[PATCH v4] drm/xe/guc: Configure TLB timeout based on CT buffer size

[Nirmoy Das]

GuC TLB invalidation depends on GuC to process the request from the CT
queue and then the real time to invalidate TLB. Add a function to return
overestimated possible time a TLB inval H2G might take which can be used
as timeout value for TLB invalidation wait time.

v4: Make sure CTB is in 4K blocks(Michal) and other doc fixes
v3: Pass CT to xe_guc_ct_queue_proc_time_jiffies() (Michal)
    Add tlb_timeout_jiffies() that replaces TLB_TIMEOUT(Michal)
v2: Address reviews from Michal.

Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1622
Cc: Matthew Brost <matthew.brost at intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko at intel.com>
Suggested-by: Daniele Ceraolo Spurio <daniele.ceraolospurio at intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das at intel.com>
Acked-by: Matthew Brost <matthew.brost at intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko at intel.com>
---
 drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c | 30 +++++++++++++++------
 drivers/gpu/drm/xe/xe_guc_ct.c              | 17 ++++++++++++
 drivers/gpu/drm/xe/xe_guc_ct.h              |  2 ++
 3 files changed, 41 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
index e1f1ccb01143..d9359976ab8b 100644
--- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
@@ -17,7 +17,22 @@
 #include "xe_trace.h"
 #include "regs/xe_guc_regs.h"
 
-#define TLB_TIMEOUT	(HZ / 4)
+/*
+ * TLB inval depends on pending commands in the CT queue and then the real
+ * invalidation time. Double up the time to process full CT queue
+ * just to be on the safe side.
+ */
+static long tlb_timeout_jiffies(struct xe_gt *gt)
+{
+	/* this reflects what HW/GuC needs to process TLB inv request */
+	const long hw_tlb_timeout = HZ / 4;
+
+	/* this estimates actual delay caused by the CTB transport */
+	long delay = xe_guc_ct_queue_proc_time_jiffies(&gt->uc.guc.ct);
+
+	return hw_tlb_timeout + 2 * delay;
+}
+
 
 static void xe_gt_tlb_fence_timeout(struct work_struct *work)
 {
@@ -32,7 +47,7 @@ static void xe_gt_tlb_fence_timeout(struct work_struct *work)
 		s64 since_inval_ms = ktime_ms_delta(ktime_get(),
 						    fence->invalidation_time);
 
-		if (msecs_to_jiffies(since_inval_ms) < TLB_TIMEOUT)
+		if (msecs_to_jiffies(since_inval_ms) < tlb_timeout_jiffies(gt))
 			break;
 
 		trace_xe_gt_tlb_invalidation_fence_timeout(xe, fence);
@@ -47,7 +62,7 @@ static void xe_gt_tlb_fence_timeout(struct work_struct *work)
 	if (!list_empty(&gt->tlb_invalidation.pending_fences))
 		queue_delayed_work(system_wq,
 				   &gt->tlb_invalidation.fence_tdr,
-				   TLB_TIMEOUT);
+				   tlb_timeout_jiffies(gt));
 	spin_unlock_irq(&gt->tlb_invalidation.pending_lock);
 }
 
@@ -183,7 +198,7 @@ static int send_tlb_invalidation(struct xe_guc *guc,
 			if (list_is_singular(&gt->tlb_invalidation.pending_fences))
 				queue_delayed_work(system_wq,
 						   &gt->tlb_invalidation.fence_tdr,
-						   TLB_TIMEOUT);
+						   tlb_timeout_jiffies(gt));
 		}
 		spin_unlock_irq(&gt->tlb_invalidation.pending_lock);
 	} else if (ret < 0 && fence) {
@@ -390,8 +405,7 @@ int xe_gt_tlb_invalidation_vma(struct xe_gt *gt,
  * @gt: graphics tile
  * @seqno: seqno to wait which was returned from xe_gt_tlb_invalidation
  *
- * Wait for 200ms for a TLB invalidation to complete, in practice we always
- * should receive the TLB invalidation within 200ms.
+ * Wait for tlb_timeout_jiffies() for a TLB invalidation to complete.
  *
  * Return: 0 on success, -ETIME on TLB invalidation timeout
  */
@@ -410,7 +424,7 @@ int xe_gt_tlb_invalidation_wait(struct xe_gt *gt, int seqno)
 	 */
 	ret = wait_event_timeout(guc->ct.wq,
 				 tlb_invalidation_seqno_past(gt, seqno),
-				 TLB_TIMEOUT);
+				 tlb_timeout_jiffies(gt));
 	if (!ret) {
 		struct drm_printer p = xe_gt_err_printer(gt);
 
@@ -486,7 +500,7 @@ int xe_guc_tlb_invalidation_done_handler(struct xe_guc *guc, u32 *msg, u32 len)
 	if (!list_empty(&gt->tlb_invalidation.pending_fences))
 		mod_delayed_work(system_wq,
 				 &gt->tlb_invalidation.fence_tdr,
-				 TLB_TIMEOUT);
+				 tlb_timeout_jiffies(gt));
 	else
 		cancel_delayed_work(&gt->tlb_invalidation.fence_tdr);
 
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
index 873d1bcbedd7..7d2e937da1d8 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct.c
+++ b/drivers/gpu/drm/xe/xe_guc_ct.c
@@ -112,6 +112,23 @@ ct_to_xe(struct xe_guc_ct *ct)
 #define CTB_G2H_BUFFER_SIZE	(4 * CTB_H2G_BUFFER_SIZE)
 #define G2H_ROOM_BUFFER_SIZE	(CTB_G2H_BUFFER_SIZE / 4)
 
+/**
+ * xe_guc_ct_queue_proc_time_jiffies - Return maximum time to process a full
+ * CT command queue
+ * @ct: the &xe_guc_ct. Unused at this moment but will be used in the future.
+ *
+ * Observation is that a 4KiB buffer full of commands takes a little over a
+ * second to process. Use that to calculate maximum time to process a full CT
+ * command queue.
+ *
+ * Return: Maximum time to process a full CT queue in jiffies.
+ */
+long xe_guc_ct_queue_proc_time_jiffies(struct xe_guc_ct *ct)
+{
+	BUILD_BUG_ON(!IS_ALIGNED(CTB_H2G_BUFFER_SIZE, SZ_4));
+	return (CTB_H2G_BUFFER_SIZE / SZ_4K) * HZ;
+}
+
 static size_t guc_ct_size(void)
 {
 	return 2 * CTB_DESC_SIZE + CTB_H2G_BUFFER_SIZE +
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.h b/drivers/gpu/drm/xe/xe_guc_ct.h
index 105bb8e99a8d..190202fce2d0 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct.h
+++ b/drivers/gpu/drm/xe/xe_guc_ct.h
@@ -64,4 +64,6 @@ xe_guc_ct_send_block_no_fail(struct xe_guc_ct *ct, const u32 *action, u32 len)
 	return xe_guc_ct_send_recv_no_fail(ct, action, len, NULL);
 }
 
+long xe_guc_ct_queue_proc_time_jiffies(struct xe_guc_ct *ct);
+
 #endif
-- 
2.42.0