drm/etnaviv: rework MMU handling

This reworks the MMU handling to make it possible to have multiple MMU contexts.
A context is basically one instance of GPU page tables. Currently we have one
set of page tables per GPU, which isn't all that clever, as it has the
following two consequences:

1. All GPU clients (aka processes) are sharing the same pagetables, which means
there is no isolation between clients, but only between GPU assigned memory
spaces and the rest of the system. Better than nothing, but also not great.

2. Clients operating on the same set of buffers with different etnaviv GPU
cores, e.g. a workload using both the 2D and 3D GPU, need to map the used
buffers into the pagetable sets of each used GPU.

This patch reworks all the MMU handling to introduce the abstraction of the
MMU context. A context can be shared across different GPU cores, as long as
they have compatible MMU implementations, which is the case for all systems
with Vivante GPUs seen in the wild.

As MMUv1 is not able to change pagetables on the fly, without a
"stop the world" operation, which stops GPU, changes pagetables via CPU
interaction, restarts GPU, the implementation introduces a shared context on
MMUv1, which is returned whenever there is a request for a new context.

This patch assigns a MMU context to each GPU, so on MMUv2 systems there is
still one set of pagetables per GPU, but due to the shared context MMUv1
systems see a change in behavior as now a single pagetable set is used
across all GPU cores.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Guido Günther <agx@sigxcpu.org>

1 files changed, 108 insertions, 156 deletions

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_iommu_v2.c b/drivers/gpu/drm/etnaviv/etnaviv_iommu_v2.c
index d7cc184da571..5ca2077c148d 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_iommu_v2.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_iommu_v2.c
@@ -13,7 +13,6 @@
 #include "etnaviv_cmdbuf.h"
 #include "etnaviv_gpu.h"
 #include "etnaviv_mmu.h"
-#include "etnaviv_iommu.h"
 #include "state.xml.h"
 #include "state_hi.xml.h"
 
@@ -28,11 +27,9 @@
 
 #define MMUv2_MAX_STLB_ENTRIES		1024
 
-struct etnaviv_iommuv2_domain {
-	struct etnaviv_iommu_domain base;
-	/* P(age) T(able) A(rray) */
-	u64 *pta_cpu;
-	dma_addr_t pta_dma;
+struct etnaviv_iommuv2_context {
+	struct etnaviv_iommu_context base;
+	unsigned short id;
 	/* M(aster) TLB aka first level pagetable */
 	u32 *mtlb_cpu;
 	dma_addr_t mtlb_dma;
@@ -41,41 +38,62 @@ struct etnaviv_iommuv2_domain {
 	dma_addr_t stlb_dma[MMUv2_MAX_STLB_ENTRIES];
 };
 
-static struct etnaviv_iommuv2_domain *
-to_etnaviv_domain(struct etnaviv_iommu_domain *domain)
+static struct etnaviv_iommuv2_context *
+to_v2_context(struct etnaviv_iommu_context *context)
 {
-	return container_of(domain, struct etnaviv_iommuv2_domain, base);
+	return container_of(context, struct etnaviv_iommuv2_context, base);
 }
 
+static void etnaviv_iommuv2_free(struct etnaviv_iommu_context *context)
+{
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
+	int i;
+
+	drm_mm_takedown(&context->mm);
+
+	for (i = 0; i < MMUv2_MAX_STLB_ENTRIES; i++) {
+		if (v2_context->stlb_cpu[i])
+			dma_free_wc(context->global->dev, SZ_4K,
+				    v2_context->stlb_cpu[i],
+				    v2_context->stlb_dma[i]);
+	}
+
+	dma_free_wc(context->global->dev, SZ_4K, v2_context->mtlb_cpu,
+		    v2_context->mtlb_dma);
+
+	clear_bit(v2_context->id, context->global->v2.pta_alloc);
+
+	vfree(v2_context);
+}
 static int
-etnaviv_iommuv2_ensure_stlb(struct etnaviv_iommuv2_domain *etnaviv_domain,
+etnaviv_iommuv2_ensure_stlb(struct etnaviv_iommuv2_context *v2_context,
 			    int stlb)
 {
-	if (etnaviv_domain->stlb_cpu[stlb])
+	if (v2_context->stlb_cpu[stlb])
 		return 0;
 
-	etnaviv_domain->stlb_cpu[stlb] =
-			dma_alloc_wc(etnaviv_domain->base.dev, SZ_4K,
-				     &etnaviv_domain->stlb_dma[stlb],
+	v2_context->stlb_cpu[stlb] =
+			dma_alloc_wc(v2_context->base.global->dev, SZ_4K,
+				     &v2_context->stlb_dma[stlb],
 				     GFP_KERNEL);
 
-	if (!etnaviv_domain->stlb_cpu[stlb])
+	if (!v2_context->stlb_cpu[stlb])
 		return -ENOMEM;
 
-	memset32(etnaviv_domain->stlb_cpu[stlb], MMUv2_PTE_EXCEPTION,
+	memset32(v2_context->stlb_cpu[stlb], MMUv2_PTE_EXCEPTION,
 		 SZ_4K / sizeof(u32));
 
-	etnaviv_domain->mtlb_cpu[stlb] = etnaviv_domain->stlb_dma[stlb] |
-						      MMUv2_PTE_PRESENT;
+	v2_context->mtlb_cpu[stlb] =
+			v2_context->stlb_dma[stlb] | MMUv2_PTE_PRESENT;
+
 	return 0;
 }
 
-static int etnaviv_iommuv2_map(struct etnaviv_iommu_domain *domain,
+static int etnaviv_iommuv2_map(struct etnaviv_iommu_context *context,
 			       unsigned long iova, phys_addr_t paddr,
 			       size_t size, int prot)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(domain);
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
 	int mtlb_entry, stlb_entry, ret;
 	u32 entry = lower_32_bits(paddr) | MMUv2_PTE_PRESENT;
 
@@ -91,20 +109,19 @@ 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);
+	ret = etnaviv_iommuv2_ensure_stlb(v2_context, mtlb_entry);
 	if (ret)
 		return ret;
 
-	etnaviv_domain->stlb_cpu[mtlb_entry][stlb_entry] = entry;
+	v2_context->stlb_cpu[mtlb_entry][stlb_entry] = entry;
 
 	return 0;
 }
 
-static size_t etnaviv_iommuv2_unmap(struct etnaviv_iommu_domain *domain,
+static size_t etnaviv_iommuv2_unmap(struct etnaviv_iommu_context *context,
 				    unsigned long iova, size_t size)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(domain);
+	struct etnaviv_iommuv2_context *etnaviv_domain = to_v2_context(context);
 	int mtlb_entry, stlb_entry;
 
 	if (size != SZ_4K)
@@ -118,118 +135,35 @@ static size_t etnaviv_iommuv2_unmap(struct etnaviv_iommu_domain *domain,
 	return SZ_4K;
 }
 
-static int etnaviv_iommuv2_init(struct etnaviv_iommuv2_domain *etnaviv_domain)
-{
-	int ret;
-
-	/* allocate scratch page */
-	etnaviv_domain->base.bad_page_cpu =
-			dma_alloc_wc(etnaviv_domain->base.dev, SZ_4K,
-				     &etnaviv_domain->base.bad_page_dma,
-				     GFP_KERNEL);
-	if (!etnaviv_domain->base.bad_page_cpu) {
-		ret = -ENOMEM;
-		goto fail_mem;
-	}
-
-	memset32(etnaviv_domain->base.bad_page_cpu, 0xdead55aa,
-		 SZ_4K / sizeof(u32));
-
-	etnaviv_domain->pta_cpu = dma_alloc_wc(etnaviv_domain->base.dev,
-					       SZ_4K, &etnaviv_domain->pta_dma,
-					       GFP_KERNEL);
-	if (!etnaviv_domain->pta_cpu) {
-		ret = -ENOMEM;
-		goto fail_mem;
-	}
-
-	etnaviv_domain->mtlb_cpu = dma_alloc_wc(etnaviv_domain->base.dev,
-						SZ_4K, &etnaviv_domain->mtlb_dma,
-						GFP_KERNEL);
-	if (!etnaviv_domain->mtlb_cpu) {
-		ret = -ENOMEM;
-		goto fail_mem;
-	}
-
-	memset32(etnaviv_domain->mtlb_cpu, MMUv2_PTE_EXCEPTION,
-		 MMUv2_MAX_STLB_ENTRIES);
-
-	return 0;
-
-fail_mem:
-	if (etnaviv_domain->base.bad_page_cpu)
-		dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-			    etnaviv_domain->base.bad_page_cpu,
-			    etnaviv_domain->base.bad_page_dma);
-
-	if (etnaviv_domain->pta_cpu)
-		dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-			    etnaviv_domain->pta_cpu, etnaviv_domain->pta_dma);
-
-	if (etnaviv_domain->mtlb_cpu)
-		dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-			    etnaviv_domain->mtlb_cpu, etnaviv_domain->mtlb_dma);
-
-	return ret;
-}
-
-static void etnaviv_iommuv2_domain_free(struct etnaviv_iommu_domain *domain)
-{
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(domain);
-	int i;
-
-	dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-		    etnaviv_domain->base.bad_page_cpu,
-		    etnaviv_domain->base.bad_page_dma);
-
-	dma_free_wc(etnaviv_domain->base.dev, SZ_4K,
-		    etnaviv_domain->pta_cpu, etnaviv_domain->pta_dma);
-
-	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]);
-	}
-
-	vfree(etnaviv_domain);
-}
-
-static size_t etnaviv_iommuv2_dump_size(struct etnaviv_iommu_domain *domain)
+static size_t etnaviv_iommuv2_dump_size(struct etnaviv_iommu_context *context)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(domain);
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
 	size_t dump_size = SZ_4K;
 	int i;
 
 	for (i = 0; i < MMUv2_MAX_STLB_ENTRIES; i++)
-		if (etnaviv_domain->mtlb_cpu[i] & MMUv2_PTE_PRESENT)
+		if (v2_context->mtlb_cpu[i] & MMUv2_PTE_PRESENT)
 			dump_size += SZ_4K;
 
 	return dump_size;
 }
 
-static void etnaviv_iommuv2_dump(struct etnaviv_iommu_domain *domain, void *buf)
+static void etnaviv_iommuv2_dump(struct etnaviv_iommu_context *context, void *buf)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(domain);
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
 	int i;
 
-	memcpy(buf, etnaviv_domain->mtlb_cpu, SZ_4K);
+	memcpy(buf, v2_context->mtlb_cpu, SZ_4K);
 	buf += SZ_4K;
 	for (i = 0; i < MMUv2_MAX_STLB_ENTRIES; i++, buf += SZ_4K)
-		if (etnaviv_domain->mtlb_cpu[i] & MMUv2_PTE_PRESENT)
-			memcpy(buf, etnaviv_domain->stlb_cpu[i], SZ_4K);
+		if (v2_context->mtlb_cpu[i] & MMUv2_PTE_PRESENT)
+			memcpy(buf, v2_context->stlb_cpu[i], SZ_4K);
 }
 
-static void etnaviv_iommuv2_restore_nonsec(struct etnaviv_gpu *gpu)
+static void etnaviv_iommuv2_restore_nonsec(struct etnaviv_gpu *gpu,
+	struct etnaviv_iommu_context *context)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-			to_etnaviv_domain(gpu->mmu->domain);
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
 	u16 prefetch;
 
 	/* If the MMU is already enabled the state is still there. */
@@ -237,8 +171,8 @@ static void etnaviv_iommuv2_restore_nonsec(struct etnaviv_gpu *gpu)
 		return;
 
 	prefetch = etnaviv_buffer_config_mmuv2(gpu,
-				(u32)etnaviv_domain->mtlb_dma,
-				(u32)etnaviv_domain->base.bad_page_dma);
+				(u32)v2_context->mtlb_dma,
+				(u32)context->global->bad_page_dma);
 	etnaviv_gpu_start_fe(gpu, (u32)etnaviv_cmdbuf_get_pa(&gpu->buffer),
 			     prefetch);
 	etnaviv_gpu_wait_idle(gpu, 100);
@@ -246,10 +180,10 @@ static void etnaviv_iommuv2_restore_nonsec(struct etnaviv_gpu *gpu)
 	gpu_write(gpu, VIVS_MMUv2_CONTROL, VIVS_MMUv2_CONTROL_ENABLE);
 }
 
-static void etnaviv_iommuv2_restore_sec(struct etnaviv_gpu *gpu)
+static void etnaviv_iommuv2_restore_sec(struct etnaviv_gpu *gpu,
+	struct etnaviv_iommu_context *context)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain =
-				to_etnaviv_domain(gpu->mmu->domain);
+	struct etnaviv_iommuv2_context *v2_context = to_v2_context(context);
 	u16 prefetch;
 
 	/* If the MMU is already enabled the state is still there. */
@@ -257,26 +191,26 @@ static void etnaviv_iommuv2_restore_sec(struct etnaviv_gpu *gpu)
 		return;
 
 	gpu_write(gpu, VIVS_MMUv2_PTA_ADDRESS_LOW,
-		  lower_32_bits(etnaviv_domain->pta_dma));
+		  lower_32_bits(context->global->v2.pta_dma));
 	gpu_write(gpu, VIVS_MMUv2_PTA_ADDRESS_HIGH,
-		  upper_32_bits(etnaviv_domain->pta_dma));
+		  upper_32_bits(context->global->v2.pta_dma));
 	gpu_write(gpu, VIVS_MMUv2_PTA_CONTROL, VIVS_MMUv2_PTA_CONTROL_ENABLE);
 
 	gpu_write(gpu, VIVS_MMUv2_NONSEC_SAFE_ADDR_LOW,
-		  lower_32_bits(etnaviv_domain->base.bad_page_dma));
+		  lower_32_bits(context->global->bad_page_dma));
 	gpu_write(gpu, VIVS_MMUv2_SEC_SAFE_ADDR_LOW,
-		  lower_32_bits(etnaviv_domain->base.bad_page_dma));
+		  lower_32_bits(context->global->bad_page_dma));
 	gpu_write(gpu, VIVS_MMUv2_SAFE_ADDRESS_CONFIG,
 		  VIVS_MMUv2_SAFE_ADDRESS_CONFIG_NON_SEC_SAFE_ADDR_HIGH(
-		  upper_32_bits(etnaviv_domain->base.bad_page_dma)) |
+		  upper_32_bits(context->global->bad_page_dma)) |
 		  VIVS_MMUv2_SAFE_ADDRESS_CONFIG_SEC_SAFE_ADDR_HIGH(
-		  upper_32_bits(etnaviv_domain->base.bad_page_dma)));
+		  upper_32_bits(context->global->bad_page_dma)));
 
-	etnaviv_domain->pta_cpu[0] = etnaviv_domain->mtlb_dma |
-				     VIVS_MMUv2_CONFIGURATION_MODE_MODE4_K;
+	context->global->v2.pta_cpu[0] = v2_context->mtlb_dma |
+				 	 VIVS_MMUv2_CONFIGURATION_MODE_MODE4_K;
 
 	/* trigger a PTA load through the FE */
-	prefetch = etnaviv_buffer_config_pta(gpu);
+	prefetch = etnaviv_buffer_config_pta(gpu, v2_context->id);
 	etnaviv_gpu_start_fe(gpu, (u32)etnaviv_cmdbuf_get_pa(&gpu->buffer),
 			     prefetch);
 	etnaviv_gpu_wait_idle(gpu, 100);
@@ -284,14 +218,15 @@ static void etnaviv_iommuv2_restore_sec(struct etnaviv_gpu *gpu)
 	gpu_write(gpu, VIVS_MMUv2_SEC_CONTROL, VIVS_MMUv2_SEC_CONTROL_ENABLE);
 }
 
-void etnaviv_iommuv2_restore(struct etnaviv_gpu *gpu)
+static void etnaviv_iommuv2_restore(struct etnaviv_gpu *gpu,
+				    struct etnaviv_iommu_context *context)
 {
 	switch (gpu->sec_mode) {
 	case ETNA_SEC_NONE:
-		etnaviv_iommuv2_restore_nonsec(gpu);
+		etnaviv_iommuv2_restore_nonsec(gpu, context);
 		break;
 	case ETNA_SEC_KERNEL:
-		etnaviv_iommuv2_restore_sec(gpu);
+		etnaviv_iommuv2_restore_sec(gpu, context);
 		break;
 	default:
 		WARN(1, "unhandled GPU security mode\n");
@@ -299,39 +234,56 @@ void etnaviv_iommuv2_restore(struct etnaviv_gpu *gpu)
 	}
 }
 
-static const struct etnaviv_iommu_domain_ops etnaviv_iommuv2_ops = {
-	.free = etnaviv_iommuv2_domain_free,
+const struct etnaviv_iommu_ops etnaviv_iommuv2_ops = {
+	.free = etnaviv_iommuv2_free,
 	.map = etnaviv_iommuv2_map,
 	.unmap = etnaviv_iommuv2_unmap,
 	.dump_size = etnaviv_iommuv2_dump_size,
 	.dump = etnaviv_iommuv2_dump,
+	.restore = etnaviv_iommuv2_restore,
 };
 
-struct etnaviv_iommu_domain *
-etnaviv_iommuv2_domain_alloc(struct etnaviv_gpu *gpu)
+struct etnaviv_iommu_context *
+etnaviv_iommuv2_context_alloc(struct etnaviv_iommu_global *global)
 {
-	struct etnaviv_iommuv2_domain *etnaviv_domain;
-	struct etnaviv_iommu_domain *domain;
-	int ret;
+	struct etnaviv_iommuv2_context *v2_context;
+	struct etnaviv_iommu_context *context;
 
-	etnaviv_domain = vzalloc(sizeof(*etnaviv_domain));
-	if (!etnaviv_domain)
+	v2_context = vzalloc(sizeof(*v2_context));
+	if (!v2_context)
 		return NULL;
 
-	domain = &etnaviv_domain->base;
+	mutex_lock(&global->lock);
+	v2_context->id = find_first_zero_bit(global->v2.pta_alloc,
+					     ETNAVIV_PTA_ENTRIES);
+	if (v2_context->id < ETNAVIV_PTA_ENTRIES) {
+		set_bit(v2_context->id, global->v2.pta_alloc);
+	} else {
+		mutex_unlock(&global->lock);
+		goto out_free;
+	}
+	mutex_unlock(&global->lock);
 
-	domain->dev = gpu->dev;
-	domain->base = SZ_4K;
-	domain->size = (u64)SZ_1G * 4 - SZ_4K;
-	domain->ops = &etnaviv_iommuv2_ops;
+	v2_context->mtlb_cpu = dma_alloc_wc(global->dev, SZ_4K,
+					    &v2_context->mtlb_dma, GFP_KERNEL);
+	if (!v2_context->mtlb_cpu)
+		goto out_free_id;
 
-	ret = etnaviv_iommuv2_init(etnaviv_domain);
-	if (ret)
-		goto out_free;
+	memset32(v2_context->mtlb_cpu, MMUv2_PTE_EXCEPTION,
+		 MMUv2_MAX_STLB_ENTRIES);
+
+	context = &v2_context->base;
+	context->global = global;
+	kref_init(&context->refcount);
+	mutex_init(&context->lock);
+	INIT_LIST_HEAD(&context->mappings);
+	drm_mm_init(&context->mm, SZ_4K, (u64)SZ_1G * 4 - SZ_4K);
 
-	return &etnaviv_domain->base;
+	return context;
 
+out_free_id:
+	clear_bit(v2_context->id, global->v2.pta_alloc);
 out_free:
-	vfree(etnaviv_domain);
+	vfree(v2_context);
 	return NULL;
 }