Skip to contentSkip to conversation

drm/i915: Init PPGTT before context enable

PATCH In review David Woodhouse 4 participants 5 messages Jan 20, 2015 – Jan 21, 2015

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Init PPGTT before context enable

Commit 82460d972 ("drm/i915: Rework ppgtt init to no require an aliasing ppgtt") introduced a regression on Broadwell, triggering the following IOMMU fault at startup:

vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
dmar: DRHD: handling fault status reg 2
dmar: DMAR:[DMA Write] Request device [00:02.0] fault addr 880000
DMAR:[fault reason 23] Unknown
fbcon: inteldrmfb (fb0) is primary device

---
[17:01] <danvet> can you pls submit this as a patch to intel-gfx?
[17:01] <danvet> with the comment above context_enable binned, since that's outdated now anyway

 drivers/gpu/drm/i915/i915_gem.c | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -13
index c11603b..b02a3f3 100644
@@ -4884,25 +4884,18 @@ i915_gem_init_hw(struct drm_device *dev)
 	for (i = 0; i < NUM_L3_SLICES(dev); i++)
 		i915_gem_l3_remap(&dev_priv->ring[RCS], i);
 
-	/*
-	 * XXX: Contexts should only be initialized once. Doing a switch to the
-	 * default context switch however is something we'd like to do after
-	 * reset or thaw (the latter may not actually be necessary for HW, but
-	 * goes with our code better). Context switching requires rings (for
-	 * the do_switch), but before enabling PPGTT. So don't move this.
-	 */
-	ret = i915_gem_context_enable(dev_priv);
+	ret = i915_ppgtt_init_hw(dev);
 	if (ret && ret != -EIO) {
-		DRM_ERROR("Context enable failed %d\n", ret);
+		DRM_ERROR("PPGTT enable failed %d\n", ret);
 		i915_gem_cleanup_ringbuffer(dev);
-
-		return ret;
 	}
 
-	ret = i915_ppgtt_init_hw(dev);
+	ret = i915_gem_context_enable(dev_priv);
 	if (ret && ret != -EIO) {
-		DRM_ERROR("PPGTT enable failed %d\n", ret);
+		DRM_ERROR("Context enable failed %d\n", ret);
 		i915_gem_cleanup_ringbuffer(dev);
+
+		return ret;
 	}
 
 	return ret;
Signature · 8 lines
2.1.0




-- 
David Woodhouse                            Open Source Technology Centre
[email protected]                              Intel Corporation

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reviewed-by on v1 1/1
#1 lore raw
Quoted 4 lines · David Woodhouse

Commit 82460d972 ("drm/i915: Rework ppgtt init to no require an aliasing ppgtt") introduced a regression on Broadwell, triggering the following IOMMU fault at startup:

Quoted 8 lines

vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
dmar: DRHD: handling fault status reg 2
dmar: DMAR:[DMA Write] Request device [00:02.0] fault addr 880000
DMAR:[fault reason 23] Unknown
fbcon: inteldrmfb (fb0) is primary device

I sugggested this change to David after staring at the offending patch for a while. I have no idea and theory whatsoever why this would upset the gpu less than the other way round. But it seems to work. David promised to chase hw people a bit more to get a more meaningful answer.

Wrt the comment that this deletes: I've done some digging and afaict loading context before ppgtt enable was once required before our recent restructuring of the context/ppgtt init code: Before that context sw setup (i.e. allocating the default context) and hw setup was smashed together. Also the setup of the default context was the bit that actually allocated the aliasing ppgtt structures. Which is the reason for the context before ppgtt depency.

Or was, since with all the untangling there's no no real depency any more (functional, who knows what the hw is doing), so the comment is just stale.

Jani, can you pls paste my elaboration into the commit message when applying? With that this is:

Quoted 53 lines

---
[17:01] <danvet> can you pls submit this as a patch to intel-gfx?
[17:01] <danvet> with the comment above context_enable binned, since that's outdated now anyway

 drivers/gpu/drm/i915/i915_gem.c | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -13
index c11603b..b02a3f3 100644
@@ -4884,25 +4884,18 @@ i915_gem_init_hw(struct drm_device *dev)
 	for (i = 0; i < NUM_L3_SLICES(dev); i++)
 		i915_gem_l3_remap(&dev_priv->ring[RCS], i);
 
-	/*
-	 * XXX: Contexts should only be initialized once. Doing a switch to the
-	 * default context switch however is something we'd like to do after
-	 * reset or thaw (the latter may not actually be necessary for HW, but
-	 * goes with our code better). Context switching requires rings (for
-	 * the do_switch), but before enabling PPGTT. So don't move this.
-	 */
-	ret = i915_gem_context_enable(dev_priv);
+	ret = i915_ppgtt_init_hw(dev);
 	if (ret && ret != -EIO) {
-		DRM_ERROR("Context enable failed %d\n", ret);
+		DRM_ERROR("PPGTT enable failed %d\n", ret);
 		i915_gem_cleanup_ringbuffer(dev);
-
-		return ret;
 	}
 
-	ret = i915_ppgtt_init_hw(dev);
+	ret = i915_gem_context_enable(dev_priv);
 	if (ret && ret != -EIO) {
-		DRM_ERROR("PPGTT enable failed %d\n", ret);
+		DRM_ERROR("Context enable failed %d\n", ret);
 		i915_gem_cleanup_ringbuffer(dev);
+
+		return ret;
 	}
 
 	return ret;
Signature · 8 lines
2.1.0




-- 
David Woodhouse                            Open Source Technology Centre
[email protected]                              Intel Corporation
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw
Quoted 11 lines · Daniel Vetter
Quoted 10 lines · David Woodhouse

Commit 82460d972 ("drm/i915: Rework ppgtt init to no require an aliasing ppgtt") introduced a regression on Broadwell, triggering the following IOMMU fault at startup:

vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
dmar: DRHD: handling fault status reg 2
dmar: DMAR:[DMA Write] Request device [00:02.0] fault addr 880000
DMAR:[fault reason 23] Unknown
fbcon: inteldrmfb (fb0) is primary device

Quoted 7 lines
Quoted 2 lines

I sugggested this change to David after staring at the offending patch for a while. I have no idea and theory whatsoever why this would upset the gpu less than the other way round. But it seems to work. David promised to chase hw people a bit more to get a more meaningful answer.

The issue is likely the execution of the golden render state batch concurrently with the flip over to ppgtt. The GPU throws a pagefault and we get an ERROR reported.

http://patchwork.freedesktop.org/patch/38270/
http://patchwork.freedesktop.org/patch/38269/
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#3 lore raw
Quoted 20 lines · Chris Wilson
Quoted 11 lines · Daniel Vetter
Quoted 10 lines · David Woodhouse

Commit 82460d972 ("drm/i915: Rework ppgtt init to no require an aliasing ppgtt") introduced a regression on Broadwell, triggering the following IOMMU fault at startup:

vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
dmar: DRHD: handling fault status reg 2
dmar: DMAR:[DMA Write] Request device [00:02.0] fault addr 880000
DMAR:[fault reason 23] Unknown
fbcon: inteldrmfb (fb0) is primary device

Quoted 7 lines
Quoted 2 lines

I sugggested this change to David after staring at the offending patch for a while. I have no idea and theory whatsoever why this would upset the gpu less than the other way round. But it seems to work. David promised to chase hw people a bit more to get a more meaningful answer.

Quoted 6 lines

The issue is likely the execution of the golden render state batch concurrently with the flip over to ppgtt. The GPU throws a pagefault and we get an ERROR reported.

http://patchwork.freedesktop.org/patch/38270/
http://patchwork.freedesktop.org/patch/38269/

Pushed the revert to drm-intel-fixes, it's cc: stable and it's getting late in the rc's too. Thanks for the patch and review.

BR,
Jani.

Quoted 8 lines

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#4 lore raw
Quoted 25 lines · Jani Nikula
Quoted 24 lines · Chris Wilson
Quoted 11 lines · Daniel Vetter
Quoted 10 lines · David Woodhouse

Commit 82460d972 ("drm/i915: Rework ppgtt init to no require an aliasing ppgtt") introduced a regression on Broadwell, triggering the following IOMMU fault at startup:

vgaarb: device changed decodes: PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
dmar: DRHD: handling fault status reg 2
dmar: DMAR:[DMA Write] Request device [00:02.0] fault addr 880000
DMAR:[fault reason 23] Unknown
fbcon: inteldrmfb (fb0) is primary device

Quoted 7 lines
Quoted 2 lines

I sugggested this change to David after staring at the offending patch for a while. I have no idea and theory whatsoever why this would upset the gpu less than the other way round. But it seems to work. David promised to chase hw people a bit more to get a more meaningful answer.

The issue is likely the execution of the golden render state batch concurrently with the flip over to ppgtt. The GPU throws a pagefault and we get an ERROR reported.

Quoted 5 lines
Quoted 2 lines

http://patchwork.freedesktop.org/patch/38270/
http://patchwork.freedesktop.org/patch/38269/

Pushed the revert to drm-intel-fixes, it's cc: stable and it's getting late in the rc's too. Thanks for the patch and review.

s/revert/David's patch/ to be clear.

BR,
Jani.

Quoted 16 lines

BR,
Jani.

Quoted 8 lines

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 1 line
Jani Nikula, Intel Open Source Technology Center
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx