Re: [PATCH] Add 2-level GPUVM pagetables support to radeon driver.

From: Christian König
Date: Thu Sep 13 2012 - 11:28:49 EST


This is a multi-part message in MIME format.Hi Dmitry,

as I already noted in our internal discussion, the first step to hierarchical page table support should be to cleanup the set_page interface. Please see the attached patch, it does exactly this. I suggest that you rebase on it and try to don't touch the chipset specific code in ni.c any more (if you find that you need to change something there move that into this cleanup patch instead).

Additionally to that see the notes below:

On 13.09.2012 16:13, Dmitry Cherkasov wrote:
PDE/PTE update code uses CP ring for memory writes.
All page table entries are preallocated for now in alloc_pt().

It is made as whole because it's hard to divide it to several patches
that compile and doesn't break anything being applied separately.

Tested on cayman card.

Signed-off-by: Dmitry Cherkasov <Dmitrii.Cherkasov@xxxxxxx>
---
I couldn't test in on SI card, so would be happy if someone could check it there.

drivers/gpu/drm/radeon/ni.c | 64 ++++++++++++++++++-------
drivers/gpu/drm/radeon/radeon.h | 79 +++++++++++++++++++++++++++---
drivers/gpu/drm/radeon/radeon_asic.c | 1 +
drivers/gpu/drm/radeon/radeon_asic.h | 13 +++--
drivers/gpu/drm/radeon/radeon_gart.c | 88 ++++++++++++++++++++++++++++++----
drivers/gpu/drm/radeon/si.c | 4 +-
6 files changed, 210 insertions(+), 39 deletions(-)

diff --git a/drivers/gpu/drm/radeon/ni.c b/drivers/gpu/drm/radeon/ni.c
index b238216..31d9027 100644
--- a/drivers/gpu/drm/radeon/ni.c
+++ b/drivers/gpu/drm/radeon/ni.c
@@ -782,7 +782,7 @@ static int cayman_pcie_gart_enable(struct radeon_device *rdev)
(u32)(rdev->dummy_page.addr >> 12));
WREG32(VM_CONTEXT1_CNTL2, 0);
WREG32(VM_CONTEXT1_CNTL, 0);
- WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
+ WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
cayman_pcie_gart_tlb_flush(rdev);
@@ -1517,37 +1517,65 @@ uint32_t cayman_vm_page_flags(struct radeon_device *rdev, uint32_t flags)
return r600_flags;
}
-/**
- * cayman_vm_set_page - update the page tables using the CP
- *
- * @rdev: radeon_device pointer
- *
- * Update the page tables using the CP (cayman-si).
- */
It's probably just a rebase artifact, please never remove those documentation lines. Instead fix them if the need arise.

-void cayman_vm_set_page(struct radeon_device *rdev, struct radeon_vm *vm,
- unsigned pfn, struct ttm_mem_reg *mem,
- unsigned npages, uint32_t flags)
+void cayman_vm_set_page(struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ uint64_t pte_gpu_addr,
+ uint64_t mem_pfn_offset,
+ unsigned count, uint32_t flags)
{
struct radeon_ring *ring = &rdev->ring[rdev->asic->vm.pt_ring_index];
- uint64_t addr, pt = vm->pt_gpu_addr + pfn * 8;
+ uint64_t addr;
int i;
addr = flags = cayman_vm_page_flags(rdev, flags);
- radeon_ring_write(ring, PACKET3(PACKET3_ME_WRITE, 1 + npages * 2));
- radeon_ring_write(ring, pt & 0xffffffff);
- radeon_ring_write(ring, (pt >> 32) & 0xff);
- for (i = 0; i < npages; ++i) {
+ radeon_ring_write(ring, PACKET3(PACKET3_ME_WRITE, 1 + count * 2));
+ radeon_ring_write(ring, pte_gpu_addr & 0xffffffff);
+ radeon_ring_write(ring, (pte_gpu_addr >> 32) & 0xff);
+ for (i = 0; i < count; ++i) {
if (mem) {
- addr = radeon_vm_get_addr(rdev, mem, i);
+ addr = radeon_vm_get_addr(rdev, mem,
+ i + mem_pfn_offset);
addr = addr & 0xFFFFFFFFFFFFF000ULL;
addr |= flags;
}
+
+ radeon_ring_write(ring, addr & 0xffffffff);
+ radeon_ring_write(ring, (addr >> 32) & 0xffffffff);
+ }
+}
+
+void cayman_vm_set_pdes(struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ struct radeon_vm *vm,
+ uint64_t pde_index,
+ unsigned pde_count, uint32_t flags)
+{
+ struct radeon_ring *ring = &rdev->ring[rdev->asic->vm.pt_ring_index];
+ uint64_t addr;
+ int i;
+ uint64_t pde_gpu_addr = RADEON_BASE_GPU_ADDR(vm) +
+ PDE_OFFSET(rdev, pde_index);
+
+ addr = 0;
+
+ radeon_ring_write(ring, PACKET3(PACKET3_ME_WRITE, 1 + pde_count * 2));
+ radeon_ring_write(ring, pde_gpu_addr & 0xffffffff);
+ radeon_ring_write(ring, (pde_gpu_addr >> 32) & 0xff);
+
+ for (i = 0; i < pde_count; ++i) {
+ addr = RADEON_BASE_GPU_ADDR(vm) +
+ PTE_OFFSET(rdev, i + pde_index, 0);
+ addr = addr & 0xFFFFFFFFFFFFF000ULL;
+ addr |= flags;
+
radeon_ring_write(ring, addr & 0xffffffff);
radeon_ring_write(ring, (addr >> 32) & 0xffffffff);
}
}
+
+
Splitting that into two functions should be made unnecessary by the interface cleanup patch.

/**
* cayman_vm_flush - vm flush using the CP
*
@@ -1571,7 +1599,7 @@ void cayman_vm_flush(struct radeon_device *rdev, struct radeon_ib *ib)
radeon_ring_write(ring, vm->last_pfn);
radeon_ring_write(ring, PACKET0(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2), 0));
- radeon_ring_write(ring, vm->pt_gpu_addr >> 12);
+ radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
/* flush hdp cache */
radeon_ring_write(ring, PACKET0(HDP_MEM_COHERENCY_FLUSH_CNTL, 0));
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 4d67f0f..062896f 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -655,8 +655,8 @@ struct radeon_vm {
struct list_head va;
unsigned id;
unsigned last_pfn;
- u64 pt_gpu_addr;
- u64 *pt;
+ u64 pd_gpu_addr;
+ u64 __iomem *pd_addr;
We probably never need the CPU address of the page directory, except for initial clearing it.

struct radeon_sa_bo *sa_bo;
struct mutex mutex;
/* last fence for cs using this vm */
@@ -679,6 +679,57 @@ struct radeon_vm_manager {
bool enabled;
};
+
+
+/* We consider the case where BLOCK_SIZE is 0 */
+/* So PDE is 19 bits long, PTE is 9 and OFFSET is 12 */
+#define RADEON_BLOCK_SIZE 0
+
+/* By default there are 512 entries in Page Table */
+#define RADEON_DEFAULT_PTE_COUNT (1 << 9)
+
+/* number of PTEs in Page Table */
+#define RADEON_PTE_COUNT (RADEON_DEFAULT_PTE_COUNT << RADEON_BLOCK_SIZE)
+
+/* Get last PDE number containing nth PTE */
+#define RADEON_GET_LAST_PDE_FOR_PFN(_n) ((_n) / RADEON_PTE_COUNT)
+
+/* Get PTE number to containing nth pfn */
+#define RADEON_GET_PTE_FOR_PFN(_n) ((_n) % RADEON_PTE_COUNT)
+
+/* Number of PDE tables to cover n PTEs */
+#define RADEON_PDE_COUNT_FOR_N_PAGES(_n) \
+ (((_n) + RADEON_PTE_COUNT - 1) / RADEON_PTE_COUNT)
+
+/* Number of PDE tables to cover max_pfn (maximum number of PTEs) */
+#define RADEON_TOTAL_PDE_COUNT(rdev) \
+ RADEON_PDE_COUNT_FOR_N_PAGES(rdev->vm_manager.max_pfn)
+
+#define PTE_SIZE 8
+#define PDE_SIZE 8
+
+/* offset for nth PDE starting from beginning of PDE table */
+#define PDE_OFFSET(_rdev, _npde) ((_npde) * PDE_SIZE)
+
+#define PT_OFFSET(_rdev) \
+ (RADEON_GPU_PAGE_ALIGN(RADEON_TOTAL_PDE_COUNT(rdev) * PDE_SIZE))
+
+/* offset for nth PTE of mth PDE starting from beginning of PDE table */
+#define PTE_OFFSET(_rdev, _npde, _npte) \
+ (PT_OFFSET(_rdev) + \
+ (_npde) * RADEON_PTE_COUNT * PTE_SIZE + \
+ (_npte) * PTE_SIZE)
All defines should have RADEON_ prefix, also I'm not 100% sure why we need that as a global define, only the vm code should deal with that.

+
+/* cpu address of gpuvm page table */
+#define RADEON_BASE_CPU_ADDR(_vm) \
+ radeon_sa_bo_cpu_addr(vm->sa_bo)
+
+/* gpu address of gpuvm page table */
+#define RADEON_BASE_GPU_ADDR(_vm) \
+ radeon_sa_bo_gpu_addr(vm->sa_bo)
+
+#define RADEON_PDE_FLAG_VALID 1
+
/*
* file private structure
*/
@@ -1141,9 +1192,20 @@ struct radeon_asic {
void (*fini)(struct radeon_device *rdev);
u32 pt_ring_index;
- void (*set_page)(struct radeon_device *rdev, struct radeon_vm *vm,
- unsigned pfn, struct ttm_mem_reg *mem,
- unsigned npages, uint32_t flags);
+ void (*set_page)(
+ struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ uint64_t pte_gpu_addr, uint64_t mem_pfn_offset,
+ unsigned count, uint32_t flags
+);
+
+ void (*set_pdes)(
+ struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ struct radeon_vm *vm,
+ uint64_t pde_index,
+ unsigned pde_count, uint32_t flags);
+
} vm;
/* ring specific callbacks */
struct {
@@ -1755,7 +1817,12 @@ void radeon_ring_write(struct radeon_ring *ring, uint32_t v);
#define radeon_gart_set_page(rdev, i, p) (rdev)->asic->gart.set_page((rdev), (i), (p))
#define radeon_asic_vm_init(rdev) (rdev)->asic->vm.init((rdev))
#define radeon_asic_vm_fini(rdev) (rdev)->asic->vm.fini((rdev))
-#define radeon_asic_vm_set_page(rdev, v, pfn, mem, npages, flags) (rdev)->asic->vm.set_page((rdev), (v), (pfn), (mem), (npages), (flags))
+#define radeon_asic_vm_set_page(rdev, mem, start_pte, start_pfn, count, flags) \
+((rdev)->asic->vm.set_page((rdev), (mem), (start_pte), (start_pfn), \
+ (count), (flags)))
+#define radeon_asic_vm_set_pdes(rdev, mem, vm, pde_index, pde_count, flags) \
+((rdev)->asic->vm.set_pdes((rdev), (mem), (vm), (pde_index), \
+ (pde_count), (flags)))
#define radeon_ring_start(rdev, r, cp) (rdev)->asic->ring[(r)].ring_start((rdev), (cp))
#define radeon_ring_test(rdev, r, cp) (rdev)->asic->ring[(r)].ring_test((rdev), (cp))
#define radeon_ib_test(rdev, r, cp) (rdev)->asic->ring[(r)].ib_test((rdev), (cp))
diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index 2f7adea..0df6a55 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -1377,6 +1377,7 @@ static struct radeon_asic cayman_asic = {
.fini = &cayman_vm_fini,
.pt_ring_index = RADEON_RING_TYPE_GFX_INDEX,
.set_page = &cayman_vm_set_page,
+ .set_pdes = &cayman_vm_set_pdes,
},
.ring = {
[RADEON_RING_TYPE_GFX_INDEX] = {
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index 29b3d05..7a694c0 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -442,9 +442,16 @@ int cayman_vm_init(struct radeon_device *rdev);
void cayman_vm_fini(struct radeon_device *rdev);
void cayman_vm_flush(struct radeon_device *rdev, struct radeon_ib *ib);
uint32_t cayman_vm_page_flags(struct radeon_device *rdev, uint32_t flags);
-void cayman_vm_set_page(struct radeon_device *rdev, struct radeon_vm *vm,
- unsigned pfn, struct ttm_mem_reg *mem,
- unsigned npages, uint32_t flags);
+void cayman_vm_set_page(struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ uint64_t pte_gpu_addr, uint64_t mem_pfn_offset,
+ unsigned count, uint32_t flags);
+void cayman_vm_set_pdes(struct radeon_device *rdev,
+ struct ttm_mem_reg *mem,
+ struct radeon_vm *vm,
+ uint64_t pde_index,
+ unsigned pde_count, uint32_t flags);
+
int evergreen_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib);
/* DCE6 - SI */
diff --git a/drivers/gpu/drm/radeon/radeon_gart.c b/drivers/gpu/drm/radeon/radeon_gart.c
index 2f28ff3..f9bda6e 100644
--- a/drivers/gpu/drm/radeon/radeon_gart.c
+++ b/drivers/gpu/drm/radeon/radeon_gart.c
@@ -490,7 +490,6 @@ static void radeon_vm_free_pt(struct radeon_device *rdev,
list_del_init(&vm->list);
radeon_sa_bo_free(rdev, &vm->sa_bo, vm->fence);
- vm->pt = NULL;
list_for_each_entry(bo_va, &vm->va, vm_list) {
bo_va->valid = false;
@@ -547,6 +546,10 @@ int radeon_vm_alloc_pt(struct radeon_device *rdev, struct radeon_vm *vm)
struct radeon_vm *vm_evict;
int r;
+ int gpuvm_tables_sz =
+ RADEON_GPU_PAGE_ALIGN(RADEON_TOTAL_PDE_COUNT(rdev) * PDE_SIZE) +
+ vm->last_pfn * PTE_SIZE;
+
if (vm == NULL) {
return -EINVAL;
}
@@ -560,7 +563,7 @@ int radeon_vm_alloc_pt(struct radeon_device *rdev, struct radeon_vm *vm)
retry:
r = radeon_sa_bo_new(rdev, &rdev->vm_manager.sa_manager, &vm->sa_bo,
- RADEON_GPU_PAGE_ALIGN(vm->last_pfn * 8),
+ gpuvm_tables_sz,
RADEON_GPU_PAGE_SIZE, false);
if (r == -ENOMEM) {
if (list_empty(&rdev->vm_manager.lru_vm)) {
@@ -576,9 +579,13 @@ retry:
return r;
}
- vm->pt = radeon_sa_bo_cpu_addr(vm->sa_bo);
- vm->pt_gpu_addr = radeon_sa_bo_gpu_addr(vm->sa_bo);
- memset(vm->pt, 0, RADEON_GPU_PAGE_ALIGN(vm->last_pfn * 8));
+ DRM_INFO("radeon: reserved %d kb pd & pt tables\n",
+ gpuvm_tables_sz/1024);
+
+ vm->pd_addr = radeon_sa_bo_cpu_addr(vm->sa_bo);
+ vm->pd_gpu_addr = radeon_sa_bo_gpu_addr(vm->sa_bo);
+ memset(vm->pd_addr, 0, gpuvm_tables_sz);
+
list_add_tail(&vm->list, &rdev->vm_manager.lru_vm);
return radeon_vm_bo_update_pte(rdev, vm, rdev->ring_tmp_bo.bo,
@@ -861,6 +868,63 @@ u64 radeon_vm_get_addr(struct radeon_device *rdev,
}
/**
+ * radeon_vm_bo_set_pages - update PDE and PTE for pfn
+ *
+ * @rdev: radeon_device pointer
+ * @first_pfn: first pfn in bo address space to start mapping from
+ * @mem: ttm mem
+ * @vm: requested vm
+ * @bo: radeon buffer object
+ *
+ * Update page directory and table entries for pfn (cayman+).
+ */
+
+int radeon_vm_bo_set_pages(struct radeon_device *rdev, struct radeon_vm *vm,
+ unsigned first_pfn, struct ttm_mem_reg *mem,
+ unsigned npages, uint32_t flags)
Shouldn't that be a static function?

+{
+ u64 pde_num, pte_num, start_pde, pde_count = 0;
+
+ unsigned pt_total, count, pfn_idx;
+ unsigned last_pfn = first_pfn + npages, pfns_to_pt_edge, pfns_to_end;
+ unsigned mem_pfn_offset;
+
+ pt_total = RADEON_PDE_COUNT_FOR_N_PAGES(npages);
+ count = RADEON_PTE_COUNT;
+
+ pfn_idx = first_pfn;
+
+ for (mem_pfn_offset = 0; mem_pfn_offset < npages;) {
+
+ pfns_to_end = last_pfn - pfn_idx;
+ pfns_to_pt_edge = RADEON_PTE_COUNT -
+ (pfn_idx % RADEON_PTE_COUNT);
+
+ count = pfns_to_pt_edge < pfns_to_end ?
+ pfns_to_pt_edge : pfns_to_end;
+
+ pde_num = RADEON_GET_LAST_PDE_FOR_PFN(pfn_idx);
+ pte_num = RADEON_GET_PTE_FOR_PFN(pfn_idx);
+
+ radeon_asic_vm_set_page(rdev, mem,
+ vm->pd_gpu_addr +
+ PTE_OFFSET(rdev, pde_num, pte_num),
+ mem_pfn_offset, count, flags);
+ pfn_idx += count;
+ mem_pfn_offset += count;
+
+ pde_count++;
+ }
+
+ start_pde = RADEON_GET_LAST_PDE_FOR_PFN(first_pfn);
+
+ radeon_asic_vm_set_pdes(rdev, mem, vm,
+ start_pde, pde_count, RADEON_PDE_FLAG_VALID);
+
+ return 0;
+}
+
+/**
* radeon_vm_bo_update_pte - map a bo into the vm page table
*
* @rdev: radeon_device pointer
@@ -882,7 +946,7 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
struct radeon_ring *ring = &rdev->ring[ridx];
struct radeon_semaphore *sem = NULL;
struct radeon_bo_va *bo_va;
- unsigned ngpu_pages, ndw;
+ unsigned ngpu_pages, ndw, npdes;
uint64_t pfn;
int r;
@@ -935,10 +999,12 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
}
}
- /* estimate number of dw needed */
+ npdes = (RADEON_PDE_COUNT_FOR_N_PAGES(pfn + ngpu_pages) -
+ RADEON_GET_LAST_PDE_FOR_PFN(pfn));
+
ndw = 32;
- ndw += (ngpu_pages >> 12) * 3;
- ndw += ngpu_pages * 2;
+ ndw += ngpu_pages * 2 + 3 * npdes;
+ ndw += npdes * 2 + 3;
r = radeon_ring_lock(rdev, ring, ndw);
if (r) {
@@ -950,7 +1016,7 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
radeon_fence_note_sync(vm->fence, ridx);
}
- radeon_asic_vm_set_page(rdev, vm, pfn, mem, ngpu_pages, bo_va->flags);
+ radeon_vm_bo_set_pages(rdev, vm, pfn, mem, ngpu_pages, bo_va->flags);
radeon_fence_unref(&vm->fence);
r = radeon_fence_emit(rdev, &vm->fence, ridx);
@@ -958,9 +1024,11 @@ int radeon_vm_bo_update_pte(struct radeon_device *rdev,
radeon_ring_unlock_undo(rdev, ring);
return r;
}
+
radeon_ring_unlock_commit(rdev, ring);
radeon_semaphore_free(rdev, &sem, vm->fence);
radeon_fence_unref(&vm->last_flush);
+
return 0;
}
diff --git a/drivers/gpu/drm/radeon/si.c b/drivers/gpu/drm/radeon/si.c
index 2a5c337..156c994 100644
--- a/drivers/gpu/drm/radeon/si.c
+++ b/drivers/gpu/drm/radeon/si.c
@@ -2426,7 +2426,7 @@ static int si_pcie_gart_enable(struct radeon_device *rdev)
WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
(u32)(rdev->dummy_page.addr >> 12));
WREG32(VM_CONTEXT1_CNTL2, 0);
- WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
+ WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
si_pcie_gart_tlb_flush(rdev);
@@ -2804,7 +2804,7 @@ void si_vm_flush(struct radeon_device *rdev, struct radeon_ib *ib)
radeon_ring_write(ring, PACKET0(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR
+ ((vm->id - 8) << 2), 0));
}
- radeon_ring_write(ring, vm->pt_gpu_addr >> 12);
+ radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
/* flush hdp cache */
radeon_ring_write(ring, PACKET0(HDP_MEM_COHERENCY_FLUSH_CNTL, 0));

Cheers,
Christian.