torvalds/linux
3
0
Fork 0
mirror of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git synced 2026-03-22 07:27:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdkfd: KFD interrupt access ih_fifo data in-place

Browse Source 
To handle 40000 to 80000 interrupts per second running CPX mode with 4
streams/queues per KFD node, KFD interrupt handler becomes the
performance bottleneck.

Remove the kfifo_out memcpy overhead by accessing ih_fifo data in-place
and updating rptr with kfifo_skip_count.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Philip Yang
2024-11-22 17:36:15 -05:00
committed by Alex Deucher
parent 11815bb0e3
commit f607b2b867
1 changed files with 14 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c
View File

@@ -114,50 +114,43 @@ void kfd_interrupt_exit(struct kfd_node *node)
*/
bool enqueue_ih_ring_entry(struct kfd_node *node, const void *ih_ring_entry)
{
int count;
count = kfifo_in(&node->ih_fifo, ih_ring_entry,
node->kfd->device_info.ih_ring_entry_size);
if (count != node->kfd->device_info.ih_ring_entry_size) {
if (kfifo_is_full(&node->ih_fifo)) {
dev_dbg_ratelimited(node->adev->dev,
"Interrupt ring overflow, dropping interrupt %d\n",
count);
"Interrupt ring overflow, dropping interrupt\n");
return false;
}
kfifo_in(&node->ih_fifo, ih_ring_entry, node->kfd->device_info.ih_ring_entry_size);
return true;
}
/*
* Assumption: single reader/writer. This function is not re-entrant
*/
static bool dequeue_ih_ring_entry(struct kfd_node *node, void *ih_ring_entry)
static bool dequeue_ih_ring_entry(struct kfd_node *node, u32 **ih_ring_entry)
{
int count;
count = kfifo_out(&node->ih_fifo, ih_ring_entry,
node->kfd->device_info.ih_ring_entry_size);
WARN_ON(count && count != node->kfd->device_info.ih_ring_entry_size);
if (kfifo_is_empty(&node->ih_fifo))
return false;
count = kfifo_out_linear_ptr(&node->ih_fifo, ih_ring_entry,
node->kfd->device_info.ih_ring_entry_size);
WARN_ON(count != node->kfd->device_info.ih_ring_entry_size);
return count == node->kfd->device_info.ih_ring_entry_size;
}
static void interrupt_wq(struct work_struct *work)
{
struct kfd_node *dev = container_of(work, struct kfd_node,
interrupt_work);
uint32_t ih_ring_entry[KFD_MAX_RING_ENTRY_SIZE];
struct kfd_node *dev = container_of(work, struct kfd_node, interrupt_work);
uint32_t *ih_ring_entry;
unsigned long start_jiffies = jiffies;
if (dev->kfd->device_info.ih_ring_entry_size > sizeof(ih_ring_entry)) {
dev_err_once(dev->adev->dev, "Ring entry too small\n");
return;
}
while (dequeue_ih_ring_entry(dev, ih_ring_entry)) {
while (dequeue_ih_ring_entry(dev, &ih_ring_entry)) {
dev->kfd->device_info.event_interrupt_class->interrupt_wq(dev,
ih_ring_entry);
kfifo_skip_count(&dev->ih_fifo, dev->kfd->device_info.ih_ring_entry_size);
if (time_is_before_jiffies(start_jiffies + HZ)) {
/* If we spent more than a second processing signals,
* reschedule the worker to avoid soft-lockup warnings