Re: [PATCH] drm/radeon: Fix screen corruption (v2)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 2022-12-15 06:27, Christian König wrote:
> Am 15.12.22 um 11:19 schrieb Luben Tuikov:
>> On 2022-12-15 04:46, Christian König wrote:
>>> Am 15.12.22 um 10:08 schrieb Luben Tuikov:
>>>> On 2022-12-15 03:07, Christian König wrote:
>>>>> Am 15.12.22 um 00:08 schrieb Robin Murphy:
>>>>>> On 2022-12-14 22:02, Alex Deucher wrote:
>>>>>>> On Wed, Dec 14, 2022 at 4:54 PM Robin Murphy <robin.murphy@xxxxxxx>
>>>>>>> wrote:
>>>>>>>> On 2022-12-12 02:08, Luben Tuikov wrote:
>>>>>>>>> Fix screen corruption on older 32-bit systems using AGP chips.
>>>>>>>>>
>>>>>>>>> On older systems with little memory, for instance 1.5 GiB, using an
>>>>>>>>> AGP chip,
>>>>>>>>> the device's DMA mask is 0xFFFFFFFF, but the memory mask is
>>>>>>>>> 0x7FFFFFF, and
>>>>>>>>> subsequently dma_addressing_limited() returns 0xFFFFFFFF < 0x7FFFFFFF,
>>>>>>>>> false. As such the result of this static inline isn't suitable for
>>>>>>>>> the last
>>>>>>>>> argument to ttm_device_init()--it simply needs to now whether to
>>>>>>>>> use GFP_DMA32
>>>>>>>>> when allocating DMA buffers.
>>>>>>>> This sounds wrong to me. If the issues happen on systems without PAE it
>>>>>>>> clearly can't have anything to with the actual DMA address size. Not to
>>>>>>>> mention that AFAICS 32-bit x86 doesn't even have ZONE_DMA32, so
>>>>>>>> GFP_DMA32 would be functionally meaningless anyway. Although the
>>>>>>>> reported symptoms initially sounded like they could be caused by DMA
>>>>>>>> going to the wrong place, that is also equally consistent with a
>>>>>>>> loss of
>>>>>>>> cache coherency.
>>>>>>>>
>>>>>>>> My (limited) understanding of AGP is that the GART can effectively
>>>>>>>> alias
>>>>>>>> memory to a second physical address, so I could well believe that
>>>>>>>> something somewhere in the driver stack needs to perform some cache
>>>>>>>> maintenance to avoid coherency issues, and that in these particular
>>>>>>>> setups whatever that is might be assuming the memory is direct-mapped
>>>>>>>> and thus going wrong for highmem pages.
>>>>>>>>
>>>>>>>> So as I said before, I really think this is not about using
>>>>>>>> GFP_DMA32 at
>>>>>>>> all, but about *not* using GFP_HIGHUSER.
>>>>>>> One of the wonderful features of AGP is that it has to be used with
>>>>>>> uncached memory.  The aperture basically just provides a remapping of
>>>>>>> physical pages into a linear aperture that you point the GPU at.  TTM
>>>>>>> has to jump through quite a few hoops to get uncached memory in the
>>>>>>> first place, so it's likely that that somehow isn't compatible with
>>>>>>> HIGHMEM.  Can you get uncached HIGHMEM?
>>>>>> I guess in principle yes, if you're careful not to use regular
>>>>>> kmap()/kmap_atomic(), and always use pgprot_noncached() for
>>>>>> userspace/vmalloc mappings, but clearly that leaves lots of scope for
>>>>>> slipping up.
>>>>> I theory we should do exactly that in TTM, but we have very few users
>>>>> who actually still exercise that functionality.
>>>>>
>>>>>> Working backwards from primitives like set_memory_uc(), I see various
>>>>>> paths in TTM where manipulating the caching state is skipped for
>>>>>> highmem pages, but I wouldn't even know where to start looking for
>>>>>> whether the right state is propagated to all the places where they
>>>>>> might eventually be mapped somewhere.
>>>>> The tt object has the caching state for the pages and
>>>>> ttm_prot_from_caching() then uses pgprot_noncached() and co for the
>>>>> userspace/vmalloc mappings.
>>>>>
>>>> The point of this patch is that dma_addressing_limited() is unsuitable as
>>>> the last parameter to ttm_pool_init(), since if it is "false"--as it is in this
>>>> particular case--then TTM ends up using HIGHUSER, and we get the screen corruption.
>>>> (gfp_flags |= GFP_HIGHUSER in in ttm_pool_alloc())
>>> Well I would rather say that dma_addressing_limited() works, but the
>>> default value from dma_get_required_mask() is broken.
>>>
>> dma_get_required_mask() for his setup of 1.5 GiB of memory returns 0x7FFFFFF.
> 
> This 0x7FFFFFF mask looks fishy to me. That would only be 128MiB 
> addressable memory (27 bits set)? Or is there another F missing?

Yeah, I'm missing an F--it is correctly described at the top of the thread above,
i.e. in the commit of v2 patch.

0x7FFF_FFFF, which seems correct, no?

>> While the dma mask is 0xFFFFFFFF, as set in radeon_device.c in radeon_device_init().
>>
>>> 32 bits only work with bounce buffers and we can't use those on graphics
>>> hardware.
>>>
>>>> Is there an objection to this patch, if it fixes the screen corruption?
>>> Not from my side, but fixing the underlying issues would be better I think.
>>>
>> Have they been identified?
> 
> I'm not 100% sure. I think by using GFP_DMA32 we just work around the 
> issue somehow.

Right. Using GFP_DMA32, we don't touch high-mem. I was looking at the DRM
code trying to understand what we do when GFP_DMA32 is not set, and the immediate
thing I see is that we set GFP_HIGHUSER when use_dma32 is unset in the device struct.
(Then I got down to the caching attributes...)

It's be nice if we can find the actual issue--what else would it show us that needs fixing...?

So what do we do with this patch?

Shouldn't leave it in a limbo--some OSes ship their kernel
with 33b3ad3788ab ("drm/radeon: handle PCIe root ports with addressing limitations") wholly
reverted.

Regards,
Luben




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux