Op 11-08-2020 om 09:34 schreef Thomas Hellström (Intel):
>
> On 8/10/20 12:30 PM, Maarten Lankhorst wrote:
>> As soon as we install fences, we should stop allocating memory
>> in order to prevent any potential deadlocks.
>>
>> This is required later on, when we start adding support for
>> dma-fence annotations, and also required for userptr.
>>
>> Signed-off-by: Maarten Lankhorst <maarten.lankho...@linux.intel.com>
>> ---
>> drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 18 ++++++++++++------
>> drivers/gpu/drm/i915/i915_vma.c | 8 +++++---
>> drivers/gpu/drm/i915/i915_vma.h | 3 +++
>> 3 files changed, 20 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
>> b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
>> index 5d08ce71f341..12397fbc0971 100644
>> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
>> @@ -48,11 +48,12 @@ enum {
>> #define DBG_FORCE_RELOC 0 /* choose one of the above! */
>> };
>> -#define __EXEC_OBJECT_HAS_PIN BIT(31)
>> -#define __EXEC_OBJECT_HAS_FENCE BIT(30)
>> -#define __EXEC_OBJECT_NEEDS_MAP BIT(29)
>> -#define __EXEC_OBJECT_NEEDS_BIAS BIT(28)
>> -#define __EXEC_OBJECT_INTERNAL_FLAGS (~0u << 28) /* all of the above */
>> +/* __EXEC_OBJECT_NO_RESERVE is BIT(31), defined in i915_vma.h */
>
> Hmm. Could we either move all these flag definitions to a header or add an
> i915_vma_move_to_active()?
I have dropped this patch for now. The real fix is a removal of that chunk in
i915_vma_move_to_active. We need to stop allocating memory that late, and only
install fences and submit.
Specifically, eb_submit() should not be allowed to allocate memory after a
certain
point and then complete without error. There are too many places that do this to
fix it up in this series, but this will definitely have to be done in the
future.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
