drm/xe/lrc: document sentinel and make CTX_TIMESTAMP read TOCTOU-safe

Problem: CTX_TIMESTAMP MMIO reads could be stale if a context
switched out between check and read; LRC stores a sentinel while
a context starts that must not be treated as a real timestamp.

Fix: Check the LRC-stored sentinel before and after the MMIO read;
return the LRC value if the context switched out to avoid TOCTOU.

Note: Keep XE_LRC_CTX_TIMESTAMP_ACTIVE in xe_lrc.h as the
canonical sentinel.

Fixes: d243ef6a39c6 ("drm/xe/lrc: Refactor xe_lrc_timestamp to simplify logic")
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/work_items/7956
Assisted-by: GitHub-Copilot:claude-sonnet-5
Signed-off-by: Gajendra Uttamchand <gajendra.uttamchand@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Acked-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patch.msgid.link/20260810071812.213358-4-gajendra.uttamchand@intel.com
(cherry picked from commit a806534474df071a730d930df479976a812b699d)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

2 files changed, 21 insertions, 8 deletions

diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
index 78969d3e5904..9f8217ff1904 100644
--- a/drivers/gpu/drm/xe/xe_lrc.c
+++ b/drivers/gpu/drm/xe/xe_lrc.c
@@ -1096,7 +1096,7 @@ static void xe_lrc_finish(struct xe_lrc *lrc)
  * on until it is scheduled, we also read the ENGINE_ID MMIO in the WA BB and
  * store it in the PPHSWP.
  */
-#define CONTEXT_ACTIVE 1ULL
+#define CONTEXT_ACTIVE XE_LRC_CTX_TIMESTAMP_ACTIVE
 static ssize_t setup_utilization_wa(struct xe_lrc *lrc,
 				    struct xe_hw_engine *hwe,
 				    u32 *batch,
@@ -2727,21 +2727,27 @@ static u64 xe_lrc_update_multi_queue_timestamp(struct xe_lrc *lrc, u64 *old_ts)
 static u64 xe_lrc_context_timestamp(struct xe_lrc *lrc)
 {
 	u64 reg_ts, new_ts = lrc->ctx_timestamp;
+	u64 stored;
 
 	/* CTX_TIMESTAMP mmio read is invalid on VF, so return the LRC value */
 	if (IS_SRIOV_VF(lrc_to_xe(lrc)))
 		return xe_lrc_ctx_timestamp(lrc);
 
-	if (context_active(lrc) &&
-	    !get_ctx_timestamp(lrc, xe_lrc_engine_id(lrc), &reg_ts))
+	/* Safely read CTX_TIMESTAMP: Avoid TOCTOU on LRC-stored CONTEXT_ACTIVE sentinel */
+	stored = xe_lrc_ctx_timestamp(lrc);
+	if (stored != CONTEXT_ACTIVE)
+		return stored;
+
+	/* Context is active: read the live timestamp from the engine's MMIO register */
+	if (!get_ctx_timestamp(lrc, xe_lrc_engine_id(lrc), &reg_ts))
 		new_ts = reg_ts;
 
-	/*
-	 * If context swicthed out while we were here, just return the latest
-	 * LRC CTX TIMESTAMP value.
+	/* If the context switched out prefer using the value
+	 * from context-save over the stale MMIO read.
 	 */
-	if (!context_active(lrc))
-		return xe_lrc_ctx_timestamp(lrc);
+	stored = xe_lrc_ctx_timestamp(lrc);
+	if (stored != CONTEXT_ACTIVE)
+		return stored;
 
 	return new_ts;
 }
diff --git a/drivers/gpu/drm/xe/xe_lrc.h b/drivers/gpu/drm/xe/xe_lrc.h
index 0a3a611391ee..7be5e3da8bc8 100644
--- a/drivers/gpu/drm/xe/xe_lrc.h
+++ b/drivers/gpu/drm/xe/xe_lrc.h
@@ -9,6 +9,13 @@
 
 #include "xe_lrc_types.h"
 
+/*
+ * Sentinel value stored in lrc->ctx_timestamp while a context is starting.
+ * The hardware hasn't yet written the real CTX_TIMESTAMP, so this is not a
+ * valid elapsed-time sample and must not be used as one.
+ */
+#define XE_LRC_CTX_TIMESTAMP_ACTIVE 1ULL
+
 struct drm_printer;
 struct xe_bb;
 struct xe_device;