]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
drm/xe: Fix early wedge on GuC load failure
authorDaniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Wed, 11 Jun 2025 21:44:54 +0000 (14:44 -0700)
committerLucas De Marchi <lucas.demarchi@intel.com>
Thu, 12 Jun 2025 22:17:12 +0000 (15:17 -0700)
When the GuC fails to load we declare the device wedged. However, the
very first GuC load attempt on GT0 (from xe_gt_init_hwconfig) is done
before the GT1 GuC objects are initialized, so things go bad when the
wedge code attempts to cleanup GT1. To fix this, check the initialization
status in the functions called during wedge.

Fixes: 7dbe8af13c18 ("drm/xe: Wedge the entire device")
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Zhanjun Dong <zhanjun.dong@intel.com>
Cc: stable@vger.kernel.org # v6.12+: 1e1981b16bb1: drm/xe: Fix taking invalid lock on wedge
Cc: stable@vger.kernel.org # v6.12+
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20250611214453.1159846-2-daniele.ceraolospurio@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
drivers/gpu/drm/xe/xe_guc_ct.c
drivers/gpu/drm/xe/xe_guc_ct.h
drivers/gpu/drm/xe/xe_guc_submit.c

index 084cbdeba8eaa581e41ffab6a6a263fef7f2fa2d..e1362e608146b66a0354071c7bb666fe78b6be77 100644 (file)
@@ -137,6 +137,14 @@ void xe_gt_tlb_invalidation_reset(struct xe_gt *gt)
        struct xe_gt_tlb_invalidation_fence *fence, *next;
        int pending_seqno;
 
+       /*
+        * we can get here before the CTs are even initialized if we're wedging
+        * very early, in which case there are not going to be any pending
+        * fences so we can bail immediately.
+        */
+       if (!xe_guc_ct_initialized(&gt->uc.guc.ct))
+               return;
+
        /*
         * CT channel is already disabled at this point. No new TLB requests can
         * appear.
index 822f4c33f73051827c5d566a745bc226ceba24c4..e303fec1817475a72645162192eacd7ed3737034 100644 (file)
@@ -517,6 +517,9 @@ void xe_guc_ct_disable(struct xe_guc_ct *ct)
  */
 void xe_guc_ct_stop(struct xe_guc_ct *ct)
 {
+       if (!xe_guc_ct_initialized(ct))
+               return;
+
        xe_guc_ct_set_state(ct, XE_GUC_CT_STATE_STOPPED);
        stop_g2h_handler(ct);
 }
@@ -788,7 +791,7 @@ static int __guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action,
        u16 seqno;
        int ret;
 
-       xe_gt_assert(gt, ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED);
+       xe_gt_assert(gt, xe_guc_ct_initialized(ct));
        xe_gt_assert(gt, !g2h_len || !g2h_fence);
        xe_gt_assert(gt, !num_g2h || !g2h_fence);
        xe_gt_assert(gt, !g2h_len || num_g2h);
@@ -1424,7 +1427,7 @@ static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
        u32 action;
        u32 *hxg;
 
-       xe_gt_assert(gt, ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED);
+       xe_gt_assert(gt, xe_guc_ct_initialized(ct));
        lockdep_assert_held(&ct->fast_lock);
 
        if (ct->state == XE_GUC_CT_STATE_DISABLED)
index 5649bda828239df591c85f66a497e833926b2a3e..99c5dec446f2325ce820a0519866f2dfee2efbfa 100644 (file)
@@ -24,6 +24,11 @@ void xe_guc_ct_print(struct xe_guc_ct *ct, struct drm_printer *p, bool want_ctb)
 
 void xe_guc_ct_fixup_messages_with_ggtt(struct xe_guc_ct *ct, s64 ggtt_shift);
 
+static inline bool xe_guc_ct_initialized(struct xe_guc_ct *ct)
+{
+       return ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED;
+}
+
 static inline bool xe_guc_ct_enabled(struct xe_guc_ct *ct)
 {
        return ct->state == XE_GUC_CT_STATE_ENABLED;
index 2b61d017eeca8ef0443cfdca67e77db855d1ea30..7170e78e5b8eaf5867f5fd7db8a972dc1da4cd0f 100644 (file)
@@ -1776,6 +1776,9 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc)
 {
        int ret;
 
+       if (!guc->submission_state.initialized)
+               return 0;
+
        /*
         * Using an atomic here rather than submission_state.lock as this
         * function can be called while holding the CT lock (engine reset