[PATCH 01/10] drm/amdgpu: use only the lower address space on GMC9

[Felix Kuehling]

The GPU always has access to the entire guest physical address space.
You can just program arbitrary addresses into page table entries and set
the system bit. That's how GART and GPUVM mappings work. They will not
go through the AGP aperture. And there is no mechanism (other than an
IOMMU) to protect system memory from GPU access.

Regards,
  Felix


On 2018-08-28 07:41 AM, Christian König wrote:
> Well that is indeed true, but we still have IOMMU between the GPU and
> the system memory.
>
> So we should still catch issues when something goes terrible wrong.
>
> Additional to that only the system domain, e.g. kernel copies, page
> table updates etc are allowed to use it.
>
> Regards,
> Christian.
>
> Am 28.08.2018 um 09:06 schrieb Xiao, Jack:
>> I mean it has risk to make GPU allowed to access to most system
>> memory without explicit claiming, it's easier to mask problem.
>>
>> Regards,
>> Jack
>>
>> -----Original Message-----
>> From: Koenig, Christian
>> Sent: Tuesday, August 28, 2018 2:46 PM
>> To: Xiao, Jack <Jack.Xiao at amd.com>; Kuehling, Felix
>> <Felix.Kuehling at amd.com>; Christian König
>> <ckoenig.leichtzumerken at gmail.com>; amd-gfx at lists.freedesktop.org
>> Subject: Re: [PATCH 01/10] drm/amdgpu: use only the lower address
>> space on GMC9
>>
>>> This series patches seems to make AGP aperture allowed to access any
>>> system memory (16GB) bypass GPU VM protection.
>> The system aperture should only be active in the system domain, or
>> otherwise applications would have access to local memory as well.
>>
>> There are some bits in the VM registers to enable/disable that, but
>> when we would have that setting incorrect we would see quite a bunch
>> of other problems.
>>
>> Might still be a good idea to double check if all the bits are setup
>> correctly.
>>
>> Regards,
>> Christian.
>>
>> Am 28.08.2018 um 07:31 schrieb Xiao, Jack:
>>> This series patches seems to make AGP aperture allowed to access any
>>> system memory (16GB) bypass GPU VM protection.
>>> If someone made a wrong logic requesting an illegal address which
>>> occasionally was located inside AGP aperture, but without any VM
>>> protection, the illegal address would be finally translated into a
>>> system memory address; if GPU read/wrote such system memory address,
>>> the system memory address might belong to kernel or any user
>>> application, the r/w operation would result in any unpredictable issue.
>>> The most important is that such kind of issue is so hard to be
>>> addressed.
>>> Is it worth doing this, but exposing risk?
>>>
>>> Regards,
>>> Jack
>>>
>>> -----Original Message-----
>>> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of
>>> Felix Kuehling
>>> Sent: Tuesday, August 28, 2018 3:03 AM
>>> To: Christian König <ckoenig.leichtzumerken at gmail.com>;
>>> amd-gfx at lists.freedesktop.org; Koenig, Christian
>>> <Christian.Koenig at amd.com>
>>> Subject: Re: [PATCH 01/10] drm/amdgpu: use only the lower address
>>> space on GMC9
>>>
>>> The point of this series seems to be to allow access to small system
>>> memory BOs (one page) without a GART mapping. I'm guessing that
>>> reduces pressure on the GART and removes the need for HDP and TLB
>>> flushes. Why does Patch 10 only enable that on GFXv9? Is there less
>>> benefit on older chips?
>>>
>>> Is this related to your recent changes to allow page tables in
>>> system memory?
>>>
>>> See my replies to patch 6 and 8. Other than that, the series is
>>> Acked-by: Felix Kuehling <Felix.Kuehling at amd.com>
>>>
>>> Regards,
>>>     Felix
>>>
>>>
>>> On 2018-08-27 12:53 PM, Christian König wrote:
>>>> Only use the lower address space on GMC9 for the system domain.
>>>> Otherwise we would need to sign extend GMC addresses.
>>>>
>>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>>> ---
>>>>    drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 7 +++----
>>>>    1 file changed, 3 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> index e44b5191735d..d982956c8329 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> @@ -938,11 +938,10 @@ static int gmc_v9_0_sw_init(void *handle)
>>>>        if (r)
>>>>            return r;
>>>>    -    /* Set the internal MC address mask
>>>> -     * This is the max address of the GPU's
>>>> -     * internal address space.
>>>> +    /* Use only the lower range for the internal MC address mask.
>>>> This is
>>>> +     * the max address of the GPU's internal address space.
>>>>         */
>>>> -    adev->gmc.mc_mask = 0xffffffffffffULL; /* 48 bit MC */
>>>> +    adev->gmc.mc_mask = 0x7fffffffffffULL;
>>>>           /* set DMA mask + need_dma32 flags.
>>>>         * PCIE - can handle 44-bits.
>>> _______________________________________________
>>> amd-gfx mailing list
>>> amd-gfx at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>