I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Dave.
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
_wc is only suitable for MMIO.
Cheers, Ben.
On Fri, 2015-10-02 at 14:45 +1000, Benjamin Herrenschmidt wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
_wc is only suitable for MMIO.
Note that I wouldn't be surprised if we weren't the only arch like that.
Playing with caching attributes of main memory smells from a system design and cache coherency protocol standpoint. x86 supports it but I wouldn't be surprised if some ARMs puke in interesting way as well.
Cheers, Ben.
On Fri, 2015-10-02 at 14:47 +1000, Benjamin Herrenschmidt wrote:
On Fri, 2015-10-02 at 14:45 +1000, Benjamin Herrenschmidt wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
_wc is only suitable for MMIO.
Note that I wouldn't be surprised if we weren't the only arch like that.
Playing with caching attributes of main memory smells from a system design and cache coherency protocol standpoint. x86 supports it but I wouldn't be surprised if some ARMs puke in interesting way as well.
Similarily I remember something in the TTM core at some point that was trying to use the same caching attribtues for memory as the original (MMIO) object when moving it.
That will blow up on POWER and possibly others. Here too, it's a complete heresy for the core to assume that it can apply MMIO attributes to system memory. I don't know if that still happens though.
Cheers, Ben.
On 2 October 2015 at 14:45, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
Except you guys screwed up and allowed AGP to exist on Power, and that requires it.
So we have to keep the AGP trapdoor in place, and at the moment only the drivers know if they are AGP, hence why I had to add this in the driver instead of in the drm core.
Dave.
On Fri, 2015-10-02 at 14:53 +1000, Dave Airlie wrote:
On 2 October 2015 at 14:45, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
Except you guys screwed up and allowed AGP to exist on Power, and that requires it.
Well, Apple did :-) Yes, AGP should never have existed in the first place, I think that's a given :)
So we have to keep the AGP trapdoor in place, and at the moment only the drivers know if they are AGP, hence why I had to add this in the driver instead of in the drm core.
Cheers, Ben.
On Fri, Oct 2, 2015 at 7:56 AM, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:53 +1000, Dave Airlie wrote:
On 2 October 2015 at 14:45, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
Except you guys screwed up and allowed AGP to exist on Power, and that requires it.
Well, Apple did :-) Yes, AGP should never have existed in the first place, I think that's a given :)
So we have to keep the AGP trapdoor in place, and at the moment only the drivers know if they are AGP, hence why I had to add this in the driver instead of in the drm core.
Cheers, Ben.
Bumping this and adding my r-b: Reviewed-by: Oded Gabbay oded.gabbay@gmail.com
+Alex
On Thu, Jan 21, 2016 at 5:24 PM, Oded Gabbay oded.gabbay@gmail.com wrote:
On Fri, Oct 2, 2015 at 7:56 AM, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:53 +1000, Dave Airlie wrote:
On 2 October 2015 at 14:45, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
Except you guys screwed up and allowed AGP to exist on Power, and that requires it.
Well, Apple did :-) Yes, AGP should never have existed in the first place, I think that's a given :)
So we have to keep the AGP trapdoor in place, and at the moment only the drivers know if they are AGP, hence why I had to add this in the driver instead of in the drm core.
Cheers, Ben.
Bumping this and adding my r-b: Reviewed-by: Oded Gabbay oded.gabbay@gmail.com
On Thu, Jan 21, 2016 at 10:39 AM, Oded Gabbay oded.gabbay@gmail.com wrote:
+Alex
No objections from me. Care to respin with amdgpu support and signed off? Would probably also be nice to split the core drm change from the driver specific ones.
Alex
On Thu, Jan 21, 2016 at 5:24 PM, Oded Gabbay oded.gabbay@gmail.com wrote:
On Fri, Oct 2, 2015 at 7:56 AM, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:53 +1000, Dave Airlie wrote:
On 2 October 2015 at 14:45, Benjamin Herrenschmidt benh@kernel.crashing.org wrote:
On Fri, 2015-10-02 at 14:42 +1000, Dave Airlie wrote:
I don't think we resolved this the last time we talked about it,
but radeon writecombine maps fail hard on ppc, I think all the fixes either did something bad to AGP systems or weren't liked.
My patch attached just fixes radeon, which is where I'm still seeing the issue.
Yes, you MUST NOT set the flags of system memory to anything other than fully cachable on any cache coherent powerpc machine. This should be bolted into the DRM core imho.
Except you guys screwed up and allowed AGP to exist on Power, and that requires it.
Well, Apple did :-) Yes, AGP should never have existed in the first place, I think that's a given :)
So we have to keep the AGP trapdoor in place, and at the moment only the drivers know if they are AGP, hence why I had to add this in the driver instead of in the drm core.
Cheers, Ben.
Bumping this and adding my r-b: Reviewed-by: Oded Gabbay oded.gabbay@gmail.com
On 22.01.2016 02:10, Alex Deucher wrote:
On Thu, Jan 21, 2016 at 10:39 AM, Oded Gabbay oded.gabbay@gmail.com wrote:
+Alex
No objections from me. Care to respin with amdgpu support and signed off? Would probably also be nice to split the core drm change from the driver specific ones.
Also, it would be better to mask out the RADEON_GEM_GTT_WC flag in radeon_bo_create, as is already done for a few other cases.
On Fri, Jan 22, 2016 at 4:32 AM, Michel Dänzer michel@daenzer.net wrote:
On 22.01.2016 02:10, Alex Deucher wrote:
On Thu, Jan 21, 2016 at 10:39 AM, Oded Gabbay oded.gabbay@gmail.com wrote:
+Alex
No objections from me. Care to respin with amdgpu support and signed off? Would probably also be nice to split the core drm change from the driver specific ones.
Also, it would be better to mask out the RADEON_GEM_GTT_WC flag in radeon_bo_create, as is already done for a few other cases.
-- Earthling Michel Dänzer | http://www.amd.com Libre software enthusiast | Mesa and X developer
ok, sent the patches now. Please check if I got all the flags correctly.
Oded
dri-devel@lists.freedesktop.org
-
Alex Deucher -
Benjamin Herrenschmidt -
Benjamin Herrenschmidt -
Dave Airlie -
Michel Dänzer -
Oded Gabbay