refactor tdr detect (#734)

Signed-off-by: Lizhi Hou <[email protected]>

Author: houlz0507

src/driver/amdxdna/aie2_ctx.c

@@ -213,11 +213,14 @@ aie2_sched_notify(struct amdxdna_sched_job *job)
 {
 	struct amdxdna_ctx *ctx = job->ctx;
 	struct dma_fence *fence = job->fence;
+	struct amdxdna_dev_hdl *ndev;
 	int idx;
 
 	amdxdna_pm_suspend_put(ctx->client->xdna);
 
 	ctx->completed++;
+	ndev = ctx->client->xdna->dev_handle;
+	WRITE_ONCE(ndev->tdr_status, AIE2_TDR_SIGNALED);
 	trace_xdna_job(&job->base, ctx->name, "signaling fence", job->seq, job->opcode);
 	job->job_done = true;
 	dma_fence_signal(fence);
@@ -358,6 +361,7 @@ aie2_sched_job_run(struct drm_sched_job *sched_job)
 	struct amdxdna_sched_job *job = drm_job_to_xdna_job(sched_job);
 	struct amdxdna_gem_obj *cmd_abo = job->cmd_bo;
 	struct amdxdna_ctx *ctx = job->ctx;
+	struct amdxdna_dev_hdl *ndev;
 	enum cmd_chain_class class;
 	struct dma_fence *fence;
 	int ret = 0;
@@ -415,6 +419,8 @@ aie2_sched_job_run(struct drm_sched_job *sched_job)
 	} else {
 		if (job->opcode != OP_NOOP)
 			amdxdna_stats_start(ctx->client);
+		ndev = ctx->client->xdna->dev_handle;
+		WRITE_ONCE(ndev->tdr_status, AIE2_TDR_SIGNALED);
 	}
 
 	return fence;
@@ -968,7 +974,12 @@ int aie2_cmd_submit(struct amdxdna_ctx *ctx, struct amdxdna_sched_job *job,
 		goto rq_yield;
 	}
 
+#if KERNEL_VERSION(6, 17, 0) <= LINUX_VERSION_CODE
+	ret = drm_sched_job_init(&job->base, &ctx->priv->entity, 1, ctx,
+				 ctx->client->filp->client_id);
+#else
 	ret = drm_sched_job_init(&job->base, &ctx->priv->entity, 1, ctx);
+#endif
 	if (ret) {
 		XDNA_ERR(xdna, "DRM job init failed, ret %d", ret);
 		goto free_chain;

src/driver/amdxdna/aie2_ctx_runqueue.c

@@ -171,33 +171,21 @@ static bool part_handle_idle_ctx(struct aie2_partition *part, bool force)
 }
 
 static bool
-part_is_all_ctx_stuck(struct aie2_partition *part)
+part_has_pending_cmd(struct aie2_partition *part)
 {
 	struct amdxdna_dev *xdna;
 	struct amdxdna_ctx *ctx;
-	int progress_cnt = 0;
-	int running_cnt = 0;
 
 	xdna = ctx_rq_to_xdna_dev(part->rq);
 	list_for_each_entry(ctx, &part->conn_list, entry) {
-		u64 completed = ctx->completed;
-		u64 last = ctx->last_completed;
-		u64 submitted = ctx->submitted;
-
-		XDNA_DBG(xdna, "%s @[%d, %d] submitted %lld completed %lld last %lld",
+		XDNA_DBG(xdna, "%s @[%d, %d] submitted %lld completed %lld",
 			 ctx->name, part->start_col, part->end_col,
-			 submitted, completed, last);
-		if (submitted == completed)
-			continue;
-
-		running_cnt++;
-		if (last != completed) {
-			ctx->last_completed = completed;
-			progress_cnt++;
-		}
+			 ctx->submitted, ctx->completed);
+		if (ctx->submitted != ctx->completed)
+			return true;
 	}
 
-	return running_cnt && !progress_cnt;
+	return false;
 }
 
 static struct aie2_partition *
@@ -812,26 +800,37 @@ static void rq_parts_work(struct work_struct *work)
  */
 bool aie2_rq_is_all_context_stuck(struct aie2_ctx_rq *rq)
 {
+	struct amdxdna_dev_hdl *ndev;
 	struct aie2_partition *part;
 	struct amdxdna_dev *xdna;
-	int active_cnt = 0;
-	int stuck_cnt = 0;
+	bool pending = false;
+	u32 tdr;
 	int i;
 
 	xdna = ctx_rq_to_xdna_dev(rq);
+	ndev = xdna->dev_handle;
 	mutex_lock(&xdna->dev_lock);
 	for (i = 0; i < rq->num_parts; i++) {
 		part = &rq->parts[i];
 		if (!part->hwctx_cnt)
 			continue;
 
-		active_cnt++;
-		if (part_is_all_ctx_stuck(part))
-			stuck_cnt++;
+		pending = part_has_pending_cmd(part);
+		if (pending)
+			break;
 	}
 	mutex_unlock(&xdna->dev_lock);
 
-	return active_cnt && active_cnt == stuck_cnt;
+	tdr = READ_ONCE(ndev->tdr_status);
+	if (pending && xdna->tdr.progress == tdr && tdr == AIE2_TDR_WAIT)
+		return true;
+
+	if (tdr != AIE2_TDR_WAIT)
+		WRITE_ONCE(ndev->tdr_status, AIE2_TDR_WAIT);
+
+	xdna->tdr.progress = tdr;
+
+	return false;
 }
 
 bool aie2_rq_handle_idle_ctx(struct aie2_ctx_rq *rq)

src/driver/amdxdna/aie2_pci.h

@@ -76,6 +76,9 @@
 #define ctx_rq_to_xdna_dev(r) \
 	(ctx_rq_to_ndev(r)->xdna)
 
+#define AIE2_TDR_WAIT		0
+#define AIE2_TDR_SIGNALED	1
+
 struct amdxdna_ctx_priv;
 struct event_trace_req_buf;
 struct start_event_trace_resp;
@@ -360,6 +363,8 @@ struct amdxdna_dev_hdl {
 	struct mutex			aie2_lock;
 
 	struct aie2_ctx_rq		ctx_rq;
+
+	u32				tdr_status;
 };
 
 #define DEFINE_BAR_OFFSET(reg_name, bar, reg_addr) \

src/driver/amdxdna/amdxdna_ctx.c

@@ -113,7 +113,6 @@ int amdxdna_drm_create_hwctx_ioctl(struct drm_device *dev, void *data, struct dr
 	}
 
 	ctx->client = client;
-	ctx->last_completed = -1;
 	ctx->num_tiles = args->num_tiles;
 	ctx->mem_size = args->mem_size;
 	ctx->max_opc = args->max_opc;

src/driver/amdxdna/amdxdna_ctx.h

@@ -151,8 +151,6 @@ struct amdxdna_ctx {
 	u64				completed ____cacheline_aligned_in_smp;
 	/* Counter for freed job */
 	atomic64_t			job_free_cnt;
-	/* For context runqueue to keep last completed. low frequency update */
-	u64				last_completed;
 	/* For command completion notification. */
 	u32				syncobj_hdl;
 	struct amdxdna_ctx_health_data	health_data;

src/driver/amdxdna/amdxdna_pci_drv.c

@@ -5,9 +5,7 @@
 
 #include <linux/module.h>
 #include <linux/version.h>
-#if KERNEL_VERSION(6, 10, 0) > LINUX_VERSION_CODE
 #include <drm/drm_managed.h>
-#endif
 
 #include "amdxdna_pci_drv.h"
 #include "amdxdna_sysfs.h"
@@ -85,11 +83,7 @@ static int amdxdna_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!xdna->dev_info)
 		return -ENODEV;
 
-#if KERNEL_VERSION(6, 10, 0) > LINUX_VERSION_CODE
 	drmm_mutex_init(&xdna->ddev, &xdna->dev_lock);
-#else
-	devm_mutex_init(dev, &xdna->dev_lock);
-#endif
 	init_rwsem(&xdna->notifier_lock);
 	INIT_LIST_HEAD(&xdna->client_list);
 	pci_set_drvdata(pdev, xdna);

src/driver/amdxdna/amdxdna_tdr.h

@@ -19,6 +19,7 @@ struct amdxdna_tdr {
 	struct work_struct	tdr_work;
 	int			tdr_counter;
 	int			started;
+	u32			progress;
 };
 
 void amdxdna_tdr_start(struct amdxdna_tdr *tdr);