Re: 答复: 答复: [PATCH 3/3] amd/amdgpu: seperate sriov fb aperture setting
Koenig, Christian
Christian.Koenig at amd.com
Mon Aug 19 10:55:38 UTC 2019
> As I point out. For SRIOV, amdgpu_gmc_agp_location() would not be called, so the value calculation would not be valid.
Yeah, that is a good point. The values are zero initialized, but that is
probably not correct in this moment.
agp_start should be set to 0xffffffff and agp_end to 0x0, that are also
the values the hardware uses to disable the AGP bar.
> Would you please give more info why we need the AGP except for the zfb?
Actually quite a bunch of things. In general we use it to access
uncached system memory from VMID0 to avoid reprogramming the GART all
the time.
This in turn is used mostly by the SDMA for keeping page table copies in
system memory, but can also be used by things like scanout from system
memory on APUs.
I don't think we have any use case for SRIOV, so disabling the feature
should be fine.
Regards,
Christian.
Am 19.08.19 um 12:24 schrieb Min, Frank:
> Hi Christian,
> As I point out. For SRIOV, amdgpu_gmc_agp_location() would not be called, so the value calculation would not be valid. Would you please give more info why we need the AGP except for the zfb?
>
> For another, whether you review the other patches?
>
> Best Regards,
> Frank
>
> -----邮件原件-----
> 发件人: Koenig, Christian <Christian.Koenig at amd.com>
> 发送时间: 2019年8月19日 15:21
> 收件人: Min, Frank <Frank.Min at amd.com>; amd-gfx at lists.freedesktop.org
> 主题: Re: 答复: [PATCH 3/3] amd/amdgpu: seperate sriov fb aperture setting
>
> Yeah, I thought so.
>
> In this case we don't need this patch or is there anything I'm still missing?
>
> The use of min/max here is exactly to avoid having a SRIOV dependency here.
>
> Regards,
> Christian.
>
> Am 19.08.19 um 09:17 schrieb Min, Frank:
>> Hi Christian,
>> Thanks for your review.
>> For SRIOV, amdgpu_gmc_agp_location() would not be called, since it do not use AGP. Also there is no need to use the min and max to judge which range is correct for using.
>>
>> Best Regards,
>> Frank
>>
>> -----邮件原件-----
>> 发件人: Christian König <ckoenig.leichtzumerken at gmail.com>
>> 发送时间: 2019年8月19日 15:07
>> 收件人: Min, Frank <Frank.Min at amd.com>; amd-gfx at lists.freedesktop.org
>> 主题: Re: [PATCH 3/3] amd/amdgpu: seperate sriov fb aperture setting
>>
>> Am 16.08.19 um 10:59 schrieb Frank.Min:
>>> sriov would not use agp, so seperate the fb aperture setting.
>> That won't work correctly. This way we don't program the AGP space into the hardware any more, but would still try to use it.
>>
>> We rather need to adjust the amdgpu_gmc_agp_location() function or it's caller to not assign an AGP space in the first place.
>>
>> Christian.
>>
>>> Change-Id: I1372cd355326731a31361bff13d79e12121b8651
>>> Signed-off-by: Frank.Min <Frank.Min at amd.com>
>>> ---
>>> drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 39 ++++++++++++++++++++------------
>>> drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 12 +++++-----
>>> drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c | 27 +++++++++++++++-------
>>> 3 files changed, 49 insertions(+), 29 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>> b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>> index 6ce37ce..ec78c8b 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>> @@ -75,23 +75,32 @@ static void gfxhub_v1_0_init_system_aperture_regs(struct amdgpu_device *adev)
>>> WREG32_SOC15_RLC(GC, 0, mmMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
>>> WREG32_SOC15_RLC(GC, 0, mmMC_VM_AGP_TOP, adev->gmc.agp_end >>
>>> 24);
>>>
>>> - /* Program the system aperture low logical page number. */
>>> - WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> - min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
>>> + if (amdgpu_sriov_vf(adev)) {
>>> + /* Program the system aperture low logical page number. */
>>> + WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> + adev->gmc.fb_start >> 18);
>>>
>>> - if (adev->asic_type == CHIP_RAVEN && adev->rev_id >= 0x8)
>>> - /*
>>> - * Raven2 has a HW issue that it is unable to use the vram which
>>> - * is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR. So here is the
>>> - * workaround that increase system aperture high address (add 1)
>>> - * to get rid of the VM fault and hardware hang.
>>> - */
>>> WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> - max((adev->gmc.fb_end >> 18) + 0x1,
>>> - adev->gmc.agp_end >> 18));
>>> - else
>>> - WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> - max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
>>> + adev->gmc.fb_end >> 18);
>>> + } else {
>>> + /* Program the system aperture low logical page number. */
>>> + WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> + min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
>>> +
>>> + if (adev->asic_type == CHIP_RAVEN && adev->rev_id >= 0x8)
>>> + /*
>>> + * Raven2 has a HW issue that it is unable to use the vram which
>>> + * is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR. So here is the
>>> + * workaround that increase system aperture high address (add 1)
>>> + * to get rid of the VM fault and hardware hang.
>>> + */
>>> + WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> + max((adev->gmc.fb_end >> 18) + 0x1,
>>> + adev->gmc.agp_end >> 18));
>>> + else
>>> + WREG32_SOC15_RLC(GC, 0, mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> + max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
>>> + }
>>>
>>> /* Set default page address. */
>>> value = adev->vram_scratch.gpu_addr - adev->gmc.vram_start diff
>>> --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>> index 6de1726..1f8bdfa 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>> @@ -920,12 +920,12 @@ static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev,
>>> struct amdgpu_gmc *mc)
>>> {
>>> u64 base = 0;
>>> - if (!amdgpu_sriov_vf(adev)) {
>>> - if (adev->asic_type == CHIP_ARCTURUS)
>>> - base = mmhub_v9_4_get_fb_location(adev);
>>> - else
>>> - base = mmhub_v1_0_get_fb_location(adev);
>>> - }
>>> +
>>> + if (adev->asic_type == CHIP_ARCTURUS)
>>> + base = mmhub_v9_4_get_fb_location(adev);
>>> + else
>>> + base = mmhub_v1_0_get_fb_location(adev);
>>> +
>>> /* add the xgmi offset of the physical node */
>>> base += adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
>>> amdgpu_gmc_vram_location(adev, mc, base); diff --git
>>> a/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
>>> b/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
>>> index 0cf7ef4..ea3359f 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v9_4.c
>>> @@ -118,14 +118,25 @@ static void mmhub_v9_4_init_system_aperture_regs(struct amdgpu_device *adev,
>>> adev->gmc.agp_start >> 24);
>>>
>>> /* Program the system aperture low logical page number. */
>>> - WREG32_SOC15_OFFSET(MMHUB, 0,
>>> - mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> - hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> - min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
>>> - WREG32_SOC15_OFFSET(MMHUB, 0,
>>> - mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> - hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> - max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
>>> + if (amdgpu_sriov_vf(adev)) {
>>> + WREG32_SOC15_OFFSET(MMHUB, 0,
>>> + mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> + hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> + adev->gmc.fb_start >> 18);
>>> + WREG32_SOC15_OFFSET(MMHUB, 0,
>>> + mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> + hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> + adev->gmc.fb_end >> 18);
>>> + } else {
>>> + WREG32_SOC15_OFFSET(MMHUB, 0,
>>> + mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_LOW_ADDR,
>>> + hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> + min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
>>> + WREG32_SOC15_OFFSET(MMHUB, 0,
>>> + mmVMSHAREDVC0_MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
>>> + hubid * MMHUB_INSTANCE_REGISTER_OFFSET,
>>> + max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
>>> + }
>>>
>>> /* Set default page address. */
>>> value = adev->vram_scratch.gpu_addr - adev->gmc.vram_start +
More information about the amd-gfx
mailing list