{
struct kfd_process_device *pdd;
struct kfd_process *p;
- int drain;
uint32_t i;
p = container_of(svms, struct kfd_process, svms);
-restart:
- drain = atomic_read(&svms->drain_pagefaults);
- if (!drain)
- return;
-
for_each_set_bit(i, svms->bitmap_supported, p->n_pdds) {
pdd = p->pdds[i];
if (!pdd)
pr_debug("drain retry fault gpu %d svms 0x%p done\n", i, svms);
}
- if (atomic_cmpxchg(&svms->drain_pagefaults, drain, 0) != drain)
- goto restart;
}
static void svm_range_deferred_list_work(struct work_struct *work)
prange->start, prange->last, prange->work_item.op);
mm = prange->work_item.mm;
-retry:
- mmap_write_lock(mm);
- /* Checking for the need to drain retry faults must be inside
- * mmap write lock to serialize with munmap notifiers.
- */
- if (unlikely(atomic_read(&svms->drain_pagefaults))) {
- mmap_write_unlock(mm);
- svm_range_drain_retry_fault(svms);
- goto retry;
- }
+ mmap_write_lock(mm);
/* Remove from deferred_list must be inside mmap write lock, for
* two race cases:
struct kfd_process *p;
unsigned long s, l;
bool unmap_parent;
+ uint32_t i;
if (atomic_read(&prange->queue_refcount)) {
int r;
pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx] [0x%lx 0x%lx]\n", svms,
prange, prange->start, prange->last, start, last);
- /* Make sure pending page faults are drained in the deferred worker
- * before the range is freed to avoid straggler interrupts on
- * unmapped memory causing "phantom faults".
+ /* calculate time stamps that are used to decide which page faults need be
+ * dropped or handled before unmap pages from gpu vm
*/
- atomic_inc(&svms->drain_pagefaults);
+ for_each_set_bit(i, svms->bitmap_supported, p->n_pdds) {
+ struct kfd_process_device *pdd;
+ struct amdgpu_device *adev;
+ struct amdgpu_ih_ring *ih;
+ uint32_t checkpoint_wptr;
+
+ pdd = p->pdds[i];
+ if (!pdd)
+ continue;
+
+ adev = pdd->dev->adev;
+
+ /* Check and drain ih1 ring if cam not available */
+ ih = &adev->irq.ih1;
+ checkpoint_wptr = amdgpu_ih_get_wptr(adev, ih);
+ if (ih->rptr != checkpoint_wptr) {
+ svms->checkpoint_ts[i] = amdgpu_ih_decode_iv_ts(adev, ih, checkpoint_wptr, -1);
+ continue;
+ }
+
+ /* check if dev->irq.ih_soft is not empty */
+ ih = &adev->irq.ih_soft;
+ checkpoint_wptr = amdgpu_ih_get_wptr(adev, ih);
+ if (ih->rptr != checkpoint_wptr)
+ svms->checkpoint_ts[i] = amdgpu_ih_decode_iv_ts(adev, ih, checkpoint_wptr, -1);
+ }
unmap_parent = start <= prange->start && last >= prange->last;
int
svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
uint32_t vmid, uint32_t node_id,
- uint64_t addr, bool write_fault)
+ uint64_t addr, uint64_t ts, bool write_fault)
{
unsigned long start, last, size;
struct mm_struct *mm = NULL;
ktime_t timestamp = ktime_get_boottime();
struct kfd_node *node;
int32_t best_loc;
- int32_t gpuidx = MAX_GPU_INSTANCE;
+ int32_t gpuid, gpuidx = MAX_GPU_INSTANCE;
bool write_locked = false;
struct vm_area_struct *vma;
bool migration = false;
pr_debug("restoring svms 0x%p fault address 0x%llx\n", svms, addr);
if (atomic_read(&svms->drain_pagefaults)) {
- pr_debug("draining retry fault, drop fault 0x%llx\n", addr);
+ pr_debug("page fault handling disabled, drop fault 0x%llx\n", addr);
r = 0;
goto out;
}
+ node = kfd_node_by_irq_ids(adev, node_id, vmid);
+ if (!node) {
+ pr_debug("kfd node does not exist node_id: %d, vmid: %d\n", node_id,
+ vmid);
+ r = -EFAULT;
+ goto out;
+ }
+
+ if (kfd_process_gpuid_from_node(p, node, &gpuid, &gpuidx)) {
+ pr_debug("failed to get gpuid/gpuidex for node_id: %d\n", node_id);
+ r = -EFAULT;
+ goto out;
+ }
+
+ /* check if this page fault time stamp is before svms->checkpoint_ts */
+ if (svms->checkpoint_ts[gpuidx] != 0) {
+ if (amdgpu_ih_ts_after(ts, svms->checkpoint_ts[gpuidx])) {
+ pr_debug("draining retry fault, drop fault 0x%llx\n", addr);
+ r = 0;
+ goto out;
+ } else
+ /* ts is after svms->checkpoint_ts now, reset svms->checkpoint_ts
+ * to zero to avoid following ts wrap around give wrong comparing
+ */
+ svms->checkpoint_ts[gpuidx] = 0;
+ }
+
if (!p->xnack_enabled) {
pr_debug("XNACK not enabled for pasid 0x%x\n", pasid);
r = -EFAULT;
goto out;
}
- node = kfd_node_by_irq_ids(adev, node_id, vmid);
- if (!node) {
- pr_debug("kfd node does not exist node_id: %d, vmid: %d\n", node_id,
- vmid);
- r = -EFAULT;
- goto out;
- }
mmap_read_lock(mm);
retry_write_locked:
mutex_lock(&svms->lock);
/*
* Ensure no retry fault comes in afterwards, as page fault handler will
* not find kfd process and take mm lock to recover fault.
+ * stop kfd page fault handing, then wait pending page faults got drained
*/
- atomic_inc(&p->svms.drain_pagefaults);
+ atomic_set(&p->svms.drain_pagefaults, 1);
svm_range_drain_retry_fault(&p->svms);
list_for_each_entry_safe(prange, next, &p->svms.list, list) {
projects / linux.git / commitdiff