accel/ivpu: Add Intel VPU MMU support

VPU Memory Management Unit is based on ARM MMU-600.
It allows the creation of multiple virtual address spaces for
the device and map noncontinuous host memory (there is no dedicated
memory on the VPU).

Address space is implemented as a struct ivpu_mmu_context, it has an ID,
drm_mm allocator for VPU addresses and struct ivpu_mmu_pgtable that
holds actual 3-level, 4KB page table.
Context with ID 0 (global context) is created upon driver initialization
and it's mainly used for mapping memory required to execute
the firmware.
Contexts with non-zero IDs are user contexts allocated each time
the devices is open()-ed and they map command buffers and other
workload-related memory.
Workloads executing in a given contexts have access only
to the memory mapped in this context.

This patch is has two main files:
  - ivpu_mmu_context.c handles MMU page tables and memory mapping
  - ivpu_mmu.c implements a driver that programs the MMU device

Co-developed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Signed-off-by: Karol Wachowski <karol.wachowski@linux.intel.com>
Co-developed-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Signed-off-by: Krystian Pradzynski <krystian.pradzynski@linux.intel.com>
Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20230117092723.60441-3-jacek.lawrynowicz@linux.intel.com

1 files changed, 398 insertions, 0 deletions

diff --git a/drivers/accel/ivpu/ivpu_mmu_context.c b/drivers/accel/ivpu/ivpu_mmu_context.c
new file mode 100644
index 000000000000..8ce9b12ac356
--- /dev/null
+++ b/drivers/accel/ivpu/ivpu_mmu_context.c
@@ -0,0 +1,398 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020-2023 Intel Corporation
+ */
+
+#include <linux/bitfield.h>
+#include <linux/highmem.h>
+
+#include "ivpu_drv.h"
+#include "ivpu_hw.h"
+#include "ivpu_mmu.h"
+#include "ivpu_mmu_context.h"
+
+#define IVPU_MMU_PGD_INDEX_MASK          GENMASK(38, 30)
+#define IVPU_MMU_PMD_INDEX_MASK          GENMASK(29, 21)
+#define IVPU_MMU_PTE_INDEX_MASK          GENMASK(20, 12)
+#define IVPU_MMU_ENTRY_FLAGS_MASK        GENMASK(11, 0)
+#define IVPU_MMU_ENTRY_FLAG_NG           BIT(11)
+#define IVPU_MMU_ENTRY_FLAG_AF           BIT(10)
+#define IVPU_MMU_ENTRY_FLAG_USER         BIT(6)
+#define IVPU_MMU_ENTRY_FLAG_LLC_COHERENT BIT(2)
+#define IVPU_MMU_ENTRY_FLAG_TYPE_PAGE    BIT(1)
+#define IVPU_MMU_ENTRY_FLAG_VALID        BIT(0)
+
+#define IVPU_MMU_PAGE_SIZE    SZ_4K
+#define IVPU_MMU_PTE_MAP_SIZE (IVPU_MMU_PGTABLE_ENTRIES * IVPU_MMU_PAGE_SIZE)
+#define IVPU_MMU_PMD_MAP_SIZE (IVPU_MMU_PGTABLE_ENTRIES * IVPU_MMU_PTE_MAP_SIZE)
+#define IVPU_MMU_PGTABLE_SIZE (IVPU_MMU_PGTABLE_ENTRIES * sizeof(u64))
+
+#define IVPU_MMU_DUMMY_ADDRESS 0xdeadb000
+#define IVPU_MMU_ENTRY_VALID   (IVPU_MMU_ENTRY_FLAG_TYPE_PAGE | IVPU_MMU_ENTRY_FLAG_VALID)
+#define IVPU_MMU_ENTRY_INVALID (IVPU_MMU_DUMMY_ADDRESS & ~IVPU_MMU_ENTRY_FLAGS_MASK)
+#define IVPU_MMU_ENTRY_MAPPED  (IVPU_MMU_ENTRY_FLAG_AF | IVPU_MMU_ENTRY_FLAG_USER | \
+				IVPU_MMU_ENTRY_FLAG_NG | IVPU_MMU_ENTRY_VALID)
+
+static int ivpu_mmu_pgtable_init(struct ivpu_device *vdev, struct ivpu_mmu_pgtable *pgtable)
+{
+	dma_addr_t pgd_dma;
+	u64 *pgd;
+
+	pgd = dma_alloc_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, &pgd_dma, GFP_KERNEL);
+	if (!pgd)
+		return -ENOMEM;
+
+	pgtable->pgd = pgd;
+	pgtable->pgd_dma = pgd_dma;
+
+	return 0;
+}
+
+static void ivpu_mmu_pgtable_free(struct ivpu_device *vdev, struct ivpu_mmu_pgtable *pgtable)
+{
+	int pgd_index, pmd_index;
+
+	for (pgd_index = 0; pgd_index < IVPU_MMU_PGTABLE_ENTRIES; ++pgd_index) {
+		u64 **pmd_entries = pgtable->pgd_cpu_entries[pgd_index];
+		u64 *pmd = pgtable->pgd_entries[pgd_index];
+
+		if (!pmd_entries)
+			continue;
+
+		for (pmd_index = 0; pmd_index < IVPU_MMU_PGTABLE_ENTRIES; ++pmd_index) {
+			if (pmd_entries[pmd_index])
+				dma_free_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE,
+					    pmd_entries[pmd_index],
+					    pmd[pmd_index] & ~IVPU_MMU_ENTRY_FLAGS_MASK);
+		}
+
+		kfree(pmd_entries);
+		dma_free_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, pgtable->pgd_entries[pgd_index],
+			    pgtable->pgd[pgd_index] & ~IVPU_MMU_ENTRY_FLAGS_MASK);
+	}
+
+	dma_free_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, pgtable->pgd,
+		    pgtable->pgd_dma & ~IVPU_MMU_ENTRY_FLAGS_MASK);
+}
+
+static u64*
+ivpu_mmu_ensure_pmd(struct ivpu_device *vdev, struct ivpu_mmu_pgtable *pgtable, u64 pgd_index)
+{
+	u64 **pmd_entries;
+	dma_addr_t pmd_dma;
+	u64 *pmd;
+
+	if (pgtable->pgd_entries[pgd_index])
+		return pgtable->pgd_entries[pgd_index];
+
+	pmd = dma_alloc_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, &pmd_dma, GFP_KERNEL);
+	if (!pmd)
+		return NULL;
+
+	pmd_entries = kzalloc(IVPU_MMU_PGTABLE_SIZE, GFP_KERNEL);
+	if (!pmd_entries)
+		goto err_free_pgd;
+
+	pgtable->pgd_entries[pgd_index] = pmd;
+	pgtable->pgd_cpu_entries[pgd_index] = pmd_entries;
+	pgtable->pgd[pgd_index] = pmd_dma | IVPU_MMU_ENTRY_VALID;
+
+	return pmd;
+
+err_free_pgd:
+	dma_free_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, pmd, pmd_dma);
+	return NULL;
+}
+
+static u64*
+ivpu_mmu_ensure_pte(struct ivpu_device *vdev, struct ivpu_mmu_pgtable *pgtable,
+		    int pgd_index, int pmd_index)
+{
+	dma_addr_t pte_dma;
+	u64 *pte;
+
+	if (pgtable->pgd_cpu_entries[pgd_index][pmd_index])
+		return pgtable->pgd_cpu_entries[pgd_index][pmd_index];
+
+	pte = dma_alloc_wc(vdev->drm.dev, IVPU_MMU_PGTABLE_SIZE, &pte_dma, GFP_KERNEL);
+	if (!pte)
+		return NULL;
+
+	pgtable->pgd_cpu_entries[pgd_index][pmd_index] = pte;
+	pgtable->pgd_entries[pgd_index][pmd_index] = pte_dma | IVPU_MMU_ENTRY_VALID;
+
+	return pte;
+}
+
+static int
+ivpu_mmu_context_map_page(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
+			  u64 vpu_addr, dma_addr_t dma_addr, int prot)
+{
+	u64 *pte;
+	int pgd_index = FIELD_GET(IVPU_MMU_PGD_INDEX_MASK, vpu_addr);
+	int pmd_index = FIELD_GET(IVPU_MMU_PMD_INDEX_MASK, vpu_addr);
+	int pte_index = FIELD_GET(IVPU_MMU_PTE_INDEX_MASK, vpu_addr);
+
+	/* Allocate PMD - second level page table if needed */
+	if (!ivpu_mmu_ensure_pmd(vdev, &ctx->pgtable, pgd_index))
+		return -ENOMEM;
+
+	/* Allocate PTE - third level page table if needed */
+	pte = ivpu_mmu_ensure_pte(vdev, &ctx->pgtable, pgd_index, pmd_index);
+	if (!pte)
+		return -ENOMEM;
+
+	/* Update PTE - third level page table with DMA address */
+	pte[pte_index] = dma_addr | prot;
+
+	return 0;
+}
+
+static void ivpu_mmu_context_unmap_page(struct ivpu_mmu_context *ctx, u64 vpu_addr)
+{
+	int pgd_index = FIELD_GET(IVPU_MMU_PGD_INDEX_MASK, vpu_addr);
+	int pmd_index = FIELD_GET(IVPU_MMU_PMD_INDEX_MASK, vpu_addr);
+	int pte_index = FIELD_GET(IVPU_MMU_PTE_INDEX_MASK, vpu_addr);
+
+	/* Update PTE with dummy physical address and clear flags */
+	ctx->pgtable.pgd_cpu_entries[pgd_index][pmd_index][pte_index] = IVPU_MMU_ENTRY_INVALID;
+}
+
+static void
+ivpu_mmu_context_flush_page_tables(struct ivpu_mmu_context *ctx, u64 vpu_addr, size_t size)
+{
+	u64 end_addr = vpu_addr + size;
+	u64 *pgd = ctx->pgtable.pgd;
+
+	/* Align to PMD entry (2 MB) */
+	vpu_addr &= ~(IVPU_MMU_PTE_MAP_SIZE - 1);
+
+	while (vpu_addr < end_addr) {
+		int pgd_index = FIELD_GET(IVPU_MMU_PGD_INDEX_MASK, vpu_addr);
+		u64 pmd_end = (pgd_index + 1) * (u64)IVPU_MMU_PMD_MAP_SIZE;
+		u64 *pmd = ctx->pgtable.pgd_entries[pgd_index];
+
+		while (vpu_addr < end_addr && vpu_addr < pmd_end) {
+			int pmd_index = FIELD_GET(IVPU_MMU_PMD_INDEX_MASK, vpu_addr);
+			u64 *pte = ctx->pgtable.pgd_cpu_entries[pgd_index][pmd_index];
+
+			clflush_cache_range(pte, IVPU_MMU_PGTABLE_SIZE);
+			vpu_addr += IVPU_MMU_PTE_MAP_SIZE;
+		}
+		clflush_cache_range(pmd, IVPU_MMU_PGTABLE_SIZE);
+	}
+	clflush_cache_range(pgd, IVPU_MMU_PGTABLE_SIZE);
+}
+
+static int
+ivpu_mmu_context_map_pages(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
+			   u64 vpu_addr, dma_addr_t dma_addr, size_t size, int prot)
+{
+	while (size) {
+		int ret = ivpu_mmu_context_map_page(vdev, ctx, vpu_addr, dma_addr, prot);
+
+		if (ret)
+			return ret;
+
+		vpu_addr += IVPU_MMU_PAGE_SIZE;
+		dma_addr += IVPU_MMU_PAGE_SIZE;
+		size -= IVPU_MMU_PAGE_SIZE;
+	}
+
+	return 0;
+}
+
+static void ivpu_mmu_context_unmap_pages(struct ivpu_mmu_context *ctx, u64 vpu_addr, size_t size)
+{
+	while (size) {
+		ivpu_mmu_context_unmap_page(ctx, vpu_addr);
+		vpu_addr += IVPU_MMU_PAGE_SIZE;
+		size -= IVPU_MMU_PAGE_SIZE;
+	}
+}
+
+int
+ivpu_mmu_context_map_sgt(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
+			 u64 vpu_addr, struct sg_table *sgt,  bool llc_coherent)
+{
+	struct scatterlist *sg;
+	int prot;
+	int ret;
+	u64 i;
+
+	if (!IS_ALIGNED(vpu_addr, IVPU_MMU_PAGE_SIZE))
+		return -EINVAL;
+	/*
+	 * VPU is only 32 bit, but DMA engine is 38 bit
+	 * Ranges < 2 GB are reserved for VPU internal registers
+	 * Limit range to 8 GB
+	 */
+	if (vpu_addr < SZ_2G || vpu_addr > SZ_8G)
+		return -EINVAL;
+
+	prot = IVPU_MMU_ENTRY_MAPPED;
+	if (llc_coherent)
+		prot |= IVPU_MMU_ENTRY_FLAG_LLC_COHERENT;
+
+	mutex_lock(&ctx->lock);
+
+	for_each_sgtable_dma_sg(sgt, sg, i) {
+		u64 dma_addr = sg_dma_address(sg) - sg->offset;
+		size_t size = sg_dma_len(sg) + sg->offset;
+
+		ret = ivpu_mmu_context_map_pages(vdev, ctx, vpu_addr, dma_addr, size, prot);
+		if (ret) {
+			ivpu_err(vdev, "Failed to map context pages\n");
+			mutex_unlock(&ctx->lock);
+			return ret;
+		}
+		ivpu_mmu_context_flush_page_tables(ctx, vpu_addr, size);
+		vpu_addr += size;
+	}
+
+	mutex_unlock(&ctx->lock);
+
+	ret = ivpu_mmu_invalidate_tlb(vdev, ctx->id);
+	if (ret)
+		ivpu_err(vdev, "Failed to invalidate TLB for ctx %u: %d\n", ctx->id, ret);
+	return ret;
+}
+
+void
+ivpu_mmu_context_unmap_sgt(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
+			   u64 vpu_addr, struct sg_table *sgt)
+{
+	struct scatterlist *sg;
+	int ret;
+	u64 i;
+
+	if (!IS_ALIGNED(vpu_addr, IVPU_MMU_PAGE_SIZE))
+		ivpu_warn(vdev, "Unaligned vpu_addr: 0x%llx\n", vpu_addr);
+
+	mutex_lock(&ctx->lock);
+
+	for_each_sgtable_dma_sg(sgt, sg, i) {
+		size_t size = sg_dma_len(sg) + sg->offset;
+
+		ivpu_mmu_context_unmap_pages(ctx, vpu_addr, size);
+		ivpu_mmu_context_flush_page_tables(ctx, vpu_addr, size);
+		vpu_addr += size;
+	}
+
+	mutex_unlock(&ctx->lock);
+
+	ret = ivpu_mmu_invalidate_tlb(vdev, ctx->id);
+	if (ret)
+		ivpu_warn(vdev, "Failed to invalidate TLB for ctx %u: %d\n", ctx->id, ret);
+}
+
+int
+ivpu_mmu_context_insert_node_locked(struct ivpu_mmu_context *ctx,
+				    const struct ivpu_addr_range *range,
+				    u64 size, struct drm_mm_node *node)
+{
+	lockdep_assert_held(&ctx->lock);
+
+	return drm_mm_insert_node_in_range(&ctx->mm, node, size, IVPU_MMU_PAGE_SIZE,
+					  0, range->start, range->end, DRM_MM_INSERT_BEST);
+}
+
+void
+ivpu_mmu_context_remove_node_locked(struct ivpu_mmu_context *ctx, struct drm_mm_node *node)
+{
+	lockdep_assert_held(&ctx->lock);
+
+	drm_mm_remove_node(node);
+}
+
+static int
+ivpu_mmu_context_init(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, u32 context_id)
+{
+	u64 start, end;
+	int ret;
+
+	mutex_init(&ctx->lock);
+	INIT_LIST_HEAD(&ctx->bo_list);
+
+	ret = ivpu_mmu_pgtable_init(vdev, &ctx->pgtable);
+	if (ret)
+		return ret;
+
+	if (!context_id) {
+		start = vdev->hw->ranges.global_low.start;
+		end = vdev->hw->ranges.global_high.end;
+	} else {
+		start = vdev->hw->ranges.user_low.start;
+		end = vdev->hw->ranges.user_high.end;
+	}
+
+	drm_mm_init(&ctx->mm, start, end - start);
+	ctx->id = context_id;
+
+	return 0;
+}
+
+static void ivpu_mmu_context_fini(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx)
+{
+	drm_WARN_ON(&vdev->drm, !ctx->pgtable.pgd);
+
+	mutex_destroy(&ctx->lock);
+	ivpu_mmu_pgtable_free(vdev, &ctx->pgtable);
+	drm_mm_takedown(&ctx->mm);
+}
+
+int ivpu_mmu_global_context_init(struct ivpu_device *vdev)
+{
+	return ivpu_mmu_context_init(vdev, &vdev->gctx, IVPU_GLOBAL_CONTEXT_MMU_SSID);
+}
+
+void ivpu_mmu_global_context_fini(struct ivpu_device *vdev)
+{
+	return ivpu_mmu_context_fini(vdev, &vdev->gctx);
+}
+
+void ivpu_mmu_user_context_mark_invalid(struct ivpu_device *vdev, u32 ssid)
+{
+	struct ivpu_file_priv *file_priv;
+
+	xa_lock(&vdev->context_xa);
+
+	file_priv = xa_load(&vdev->context_xa, ssid);
+	if (file_priv)
+		file_priv->has_mmu_faults = true;
+
+	xa_unlock(&vdev->context_xa);
+}
+
+int ivpu_mmu_user_context_init(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, u32 ctx_id)
+{
+	int ret;
+
+	drm_WARN_ON(&vdev->drm, !ctx_id);
+
+	ret = ivpu_mmu_context_init(vdev, ctx, ctx_id);
+	if (ret) {
+		ivpu_err(vdev, "Failed to initialize context: %d\n", ret);
+		return ret;
+	}
+
+	ret = ivpu_mmu_set_pgtable(vdev, ctx_id, &ctx->pgtable);
+	if (ret) {
+		ivpu_err(vdev, "Failed to set page table: %d\n", ret);
+		goto err_context_fini;
+	}
+
+	return 0;
+
+err_context_fini:
+	ivpu_mmu_context_fini(vdev, ctx);
+	return ret;
+}
+
+void ivpu_mmu_user_context_fini(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx)
+{
+	drm_WARN_ON(&vdev->drm, !ctx->id);
+
+	ivpu_mmu_clear_pgtable(vdev, ctx->id);
+	ivpu_mmu_context_fini(vdev, ctx);
+}