Merge tag 'drm-fixes-2023-09-29' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Regular pull, this feel suspiciously light so I expect next week might
  be a bit heavier? Let's see how we go. This is from a code point of
  view ivpu and i915 fixes.

  The only other patch is adding Danilo Krummrich to the nouveau
  maintainers, he's agreed to take on more of the roll after Ben
  retired.

  MAINTAINERS:
   - add Danilo for nouveau

  ivpu:
   - Add PCI ids for Arrow Lake
   - Fix memory corruption during IPC
   - Avoid dmesg flooding
   - 40xx: Wait for clock resource
   - 40xx: Fix interrupt usage
   - 40xx: Support caching when loading firmware

  i915:
   - Fix a panic regression on gen8_ggtt_insert_entries
   - Fix load issue due to reservation address in ggtt_reserve_guc_top
   - Fix a possible deadlock with guc busyness worker"

* tag 'drm-fixes-2023-09-29' of git://anongit.freedesktop.org/drm/drm:
  accel/ivpu: Use cached buffers for FW loading
  accel/ivpu/40xx: Fix missing VPUIP interrupts
  accel/ivpu/40xx: Disable frequency change interrupt
  accel/ivpu/40xx: Ensure clock resource ownership Ack before Power-Up
  accel/ivpu: Don't flood dmesg with VPU ready message
  accel/ivpu: Do not use wait event interruptible
  MAINTAINERS: update nouveau maintainers
  i915/guc: Get runtime pm in busyness worker only if already active
  drm/i915/gt: Fix reservation address in ggtt_reserve_guc_top
  i915: Limit the length of an sg list to the requested length
  accel/ivpu: Add Arrow Lake pci id

3 files changed, 59 insertions, 13 deletions

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
index 8f1633c3fb93..73a4a4eb29e0 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
@@ -100,6 +100,7 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st,
 	st->nents = 0;
 	for (i = 0; i < page_count; i++) {
 		struct folio *folio;
+		unsigned long nr_pages;
 		const unsigned int shrink[] = {
 			I915_SHRINK_BOUND | I915_SHRINK_UNBOUND,
 			0,
@@ -150,6 +151,8 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st,
 			}
 		} while (1);
 
+		nr_pages = min_t(unsigned long,
+				folio_nr_pages(folio), page_count - i);
 		if (!i ||
 		    sg->length >= max_segment ||
 		    folio_pfn(folio) != next_pfn) {
@@ -157,13 +160,13 @@ int shmem_sg_alloc_table(struct drm_i915_private *i915, struct sg_table *st,
 				sg = sg_next(sg);
 
 			st->nents++;
-			sg_set_folio(sg, folio, folio_size(folio), 0);
+			sg_set_folio(sg, folio, nr_pages * PAGE_SIZE, 0);
 		} else {
 			/* XXX: could overflow? */
-			sg->length += folio_size(folio);
+			sg->length += nr_pages * PAGE_SIZE;
 		}
-		next_pfn = folio_pfn(folio) + folio_nr_pages(folio);
-		i += folio_nr_pages(folio) - 1;
+		next_pfn = folio_pfn(folio) + nr_pages;
+		i += nr_pages - 1;
 
 		/* Check that the i965g/gm workaround works. */
 		GEM_BUG_ON(gfp & __GFP_DMA32 && next_pfn >= 0x00100000UL);
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index dd0ed941441a..da21f2786b5d 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -511,20 +511,31 @@ void intel_ggtt_unbind_vma(struct i915_address_space *vm,
 	vm->clear_range(vm, vma_res->start, vma_res->vma_size);
 }
 
+/*
+ * Reserve the top of the GuC address space for firmware images. Addresses
+ * beyond GUC_GGTT_TOP in the GuC address space are inaccessible by GuC,
+ * which makes for a suitable range to hold GuC/HuC firmware images if the
+ * size of the GGTT is 4G. However, on a 32-bit platform the size of the GGTT
+ * is limited to 2G, which is less than GUC_GGTT_TOP, but we reserve a chunk
+ * of the same size anyway, which is far more than needed, to keep the logic
+ * in uc_fw_ggtt_offset() simple.
+ */
+#define GUC_TOP_RESERVE_SIZE (SZ_4G - GUC_GGTT_TOP)
+
 static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
 {
-	u64 size;
+	u64 offset;
 	int ret;
 
 	if (!intel_uc_uses_guc(&ggtt->vm.gt->uc))
 		return 0;
 
-	GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
-	size = ggtt->vm.total - GUC_GGTT_TOP;
+	GEM_BUG_ON(ggtt->vm.total <= GUC_TOP_RESERVE_SIZE);
+	offset = ggtt->vm.total - GUC_TOP_RESERVE_SIZE;
 
-	ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size,
-				   GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
-				   PIN_NOEVICT);
+	ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw,
+				   GUC_TOP_RESERVE_SIZE, offset,
+				   I915_COLOR_UNEVICTABLE, PIN_NOEVICT);
 	if (ret)
 		drm_dbg(&ggtt->vm.i915->drm,
 			"Failed to reserve top of GGTT for GuC\n");
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index b5b7f2fe8c78..dc7b40e06e38 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1433,6 +1433,36 @@ static void guc_timestamp_ping(struct work_struct *wrk)
 	int srcu, ret;
 
 	/*
+	 * Ideally the busyness worker should take a gt pm wakeref because the
+	 * worker only needs to be active while gt is awake. However, the
+	 * gt_park path cancels the worker synchronously and this complicates
+	 * the flow if the worker is also running at the same time. The cancel
+	 * waits for the worker and when the worker releases the wakeref, that
+	 * would call gt_park and would lead to a deadlock.
+	 *
+	 * The resolution is to take the global pm wakeref if runtime pm is
+	 * already active. If not, we don't need to update the busyness stats as
+	 * the stats would already be updated when the gt was parked.
+	 *
+	 * Note:
+	 * - We do not requeue the worker if we cannot take a reference to runtime
+	 *   pm since intel_guc_busyness_unpark would requeue the worker in the
+	 *   resume path.
+	 *
+	 * - If the gt was parked longer than time taken for GT timestamp to roll
+	 *   over, we ignore those rollovers since we don't care about tracking
+	 *   the exact GT time. We only care about roll overs when the gt is
+	 *   active and running workloads.
+	 *
+	 * - There is a window of time between gt_park and runtime suspend,
+	 *   where the worker may run. This is acceptable since the worker will
+	 *   not find any new data to update busyness.
+	 */
+	wakeref = intel_runtime_pm_get_if_active(&gt->i915->runtime_pm);
+	if (!wakeref)
+		return;
+
+	/*
 	 * Synchronize with gt reset to make sure the worker does not
 	 * corrupt the engine/guc stats. NB: can't actually block waiting
 	 * for a reset to complete as the reset requires flushing out
@@ -1440,10 +1470,9 @@ static void guc_timestamp_ping(struct work_struct *wrk)
 	 */
 	ret = intel_gt_reset_trylock(gt, &srcu);
 	if (ret)
-		return;
+		goto err_trylock;
 
-	with_intel_runtime_pm(&gt->i915->runtime_pm, wakeref)
-		__update_guc_busyness_stats(guc);
+	__update_guc_busyness_stats(guc);
 
 	/* adjust context stats for overflow */
 	xa_for_each(&guc->context_lookup, index, ce)
@@ -1452,6 +1481,9 @@ static void guc_timestamp_ping(struct work_struct *wrk)
 	intel_gt_reset_unlock(gt, srcu);
 
 	guc_enable_busyness_worker(guc);
+
+err_trylock:
+	intel_runtime_pm_put(&gt->i915->runtime_pm, wakeref);
 }
 
 static int guc_action_enable_usage_stats(struct intel_guc *guc)