* Alex Deucher
* Jerome Glisse
*/
+#include <linux/fence-array.h>
#include <drm/drmP.h>
#include <drm/amdgpu_drm.h>
#include "amdgpu.h"
* SI supports 16.
*/
-/* Special value that no flush is necessary */
-#define AMDGPU_VM_NO_FLUSH (~0ll)
-
/* Local structure. Encapsulate some VM table update parameters to reduce
* the number of function parameters
*/
-struct amdgpu_vm_update_params {
+struct amdgpu_pte_update_params {
+ /* amdgpu device we do this update for */
+ struct amdgpu_device *adev;
/* address where to copy page table entries from */
uint64_t src;
- /* DMA addresses to use for mapping */
- dma_addr_t *pages_addr;
/* indirect buffer to fill with commands */
struct amdgpu_ib *ib;
+ /* Function which actually does the update */
+ void (*func)(struct amdgpu_pte_update_params *params, uint64_t pe,
+ uint64_t addr, unsigned count, uint32_t incr,
+ uint32_t flags);
+ /* indicate update pt or its shadow */
+ bool shadow;
};
/**
/**
* amdgpu_vm_get_bos - add the vm BOs to a duplicates list
*
+ * @adev: amdgpu device pointer
* @vm: vm providing the BOs
* @duplicates: head of duplicates list
*
* Add the page directory to the BO duplicates list
* for command submission.
*/
-void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates)
+void amdgpu_vm_get_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ struct list_head *duplicates)
{
+ uint64_t num_evictions;
unsigned i;
+ /* We only need to validate the page tables
+ * if they aren't already valid.
+ */
+ num_evictions = atomic64_read(&adev->num_evictions);
+ if (num_evictions == vm->last_eviction_counter)
+ return;
+
/* add the vm page table to the list */
for (i = 0; i <= vm->max_pde_used; ++i) {
struct amdgpu_bo_list_entry *entry = &vm->page_tables[i].entry;
spin_unlock(&glob->lru_lock);
}
+static bool amdgpu_vm_is_gpu_reset(struct amdgpu_device *adev,
+ struct amdgpu_vm_id *id)
+{
+ return id->current_gpu_reset_count !=
+ atomic_read(&adev->gpu_reset_counter) ? true : false;
+}
+
/**
* amdgpu_vm_grab_id - allocate the next free VMID
*
*/
int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
struct amdgpu_sync *sync, struct fence *fence,
- unsigned *vm_id, uint64_t *vm_pd_addr)
+ struct amdgpu_job *job)
{
- uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory);
struct amdgpu_device *adev = ring->adev;
+ uint64_t fence_context = adev->fence_context + ring->idx;
struct fence *updates = sync->last_vm_update;
- struct amdgpu_vm_id *id;
- unsigned i = ring->idx;
- int r;
+ struct amdgpu_vm_id *id, *idle;
+ struct fence **fences;
+ unsigned i;
+ int r = 0;
+
+ fences = kmalloc_array(sizeof(void *), adev->vm_manager.num_ids,
+ GFP_KERNEL);
+ if (!fences)
+ return -ENOMEM;
mutex_lock(&adev->vm_manager.lock);
+ /* Check if we have an idle VMID */
+ i = 0;
+ list_for_each_entry(idle, &adev->vm_manager.ids_lru, list) {
+ fences[i] = amdgpu_sync_peek_fence(&idle->active, ring);
+ if (!fences[i])
+ break;
+ ++i;
+ }
+
+ /* If we can't find a idle VMID to use, wait till one becomes available */
+ if (&idle->list == &adev->vm_manager.ids_lru) {
+ u64 fence_context = adev->vm_manager.fence_context + ring->idx;
+ unsigned seqno = ++adev->vm_manager.seqno[ring->idx];
+ struct fence_array *array;
+ unsigned j;
+
+ for (j = 0; j < i; ++j)
+ fence_get(fences[j]);
+
+ array = fence_array_create(i, fences, fence_context,
+ seqno, true);
+ if (!array) {
+ for (j = 0; j < i; ++j)
+ fence_put(fences[j]);
+ kfree(fences);
+ r = -ENOMEM;
+ goto error;
+ }
+
+
+ r = amdgpu_sync_fence(ring->adev, sync, &array->base);
+ fence_put(&array->base);
+ if (r)
+ goto error;
+
+ mutex_unlock(&adev->vm_manager.lock);
+ return 0;
+
+ }
+ kfree(fences);
+
+ job->vm_needs_flush = true;
/* Check if we can use a VMID already assigned to this VM */
+ i = ring->idx;
do {
struct fence *flushed;
/* Check all the prerequisites to using this VMID */
if (!id)
continue;
+ if (amdgpu_vm_is_gpu_reset(adev, id))
+ continue;
if (atomic64_read(&id->owner) != vm->client_id)
continue;
- if (pd_addr != id->pd_gpu_addr)
+ if (job->vm_pd_addr != id->pd_gpu_addr)
continue;
- if (id->last_user != ring &&
- (!id->last_flush || !fence_is_signaled(id->last_flush)))
+ if (!id->last_flush)
continue;
- flushed = id->flushed_updates;
- if (updates && (!flushed || fence_is_later(updates, flushed)))
+ if (id->last_flush->context != fence_context &&
+ !fence_is_signaled(id->last_flush))
continue;
- /* Good we can use this VMID */
- if (id->last_user == ring) {
- r = amdgpu_sync_fence(ring->adev, sync,
- id->first);
- if (r)
- goto error;
- }
+ flushed = id->flushed_updates;
+ if (updates &&
+ (!flushed || fence_is_later(updates, flushed)))
+ continue;
- /* And remember this submission as user of the VMID */
+ /* Good we can use this VMID. Remember this submission as
+ * user of the VMID.
+ */
r = amdgpu_sync_fence(ring->adev, &id->active, fence);
if (r)
goto error;
+ id->current_gpu_reset_count = atomic_read(&adev->gpu_reset_counter);
list_move_tail(&id->list, &adev->vm_manager.ids_lru);
vm->ids[ring->idx] = id;
- *vm_id = id - adev->vm_manager.ids;
- *vm_pd_addr = AMDGPU_VM_NO_FLUSH;
- trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, *vm_pd_addr);
+ job->vm_id = id - adev->vm_manager.ids;
+ job->vm_needs_flush = false;
+ trace_amdgpu_vm_grab_id(vm, ring->idx, job);
mutex_unlock(&adev->vm_manager.lock);
return 0;
} while (i != ring->idx);
- id = list_first_entry(&adev->vm_manager.ids_lru,
- struct amdgpu_vm_id,
- list);
-
- if (!amdgpu_sync_is_idle(&id->active)) {
- struct list_head *head = &adev->vm_manager.ids_lru;
- struct amdgpu_vm_id *tmp;
+ /* Still no ID to use? Then use the idle one found earlier */
+ id = idle;
- list_for_each_entry_safe(id, tmp, &adev->vm_manager.ids_lru,
- list) {
- if (amdgpu_sync_is_idle(&id->active)) {
- list_move(&id->list, head);
- head = &id->list;
- }
- }
- id = list_first_entry(&adev->vm_manager.ids_lru,
- struct amdgpu_vm_id,
- list);
- }
-
- r = amdgpu_sync_cycle_fences(sync, &id->active, fence);
+ /* Remember this submission as user of the VMID */
+ r = amdgpu_sync_fence(ring->adev, &id->active, fence);
if (r)
goto error;
fence_put(id->flushed_updates);
id->flushed_updates = fence_get(updates);
- id->pd_gpu_addr = pd_addr;
-
+ id->pd_gpu_addr = job->vm_pd_addr;
+ id->current_gpu_reset_count = atomic_read(&adev->gpu_reset_counter);
list_move_tail(&id->list, &adev->vm_manager.ids_lru);
- id->last_user = ring;
atomic64_set(&id->owner, vm->client_id);
vm->ids[ring->idx] = id;
- *vm_id = id - adev->vm_manager.ids;
- *vm_pd_addr = pd_addr;
- trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, *vm_pd_addr);
+ job->vm_id = id - adev->vm_manager.ids;
+ trace_amdgpu_vm_grab_id(vm, ring->idx, job);
error:
mutex_unlock(&adev->vm_manager.lock);
return r;
}
+static bool amdgpu_vm_ring_has_compute_vm_bug(struct amdgpu_ring *ring)
+{
+ struct amdgpu_device *adev = ring->adev;
+ const struct amdgpu_ip_block_version *ip_block;
+
+ if (ring->type != AMDGPU_RING_TYPE_COMPUTE)
+ /* only compute rings */
+ return false;
+
+ ip_block = amdgpu_get_ip_block(adev, AMD_IP_BLOCK_TYPE_GFX);
+ if (!ip_block)
+ return false;
+
+ if (ip_block->major <= 7) {
+ /* gfx7 has no workaround */
+ return true;
+ } else if (ip_block->major == 8) {
+ if (adev->gfx.mec_fw_version >= 673)
+ /* gfx8 is fixed in MEC firmware 673 */
+ return false;
+ else
+ return true;
+ }
+ return false;
+}
+
/**
* amdgpu_vm_flush - hardware flush the vm
*
*
* Emit a VM flush when it is necessary.
*/
-int amdgpu_vm_flush(struct amdgpu_ring *ring,
- unsigned vm_id, uint64_t pd_addr,
- uint32_t gds_base, uint32_t gds_size,
- uint32_t gws_base, uint32_t gws_size,
- uint32_t oa_base, uint32_t oa_size)
+int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job)
{
struct amdgpu_device *adev = ring->adev;
- struct amdgpu_vm_id *id = &adev->vm_manager.ids[vm_id];
+ struct amdgpu_vm_id *id = &adev->vm_manager.ids[job->vm_id];
bool gds_switch_needed = ring->funcs->emit_gds_switch && (
- id->gds_base != gds_base ||
- id->gds_size != gds_size ||
- id->gws_base != gws_base ||
- id->gws_size != gws_size ||
- id->oa_base != oa_base ||
- id->oa_size != oa_size);
+ id->gds_base != job->gds_base ||
+ id->gds_size != job->gds_size ||
+ id->gws_base != job->gws_base ||
+ id->gws_size != job->gws_size ||
+ id->oa_base != job->oa_base ||
+ id->oa_size != job->oa_size);
int r;
if (ring->funcs->emit_pipeline_sync && (
- pd_addr != AMDGPU_VM_NO_FLUSH || gds_switch_needed ||
- ring->type == AMDGPU_RING_TYPE_COMPUTE))
+ job->vm_needs_flush || gds_switch_needed ||
+ amdgpu_vm_ring_has_compute_vm_bug(ring)))
amdgpu_ring_emit_pipeline_sync(ring);
- if (ring->funcs->emit_vm_flush &&
- pd_addr != AMDGPU_VM_NO_FLUSH) {
+ if (ring->funcs->emit_vm_flush && (job->vm_needs_flush ||
+ amdgpu_vm_is_gpu_reset(adev, id))) {
struct fence *fence;
- trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id);
- amdgpu_ring_emit_vm_flush(ring, vm_id, pd_addr);
+ trace_amdgpu_vm_flush(job->vm_pd_addr, ring->idx, job->vm_id);
+ amdgpu_ring_emit_vm_flush(ring, job->vm_id, job->vm_pd_addr);
+
+ r = amdgpu_fence_emit(ring, &fence);
+ if (r)
+ return r;
mutex_lock(&adev->vm_manager.lock);
- if ((id->pd_gpu_addr == pd_addr) && (id->last_user == ring)) {
- r = amdgpu_fence_emit(ring, &fence);
- if (r) {
- mutex_unlock(&adev->vm_manager.lock);
- return r;
- }
- fence_put(id->last_flush);
- id->last_flush = fence;
- }
+ fence_put(id->last_flush);
+ id->last_flush = fence;
mutex_unlock(&adev->vm_manager.lock);
}
if (gds_switch_needed) {
- id->gds_base = gds_base;
- id->gds_size = gds_size;
- id->gws_base = gws_base;
- id->gws_size = gws_size;
- id->oa_base = oa_base;
- id->oa_size = oa_size;
- amdgpu_ring_emit_gds_switch(ring, vm_id,
- gds_base, gds_size,
- gws_base, gws_size,
- oa_base, oa_size);
+ id->gds_base = job->gds_base;
+ id->gds_size = job->gds_size;
+ id->gws_base = job->gws_base;
+ id->gws_size = job->gws_size;
+ id->oa_base = job->oa_base;
+ id->oa_size = job->oa_size;
+ amdgpu_ring_emit_gds_switch(ring, job->vm_id,
+ job->gds_base, job->gds_size,
+ job->gws_base, job->gws_size,
+ job->oa_base, job->oa_size);
}
return 0;
}
/**
- * amdgpu_vm_update_pages - helper to call the right asic function
+ * amdgpu_vm_do_set_ptes - helper to call the right asic function
*
- * @adev: amdgpu_device pointer
- * @vm_update_params: see amdgpu_vm_update_params definition
+ * @params: see amdgpu_pte_update_params definition
* @pe: addr of the page entry
* @addr: dst addr to write into pe
* @count: number of page entries to update
* Traces the parameters and calls the right asic functions
* to setup the page table using the DMA.
*/
-static void amdgpu_vm_update_pages(struct amdgpu_device *adev,
- struct amdgpu_vm_update_params
- *vm_update_params,
- uint64_t pe, uint64_t addr,
- unsigned count, uint32_t incr,
- uint32_t flags)
+static void amdgpu_vm_do_set_ptes(struct amdgpu_pte_update_params *params,
+ uint64_t pe, uint64_t addr,
+ unsigned count, uint32_t incr,
+ uint32_t flags)
{
trace_amdgpu_vm_set_page(pe, addr, count, incr, flags);
- if (vm_update_params->src) {
- amdgpu_vm_copy_pte(adev, vm_update_params->ib,
- pe, (vm_update_params->src + (addr >> 12) * 8), count);
-
- } else if (vm_update_params->pages_addr) {
- amdgpu_vm_write_pte(adev, vm_update_params->ib,
- vm_update_params->pages_addr,
- pe, addr, count, incr, flags);
-
- } else if (count < 3) {
- amdgpu_vm_write_pte(adev, vm_update_params->ib, NULL, pe, addr,
- count, incr, flags);
+ if (count < 3) {
+ amdgpu_vm_write_pte(params->adev, params->ib, pe,
+ addr | flags, count, incr);
} else {
- amdgpu_vm_set_pte_pde(adev, vm_update_params->ib, pe, addr,
+ amdgpu_vm_set_pte_pde(params->adev, params->ib, pe, addr,
count, incr, flags);
}
}
+/**
+ * amdgpu_vm_do_copy_ptes - copy the PTEs from the GART
+ *
+ * @params: see amdgpu_pte_update_params definition
+ * @pe: addr of the page entry
+ * @addr: dst addr to write into pe
+ * @count: number of page entries to update
+ * @incr: increase next addr by incr bytes
+ * @flags: hw access flags
+ *
+ * Traces the parameters and calls the DMA function to copy the PTEs.
+ */
+static void amdgpu_vm_do_copy_ptes(struct amdgpu_pte_update_params *params,
+ uint64_t pe, uint64_t addr,
+ unsigned count, uint32_t incr,
+ uint32_t flags)
+{
+ trace_amdgpu_vm_set_page(pe, addr, count, incr, flags);
+
+ amdgpu_vm_copy_pte(params->adev, params->ib, pe,
+ (params->src + (addr >> 12) * 8), count);
+}
+
/**
* amdgpu_vm_clear_bo - initially clear the page dir/table
*
struct amdgpu_ring *ring;
struct fence *fence = NULL;
struct amdgpu_job *job;
- struct amdgpu_vm_update_params vm_update_params;
+ struct amdgpu_pte_update_params params;
unsigned entries;
uint64_t addr;
int r;
- memset(&vm_update_params, 0, sizeof(vm_update_params));
ring = container_of(vm->entity.sched, struct amdgpu_ring, sched);
r = reservation_object_reserve_shared(bo->tbo.resv);
if (r)
goto error;
- vm_update_params.ib = &job->ibs[0];
- amdgpu_vm_update_pages(adev, &vm_update_params, addr, 0, entries,
- 0, 0);
+ memset(¶ms, 0, sizeof(params));
+ params.adev = adev;
+ params.ib = &job->ibs[0];
+ amdgpu_vm_do_set_ptes(¶ms, addr, 0, entries, 0, 0);
amdgpu_ring_pad_ib(ring, &job->ibs[0]);
WARN_ON(job->ibs[0].length_dw > 64);
* Look up the physical address of the page that the pte resolves
* to and return the pointer for the page table entry.
*/
-uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr)
+static uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr)
{
uint64_t result;
- if (pages_addr) {
- /* page table offset */
- result = pages_addr[addr >> PAGE_SHIFT];
+ /* page table offset */
+ result = pages_addr[addr >> PAGE_SHIFT];
- /* in case cpu page size != gpu page size*/
- result |= addr & (~PAGE_MASK);
-
- } else {
- /* No mapping required */
- result = addr;
- }
+ /* in case cpu page size != gpu page size*/
+ result |= addr & (~PAGE_MASK);
result &= 0xFFFFFFFFFFFFF000ULL;
return result;
}
-/**
- * amdgpu_vm_update_pdes - make sure that page directory is valid
- *
- * @adev: amdgpu_device pointer
- * @vm: requested vm
- * @start: start of GPU address range
- * @end: end of GPU address range
- *
- * Allocates new page tables if necessary
- * and updates the page directory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
- struct amdgpu_vm *vm)
+static int amdgpu_vm_update_pd_or_shadow(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ bool shadow)
{
struct amdgpu_ring *ring;
- struct amdgpu_bo *pd = vm->page_directory;
- uint64_t pd_addr = amdgpu_bo_gpu_offset(pd);
+ struct amdgpu_bo *pd = shadow ? vm->page_directory->shadow :
+ vm->page_directory;
+ uint64_t pd_addr;
uint32_t incr = AMDGPU_VM_PTE_COUNT * 8;
uint64_t last_pde = ~0, last_pt = ~0;
unsigned count = 0, pt_idx, ndw;
struct amdgpu_job *job;
- struct amdgpu_vm_update_params vm_update_params;
+ struct amdgpu_pte_update_params params;
struct fence *fence = NULL;
int r;
- memset(&vm_update_params, 0, sizeof(vm_update_params));
+ if (!pd)
+ return 0;
+ pd_addr = amdgpu_bo_gpu_offset(pd);
ring = container_of(vm->entity.sched, struct amdgpu_ring, sched);
/* padding, etc. */
if (r)
return r;
- vm_update_params.ib = &job->ibs[0];
+ memset(¶ms, 0, sizeof(params));
+ params.adev = adev;
+ params.ib = &job->ibs[0];
/* walk over the address space and update the page directory */
for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) {
continue;
pt = amdgpu_bo_gpu_offset(bo);
- if (vm->page_tables[pt_idx].addr == pt)
- continue;
- vm->page_tables[pt_idx].addr = pt;
+ if (!shadow) {
+ if (vm->page_tables[pt_idx].addr == pt)
+ continue;
+ vm->page_tables[pt_idx].addr = pt;
+ } else {
+ if (vm->page_tables[pt_idx].shadow_addr == pt)
+ continue;
+ vm->page_tables[pt_idx].shadow_addr = pt;
+ }
pde = pd_addr + pt_idx * 8;
if (((last_pde + 8 * count) != pde) ||
- ((last_pt + incr * count) != pt)) {
+ ((last_pt + incr * count) != pt) ||
+ (count == AMDGPU_VM_MAX_UPDATE_SIZE)) {
if (count) {
- amdgpu_vm_update_pages(adev, &vm_update_params,
- last_pde, last_pt,
- count, incr,
- AMDGPU_PTE_VALID);
+ amdgpu_vm_do_set_ptes(¶ms, last_pde,
+ last_pt, count, incr,
+ AMDGPU_PTE_VALID);
}
count = 1;
}
if (count)
- amdgpu_vm_update_pages(adev, &vm_update_params,
- last_pde, last_pt,
- count, incr, AMDGPU_PTE_VALID);
+ amdgpu_vm_do_set_ptes(¶ms, last_pde, last_pt,
+ count, incr, AMDGPU_PTE_VALID);
- if (vm_update_params.ib->length_dw != 0) {
- amdgpu_ring_pad_ib(ring, vm_update_params.ib);
+ if (params.ib->length_dw != 0) {
+ amdgpu_ring_pad_ib(ring, params.ib);
amdgpu_sync_resv(adev, &job->sync, pd->tbo.resv,
AMDGPU_FENCE_OWNER_VM);
- WARN_ON(vm_update_params.ib->length_dw > ndw);
+ WARN_ON(params.ib->length_dw > ndw);
r = amdgpu_job_submit(job, ring, &vm->entity,
AMDGPU_FENCE_OWNER_VM, &fence);
if (r)
return r;
}
+/*
+ * amdgpu_vm_update_pdes - make sure that page directory is valid
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: requested vm
+ * @start: start of GPU address range
+ * @end: end of GPU address range
+ *
+ * Allocates new page tables if necessary
+ * and updates the page directory.
+ * Returns 0 for success, error for failure.
+ */
+int amdgpu_vm_update_page_directory(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm)
+{
+ int r;
+
+ r = amdgpu_vm_update_pd_or_shadow(adev, vm, true);
+ if (r)
+ return r;
+ return amdgpu_vm_update_pd_or_shadow(adev, vm, false);
+}
+
/**
+ * amdgpu_vm_update_ptes - make sure that page tables are valid
+ *
+ * @params: see amdgpu_pte_update_params definition
+ * @vm: requested vm
+ * @start: start of GPU address range
+ * @end: end of GPU address range
+ * @dst: destination address to map to, the next dst inside the function
+ * @flags: mapping flags
+ *
+ * Update the page tables in the range @start - @end.
+ */
+static void amdgpu_vm_update_ptes(struct amdgpu_pte_update_params *params,
+ struct amdgpu_vm *vm,
+ uint64_t start, uint64_t end,
+ uint64_t dst, uint32_t flags)
+{
+ const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
+
+ uint64_t cur_pe_start, cur_nptes, cur_dst;
+ uint64_t addr; /* next GPU address to be updated */
+ uint64_t pt_idx;
+ struct amdgpu_bo *pt;
+ unsigned nptes; /* next number of ptes to be updated */
+ uint64_t next_pe_start;
+
+ /* initialize the variables */
+ addr = start;
+ pt_idx = addr >> amdgpu_vm_block_size;
+ pt = vm->page_tables[pt_idx].entry.robj;
+ if (params->shadow) {
+ if (!pt->shadow)
+ return;
+ pt = vm->page_tables[pt_idx].entry.robj->shadow;
+ }
+ if ((addr & ~mask) == (end & ~mask))
+ nptes = end - addr;
+ else
+ nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
+
+ cur_pe_start = amdgpu_bo_gpu_offset(pt);
+ cur_pe_start += (addr & mask) * 8;
+ cur_nptes = nptes;
+ cur_dst = dst;
+
+ /* for next ptb*/
+ addr += nptes;
+ dst += nptes * AMDGPU_GPU_PAGE_SIZE;
+
+ /* walk over the address space and update the page tables */
+ while (addr < end) {
+ pt_idx = addr >> amdgpu_vm_block_size;
+ pt = vm->page_tables[pt_idx].entry.robj;
+ if (params->shadow) {
+ if (!pt->shadow)
+ return;
+ pt = vm->page_tables[pt_idx].entry.robj->shadow;
+ }
+
+ if ((addr & ~mask) == (end & ~mask))
+ nptes = end - addr;
+ else
+ nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
+
+ next_pe_start = amdgpu_bo_gpu_offset(pt);
+ next_pe_start += (addr & mask) * 8;
+
+ if ((cur_pe_start + 8 * cur_nptes) == next_pe_start &&
+ ((cur_nptes + nptes) <= AMDGPU_VM_MAX_UPDATE_SIZE)) {
+ /* The next ptb is consecutive to current ptb.
+ * Don't call the update function now.
+ * Will update two ptbs together in future.
+ */
+ cur_nptes += nptes;
+ } else {
+ params->func(params, cur_pe_start, cur_dst, cur_nptes,
+ AMDGPU_GPU_PAGE_SIZE, flags);
+
+ cur_pe_start = next_pe_start;
+ cur_nptes = nptes;
+ cur_dst = dst;
+ }
+
+ /* for next ptb*/
+ addr += nptes;
+ dst += nptes * AMDGPU_GPU_PAGE_SIZE;
+ }
+
+ params->func(params, cur_pe_start, cur_dst, cur_nptes,
+ AMDGPU_GPU_PAGE_SIZE, flags);
+}
+
+/*
* amdgpu_vm_frag_ptes - add fragment information to PTEs
*
- * @adev: amdgpu_device pointer
- * @vm_update_params: see amdgpu_vm_update_params definition
- * @pe_start: first PTE to handle
- * @pe_end: last PTE to handle
- * @addr: addr those PTEs should point to
+ * @params: see amdgpu_pte_update_params definition
+ * @vm: requested vm
+ * @start: first PTE to handle
+ * @end: last PTE to handle
+ * @dst: addr those PTEs should point to
* @flags: hw mapping flags
*/
-static void amdgpu_vm_frag_ptes(struct amdgpu_device *adev,
- struct amdgpu_vm_update_params
- *vm_update_params,
- uint64_t pe_start, uint64_t pe_end,
- uint64_t addr, uint32_t flags)
+static void amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params *params,
+ struct amdgpu_vm *vm,
+ uint64_t start, uint64_t end,
+ uint64_t dst, uint32_t flags)
{
/**
* The MC L1 TLB supports variable sized pages, based on a fragment
* allocation size to the fragment size.
*/
- /* SI and newer are optimized for 64KB */
- uint64_t frag_flags = AMDGPU_PTE_FRAG_64KB;
- uint64_t frag_align = 0x80;
-
- uint64_t frag_start = ALIGN(pe_start, frag_align);
- uint64_t frag_end = pe_end & ~(frag_align - 1);
+ const uint64_t frag_align = 1 << AMDGPU_LOG2_PAGES_PER_FRAG;
- unsigned count;
+ uint64_t frag_start = ALIGN(start, frag_align);
+ uint64_t frag_end = end & ~(frag_align - 1);
- /* Abort early if there isn't anything to do */
- if (pe_start == pe_end)
- return;
+ uint32_t frag;
/* system pages are non continuously */
- if (vm_update_params->src || vm_update_params->pages_addr ||
- !(flags & AMDGPU_PTE_VALID) || (frag_start >= frag_end)) {
+ if (params->src || !(flags & AMDGPU_PTE_VALID) ||
+ (frag_start >= frag_end)) {
- count = (pe_end - pe_start) / 8;
- amdgpu_vm_update_pages(adev, vm_update_params, pe_start,
- addr, count, AMDGPU_GPU_PAGE_SIZE,
- flags);
+ amdgpu_vm_update_ptes(params, vm, start, end, dst, flags);
return;
}
+ /* use more than 64KB fragment size if possible */
+ frag = lower_32_bits(frag_start | frag_end);
+ frag = likely(frag) ? __ffs(frag) : 31;
+
/* handle the 4K area at the beginning */
- if (pe_start != frag_start) {
- count = (frag_start - pe_start) / 8;
- amdgpu_vm_update_pages(adev, vm_update_params, pe_start, addr,
- count, AMDGPU_GPU_PAGE_SIZE, flags);
- addr += AMDGPU_GPU_PAGE_SIZE * count;
+ if (start != frag_start) {
+ amdgpu_vm_update_ptes(params, vm, start, frag_start,
+ dst, flags);
+ dst += (frag_start - start) * AMDGPU_GPU_PAGE_SIZE;
}
/* handle the area in the middle */
- count = (frag_end - frag_start) / 8;
- amdgpu_vm_update_pages(adev, vm_update_params, frag_start, addr, count,
- AMDGPU_GPU_PAGE_SIZE, flags | frag_flags);
+ amdgpu_vm_update_ptes(params, vm, frag_start, frag_end, dst,
+ flags | AMDGPU_PTE_FRAG(frag));
/* handle the 4K area at the end */
- if (frag_end != pe_end) {
- addr += AMDGPU_GPU_PAGE_SIZE * count;
- count = (pe_end - frag_end) / 8;
- amdgpu_vm_update_pages(adev, vm_update_params, frag_end, addr,
- count, AMDGPU_GPU_PAGE_SIZE, flags);
- }
-}
-
-/**
- * amdgpu_vm_update_ptes - make sure that page tables are valid
- *
- * @adev: amdgpu_device pointer
- * @vm_update_params: see amdgpu_vm_update_params definition
- * @vm: requested vm
- * @start: start of GPU address range
- * @end: end of GPU address range
- * @dst: destination address to map to
- * @flags: mapping flags
- *
- * Update the page tables in the range @start - @end.
- */
-static void amdgpu_vm_update_ptes(struct amdgpu_device *adev,
- struct amdgpu_vm_update_params
- *vm_update_params,
- struct amdgpu_vm *vm,
- uint64_t start, uint64_t end,
- uint64_t dst, uint32_t flags)
-{
- const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
-
- uint64_t last_pe_start = ~0, last_pe_end = ~0, last_dst = ~0;
- uint64_t addr;
-
- /* walk over the address space and update the page tables */
- for (addr = start; addr < end; ) {
- uint64_t pt_idx = addr >> amdgpu_vm_block_size;
- struct amdgpu_bo *pt = vm->page_tables[pt_idx].entry.robj;
- unsigned nptes;
- uint64_t pe_start;
-
- if ((addr & ~mask) == (end & ~mask))
- nptes = end - addr;
- else
- nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
-
- pe_start = amdgpu_bo_gpu_offset(pt);
- pe_start += (addr & mask) * 8;
-
- if (last_pe_end != pe_start) {
-
- amdgpu_vm_frag_ptes(adev, vm_update_params,
- last_pe_start, last_pe_end,
- last_dst, flags);
-
- last_pe_start = pe_start;
- last_pe_end = pe_start + 8 * nptes;
- last_dst = dst;
- } else {
- last_pe_end += 8 * nptes;
- }
-
- addr += nptes;
- dst += nptes * AMDGPU_GPU_PAGE_SIZE;
+ if (frag_end != end) {
+ dst += (frag_end - frag_start) * AMDGPU_GPU_PAGE_SIZE;
+ amdgpu_vm_update_ptes(params, vm, frag_end, end, dst, flags);
}
-
- amdgpu_vm_frag_ptes(adev, vm_update_params, last_pe_start,
- last_pe_end, last_dst, flags);
}
/**
* amdgpu_vm_bo_update_mapping - update a mapping in the vm page table
*
* @adev: amdgpu_device pointer
+ * @exclusive: fence we need to sync to
* @src: address where to copy page table entries from
* @pages_addr: DMA addresses to use for mapping
* @vm: requested vm
* Returns 0 for success, -EINVAL for failure.
*/
static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
+ struct fence *exclusive,
uint64_t src,
dma_addr_t *pages_addr,
struct amdgpu_vm *vm,
void *owner = AMDGPU_FENCE_OWNER_VM;
unsigned nptes, ncmds, ndw;
struct amdgpu_job *job;
- struct amdgpu_vm_update_params vm_update_params;
+ struct amdgpu_pte_update_params params;
struct fence *f = NULL;
int r;
+ memset(¶ms, 0, sizeof(params));
+ params.adev = adev;
+ params.src = src;
+
ring = container_of(vm->entity.sched, struct amdgpu_ring, sched);
- memset(&vm_update_params, 0, sizeof(vm_update_params));
- vm_update_params.src = src;
- vm_update_params.pages_addr = pages_addr;
+
+ memset(¶ms, 0, sizeof(params));
+ params.adev = adev;
+ params.src = src;
/* sync to everything on unmapping */
if (!(flags & AMDGPU_PTE_VALID))
/* padding, etc. */
ndw = 64;
- if (vm_update_params.src) {
+ if (src) {
/* only copy commands needed */
ndw += ncmds * 7;
- } else if (vm_update_params.pages_addr) {
- /* header for write data commands */
- ndw += ncmds * 4;
+ params.func = amdgpu_vm_do_copy_ptes;
- /* body of write data command */
+ } else if (pages_addr) {
+ /* copy commands needed */
+ ndw += ncmds * 7;
+
+ /* and also PTEs */
ndw += nptes * 2;
+ params.func = amdgpu_vm_do_copy_ptes;
+
} else {
/* set page commands needed */
ndw += ncmds * 10;
/* two extra commands for begin/end of fragment */
ndw += 2 * 10;
+
+ params.func = amdgpu_vm_do_set_ptes;
}
r = amdgpu_job_alloc_with_ib(adev, ndw * 4, &job);
if (r)
return r;
- vm_update_params.ib = &job->ibs[0];
+ params.ib = &job->ibs[0];
+
+ if (!src && pages_addr) {
+ uint64_t *pte;
+ unsigned i;
+
+ /* Put the PTEs at the end of the IB. */
+ i = ndw - nptes * 2;
+ pte= (uint64_t *)&(job->ibs->ptr[i]);
+ params.src = job->ibs->gpu_addr + i * 4;
+
+ for (i = 0; i < nptes; ++i) {
+ pte[i] = amdgpu_vm_map_gart(pages_addr, addr + i *
+ AMDGPU_GPU_PAGE_SIZE);
+ pte[i] |= flags;
+ }
+ }
+
+ r = amdgpu_sync_fence(adev, &job->sync, exclusive);
+ if (r)
+ goto error_free;
r = amdgpu_sync_resv(adev, &job->sync, vm->page_directory->tbo.resv,
owner);
if (r)
goto error_free;
- amdgpu_vm_update_ptes(adev, &vm_update_params, vm, start,
- last + 1, addr, flags);
+ params.shadow = true;
+ amdgpu_vm_frag_ptes(¶ms, vm, start, last + 1, addr, flags);
+ params.shadow = false;
+ amdgpu_vm_frag_ptes(¶ms, vm, start, last + 1, addr, flags);
- amdgpu_ring_pad_ib(ring, vm_update_params.ib);
- WARN_ON(vm_update_params.ib->length_dw > ndw);
+ amdgpu_ring_pad_ib(ring, params.ib);
+ WARN_ON(params.ib->length_dw > ndw);
r = amdgpu_job_submit(job, ring, &vm->entity,
AMDGPU_FENCE_OWNER_VM, &f);
if (r)
* amdgpu_vm_bo_split_mapping - split a mapping into smaller chunks
*
* @adev: amdgpu_device pointer
+ * @exclusive: fence we need to sync to
* @gtt_flags: flags as they are used for GTT
* @pages_addr: DMA addresses to use for mapping
* @vm: requested vm
* Returns 0 for success, -EINVAL for failure.
*/
static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev,
+ struct fence *exclusive,
uint32_t gtt_flags,
dma_addr_t *pages_addr,
struct amdgpu_vm *vm,
addr += mapping->offset;
if (!pages_addr || src)
- return amdgpu_vm_bo_update_mapping(adev, src, pages_addr, vm,
+ return amdgpu_vm_bo_update_mapping(adev, exclusive,
+ src, pages_addr, vm,
start, mapping->it.last,
flags, addr, fence);
uint64_t last;
last = min((uint64_t)mapping->it.last, start + max_size - 1);
- r = amdgpu_vm_bo_update_mapping(adev, src, pages_addr, vm,
+ r = amdgpu_vm_bo_update_mapping(adev, exclusive,
+ src, pages_addr, vm,
start, last, flags, addr,
fence);
if (r)
struct amdgpu_bo_va_mapping *mapping;
dma_addr_t *pages_addr = NULL;
uint32_t gtt_flags, flags;
+ struct fence *exclusive;
uint64_t addr;
int r;
default:
break;
}
+
+ exclusive = reservation_object_get_excl(bo_va->bo->tbo.resv);
} else {
addr = 0;
+ exclusive = NULL;
}
flags = amdgpu_ttm_tt_pte_flags(adev, bo_va->bo->tbo.ttm, mem);
spin_unlock(&vm->status_lock);
list_for_each_entry(mapping, &bo_va->invalids, list) {
- r = amdgpu_vm_bo_split_mapping(adev, gtt_flags, pages_addr, vm,
+ r = amdgpu_vm_bo_split_mapping(adev, exclusive,
+ gtt_flags, pages_addr, vm,
mapping, flags, addr,
&bo_va->last_pt_update);
if (r)
struct amdgpu_bo_va_mapping, list);
list_del(&mapping->list);
- r = amdgpu_vm_bo_split_mapping(adev, 0, NULL, vm, mapping,
+ r = amdgpu_vm_bo_split_mapping(adev, NULL, 0, NULL, vm, mapping,
0, 0, NULL);
kfree(mapping);
if (r)
r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8,
AMDGPU_GPU_PAGE_SIZE, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
+ AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
+ AMDGPU_GEM_CREATE_SHADOW,
NULL, resv, &pt);
if (r)
goto error_free;
r = amdgpu_bo_create(adev, pd_size, align, true,
AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_NO_CPU_ACCESS,
+ AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
+ AMDGPU_GEM_CREATE_SHADOW,
NULL, NULL, &vm->page_directory);
if (r)
goto error_free_sched_entity;
amdgpu_bo_unreserve(vm->page_directory);
if (r)
goto error_free_page_directory;
+ vm->last_eviction_counter = atomic64_read(&adev->num_evictions);
return 0;
kfree(mapping);
}
- for (i = 0; i < amdgpu_vm_num_pdes(adev); i++)
+ for (i = 0; i < amdgpu_vm_num_pdes(adev); i++) {
+ if (vm->page_tables[i].entry.robj &&
+ vm->page_tables[i].entry.robj->shadow)
+ amdgpu_bo_unref(&vm->page_tables[i].entry.robj->shadow);
amdgpu_bo_unref(&vm->page_tables[i].entry.robj);
+ }
drm_free_large(vm->page_tables);
+ if (vm->page_directory->shadow)
+ amdgpu_bo_unref(&vm->page_directory->shadow);
amdgpu_bo_unref(&vm->page_directory);
fence_put(vm->page_directory_fence);
}
&adev->vm_manager.ids_lru);
}
+ adev->vm_manager.fence_context = fence_context_alloc(AMDGPU_MAX_RINGS);
+ for (i = 0; i < AMDGPU_MAX_RINGS; ++i)
+ adev->vm_manager.seqno[i] = 0;
+
atomic_set(&adev->vm_manager.vm_pte_next_ring, 0);
atomic64_set(&adev->vm_manager.client_counter, 0);
}
projects / karo-tx-linux.git / blobdiff