[PATCH v6 1/3] drm/xe/guc_ads: allocate UM queues in a separate BO

[Jia Yao <jia.yao@intel.com>]

Move the UM queues into a dedicated UC BO (ads->um_queue_bo) and avoid
CPU memset operations on it, which eliminates the CPU as a potential
cacheline-polluting agent and helps maintain consistency between GAM
writes and GuC reads.

Fixes: 9c57bc08652a ("drm/xe/lnl: Drop force_probe requirement")
Cc: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: <stable@vger.kernel.org> # v6.12+
Signed-off-by: Jia Yao <jia.yao@intel.com>
---
 drivers/gpu/drm/xe/xe_guc.c           |  6 ++
 drivers/gpu/drm/xe/xe_guc_ads.c       | 91 +++++++++++++++++----------
 drivers/gpu/drm/xe/xe_guc_ads_types.h |  5 ++
 3 files changed, 70 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
index 4286bd05c686..4f132bf8539d 100644
--- a/drivers/gpu/drm/xe/xe_guc.c
+++ b/drivers/gpu/drm/xe/xe_guc.c
@@ -732,6 +732,12 @@ static int xe_guc_realloc_post_hwconfig(struct xe_guc *guc)
 	if (ret)
 		return ret;
 
+	if (guc->ads.um_queue_bo) {
+		ret = xe_managed_bo_reinit_in_vram(xe, tile, &guc->ads.um_queue_bo);
+		if (ret)
+			return ret;
+	}
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c
index f0ac00586d3a..46c4d702d1f3 100644
--- a/drivers/gpu/drm/xe/xe_guc_ads.c
+++ b/drivers/gpu/drm/xe/xe_guc_ads.c
@@ -8,6 +8,7 @@
 #include <linux/fault-inject.h>
 
 #include <drm/drm_managed.h>
+#include <drm/drm_prime.h>
 
 #include <generated/xe_wa_oob.h>
 
@@ -64,10 +65,14 @@ ads_to_map(struct xe_guc_ads *ads)
 
 /*
  * The Additional Data Struct (ADS) has pointers for different buffers used by
- * the GuC. One single gem object contains the ADS struct itself (guc_ads) and
- * all the extra buffers indirectly linked via the ADS struct's entries.
+ * the GuC. One gem object (ads->bo) contains the ADS struct itself (guc_ads)
+ * and most of the extra buffers linked via the ADS struct's entries.  The UM
+ * fault queues (PAGE_FAULT, PAGE_FAULT_RESPONSE, ACCESS_COUNTER rings) are
+ * kept in a separate BO (ads->um_queue_bo) so that the full memset of ads->bo
+ * performed on every GT reset does not discard fault descriptors already
+ * written into the rings by the GPU.
  *
- * Layout of the ADS blob allocated for the GuC:
+ * Layout of the ADS blob (ads->bo):
  *
  *      +---------------------------------------+ <== base
  *      | guc_ads                               |
@@ -99,10 +104,6 @@ ads_to_map(struct xe_guc_ads *ads)
  *      +---------------------------------------+
  *      | padding                               |
  *      +---------------------------------------+ <== 4K aligned
- *      | UM queues                             |
- *      +---------------------------------------+
- *      | padding                               |
- *      +---------------------------------------+ <== 4K aligned
  *      | private data                          |
  *      +---------------------------------------+
  *      | padding                               |
@@ -156,16 +157,6 @@ static size_t guc_ads_capture_size(struct xe_guc_ads *ads)
 	return PAGE_ALIGN(ads->capture_size);
 }
 
-static size_t guc_ads_um_queues_size(struct xe_guc_ads *ads)
-{
-	struct xe_device *xe = ads_to_xe(ads);
-
-	if (!xe->info.has_usm)
-		return 0;
-
-	return GUC_UM_QUEUE_SIZE * GUC_UM_HW_QUEUE_MAX;
-}
-
 static size_t guc_ads_private_data_size(struct xe_guc_ads *ads)
 {
 	return PAGE_ALIGN(ads_to_guc(ads)->fw.private_data_size);
@@ -206,22 +197,12 @@ static size_t guc_ads_capture_offset(struct xe_guc_ads *ads)
 	return PAGE_ALIGN(offset);
 }
 
-static size_t guc_ads_um_queues_offset(struct xe_guc_ads *ads)
-{
-	u32 offset;
-
-	offset = guc_ads_capture_offset(ads) +
-		 guc_ads_capture_size(ads);
-
-	return PAGE_ALIGN(offset);
-}
-
 static size_t guc_ads_private_data_offset(struct xe_guc_ads *ads)
 {
 	size_t offset;
 
-	offset = guc_ads_um_queues_offset(ads) +
-		guc_ads_um_queues_size(ads);
+	offset = guc_ads_capture_offset(ads) +
+		guc_ads_capture_size(ads);
 
 	return PAGE_ALIGN(offset);
 }
@@ -461,6 +442,46 @@ int xe_guc_ads_init(struct xe_guc_ads *ads)
 
 	ads->bo = bo;
 
+	if (xe->info.has_usm) {
+		/*
+		 * Allocate a separate BO for the HW fault ring (UM queues).
+		 *
+		 * Round the size up to the next power of two so that on iGPU
+		 * (system memory, no IOMMU) the TTM pool issues a single
+		 * alloc_pages(order=N) call, maximising the chance of getting
+		 * a physically contiguous block.  GuC requires contiguous DPA.
+		 */
+		size_t um_size = IS_DGFX(xe) ?
+				 GUC_UM_QUEUE_SIZE * GUC_UM_HW_QUEUE_MAX :
+				 roundup_pow_of_two(GUC_UM_QUEUE_SIZE *
+						    GUC_UM_HW_QUEUE_MAX);
+
+		bo = xe_managed_bo_create_pin_map(xe, tile, um_size,
+						  XE_BO_FLAG_SYSTEM |
+						  XE_BO_FLAG_GGTT |
+						  XE_BO_FLAG_GGTT_INVALIDATE |
+						  XE_BO_FLAG_PINNED_NORESTORE);
+		if (IS_ERR(bo))
+			return PTR_ERR(bo);
+
+		/*
+		 * On pre-Xe3p platforms, GAM (not GuC) accesses the UM queue
+		 * ring via base_dpa, which must be a contiguous DMA address
+		 * range.  Verify that the allocated pages are contiguous in
+		 * DMA address space.
+		 */
+		if (xe_bo_sg(bo) &&
+		    !xe_guc_using_main_gamctrl_queues(ads_to_guc(ads)) &&
+		    unlikely(drm_prime_get_contiguous_size(xe_bo_sg(bo)) <
+			     GUC_UM_QUEUE_SIZE * GUC_UM_HW_QUEUE_MAX)) {
+			drm_err(&xe->drm,
+				"UM fault queue memory is not contiguous in DMA address space; GAM requires contiguous DPA\n");
+			return -ENOMEM;
+		}
+
+		ads->um_queue_bo = bo;
+	}
+
 	return 0;
 }
 ALLOW_ERROR_INJECTION(xe_guc_ads_init, ERRNO); /* See xe_pci_probe() */
@@ -893,7 +914,7 @@ static void guc_mmio_reg_state_init(struct xe_guc_ads *ads)
 
 static void guc_um_init_params(struct xe_guc_ads *ads)
 {
-	u32 um_queue_offset = guc_ads_um_queues_offset(ads);
+	struct xe_bo *um_bo = ads->um_queue_bo;
 	struct xe_guc *guc = ads_to_guc(ads);
 	struct xe_device *xe = ads_to_xe(ads);
 	u64 base_dpa;
@@ -903,8 +924,14 @@ static void guc_um_init_params(struct xe_guc_ads *ads)
 
 	with_dpa = !xe_guc_using_main_gamctrl_queues(guc);
 
-	base_ggtt = xe_bo_ggtt_addr(ads->bo) + um_queue_offset;
-	base_dpa = xe_bo_main_addr(ads->bo, PAGE_SIZE) + um_queue_offset;
+	if (um_bo) {
+		/* All USM platforms: UM queues in dedicated um_queue_bo */
+		base_ggtt = xe_bo_ggtt_addr(um_bo);
+		base_dpa = xe_bo_main_addr(um_bo, PAGE_SIZE);
+	} else {
+		/* Platform does not support USM: no UM queues, nothing to do */
+		return;
+	}
 
 	for (i = 0; i < GUC_UM_HW_QUEUE_MAX; ++i) {
 		/*
diff --git a/drivers/gpu/drm/xe/xe_guc_ads_types.h b/drivers/gpu/drm/xe/xe_guc_ads_types.h
index 48a8e092023f..845c1fbd93a4 100644
--- a/drivers/gpu/drm/xe/xe_guc_ads_types.h
+++ b/drivers/gpu/drm/xe/xe_guc_ads_types.h
@@ -16,6 +16,11 @@ struct xe_bo;
 struct xe_guc_ads {
 	/** @bo: Xe BO for GuC ads blob */
 	struct xe_bo *bo;
+	/**
+	 * @um_queue_bo: Dedicated BO for the HW fault ring (UM queues).
+	 * NULL if the platform does not support USM.
+	 */
+	struct xe_bo *um_queue_bo;
 	/** @golden_lrc_size: golden LRC size */
 	size_t golden_lrc_size;
 	/** @regset_size: size of register set passed to GuC for save/restore */
-- 
2.43.0