On 6/6/25 15:14, Jocelyn Falempe wrote:
> On 06/06/2025 14:28, Christian König wrote:
>> On 6/6/25 13:48, Jocelyn Falempe wrote:
>>> If the ttm bo is backed by pages, then it's possible to safely kmap
>>> one page at a time, using kmap_try_from_panic().
>>
>> I strongly assume that we don't care about locking anything in this case,
>> don't we?
>
> Yes, normally it's called for the current framebuffer, so I assume it's
> properly allocated, and isn't growing/shrinking while being displayed.
>
>>
>>> Unfortunately there is no way to do the same with ioremap, so it
>>> only supports the kmap case.
>>
>> Oh, there actually is on most modern systems.
>>
>> At least on 64bit systems amdgpu maps the whole VRAM BAR into kernel address
>> space on driver load.
>>
>> So as long as you have a large BAR system you can trivially have access to
>> the MMIO memory.
>
> For amdgpu, I used the indirect MMIO access, so I didn't need to ioremap
> https://elixir.bootlin.com/linux/v6.15/source/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c#L1800
Good point. That is probably quite slow, but works under really all
circumstances as long as the device hasn't fallen of the bus.
> For the xe driver, I only tested on integrated GPU, using system RAM, so this
> first approach is good enough.
> But I'm still interested to find a solution, is there a way to get the
> current io-mapping if it exists?
You need to ask that the XE guys. There is TTMs bdev->funcs->access_memory()
callback which should allow doing that, but I have no idea how that is
implemented for XE.
Regards,
Christian.
>
>
>>
>>> This is needed for proper drm_panic support with xe driver.
>>>
>>> Signed-off-by: Jocelyn Falempe <jfale...@redhat.com>
>>> ---
>>>
>>> v8:
>>> * Added in v8
>>>
>>> drivers/gpu/drm/ttm/ttm_bo_util.c | 27 +++++++++++++++++++++++++++
>>> include/drm/ttm/ttm_bo.h | 1 +
>>> 2 files changed, 28 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c
>>> b/drivers/gpu/drm/ttm/ttm_bo_util.c
>>> index 15cab9bda17f..9c3f3b379c2a 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo_util.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
>>> @@ -377,6 +377,33 @@ static int ttm_bo_kmap_ttm(struct ttm_buffer_object
>>> *bo,
>>> return (!map->virtual) ? -ENOMEM : 0;
>>> }
>>> +/**
>>> + *
>>> + * ttm_bo_kmap_try_from_panic
>>> + *
>>> + * @bo: The buffer object
>>> + * @page: The page to map
>>> + *
>>> + * Sets up a kernel virtual mapping using kmap_local_page_try_from_panic().
>>> + * This can safely be called from the panic handler, if you make sure the
>>> bo
>>
>> "This can *only* be called from the panic handler..."
>
> Yes, I will fix that, it shouldn't be called for normal operations.
>
>>
>> Apart from those open questions, looks sane to me.
>>
>> Regards,
>> Christian.
>>
>>> + * is the one being displayed, so is properly allocated, and won't be
>>> modified.
>>> + *
>>> + * Returns the vaddr, that you can use to write to the bo, and that you
>>> should
>>> + * pass to kunmap_local() when you're done with this page, or NULL if the
>>> bo
>>> + * is in iomem.
>>> + */
>>> +void *ttm_bo_kmap_try_from_panic(struct ttm_buffer_object *bo, unsigned
>>> long page)
>>> +{
>>> + if (page + 1 > PFN_UP(bo->resource->size))
>>> + return NULL;
>>> +
>>> + if (!bo->resource->bus.is_iomem && bo->ttm->pages &&
>>> bo->ttm->pages[page])
>>> + return kmap_local_page_try_from_panic(bo->ttm->pages[page]);
>>> +
>>> + return NULL;
>>> +}
>>> +EXPORT_SYMBOL(ttm_bo_kmap_try_from_panic);
>>> +
>>> /**
>>> * ttm_bo_kmap
>>> *
>>> diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h
>>> index cf027558b6db..8c0ce3fa077f 100644
>>> --- a/include/drm/ttm/ttm_bo.h
>>> +++ b/include/drm/ttm/ttm_bo.h
>>> @@ -429,6 +429,7 @@ int ttm_bo_init_validate(struct ttm_device *bdev,
>>> struct ttm_buffer_object *bo,
>>> int ttm_bo_kmap(struct ttm_buffer_object *bo, unsigned long start_page,
>>> unsigned long num_pages, struct ttm_bo_kmap_obj *map);
>>> void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map);
>>> +void *ttm_bo_kmap_try_from_panic(struct ttm_buffer_object *bo, unsigned
>>> long page);
>>> int ttm_bo_vmap(struct ttm_buffer_object *bo, struct iosys_map *map);
>>> void ttm_bo_vunmap(struct ttm_buffer_object *bo, struct iosys_map *map);
>>> int ttm_bo_mmap_obj(struct vm_area_struct *vma, struct ttm_buffer_object
>>> *bo);
>>
>
