Merge tag 'drm-xe-fixes-2025-05-09' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes

Driver Changes:
- Prevent PF queue overflow
- Hold all forcewake during mocs test
- Remove GSC flush on reset path
- Fix forcewake put on error path
- Fix runtime warning when building without svm

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/jffqa56f2zp4i5ztz677cdspgxhnw7qfop3dd3l2epykfpfvza@q2nw6wapsphz

Author: airlied

drivers/gpu/drm/xe/tests/xe_mocs.c

@@ -46,8 +46,11 @@ static void read_l3cc_table(struct xe_gt *gt,
 	unsigned int fw_ref, i;
 	u32 reg_val;
 
-	fw_ref = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
-	KUNIT_ASSERT_NE_MSG(test, fw_ref, 0, "Forcewake Failed.\n");
+	fw_ref = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+	if (!xe_force_wake_ref_has_domain(fw_ref, XE_FORCEWAKE_ALL)) {
+		xe_force_wake_put(gt_to_fw(gt), fw_ref);
+		KUNIT_ASSERT_TRUE_MSG(test, true, "Forcewake Failed.\n");
+	}
 
 	for (i = 0; i < info->num_mocs_regs; i++) {
 		if (!(i & 1)) {

drivers/gpu/drm/xe/xe_gsc.c

@@ -555,6 +555,28 @@ void xe_gsc_wait_for_worker_completion(struct xe_gsc *gsc)
 		flush_work(&gsc->work);
 }
 
+void xe_gsc_stop_prepare(struct xe_gsc *gsc)
+{
+	struct xe_gt *gt = gsc_to_gt(gsc);
+	int ret;
+
+	if (!xe_uc_fw_is_loadable(&gsc->fw) || xe_uc_fw_is_in_error_state(&gsc->fw))
+		return;
+
+	xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GSC);
+
+	/*
+	 * If the GSC FW load or the proxy init are interrupted, the only way
+	 * to recover it is to do an FLR and reload the GSC from scratch.
+	 * Therefore, let's wait for the init to complete before stopping
+	 * operations. The proxy init is the last step, so we can just wait on
+	 * that
+	 */
+	ret = xe_gsc_wait_for_proxy_init_done(gsc);
+	if (ret)
+		xe_gt_err(gt, "failed to wait for GSC init completion before uc stop\n");
+}
+
 /*
  * wa_14015076503: if the GSC FW is loaded, we need to alert it before doing a
  * GSC engine reset by writing a notification bit in the GS1 register and then

drivers/gpu/drm/xe/xe_gsc.h

@@ -16,6 +16,7 @@ struct xe_hw_engine;
 int xe_gsc_init(struct xe_gsc *gsc);
 int xe_gsc_init_post_hwconfig(struct xe_gsc *gsc);
 void xe_gsc_wait_for_worker_completion(struct xe_gsc *gsc);
+void xe_gsc_stop_prepare(struct xe_gsc *gsc);
 void xe_gsc_load_start(struct xe_gsc *gsc);
 void xe_gsc_hwe_irq_handler(struct xe_hw_engine *hwe, u16 intr_vec);
 

drivers/gpu/drm/xe/xe_gsc_proxy.c

@@ -71,6 +71,17 @@ bool xe_gsc_proxy_init_done(struct xe_gsc *gsc)
 	       HECI1_FWSTS1_PROXY_STATE_NORMAL;
 }
 
+int xe_gsc_wait_for_proxy_init_done(struct xe_gsc *gsc)
+{
+	struct xe_gt *gt = gsc_to_gt(gsc);
+
+	/* Proxy init can take up to 500ms, so wait double that for safety */
+	return xe_mmio_wait32(&gt->mmio, HECI_FWSTS1(MTL_GSC_HECI1_BASE),
+			      HECI1_FWSTS1_CURRENT_STATE,
+			      HECI1_FWSTS1_PROXY_STATE_NORMAL,
+			      USEC_PER_SEC, NULL, false);
+}
+
 static void __gsc_proxy_irq_rmw(struct xe_gsc *gsc, u32 clr, u32 set)
 {
 	struct xe_gt *gt = gsc_to_gt(gsc);

drivers/gpu/drm/xe/xe_gsc_proxy.h

@@ -12,6 +12,7 @@ struct xe_gsc;
 
 int xe_gsc_proxy_init(struct xe_gsc *gsc);
 bool xe_gsc_proxy_init_done(struct xe_gsc *gsc);
+int xe_gsc_wait_for_proxy_init_done(struct xe_gsc *gsc);
 int xe_gsc_proxy_start(struct xe_gsc *gsc);
 
 int xe_gsc_proxy_request_handler(struct xe_gsc *gsc);

drivers/gpu/drm/xe/xe_gt.c

@@ -857,7 +857,7 @@ void xe_gt_suspend_prepare(struct xe_gt *gt)
 
 	fw_ref = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
 
-	xe_uc_stop_prepare(&gt->uc);
+	xe_uc_suspend_prepare(&gt->uc);
 
 	xe_force_wake_put(gt_to_fw(gt), fw_ref);
 }

drivers/gpu/drm/xe/xe_gt_debugfs.c

@@ -92,22 +92,23 @@ static int hw_engines(struct xe_gt *gt, struct drm_printer *p)
 	struct xe_hw_engine *hwe;
 	enum xe_hw_engine_id id;
 	unsigned int fw_ref;
+	int ret = 0;
 
 	xe_pm_runtime_get(xe);
 	fw_ref = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
 	if (!xe_force_wake_ref_has_domain(fw_ref, XE_FORCEWAKE_ALL)) {
-		xe_pm_runtime_put(xe);
-		xe_force_wake_put(gt_to_fw(gt), fw_ref);
-		return -ETIMEDOUT;
+		ret = -ETIMEDOUT;
+		goto fw_put;
 	}
 
 	for_each_hw_engine(hwe, gt, id)
 		xe_hw_engine_print(hwe, p);
 
+fw_put:
 	xe_force_wake_put(gt_to_fw(gt), fw_ref);
 	xe_pm_runtime_put(xe);
 
-	return 0;
+	return ret;
 }
 
 static int powergate_info(struct xe_gt *gt, struct drm_printer *p)

drivers/gpu/drm/xe/xe_gt_pagefault.c

@@ -435,9 +435,16 @@ static int xe_alloc_pf_queue(struct xe_gt *gt, struct pf_queue *pf_queue)
 	num_eus = bitmap_weight(gt->fuse_topo.eu_mask_per_dss,
 				XE_MAX_EU_FUSE_BITS) * num_dss;
 
-	/* user can issue separate page faults per EU and per CS */
+	/*
+	 * user can issue separate page faults per EU and per CS
+	 *
+	 * XXX: Multiplier required as compute UMD are getting PF queue errors
+	 * without it. Follow on why this multiplier is required.
+	 */
+#define PF_MULTIPLIER	8
 	pf_queue->num_dw =
-		(num_eus + XE_NUM_HW_ENGINES) * PF_MSG_LEN_DW;
+		(num_eus + XE_NUM_HW_ENGINES) * PF_MSG_LEN_DW * PF_MULTIPLIER;
+#undef PF_MULTIPLIER
 
 	pf_queue->gt = gt;
 	pf_queue->data = devm_kcalloc(xe->drm.dev, pf_queue->num_dw,

drivers/gpu/drm/xe/xe_svm.c

@@ -947,3 +947,15 @@ int xe_devm_add(struct xe_tile *tile, struct xe_vram_region *vr)
 	return 0;
 }
 #endif
+
+/**
+ * xe_svm_flush() - SVM flush
+ * @vm: The VM.
+ *
+ * Flush all SVM actions.
+ */
+void xe_svm_flush(struct xe_vm *vm)
+{
+	if (xe_vm_in_fault_mode(vm))
+		flush_work(&vm->svm.garbage_collector.work);
+}

drivers/gpu/drm/xe/xe_svm.h

@@ -72,6 +72,9 @@ bool xe_svm_has_mapping(struct xe_vm *vm, u64 start, u64 end);
 int xe_svm_bo_evict(struct xe_bo *bo);
 
 void xe_svm_range_debug(struct xe_svm_range *range, const char *operation);
+
+void xe_svm_flush(struct xe_vm *vm);
+
 #else
 static inline bool xe_svm_range_pages_valid(struct xe_svm_range *range)
 {
@@ -124,6 +127,11 @@ static inline
 void xe_svm_range_debug(struct xe_svm_range *range, const char *operation)
 {
 }
+
+static inline void xe_svm_flush(struct xe_vm *vm)
+{
+}
+
 #endif
 
 /**