diff options
Diffstat (limited to 'drivers/gpu/drm/v3d/v3d_sched.c')
| -rw-r--r-- | drivers/gpu/drm/v3d/v3d_sched.c | 52 |
1 files changed, 34 insertions, 18 deletions
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 34c42d6e12cd..cb9df8822472 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -199,7 +199,6 @@ v3d_job_update_stats(struct v3d_job *job, enum v3d_queue queue) struct v3d_dev *v3d = job->v3d; struct v3d_file_priv *file = job->file->driver_priv; struct v3d_stats *global_stats = &v3d->queue[queue].stats; - struct v3d_stats *local_stats = &file->stats[queue]; u64 now = local_clock(); unsigned long flags; @@ -209,7 +208,12 @@ v3d_job_update_stats(struct v3d_job *job, enum v3d_queue queue) else preempt_disable(); - v3d_stats_update(local_stats, now); + /* Don't update the local stats if the file context has already closed */ + if (file) + v3d_stats_update(&file->stats[queue], now); + else + drm_dbg(&v3d->drm, "The file descriptor was closed before job completion\n"); + v3d_stats_update(global_stats, now); if (IS_ENABLED(CONFIG_LOCKDEP)) @@ -357,11 +361,11 @@ v3d_tfu_job_run(struct drm_sched_job *sched_job) V3D_WRITE(V3D_TFU_ICA(v3d->ver), job->args.ica); V3D_WRITE(V3D_TFU_IUA(v3d->ver), job->args.iua); V3D_WRITE(V3D_TFU_IOA(v3d->ver), job->args.ioa); - if (v3d->ver >= 71) + if (v3d->ver >= V3D_GEN_71) V3D_WRITE(V3D_V7_TFU_IOC, job->args.v71.ioc); V3D_WRITE(V3D_TFU_IOS(v3d->ver), job->args.ios); V3D_WRITE(V3D_TFU_COEF0(v3d->ver), job->args.coef[0]); - if (v3d->ver >= 71 || (job->args.coef[0] & V3D_TFU_COEF0_USECOEF)) { + if (v3d->ver >= V3D_GEN_71 || (job->args.coef[0] & V3D_TFU_COEF0_USECOEF)) { V3D_WRITE(V3D_TFU_COEF1(v3d->ver), job->args.coef[1]); V3D_WRITE(V3D_TFU_COEF2(v3d->ver), job->args.coef[2]); V3D_WRITE(V3D_TFU_COEF3(v3d->ver), job->args.coef[3]); @@ -412,7 +416,7 @@ v3d_csd_job_run(struct drm_sched_job *sched_job) * * XXX: Set the CFG7 register */ - if (v3d->ver >= 71) + if (v3d->ver >= V3D_GEN_71) V3D_CORE_WRITE(0, V3D_V7_CSD_QUEUED_CFG7, 0); /* CFG0 write kicks off the job. */ @@ -428,7 +432,8 @@ v3d_rewrite_csd_job_wg_counts_from_indirect(struct v3d_cpu_job *job) struct v3d_bo *bo = to_v3d_bo(job->base.bo[0]); struct v3d_bo *indirect = to_v3d_bo(indirect_csd->indirect); struct drm_v3d_submit_csd *args = &indirect_csd->job->args; - u32 *wg_counts; + struct v3d_dev *v3d = job->base.v3d; + u32 num_batches, *wg_counts; v3d_get_bo_vaddr(bo); v3d_get_bo_vaddr(indirect); @@ -441,8 +446,17 @@ v3d_rewrite_csd_job_wg_counts_from_indirect(struct v3d_cpu_job *job) args->cfg[0] = wg_counts[0] << V3D_CSD_CFG012_WG_COUNT_SHIFT; args->cfg[1] = wg_counts[1] << V3D_CSD_CFG012_WG_COUNT_SHIFT; args->cfg[2] = wg_counts[2] << V3D_CSD_CFG012_WG_COUNT_SHIFT; - args->cfg[4] = DIV_ROUND_UP(indirect_csd->wg_size, 16) * - (wg_counts[0] * wg_counts[1] * wg_counts[2]) - 1; + + num_batches = DIV_ROUND_UP(indirect_csd->wg_size, 16) * + (wg_counts[0] * wg_counts[1] * wg_counts[2]); + + /* V3D 7.1.6 and later don't subtract 1 from the number of batches */ + if (v3d->ver < 71 || (v3d->ver == 71 && v3d->rev < 6)) + args->cfg[4] = num_batches - 1; + else + args->cfg[4] = num_batches; + + WARN_ON(args->cfg[4] == ~0); for (int i = 0; i < 3; i++) { /* 0xffffffff indicates that the uniform rewrite is not needed */ @@ -731,14 +745,9 @@ v3d_gpu_reset_for_timeout(struct v3d_dev *v3d, struct drm_sched_job *sched_job) mutex_unlock(&v3d->reset_lock); - return DRM_GPU_SCHED_STAT_NOMINAL; + return DRM_GPU_SCHED_STAT_RESET; } -/* If the current address or return address have changed, then the GPU - * has probably made progress and we should delay the reset. This - * could fail if the GPU got in an infinite loop in the CL, but that - * is pretty unlikely outside of an i-g-t testcase. - */ static enum drm_gpu_sched_stat v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q, u32 *timedout_ctca, u32 *timedout_ctra) @@ -748,10 +757,16 @@ v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q, u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(q)); u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(q)); + /* If the current address or return address have changed, then the GPU + * has probably made progress and we should delay the reset. This + * could fail if the GPU got in an infinite loop in the CL, but that + * is pretty unlikely outside of an i-g-t testcase. + */ if (*timedout_ctca != ctca || *timedout_ctra != ctra) { *timedout_ctca = ctca; *timedout_ctra = ctra; - return DRM_GPU_SCHED_STAT_NOMINAL; + + return DRM_GPU_SCHED_STAT_NO_HANG; } return v3d_gpu_reset_for_timeout(v3d, sched_job); @@ -790,12 +805,13 @@ v3d_csd_job_timedout(struct drm_sched_job *sched_job) struct v3d_dev *v3d = job->base.v3d; u32 batches = V3D_CORE_READ(0, V3D_CSD_CURRENT_CFG4(v3d->ver)); - /* If we've made progress, skip reset and let the timer get - * rearmed. + /* If we've made progress, skip reset, add the job to the pending + * list, and let the timer get rearmed. */ if (job->timedout_batches != batches) { job->timedout_batches = batches; - return DRM_GPU_SCHED_STAT_NOMINAL; + + return DRM_GPU_SCHED_STAT_NO_HANG; } return v3d_gpu_reset_for_timeout(v3d, sched_job); |