[PATCH v5 4/7] drm/msm: Fix cx collapse issue during recovery

From: Akhil P Oommen
Date: Thu Aug 18 2022 - 16:23:09 EST


There are some hardware logic under CX domain. For a successful
recovery, we should ensure cx headswitch collapses to ensure all the
stale states are cleard out. This is especially true to for a6xx family
where we can GMU co-processor.

Currently, cx doesn't collapse due to a devlink between gpu and its
smmu. So the *struct gpu device* needs to be runtime suspended to ensure
that the iommu driver removes its vote on cx gdsc.

Signed-off-by: Akhil P Oommen <quic_akhilpo@xxxxxxxxxxx>
---

(no changes since v4)

Changes in v4:
- Keep active_submit lock across the suspend & resume (Rob)
- Clear gpu->active_submits to silence a WARN() during runpm suspend (Rob)

Changes in v3:
- Simplied the pm refcount drop since we have just a single refcount now
for all active submits

drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 32 +++++++++++++++++++++++++++++---
drivers/gpu/drm/msm/msm_gpu.c | 4 +---
2 files changed, 30 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index 42ed9a3..0c8f19e 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -1193,7 +1193,7 @@ static void a6xx_recover(struct msm_gpu *gpu)
{
struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
struct a6xx_gpu *a6xx_gpu = to_a6xx_gpu(adreno_gpu);
- int i;
+ int i, active_submits;

adreno_dump_info(gpu);

@@ -1210,8 +1210,34 @@ static void a6xx_recover(struct msm_gpu *gpu)
*/
gmu_write(&a6xx_gpu->gmu, REG_A6XX_GMU_GMU_PWR_COL_KEEPALIVE, 0);

- gpu->funcs->pm_suspend(gpu);
- gpu->funcs->pm_resume(gpu);
+ pm_runtime_dont_use_autosuspend(&gpu->pdev->dev);
+
+ /* active_submit won't change until we make a submission */
+ mutex_lock(&gpu->active_lock);
+ active_submits = gpu->active_submits;
+
+ /*
+ * Temporarily clear active_submits count to silence a WARN() in the
+ * runtime suspend cb
+ */
+ gpu->active_submits = 0;
+
+ /* Drop the rpm refcount from active submits */
+ if (active_submits)
+ pm_runtime_put(&gpu->pdev->dev);
+
+ /* And the final one from recover worker */
+ pm_runtime_put_sync(&gpu->pdev->dev);
+
+ pm_runtime_use_autosuspend(&gpu->pdev->dev);
+
+ if (active_submits)
+ pm_runtime_get(&gpu->pdev->dev);
+
+ pm_runtime_get_sync(&gpu->pdev->dev);
+
+ gpu->active_submits = active_submits;
+ mutex_unlock(&gpu->active_lock);

msm_gpu_hw_init(gpu);
}
diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c
index 1945efb..07e55a6 100644
--- a/drivers/gpu/drm/msm/msm_gpu.c
+++ b/drivers/gpu/drm/msm/msm_gpu.c
@@ -426,9 +426,7 @@ static void recover_worker(struct kthread_work *work)
/* retire completed submits, plus the one that hung: */
retire_submits(gpu);

- pm_runtime_get_sync(&gpu->pdev->dev);
gpu->funcs->recover(gpu);
- pm_runtime_put_sync(&gpu->pdev->dev);

/*
* Replay all remaining submits starting with highest priority
@@ -445,7 +443,7 @@ static void recover_worker(struct kthread_work *work)
}
}

- pm_runtime_put_sync(&gpu->pdev->dev);
+ pm_runtime_put(&gpu->pdev->dev);

mutex_unlock(&gpu->lock);

--
2.7.4