That looks like it would, but I think it's still confusing to reference LRC state when we haven't initialized execlists at all...
Jesse
Quoted 78 lines · Yu Dai
Jesse,
Will the patch here fix the issue? It should help other cases where
LRC_PPHWSP_PNis referenced on non-execlist / guc platforms.drivers/gpu/drm/i915/intel_lrc.h +1
-1@@ -71,7 +71,7 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf,/* One extra page is added before LRC for GuC as shared data */ #define LRC_GUCSHR_PN (0) -#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + i915.enable_guc_submission ? 1 : 0)#define LRC_STATE_PN (LRC_PPHWSP_PN + 1)void intel_lr_context_free (struct intel_context *ctx);Thanks,
AlexQuoted 54 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
Cc Alex Dai <[email protected]>Signed-off-by Jesse Barnes <[email protected]>--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_gpu_error.c +13
-6@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset); +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) {err_printf (m,"[%04x] %08x %08x %08x %08x\n" , offset, - obj->pages[LRC_PPHWSP_PN ][elt], - obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN ][elt+3 ]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3 ]); offset +=16 ; } }
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 1 line · Yu Dai
Agree. The LRC prefix is confusing. Thanks for the patch. -Alex
Care to do an official r-b?
Thanks, Daniel
Quoted 91 lines · Jesse Barnes
Quoted 84 lines
That looks like it would, but I think it's still confusing to reference LRC state when we haven't initialized execlists at all...
Jesse
Quoted 78 lines · Yu Dai
Jesse,
Will the patch here fix the issue? It should help other cases where
LRC_PPHWSP_PNis referenced on non-execlist / guc platforms.drivers/gpu/drm/i915/intel_lrc.h +1
-1@@ -71,7 +71,7 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf,/* One extra page is added before LRC for GuC as shared data */ #define LRC_GUCSHR_PN (0) -#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + i915.enable_guc_submission ? 1 : 0)#define LRC_STATE_PN (LRC_PPHWSP_PN + 1)void intel_lr_context_free (struct intel_context *ctx);Thanks,
AlexQuoted 54 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
Cc Alex Dai <[email protected]>Signed-off-by Jesse Barnes <[email protected]>--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_gpu_error.c +13
-6@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset); +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) {err_printf (m,"[%04x] %08x %08x %08x %08x\n" , offset, - obj->pages[LRC_PPHWSP_PN ][elt], - obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN ][elt+3 ]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3 ]); offset +=16 ; } }_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 24 lines · Daniel Vetter
Quoted 1 line · Yu Dai
Agree. The LRC prefix is confusing. Thanks for the patch. -Alex
Care to do an official r-b?
Thanks, Daniel
Quoted 17 lines · Jesse Barnes
Quoted 15 lines
That looks like it would, but I think it's still confusing to reference LRC state when we haven't initialized execlists at all...
Jesse
Quoted 10 lines · Yu Dai
Jesse,
Will the patch here fix the issue? It should help other cases where
LRC_PPHWSP_PNis referenced on non-execlist / guc platforms.drivers/gpu/drm/i915/intel_lrc.h +0
-0@@ -71,7 +71,7 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf,
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
drivers/gpu/drm/i915/intel_lrc.h static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf, · line 75
Quoted 5 lines
drivers/gpu/drm/i915/intel_lrc.h +1
-1/* One extra page is added before LRC for GuC as shared data */ #define LRC_GUCSHR_PN (0) -#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + i915.enable_guc_submission ? 1 : 0)#define LRC_STATE_PN (LRC_PPHWSP_PN + 1)
I don't like this approach of hiding the runtime conditional offset in the macro. I think it would be better to leave this alone and use the technique in the code below where the (const) macros named 'LRC_*' are used only inside clauses guarded by an "if (i915.enable_execlists)". That way the code won't mention the LRC_* symbols except where they're known tp be meaningful.
Of these symbols, LRC_GUCSHR_PN is not used except in GuC-specific code, and LRC_STATE_PN is used only in GuC and/or LRC (execlist-mode) specific code. So it's only LRC_PPHWSP_PN that needs to be used selectively (specifically, added to the base of a context object when using execlist mode). And the code below already accomplishes that without changing the macro.
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
drivers/gpu/drm/i915/intel_lrc.h +0
-0void intel_lr_context_free (struct intel_context *ctx);Thanks,
AlexQuoted 20 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
Cc Alex Dai <[email protected]>Signed-off-by Jesse Barnes <[email protected]>--- drivers/gpu/drm/i915/i915_gpu_error.c | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-)
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c int i915_error_state_to_str(struct drm_i915_error_state_buf *m, · line 459
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +1
-0@@ -457,17 +457,24 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset =lower_32_bits (obj->gtt_offset);
No need to take only lower 32 bits here, the variable is already u64 and the format below is 0x%08llx, so we'll get a 32-bit value in the output (unless it actually is bigger than 32 bits, which shouldn't happen and which we'd want to know about!).
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
drivers/gpu/drm/i915/i915_gpu_error.c int i915_error_state_to_str(struct drm_i915_error_state_buf *m, · line 464
Quoted 5 lines
drivers/gpu/drm/i915/i915_gpu_error.c +5
-0+u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset = obj->gtt_offset +LRC_PPHWSP_PN * +PAGE_SIZE ;
If we keep the full 64-bit value (or even if we don't), this should simplify to "hws_offset += LRC_PPHWSP_PN*PAGE_SIZE;"
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
drivers/gpu/drm/i915/i915_gpu_error.c +3
-5+ hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) { err_printf(m, "[%04x] %08x %08x %08x %08x\n", offset, - obj->pages[LRC_PPHWSP_PN][elt], - obj->pages[LRC_PPHWSP_PN][elt+1], - obj->pages[LRC_PPHWSP_PN][elt+2],
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c int i915_error_state_to_str(struct drm_i915_error_state_buf *m, · line 479
Quoted 8 lines
drivers/gpu/drm/i915/i915_gpu_error.c +4
-1- obj->pages[LRC_PPHWSP_PN][elt+3]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3]); offset +=16 ; } }
With the 64-bit handling changed, I'd be happy to r-b Jesse's original patch, but not Alex's redefinition of LRC_PPHWSP_PN.
Cheers,
.Dave.
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 1/1 drm/i915: fix crash in error state readout on non-execlist platforms v2
In reply to Jesse BarnesLooks like this was introduced in:
commit d1675198ed1f21aec6e036336e4340c40b726497
Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100
drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
I don't see a bug filed for this, but there may be one that I haven't found.
v2: fixup offset handling for error capture fix (Dave)
--- drivers/gpu/drm/i915/i915_gpu_error.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/i915_gpu_error.c +12 -6
@@ -457,17 +457,23 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset = obj->gtt_offset; +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset +=LRC_PPHWSP_PN *PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) {err_printf (m,"[%04x] %08x %08x %08x %08x\n" , offset, - obj->pages[LRC_PPHWSP_PN ][elt], - obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN ][elt+3 ]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3 ]); offset +=16 ; } }
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Daniel Vetter
AppliedRe: [PATCH] drm/i915: fix crash in error state readout on non-execlist platforms v2
Quoted 11 lines · Jesse Barnes
Looks like this was introduced in:
commitd1675198ed1f21aec6e036336e4340c40b726497Author: Alex Dai <[email protected]> Date: Wed Aug 12 15:43:43 2015 +0100drm/i915: Integrate GuC-based command submission
This patch assumed LRC contexts and HWS layout, which is incorrect on platforms without execlists. This can lead to a crash in GPU error state readout on those platforms.
Quoted 8 lines
I don't see a bug filed for this, but there may be one that I haven't found.
v2: fixup offset handling for error capture fix (Dave)
Cc Alex Dai <[email protected]>Reviewed-by Dave Gordon <[email protected]>Signed-off-by Jesse Barnes <[email protected]>
Queued for -next, thanks for the patch.
-Daniel
Quoted 45 lines
--- drivers/gpu/drm/i915/i915_gpu_error.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-)drivers/gpu/drm/i915/i915_gpu_error.c +12
-6@@ -457,17 +457,23 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, }if ((obj = error->ring[i].hws_page)) { +u64 hws_offset = obj->gtt_offset; +u32 *hws_page = &obj->pages[0 ][0 ]; + +if (i915.enable_execlists) { + hws_offset +=LRC_PPHWSP_PN *PAGE_SIZE ; + hws_page = &obj->pages[LRC_PPHWSP_PN ][0 ]; + }err_printf (m,"%s --- HW Status = 0x%08llx\n" , - dev_priv->ring[i].name, - obj->gtt_offset +LRC_PPHWSP_PN *PAGE_SIZE ); + dev_priv->ring[i].name, hws_offset); offset =0 ;for (elt =0 ; elt <PAGE_SIZE /16 ; elt +=4 ) {err_printf (m,"[%04x] %08x %08x %08x %08x\n" , offset, - obj->pages[LRC_PPHWSP_PN ][elt], - obj->pages[LRC_PPHWSP_PN ][elt+1 ], - obj->pages[LRC_PPHWSP_PN ][elt+2 ], - obj->pages[LRC_PPHWSP_PN ][elt+3 ]); + hws_page[elt], + hws_page[elt+1 ], + hws_page[elt+2 ], + hws_page[elt+3 ]); offset +=16 ; } }Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx