drm/etnaviv: mmuv2: allocate 2nd level page tables on demand

With etnaviv not being tied into the IOMMU framework anymore, the MMU
functions will only be called under sleeping locks. Thus we are able
to allocate the memory for the 2nd level page tables on demand without
having to deal with memory allocation in atomic context.

This speeds up driver intitialization on MMUv2 GPU cores, as we don't
need to preallocate all the page table memory and also reduces memory
consumption for most workloads, as most of them won't use the full
GPU virtual address space.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

drivers/gpu/drm/etnaviv/etnaviv_iommu_v2.c

@@ -47,8 +47,8 @@ struct etnaviv_iommuv2_domain {
 	u32 *mtlb_cpu;
 	dma_addr_t mtlb_dma;
 	/* S(lave) TLB aka second level pagetable */
-	u32 *stlb_cpu[1024];
-	dma_addr_t stlb_dma[1024];
+	u32 *stlb_cpu[MMUv2_MAX_STLB_ENTRIES];
+	dma_addr_t stlb_dma[MMUv2_MAX_STLB_ENTRIES];
 };
 
 static struct etnaviv_iommuv2_domain *
@@ -57,13 +57,36 @@ to_etnaviv_domain(struct etnaviv_iommu_domain *domain)
 	return container_of(domain, struct etnaviv_iommuv2_domain, base);
 }
 
+static int
+etnaviv_iommuv2_ensure_stlb(struct etnaviv_iommuv2_domain *etnaviv_domain,
+			    int stlb)
+{
+	if (etnaviv_domain->stlb_cpu[stlb])
+		return 0;
+
+	etnaviv_domain->stlb_cpu[stlb] =
+			dma_alloc_wc(etnaviv_domain->base.dev, SZ_4K,
+				     &etnaviv_domain->stlb_dma[stlb],
+				     GFP_KERNEL);
+
+	if (!etnaviv_domain->stlb_cpu[stlb])
+		return -ENOMEM;
+
+	memset32(etnaviv_domain->stlb_cpu[stlb], MMUv2_PTE_EXCEPTION,
+		 SZ_4K / sizeof(u32));
+
+	etnaviv_domain->mtlb_cpu[stlb] = etnaviv_domain->stlb_dma[stlb] |
+						      MMUv2_PTE_PRESENT;
+	return 0;
+}
+
 static int etnaviv_iommuv2_map(struct etnaviv_iommu_domain *domain,
 			       unsigned long iova, phys_addr_t paddr,
 			       size_t size, int prot)
 {
 	struct etnaviv_iommuv2_domain *etnaviv_domain =
 			to_etnaviv_domain(domain);
-	int mtlb_entry, stlb_entry;
+	int mtlb_entry, stlb_entry, ret;
 	u32 entry = (u32)paddr | MMUv2_PTE_PRESENT;
 
 	if (size != SZ_4K)
@@ -75,6 +98,10 @@ static int etnaviv_iommuv2_map(struct etnaviv_iommu_domain *domain,
 	mtlb_entry = (iova & MMUv2_MTLB_MASK) >> MMUv2_MTLB_SHIFT;
 	stlb_entry = (iova & MMUv2_STLB_MASK) >> MMUv2_STLB_SHIFT;
 
+	ret = etnaviv_iommuv2_ensure_stlb(etnaviv_domain, mtlb_entry);
+	if (ret)
+		return ret;
+
 	etnaviv_domain->stlb_cpu[mtlb_entry][stlb_entry] = entry;
 
 	return 0;
@@ -101,7 +128,7 @@ static size_t etnaviv_iommuv2_unmap(struct etnaviv_iommu_domain *domain,
 static int etnaviv_iommuv2_init(struct etnaviv_iommuv2_domain *etnaviv_domain)
 {
 	u32 *p;
-	int ret, i, j;
+	int ret, i;
 
 	/* allocate scratch page */
 	etnaviv_domain->base.bad_page_cpu =
@@ -132,23 +159,8 @@ static int etnaviv_iommuv2_init(struct etnaviv_iommuv2_domain *etnaviv_domain)
 		goto fail_mem;
 	}
 
-	/* pre-populate STLB pages (may want to switch to on-demand later) */
-	for (i = 0; i < MMUv2_MAX_STLB_ENTRIES; i++) {
-		etnaviv_domain->stlb_cpu[i] =
-				dma_alloc_wc(etnaviv_domain->base.dev, SZ_4K,
-					     &etnaviv_domain->stlb_dma[i],
-					     GFP_KERNEL);
-		if (!etnaviv_domain->stlb_cpu[i]) {
-			ret = -ENOMEM;
-			goto fail_mem;
-		}
-		p = etnaviv_domain->stlb_cpu[i];
-		for (j = 0; j < SZ_4K / 4; j++)
-			*p++ = MMUv2_PTE_EXCEPTION;
-
-		etnaviv_domain->mtlb_cpu[i] = etnaviv_domain->stlb_dma[i] |
-					      MMUv2_PTE_PRESENT;
-	}
+	memset32(etnaviv_domain->mtlb_cpu, MMUv2_PTE_EXCEPTION,
+		 MMUv2_MAX_STLB_ENTRIES);
 
 	return 0;
 
@@ -166,13 +178,6 @@ static int etnaviv_iommuv2_init(struct etnaviv_iommuv2_domain *etnaviv_domain)
 		dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
 			    etnaviv_domain->mtlb_cpu, etnaviv_domain->mtlb_dma);
 
-	for (i = 0; i < MMUv2_MAX_STLB_ENTRIES; i++) {
-		if (etnaviv_domain->stlb_cpu[i])
-			dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-				    etnaviv_domain->stlb_cpu[i],
-				    etnaviv_domain->stlb_dma[i]);
-	}
-
 	return ret;
 }