iommu/arm-smmu-v3: Shrink the cdtab l1_desc array

The top of the 2 level CD table is (at most) 1024 entries big, and two
high order allocations are required. One of __le64 which is programmed
into the HW (8k) and one of struct arm_smmu_l1_ctx_desc which holds the
CPU pointer (16k).

There are two copies of the l2ptr_dma, one is stored in the struct
arm_smmu_l1_ctx_desc, and another is encoded in the __le64 for the HW to
use. Instead of storing two copies just decode the value from the __le64.
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/6-v4-6416877274e1+1af-smmuv3_tidy_jgg@nvidia.comSigned-off-by: Will Deacon <will@kernel.org>

drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c

@@ -1217,29 +1217,17 @@ static void arm_smmu_sync_cd(struct arm_smmu_master *master,
 	arm_smmu_cmdq_batch_submit(smmu, &cmds);
 }
 
-static int arm_smmu_alloc_cd_leaf_table(struct arm_smmu_device *smmu,
-					struct arm_smmu_l1_ctx_desc *l1_desc)
+static void arm_smmu_write_cd_l1_desc(__le64 *dst, dma_addr_t l2ptr_dma)
 {
-	size_t size = CTXDESC_L2_ENTRIES * (CTXDESC_CD_DWORDS << 3);
+	u64 val = (l2ptr_dma & CTXDESC_L1_DESC_L2PTR_MASK) | CTXDESC_L1_DESC_V;
 
-	l1_desc->l2ptr = dma_alloc_coherent(smmu->dev, size,
-					    &l1_desc->l2ptr_dma, GFP_KERNEL);
-	if (!l1_desc->l2ptr) {
-		dev_warn(smmu->dev,
-			 "failed to allocate context descriptor table\n");
-		return -ENOMEM;
-	}
-	return 0;
+	/* The HW has 64 bit atomicity with stores to the L2 CD table */
+	WRITE_ONCE(*dst, cpu_to_le64(val));
 }
 
-static void arm_smmu_write_cd_l1_desc(__le64 *dst,
-				      struct arm_smmu_l1_ctx_desc *l1_desc)
+static dma_addr_t arm_smmu_cd_l1_get_desc(const __le64 *src)
 {
-	u64 val = (l1_desc->l2ptr_dma & CTXDESC_L1_DESC_L2PTR_MASK) |
-		  CTXDESC_L1_DESC_V;
-
-	/* The HW has 64 bit atomicity with stores to the L2 CD table */
-	WRITE_ONCE(*dst, cpu_to_le64(val));
+	return le64_to_cpu(*src) & CTXDESC_L1_DESC_L2PTR_MASK;
 }
 
 struct arm_smmu_cd *arm_smmu_get_cd_ptr(struct arm_smmu_master *master,
@@ -1281,13 +1269,18 @@ static struct arm_smmu_cd *arm_smmu_alloc_cd_ptr(struct arm_smmu_master *master,
 
 		l1_desc = &cd_table->l1_desc[idx];
 		if (!l1_desc->l2ptr) {
-			__le64 *l1ptr;
-
-			if (arm_smmu_alloc_cd_leaf_table(smmu, l1_desc))
+			dma_addr_t l2ptr_dma;
+			size_t size;
+
+			size = CTXDESC_L2_ENTRIES * sizeof(struct arm_smmu_cd);
+			l1_desc->l2ptr = dma_alloc_coherent(smmu->dev, size,
+							    &l2ptr_dma,
+							    GFP_KERNEL);
+			if (!l1_desc->l2ptr)
 				return NULL;
 
-			l1ptr = cd_table->cdtab + idx * CTXDESC_L1_DESC_DWORDS;
-			arm_smmu_write_cd_l1_desc(l1ptr, l1_desc);
+			arm_smmu_write_cd_l1_desc(&cd_table->cdtab[idx],
+						  l2ptr_dma);
 			/* An invalid L1CD can be cached */
 			arm_smmu_sync_cd(master, ssid, false);
 		}
@@ -1480,7 +1473,8 @@ static void arm_smmu_free_cd_tables(struct arm_smmu_master *master)
 
 			dma_free_coherent(smmu->dev, size,
 					  cd_table->l1_desc[i].l2ptr,
-					  cd_table->l1_desc[i].l2ptr_dma);
+					  arm_smmu_cd_l1_get_desc(
+						  &cd_table->cdtab[i]));
 		}
 		kfree(cd_table->l1_desc);
 

drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h

@@ -619,7 +619,6 @@ struct arm_smmu_ctx_desc {
 
 struct arm_smmu_l1_ctx_desc {
 	struct arm_smmu_cd		*l2ptr;
-	dma_addr_t			l2ptr_dma;
 };
 
 struct arm_smmu_ctx_desc_cfg {