[PATCH 04/10] drm/amdgpu: Add reset on init handler for XGMI

Alex Deucher alexdeucher at gmail.com
Mon Sep 9 21:25:17 UTC 2024


On Mon, Sep 2, 2024 at 3:34 AM Lijo Lazar <lijo.lazar at amd.com> wrote:
>
> In some cases, device needs to be reset before first use. Add handlers
> for doing device reset during driver init sequence.
>
> Signed-off-by: Lijo Lazar <lijo.lazar at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu.h       |   1 +
>  drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 148 ++++++++++++++++++++++
>  drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h |   4 +
>  3 files changed, 153 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index e1ae898b42eb..d17506d9adae 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -562,6 +562,7 @@ enum amd_reset_method {
>         AMD_RESET_METHOD_MODE2,
>         AMD_RESET_METHOD_BACO,
>         AMD_RESET_METHOD_PCI,
> +       AMD_RESET_METHOD_ON_INIT,
>  };
>
>  struct amdgpu_video_codec_info {
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> index 66c1a868c0e1..29128d5edf14 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> @@ -26,6 +26,154 @@
>  #include "sienna_cichlid.h"
>  #include "smu_v13_0_10.h"
>
> +static int amdgpu_reset_xgmi_rst_on_init_suspend(struct amdgpu_device *adev)

maybe write out reset_on_init rather than rst_on_init?  or use the roi
or r_o_i for consistency?

> +{
> +       int i, r;
> +
> +       for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
> +               if (!adev->ip_blocks[i].status.valid)
> +                       continue;
> +               if (!adev->ip_blocks[i].status.hw)
> +                       continue;
> +               /* displays are handled in phase1 */
> +               if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)
> +                       continue;
> +
> +               /* XXX handle errors */
> +               r = adev->ip_blocks[i].version->funcs->suspend(adev);
> +               /* XXX handle errors */
> +               if (r) {
> +                       dev_err(adev->dev, "suspend of IP block <%s> failed %d",
> +                               adev->ip_blocks[i].version->funcs->name, r);
> +               }
> +               adev->ip_blocks[i].status.hw = false;
> +       }
> +
> +       return 0;
> +}
> +
> +static int
> +amdgpu_reset_xgmi_roi_prep_hwctxt(struct amdgpu_reset_control *reset_ctl,
> +                                 struct amdgpu_reset_context *reset_context)

use consistent naming; e.g., roi vs reset_on_init.

> +{
> +       struct list_head *reset_device_list = reset_context->reset_device_list;
> +       struct amdgpu_device *tmp_adev;
> +       int r;
> +
> +       list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +               amdgpu_unregister_gpu_instance(tmp_adev);
> +               r = amdgpu_reset_xgmi_rst_on_init_suspend(tmp_adev);
> +               if (r) {
> +                       dev_err(tmp_adev->dev,
> +                               "xgmi reset on init: prepare for reset failed");
> +                       return r;
> +               }
> +       }
> +
> +       return r;
> +}
> +
> +static int
> +amdgpu_reset_xgmi_roi_restore_hwctxt(struct amdgpu_reset_control *reset_ctl,
> +                                    struct amdgpu_reset_context *reset_context)
> +{
> +       struct list_head *reset_device_list = reset_context->reset_device_list;
> +       struct amdgpu_device *tmp_adev = NULL;
> +       int r;
> +
> +       r = amdgpu_device_reinit_after_reset(reset_context);
> +       if (r)
> +               return r;
> +       list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +               if (!tmp_adev->kfd.init_complete) {
> +                       kgd2kfd_init_zone_device(tmp_adev);
> +                       amdgpu_amdkfd_device_init(tmp_adev);
> +                       amdgpu_amdkfd_drm_client_create(tmp_adev);
> +               }
> +       }
> +
> +       return r;
> +}
> +
> +static int
> +amdgpu_reset_xgmi_roi_perform_reset(struct amdgpu_reset_control *reset_ctl,

Same comment here.

> +                                   struct amdgpu_reset_context *reset_context)
> +{
> +       struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
> +       struct list_head *reset_device_list = reset_context->reset_device_list;
> +       struct amdgpu_device *tmp_adev = NULL;
> +       int r;
> +
> +       dev_dbg(adev->dev, "xgmi roi - hw reset\n");
> +
> +       list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +               mutex_lock(&tmp_adev->reset_cntl->reset_lock);
> +               tmp_adev->reset_cntl->active_reset =
> +                       amdgpu_asic_reset_method(adev);
> +       }
> +       r = 0;
> +       /* Mode1 reset needs to be triggered on all devices together */
> +       list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +               /* For XGMI run all resets in parallel to speed up the process */
> +               if (!queue_work(system_unbound_wq, &tmp_adev->xgmi_reset_work))
> +                       r = -EALREADY;
> +               if (r) {
> +                       dev_err(tmp_adev->dev,
> +                               "xgmi reset on init: reset failed with error, %d",
> +                               r);
> +                       break;
> +               }
> +       }
> +
> +       /* For XGMI wait for all resets to complete before proceed */
> +       if (!r) {
> +               list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +                       flush_work(&tmp_adev->xgmi_reset_work);
> +                       r = tmp_adev->asic_reset_res;
> +                       if (r)
> +                               break;
> +               }
> +       }
> +
> +       list_for_each_entry(tmp_adev, reset_device_list, reset_list) {
> +               mutex_unlock(&tmp_adev->reset_cntl->reset_lock);
> +               tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_NONE;
> +       }
> +
> +       return r;
> +}
> +
> +int amdgpu_reset_xgmi_rst_on_init(struct amdgpu_reset_context *reset_context)

and here.

Alex

> +{
> +       struct list_head *reset_device_list = reset_context->reset_device_list;
> +       struct amdgpu_device *adev;
> +       int r;
> +
> +       if (!reset_device_list || list_empty(reset_device_list) ||
> +           list_is_singular(reset_device_list))
> +               return -EINVAL;
> +
> +       adev = list_first_entry(reset_device_list, struct amdgpu_device,
> +                               reset_list);
> +       r = amdgpu_reset_prepare_hwcontext(adev, reset_context);
> +       if (r)
> +               return r;
> +
> +       r = amdgpu_reset_perform_reset(adev, reset_context);
> +
> +       return r;
> +}
> +
> +struct amdgpu_reset_handler reset_on_init_handler = {
> +       .reset_method = AMD_RESET_METHOD_ON_INIT,
> +       .prepare_env = NULL,
> +       .prepare_hwcontext = amdgpu_reset_xgmi_roi_prep_hwctxt,
> +       .perform_reset = amdgpu_reset_xgmi_roi_perform_reset,
> +       .restore_hwcontext = amdgpu_reset_xgmi_roi_restore_hwctxt,
> +       .restore_env = NULL,
> +       .do_reset = NULL,
> +};
> +
>  int amdgpu_reset_init(struct amdgpu_device *adev)
>  {
>         int ret = 0;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
> index 1cb920abc2fe..d89929f412fc 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
> @@ -153,4 +153,8 @@ void amdgpu_reset_get_desc(struct amdgpu_reset_context *rst_ctxt, char *buf,
>         for (i = 0; (i < AMDGPU_RESET_MAX_HANDLERS) &&           \
>                     (handler = (*reset_ctl->reset_handlers)[i]); \
>              ++i)
> +
> +extern struct amdgpu_reset_handler reset_on_init_handler;
> +int amdgpu_reset_xgmi_rst_on_init(struct amdgpu_reset_context *reset_context);
> +
>  #endif
> --
> 2.25.1
>


More information about the amd-gfx mailing list