drm/xe: Fix pagefault and access counter worker functions

When processing G2H messages for pagefault or access counters, we queue a
work item and call queue_work(). This fails if the worker thread is already
queued to run.
The expectation is that the worker function will do more than process a
single item and return. It needs to either process all pending items or
requeue itself if items are pending. But requeuing will add latency and
potential context switch can occur.

We don't want to add unnecessary latency and so the worker should process
as many faults as it can within a reasonable duration of time.
We also do not want to hog the cpu core, so here we execute in a loop
and requeue if still running after more than 20 ms.
This seems reasonable framework and easy to tune this futher if needed.

This resolves issues seen with several igt@xe_exec_fault_mode subtests
where the GPU will hang when KMD ignores a pending pagefault.

v2: requeue the worker instead of having an internal processing loop.
v3: implement hybrid model of v1 and v2
now, run for 20 msec before we will requeue if still running
v4: only requeue in worker if queue is non-empty (Matt B)

Signed-off-by: Brian Welty <[e-mail]>
Reviewed-by: Matthew Brost <[e-mail]>
Reviewed-by: Stuart Summers <[e-mail]>
Signed-off-by: Rodrigo Vivi <[e-mail]>

drivers/gpu/drm/xe/xe_gt_pagefault.c

···
 
 #define PF_MSG_LEN_DW	4
 
-static int get_pagefault(struct pf_queue *pf_queue, struct pagefault *pf)
+static bool get_pagefault(struct pf_queue *pf_queue, struct pagefault *pf)
 {
 	const struct xe_guc_pagefault_desc *desc;
-	int ret = 0;
+	bool ret = false;
 
 	spin_lock_irq(&pf_queue->lock);
 	if (pf_queue->head != pf_queue->tail) {
···
 
 		pf_queue->head = (pf_queue->head + PF_MSG_LEN_DW) %
 			PF_QUEUE_NUM_DW;
-	} else {
-		ret = -1;
+		ret = true;
 	}
 	spin_unlock_irq(&pf_queue->lock);
 
···
 	return full ? -ENOSPC : 0;
 }
 
+#define USM_QUEUE_MAX_RUNTIME_MS	20
+
 static void pf_queue_work_func(struct work_struct *w)
 {
 	struct pf_queue *pf_queue = container_of(w, struct pf_queue, worker);
···
 	struct xe_device *xe = gt_to_xe(gt);
 	struct xe_guc_pagefault_reply reply = {};
 	struct pagefault pf = {};
+	unsigned long threshold;
 	int ret;
 
-	ret = get_pagefault(pf_queue, &pf);
-	if (ret)
-		return;
+	threshold = jiffies + msecs_to_jiffies(USM_QUEUE_MAX_RUNTIME_MS);
 
-	ret = handle_pagefault(gt, &pf);
-	if (unlikely(ret)) {
-		print_pagefault(xe, &pf);
-		pf.fault_unsuccessful = 1;
-		drm_dbg(&xe->drm, "Fault response: Unsuccessful %d\n", ret);
+	while (get_pagefault(pf_queue, &pf)) {
+		ret = handle_pagefault(gt, &pf);
+		if (unlikely(ret)) {
+			print_pagefault(xe, &pf);
+			pf.fault_unsuccessful = 1;
+			drm_dbg(&xe->drm, "Fault response: Unsuccessful %d\n", ret);
+		}
+
+		reply.dw0 = FIELD_PREP(PFR_VALID, 1) |
+			FIELD_PREP(PFR_SUCCESS, pf.fault_unsuccessful) |
+			FIELD_PREP(PFR_REPLY, PFR_ACCESS) |
+			FIELD_PREP(PFR_DESC_TYPE, FAULT_RESPONSE_DESC) |
+			FIELD_PREP(PFR_ASID, pf.asid);
+
+		reply.dw1 = FIELD_PREP(PFR_VFID, pf.vfid) |
+			FIELD_PREP(PFR_ENG_INSTANCE, pf.engine_instance) |
+			FIELD_PREP(PFR_ENG_CLASS, pf.engine_class) |
+			FIELD_PREP(PFR_PDATA, pf.pdata);
+
+		send_pagefault_reply(&gt->uc.guc, &reply);
+
+		if (time_after(jiffies, threshold) &&
+		    pf_queue->head != pf_queue->tail) {
+			queue_work(gt->usm.pf_wq, w);
+			break;
+		}
 	}
-
-	reply.dw0 = FIELD_PREP(PFR_VALID, 1) |
-		FIELD_PREP(PFR_SUCCESS, pf.fault_unsuccessful) |
-		FIELD_PREP(PFR_REPLY, PFR_ACCESS) |
-		FIELD_PREP(PFR_DESC_TYPE, FAULT_RESPONSE_DESC) |
-		FIELD_PREP(PFR_ASID, pf.asid);
-
-	reply.dw1 = FIELD_PREP(PFR_VFID, pf.vfid) |
-		FIELD_PREP(PFR_ENG_INSTANCE, pf.engine_instance) |
-		FIELD_PREP(PFR_ENG_CLASS, pf.engine_class) |
-		FIELD_PREP(PFR_PDATA, pf.pdata);
-
-	send_pagefault_reply(&gt->uc.guc, &reply);
 }
 
 static void acc_queue_work_func(struct work_struct *w);
···
 
 #define ACC_MSG_LEN_DW        4
 
-static int get_acc(struct acc_queue *acc_queue, struct acc *acc)
+static bool get_acc(struct acc_queue *acc_queue, struct acc *acc)
 {
 	const struct xe_guc_acc_desc *desc;
-	int ret = 0;
+	bool ret = false;
 
 	spin_lock(&acc_queue->lock);
 	if (acc_queue->head != acc_queue->tail) {
···
 
 		acc_queue->head = (acc_queue->head + ACC_MSG_LEN_DW) %
 				  ACC_QUEUE_NUM_DW;
-	} else {
-		ret = -1;
+		ret = true;
 	}
 	spin_unlock(&acc_queue->lock);
 
···
 	struct xe_gt *gt = acc_queue->gt;
 	struct xe_device *xe = gt_to_xe(gt);
 	struct acc acc = {};
+	unsigned long threshold;
 	int ret;
 
-	ret = get_acc(acc_queue, &acc);
-	if (ret)
-		return;
+	threshold = jiffies + msecs_to_jiffies(USM_QUEUE_MAX_RUNTIME_MS);
 
-	ret = handle_acc(gt, &acc);
-	if (unlikely(ret)) {
-		print_acc(xe, &acc);
-		drm_warn(&xe->drm, "ACC: Unsuccessful %d\n", ret);
+	while (get_acc(acc_queue, &acc)) {
+		ret = handle_acc(gt, &acc);
+		if (unlikely(ret)) {
+			print_acc(xe, &acc);
+			drm_warn(&xe->drm, "ACC: Unsuccessful %d\n", ret);
+		}
+
+		if (time_after(jiffies, threshold) &&
+		    acc_queue->head != acc_queue->tail) {
+			queue_work(gt->usm.acc_wq, w);
+			break;
+		}
 	}
 }