amdgpu_cs.o amdgpu_bios.o amdgpu_benchmark.o \
atombios_dp.o amdgpu_afmt.o amdgpu_trace_points.o \
atombios_encoders.o amdgpu_sa.o atombios_i2c.o \
- amdgpu_dma_buf.o amdgpu_vm.o amdgpu_vm_pt.o amdgpu_ib.o amdgpu_pll.o \
+ amdgpu_dma_buf.o amdgpu_vm.o amdgpu_vm_pt.o amdgpu_vm_tlb_fence.o \
+ amdgpu_ib.o amdgpu_pll.o \
amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o \
amdgpu_gtt_mgr.o amdgpu_preempt_mgr.o amdgpu_vram_mgr.o amdgpu_virt.o \
amdgpu_atomfirmware.o amdgpu_vf_error.o amdgpu_sched.o \
kfree(tlb_cb);
}
+/**
+ * amdgpu_vm_tlb_flush - prepare TLB flush
+ *
+ * @params: parameters for update
+ * @fence: input fence to sync TLB flush with
+ * @tlb_cb: the callback structure
+ *
+ * Increments the tlb sequence to make sure that future CS execute a VM flush.
+ */
+static void
+amdgpu_vm_tlb_flush(struct amdgpu_vm_update_params *params,
+ struct dma_fence **fence,
+ struct amdgpu_vm_tlb_seq_struct *tlb_cb)
+{
+ struct amdgpu_vm *vm = params->vm;
+
+ if (!fence || !*fence)
+ return;
+
+ tlb_cb->vm = vm;
+ if (!dma_fence_add_callback(*fence, &tlb_cb->cb,
+ amdgpu_vm_tlb_seq_cb)) {
+ dma_fence_put(vm->last_tlb_flush);
+ vm->last_tlb_flush = dma_fence_get(*fence);
+ } else {
+ amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
+ }
+
+ /* Prepare a TLB flush fence to be attached to PTs */
+ if (!params->unlocked && vm->is_compute_context) {
+ amdgpu_vm_tlb_fence_create(params->adev, vm, fence);
+
+ /* Makes sure no PD/PT is freed before the flush */
+ dma_resv_add_fence(vm->root.bo->tbo.base.resv, *fence,
+ DMA_RESV_USAGE_BOOKKEEP);
+ }
+}
+
/**
* amdgpu_vm_update_range - update a range in the vm page table
*
struct ttm_resource *res, dma_addr_t *pages_addr,
struct dma_fence **fence)
{
- struct amdgpu_vm_update_params params;
struct amdgpu_vm_tlb_seq_struct *tlb_cb;
+ struct amdgpu_vm_update_params params;
struct amdgpu_res_cursor cursor;
enum amdgpu_sync_mode sync_mode;
int r, idx;
tlb_cb = kmalloc(sizeof(*tlb_cb), GFP_KERNEL);
if (!tlb_cb) {
- r = -ENOMEM;
- goto error_unlock;
+ drm_dev_exit(idx);
+ return -ENOMEM;
}
/* Vega20+XGMI where PTEs get inadvertently cached in L2 texture cache,
params.immediate = immediate;
params.pages_addr = pages_addr;
params.unlocked = unlocked;
+ params.needs_flush = flush_tlb;
params.allow_override = allow_override;
/* Implicitly sync to command submissions in the same VM before
}
r = vm->update_funcs->commit(¶ms, fence);
+ if (r)
+ goto error_free;
- if (flush_tlb || params.table_freed) {
- tlb_cb->vm = vm;
- if (fence && *fence &&
- !dma_fence_add_callback(*fence, &tlb_cb->cb,
- amdgpu_vm_tlb_seq_cb)) {
- dma_fence_put(vm->last_tlb_flush);
- vm->last_tlb_flush = dma_fence_get(*fence);
- } else {
- amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
- }
+ if (params.needs_flush) {
+ amdgpu_vm_tlb_flush(¶ms, fence, tlb_cb);
tlb_cb = NULL;
}
error_free:
kfree(tlb_cb);
-
-error_unlock:
amdgpu_vm_eviction_unlock(vm);
drm_dev_exit(idx);
return r;
mutex_init(&vm->eviction_lock);
vm->evicting = false;
+ vm->tlb_fence_context = dma_fence_context_alloc(1);
r = amdgpu_vm_pt_create(adev, vm, adev->vm_manager.root_level,
false, &root, xcp_id);
unsigned int num_dw_left;
/**
- * @table_freed: return true if page table is freed when updating
+ * @needs_flush: true whenever we need to invalidate the TLB
*/
- bool table_freed;
+ bool needs_flush;
/**
* @allow_override: true for memory that is not uncached: allows MTYPE
atomic64_t tlb_seq;
struct dma_fence *last_tlb_flush;
atomic64_t kfd_last_flushed_seq;
+ uint64_t tlb_fence_context;
/* How many times we had to re-generate the page tables */
uint64_t generation;
uint64_t addr,
uint32_t status,
unsigned int vmhub);
+void amdgpu_vm_tlb_fence_create(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ struct dma_fence **fence);
#endif
static int amdgpu_vm_cpu_commit(struct amdgpu_vm_update_params *p,
struct dma_fence **fence)
{
- /* Flush HDP */
+ if (p->needs_flush)
+ atomic64_inc(&p->vm->tlb_seq);
+
mb();
amdgpu_device_flush_hdp(p->adev, NULL);
return 0;
while (cursor.pfn < frag_start) {
/* Make sure previous mapping is freed */
if (cursor.entry->bo) {
- params->table_freed = true;
+ params->needs_flush = true;
amdgpu_vm_pt_free_dfs(adev, params->vm,
&cursor,
params->unlocked);
WARN_ON(ib->length_dw == 0);
amdgpu_ring_pad_ib(ring, ib);
+
+ if (p->needs_flush)
+ atomic64_inc(&p->vm->tlb_seq);
+
WARN_ON(ib->length_dw > p->num_dw_left);
f = amdgpu_job_submit(p->job);
--- /dev/null
+// SPDX-License-Identifier: GPL-2.0 OR MIT
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <linux/dma-fence.h>
+#include <linux/workqueue.h>
+
+#include "amdgpu.h"
+#include "amdgpu_vm.h"
+#include "amdgpu_gmc.h"
+
+struct amdgpu_tlb_fence {
+ struct dma_fence base;
+ struct amdgpu_device *adev;
+ struct dma_fence *dependency;
+ struct work_struct work;
+ spinlock_t lock;
+ uint16_t pasid;
+
+};
+
+static const char *amdgpu_tlb_fence_get_driver_name(struct dma_fence *fence)
+{
+ return "amdgpu tlb fence";
+}
+
+static const char *amdgpu_tlb_fence_get_timeline_name(struct dma_fence *f)
+{
+ return "amdgpu tlb timeline";
+}
+
+static void amdgpu_tlb_fence_work(struct work_struct *work)
+{
+ struct amdgpu_tlb_fence *f = container_of(work, typeof(*f), work);
+ int r;
+
+ if (f->dependency) {
+ dma_fence_wait(f->dependency, false);
+ dma_fence_put(f->dependency);
+ f->dependency = NULL;
+ }
+
+ r = amdgpu_gmc_flush_gpu_tlb_pasid(f->adev, f->pasid, 2, true, 0);
+ if (r) {
+ dev_err(f->adev->dev, "TLB flush failed for PASID %d.\n",
+ f->pasid);
+ dma_fence_set_error(&f->base, r);
+ }
+
+ dma_fence_signal(&f->base);
+ dma_fence_put(&f->base);
+}
+
+static const struct dma_fence_ops amdgpu_tlb_fence_ops = {
+ .use_64bit_seqno = true,
+ .get_driver_name = amdgpu_tlb_fence_get_driver_name,
+ .get_timeline_name = amdgpu_tlb_fence_get_timeline_name
+};
+
+void amdgpu_vm_tlb_fence_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ struct dma_fence **fence)
+{
+ struct amdgpu_tlb_fence *f;
+
+ f = kmalloc(sizeof(*f), GFP_KERNEL);
+ if (!f) {
+ /*
+ * We can't fail since the PDEs and PTEs are already updated, so
+ * just block for the dependency and execute the TLB flush
+ */
+ if (*fence)
+ dma_fence_wait(*fence, false);
+
+ amdgpu_gmc_flush_gpu_tlb_pasid(adev, vm->pasid, 2, true, 0);
+ *fence = dma_fence_get_stub();
+ return;
+ }
+
+ f->adev = adev;
+ f->dependency = *fence;
+ f->pasid = vm->pasid;
+ INIT_WORK(&f->work, amdgpu_tlb_fence_work);
+ spin_lock_init(&f->lock);
+
+ dma_fence_init(&f->base, &amdgpu_tlb_fence_ops, &f->lock,
+ vm->tlb_fence_context, atomic64_read(&vm->tlb_seq));
+
+ /* TODO: We probably need a separate wq here */
+ dma_fence_get(&f->base);
+ schedule_work(&f->work);
+
+ *fence = &f->base;
+}
projects / users / hch / dma-mapping.git / commitdiff