[PATCH v3 1/7] drm/amdgpu: update XGMI physical node id and GMC configs on resume

Zhang, GuoQing (Sam) GuoQing.Zhang at amd.com
Thu May 8 04:06:54 UTC 2025


On 2025/5/7 20:56, Christian König wrote:
> On 5/7/25 14:49, Sam wrote:
>> On 2025/5/7 20:21, Christian König wrote:
>>> On 5/7/25 13:03, Sam wrote:
>>>> On 2025/5/7 18:03, Lazar, Lijo wrote:
>>>>> On 5/7/2025 11:52 AM, Zhang, GuoQing (Sam) wrote:
>>>>>> [AMD Official Use Only - AMD Internal Distribution Only]
>>>>>>
>>>>>>
>>>>>>
>>>>>>> Please keep in mind that this is not the only scenario addressed by the
>>>>>>> driver - for ex: a resume sequence is executed after a device reset.
>>>>>>> This patch itself introduces unwanted sequences for other commonly used
>>>>>>> usecases. Please rework on the series without breaking existing usecases.
>>>>>>> Thanks,
>>>>>>> Lijo
>>>>>>    Hi @Lazar, Lijo<mailto:Lijo.Lazar at amd.com>, Thank you for the feedback.
>>>>>>
>>>>>>    I also think the new code should be inside a check so that new code is
>>>>>> executed only on resume with different VF and do not break existing
>>>>>> usecases. Following is the implementation of this approach I can think of.
>>>>>>
>>>>>> - introduce new field `prev_physical_node_id ` in `struct amdgpu_xgmi `.
>>>>>> update the fields on resume.
>>>>>>
>>>>>> - put new code inside code block `if (prev_physical_node_id  !=
>>>>>> physical_node_id )`
>>>>>>
>>>>>>
>>>>> Can this happen only with XGMI under this condition? Any other method
>>>>> possible like preparing a 'unique signature' and matching it to identify
>>>>> if it resumed on an identically configured system?
>>>> Yes, this hibernate-resume with different VF feature is only for devices with XGMI. Detecting XGMI node id change is the only way I can think of to identify the case. It's also a very simple way.
>>>>
>>>> @Koenig, Christian<mailto:Christian.Koenig at amd.com> Are you OK with this approach, adding a check for the new code sequence?
>>> Well you still need to avoid calling gmc_v9_0_mc_init() since that is most likely incorrect.
>> Yes, I will change it to
>>
>>      if (amdgpu_xmgi_is_node_changed(adev))
>>          gmc_v9_0_vram_gtt_location(adev, &adev->gmc);
> Even that is incorrect. The VRAM and GTT location can't change on resume.
>
> What changes are the XGMI node ID and with it where inside the FB aperture our VRAM PDB0 should point to.


2 updates in `gmc_v9_0_vram_gtt_location()` is needed:
1. `vm_manager.vram_base_offset` is changed with new `xgmi.physical_node_id` at the end of `gmc_v9_0_vram_gtt_location()`.
2. `fb_start` and `fb_end` got reset in `mmhub_v1_8_get_fb_location()`  called by the new `amdgpu_bo_fb_aper_addr()`. It needs to be updated again.

For update 1, I can inline it in `gmc_v9_0_resume()`.
For update 2, I can disable reset `fb_start/fb_end` in `mmhub_v1_8_get_fb_location()` when pdb0 is enabled.
Is this OK? @Koenig, Christian <mailto:Christian.Koenig at amd.com>

Regards
Sam


> Regards,
> Christian.
>
>> And remove the change of `refresh`.
>>
>> Regards
>> Sam
>>
>>
>>> Regards,
>>> Christian.
>>>
>>>>> Regardless, instead of having a direct check, better to wrap it inside
>>>>> something like
>>>>>       if (amdgpu_virt_need_migration()) or something more appropriate.
>>>> Yes, I will do that. Thank you!
>>>>
>>>> Regards
>>>> Sam
>>>>
>>>>>     Thanks,
>>>>> Lijo
>>>>>
>>>>>> Is this approach acceptable? If not, can you suggest a better approach?
>>>>>> @Lazar, Lijo<mailto:Lijo.Lazar at amd.com> @Koenig, Christian
>>>>>> <mailto:Christian.Koenig at amd.com> Thank you!
>>>>>>
>>>>>>    Regards
>>>>>>
>>>>>> Sam
>>>>>>
>>>>>>    *From: *Lazar, Lijo<Lijo.Lazar at amd.com>
>>>>>> *Date: *Tuesday, May 6, 2025 at 19:55
>>>>>> *To: *Zhang, GuoQing (Sam)<GuoQing.Zhang at amd.com>, amd-
>>>>>> gfx at lists.freedesktop.org <amd-gfx at lists.freedesktop.org>
>>>>>> *Cc: *Zhao, Victor<Victor.Zhao at amd.com>, Chang, HaiJun
>>>>>> <HaiJun.Chang at amd.com>, Koenig, Christian<Christian.Koenig at amd.com>,
>>>>>> Deucher, Alexander<Alexander.Deucher at amd.com>, Zhang, Owen(SRDC)
>>>>>> <Owen.Zhang2 at amd.com>, Ma, Qing (Mark)<Qing.Ma at amd.com>, Jiang Liu
>>>>>> <gerry at linux.alibaba.com>
>>>>>> *Subject: *Re: [PATCH v3 1/7] drm/amdgpu: update XGMI physical node id
>>>>>> and GMC configs on resume
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 5/6/2025 3:06 PM, Samuel Zhang wrote:
>>>>>>> For virtual machine with vGPUs in SRIOV single device mode and XGMI
>>>>>>> is enabled, XGMI physical node ids may change when waking up from
>>>>>>> hiberation with different vGPU devices. So update XGMI physical node
>>>>>>> ids on resume.
>>>>>>>
>>>>>> Please keep in mind that this is not the only scenario addressed by the
>>>>>> driver - for ex: a resume sequence is executed after a device reset.
>>>>>> This patch itself introduces unwanted sequences for other commonly used
>>>>>> usecases. Please rework on the series without breaking existing usecases.
>>>>>>
>>>>>> Thanks,
>>>>>> Lijo
>>>>>>
>>>>>>> Update GPU memory controller configuration on resume if XGMI physical
>>>>>>> node ids are changed.
>>>>>>>
>>>>>>> Signed-off-by: Jiang Liu<gerry at linux.alibaba.com>
>>>>>>> Signed-off-by: Samuel Zhang<guoqing.zhang at amd.com>
>>>>>>> ---
>>>>>>>      drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 24 ++++++++++++++++++++++
>>>>>>>      drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c    |  3 +--
>>>>>>>      drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      |  4 ++++
>>>>>>>      3 files changed, 29 insertions(+), 2 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/
>>>>>> drm/amd/amdgpu/amdgpu_device.c
>>>>>>> index d477a901af84..e795af5067e5 100644
>>>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>>>>> @@ -5040,6 +5040,27 @@ int amdgpu_device_suspend(struct drm_device
>>>>>> *dev, bool notify_clients)
>>>>>>>            return 0;
>>>>>>>      }
>>>>>>>     +static int amdgpu_device_update_xgmi_info(struct amdgpu_device *adev)
>>>>>>> +{
>>>>>>> +     int r;
>>>>>>> +     unsigned int prev_physical_node_id;
>>>>>>> +
>>>>>>> +     /* Get xgmi info again for sriov to detect device changes */
>>>>>>> +     if (amdgpu_sriov_vf(adev) &&
>>>>>>> +         !(adev->flags & AMD_IS_APU) &&
>>>>>>> +         adev->gmc.xgmi.supported &&
>>>>>>> +         !adev->gmc.xgmi.connected_to_cpu) {
>>>>>>> +             prev_physical_node_id = adev->gmc.xgmi.physical_node_id;
>>>>>>> +             r = adev->gfxhub.funcs->get_xgmi_info(adev);
>>>>>>> +             if (r)
>>>>>>> +                     return r;
>>>>>>> +
>>>>>>> +             dev_info(adev->dev, "xgmi node, old id %d, new id %d\n",
>>>>>>> +                     prev_physical_node_id, adev-
>>>>>>> gmc.xgmi.physical_node_id);
>>>>>>> +     }
>>>>>>> +     return 0;
>>>>>>> +}
>>>>>>> +
>>>>>>>      /**
>>>>>>>       * amdgpu_device_resume - initiate device resume
>>>>>>>       *
>>>>>>> @@ -5059,6 +5080,9 @@ int amdgpu_device_resume(struct drm_device *dev,
>>>>>> bool notify_clients)
>>>>>>>                    r = amdgpu_virt_request_full_gpu(adev, true);
>>>>>>>                    if (r)
>>>>>>>                            return r;
>>>>>>> +             r = amdgpu_device_update_xgmi_info(adev);
>>>>>>> +             if (r)
>>>>>>> +                     return r;
>>>>>>>            }
>>>>>>>              if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
>>>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/
>>>>>> drm/amd/amdgpu/amdgpu_gmc.c
>>>>>>> index d1fa5e8e3937..a2abddf3c110 100644
>>>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
>>>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
>>>>>>> @@ -1298,8 +1298,7 @@ int amdgpu_gmc_get_nps_memranges(struct
>>>>>> amdgpu_device *adev,
>>>>>>>            if (!mem_ranges || !exp_ranges)
>>>>>>>                    return -EINVAL;
>>>>>>>     -     refresh = (adev->init_lvl->level !=
>>>>>> AMDGPU_INIT_LEVEL_MINIMAL_XGMI) &&
>>>>>>> -               (adev->gmc.reset_flags & AMDGPU_GMC_INIT_RESET_NPS);
>>>>>>> +     refresh = true;
>>>>>>>            ret = amdgpu_discovery_get_nps_info(adev, &nps_type, &ranges,
>>>>>>>                                                &range_cnt, refresh);
>>>>>>>     diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/
>>>>>> amd/amdgpu/gmc_v9_0.c
>>>>>>> index 59385da80185..1eb451a3743b 100644
>>>>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>>>>> @@ -2533,6 +2533,10 @@ static int gmc_v9_0_resume(struct
>>>>>> amdgpu_ip_block *ip_block)
>>>>>>>            struct amdgpu_device *adev = ip_block->adev;
>>>>>>>            int r;
>>>>>>>     +     r = gmc_v9_0_mc_init(adev);
>>>>>>> +     if (r)
>>>>>>> +             return r;
>>>>>>> +
>>>>>>>            /* If a reset is done for NPS mode switch, read the memory range
>>>>>>>             * information again.
>>>>>>>             */
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20250508/4ef45ab5/attachment-0001.htm>


More information about the amd-gfx mailing list