[PATCH v2 6/7] drm/amdgpu: Trim amdgpu_pci_slot_reset by reusing code.

[Alex Deucher]

On Fri, Aug 28, 2020 at 12:06 PM Andrey Grodzovsky
<andrey.grodzovsky at amd.com> wrote:
>
> Reuse exsisting functions from GPU recovery to avoid code
> duplications.
>
> Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 75 ++++++------------------------
>  1 file changed, 14 insertions(+), 61 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 7f1b970..429167b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -4115,7 +4115,8 @@ static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
>
>  static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
>                                struct list_head *device_list_handle,
> -                              bool *need_full_reset_arg)
> +                              bool *need_full_reset_arg,
> +                              bool skip_hw_reset)
>  {
>         struct amdgpu_device *tmp_adev = NULL;
>         bool need_full_reset = *need_full_reset_arg, vram_lost = false;
> @@ -4125,7 +4126,7 @@ static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
>          * ASIC reset has to be done on all HGMI hive nodes ASAP
>          * to allow proper links negotiation in FW (within 1 sec)
>          */
> -       if (need_full_reset) {
> +       if (!skip_hw_reset && need_full_reset) {
>                 list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
>                         /* For XGMI run all resets in parallel to speed up the process */
>                         if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
> @@ -4521,7 +4522,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
>                 if (r)
>                         adev->asic_reset_res = r;
>         } else {
> -               r  = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset);
> +               r  = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset, false);
>                 if (r && r == -EAGAIN)
>                         goto retry;
>         }
> @@ -4850,14 +4851,19 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
>         struct drm_device *dev = pci_get_drvdata(pdev);
>         struct amdgpu_device *adev = drm_to_adev(dev);
>         int r, i;
> -       bool vram_lost;
> +       bool need_full_reset = true;
>         u32 memsize;
> +       struct list_head device_list;
>
>         DRM_INFO("PCI error: slot reset callback!!\n");
>
> +       INIT_LIST_HEAD(&device_list);
> +       list_add_tail(&adev->gmc.xgmi.head, &device_list);
> +
>         /* wait for asic to come out of reset */
>         msleep(500);
>
> +       /* Restore PCI confspace */
>         amdgpu_device_load_pci_state(pdev);
>
>         /* confirm  ASIC came out of reset */
> @@ -4873,70 +4879,15 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
>                 goto out;
>         }
>
> -       /* TODO Call amdgpu_pre_asic_reset instead */
>         adev->in_pci_err_recovery = true;
> -       r = amdgpu_device_ip_suspend(adev);
> +       r = amdgpu_device_pre_asic_reset(adev, NULL, &need_full_reset);
>         adev->in_pci_err_recovery = false;
>         if (r)
>                 goto out;
>
> -
> -       /* post card */
> -       r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
> -       if (r)
> -               goto out;
> -
> -       r = amdgpu_device_ip_resume_phase1(adev);
> -       if (r)
> -               goto out;
> -
> -       vram_lost = amdgpu_device_check_vram_lost(adev);
> -       if (vram_lost) {
> -               DRM_INFO("VRAM is lost due to GPU reset!\n");
> -               amdgpu_inc_vram_lost(adev);
> -       }
> -
> -       r = amdgpu_gtt_mgr_recover(
> -               &adev->mman.bdev.man[TTM_PL_TT]);
> -       if (r)
> -               goto out;
> -
> -       r = amdgpu_device_fw_loading(adev);
> -       if (r)
> -               return r;
> -
> -       r = amdgpu_device_ip_resume_phase2(adev);
> -       if (r)
> -               goto out;
> -
> -       if (vram_lost)
> -               amdgpu_device_fill_reset_magic(adev);
> -
> -       /*
> -        * Add this ASIC as tracked as reset was already
> -        * complete successfully.
> -        */
> -       amdgpu_register_gpu_instance(adev);
> -
> -       r = amdgpu_device_ip_late_init(adev);
> -       if (r)
> -               goto out;
> -
> -       amdgpu_fbdev_set_suspend(adev, 0);
> -
> -       /* must succeed. */
> -       amdgpu_ras_resume(adev);
> -
> -
> -       amdgpu_irq_gpu_reset_resume_helper(adev);
> -       r = amdgpu_ib_ring_tests(adev);
> -       if (r)
> -               goto out;
> -
> -       r = amdgpu_device_recover_vram(adev);
> +       r = amdgpu_do_asic_reset(NULL, &device_list, &need_full_reset, true);
>
>  out:
> -
>         if (!r) {
>                 amdgpu_device_cache_pci_state(adev->pdev);
>                 DRM_INFO("PCIe error recovery succeeded\n");
> @@ -5022,4 +4973,6 @@ bool amdgpu_device_load_pci_state(struct pci_dev *pdev)
>                 DRM_WARN("Failed to load PCI state, err:%d\n", r);
>                 return false;
>         }
> +
> +       return true;

Looks like this should have been part of a previous patch.  With that fixed:
Reviewed-by: Alex Deucher <alexander.deucher at amd.com>


>  }
> --
> 2.7.4
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx