mirror of
https://github.com/AuxXxilium/linux_dsm_epyc7002.git
synced 2024-12-26 16:15:05 +07:00
52c0fdb25c
A few years ago, see commit688e6c7258
("drm/i915: Slaughter the thundering i915_wait_request herd"), the issue of handling multiple clients waiting in parallel was brought to our attention. The requirement was that every client should be woken immediately upon its request being signaled, without incurring any cpu overhead. To handle certain fragility of our hw meant that we could not do a simple check inside the irq handler (some generations required almost unbounded delays before we could be sure of seqno coherency) and so request completion checking required delegation. Before commit688e6c7258
, the solution was simple. Every client waiting on a request would be woken on every interrupt and each would do a heavyweight check to see if their request was complete. Commit688e6c7258
introduced an rbtree so that only the earliest waiter on the global timeline would woken, and would wake the next and so on. (Along with various complications to handle requests being reordered along the global timeline, and also a requirement for kthread to provide a delegate for fence signaling that had no process context.) The global rbtree depends on knowing the execution timeline (and global seqno). Without knowing that order, we must instead check all contexts queued to the HW to see which may have advanced. We trim that list by only checking queued contexts that are being waited on, but still we keep a list of all active contexts and their active signalers that we inspect from inside the irq handler. By moving the waiters onto the fence signal list, we can combine the client wakeup with the dma_fence signaling (a dramatic reduction in complexity, but does require the HW being coherent, the seqno must be visible from the cpu before the interrupt is raised - we keep a timer backup just in case). Having previously fixed all the issues with irq-seqno serialisation (by inserting delays onto the GPU after each request instead of random delays on the CPU after each interrupt), we can rely on the seqno state to perfom direct wakeups from the interrupt handler. This allows us to preserve our single context switch behaviour of the current routine, with the only downside that we lose the RT priority sorting of wakeups. In general, direct wakeup latency of multiple clients is about the same (about 10% better in most cases) with a reduction in total CPU time spent in the waiter (about 20-50% depending on gen). Average herd behaviour is improved, but at the cost of not delegating wakeups on task_prio. v2: Capture fence signaling state for error state and add comments to warm even the most cold of hearts. v3: Check if the request is still active before busywaiting v4: Reduce the amount of pointer misdirection with list_for_each_safe and using a local i915_request variable inside the loops v5: Add a missing pluralisation to a purely informative selftest message. References:688e6c7258
("drm/i915: Slaughter the thundering i915_wait_request herd") Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190129205230.19056-2-chris@chris-wilson.co.uk
195 lines
4.1 KiB
C
195 lines
4.1 KiB
C
/*
|
|
* SPDX-License-Identifier: MIT
|
|
*
|
|
* Copyright © 2018 Intel Corporation
|
|
*/
|
|
|
|
#include "igt_spinner.h"
|
|
|
|
int igt_spinner_init(struct igt_spinner *spin, struct drm_i915_private *i915)
|
|
{
|
|
unsigned int mode;
|
|
void *vaddr;
|
|
int err;
|
|
|
|
GEM_BUG_ON(INTEL_GEN(i915) < 8);
|
|
|
|
memset(spin, 0, sizeof(*spin));
|
|
spin->i915 = i915;
|
|
|
|
spin->hws = i915_gem_object_create_internal(i915, PAGE_SIZE);
|
|
if (IS_ERR(spin->hws)) {
|
|
err = PTR_ERR(spin->hws);
|
|
goto err;
|
|
}
|
|
|
|
spin->obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
|
|
if (IS_ERR(spin->obj)) {
|
|
err = PTR_ERR(spin->obj);
|
|
goto err_hws;
|
|
}
|
|
|
|
i915_gem_object_set_cache_level(spin->hws, I915_CACHE_LLC);
|
|
vaddr = i915_gem_object_pin_map(spin->hws, I915_MAP_WB);
|
|
if (IS_ERR(vaddr)) {
|
|
err = PTR_ERR(vaddr);
|
|
goto err_obj;
|
|
}
|
|
spin->seqno = memset(vaddr, 0xff, PAGE_SIZE);
|
|
|
|
mode = i915_coherent_map_type(i915);
|
|
vaddr = i915_gem_object_pin_map(spin->obj, mode);
|
|
if (IS_ERR(vaddr)) {
|
|
err = PTR_ERR(vaddr);
|
|
goto err_unpin_hws;
|
|
}
|
|
spin->batch = vaddr;
|
|
|
|
return 0;
|
|
|
|
err_unpin_hws:
|
|
i915_gem_object_unpin_map(spin->hws);
|
|
err_obj:
|
|
i915_gem_object_put(spin->obj);
|
|
err_hws:
|
|
i915_gem_object_put(spin->hws);
|
|
err:
|
|
return err;
|
|
}
|
|
|
|
static unsigned int seqno_offset(u64 fence)
|
|
{
|
|
return offset_in_page(sizeof(u32) * fence);
|
|
}
|
|
|
|
static u64 hws_address(const struct i915_vma *hws,
|
|
const struct i915_request *rq)
|
|
{
|
|
return hws->node.start + seqno_offset(rq->fence.context);
|
|
}
|
|
|
|
static int move_to_active(struct i915_vma *vma,
|
|
struct i915_request *rq,
|
|
unsigned int flags)
|
|
{
|
|
int err;
|
|
|
|
err = i915_vma_move_to_active(vma, rq, flags);
|
|
if (err)
|
|
return err;
|
|
|
|
if (!i915_gem_object_has_active_reference(vma->obj)) {
|
|
i915_gem_object_get(vma->obj);
|
|
i915_gem_object_set_active_reference(vma->obj);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
struct i915_request *
|
|
igt_spinner_create_request(struct igt_spinner *spin,
|
|
struct i915_gem_context *ctx,
|
|
struct intel_engine_cs *engine,
|
|
u32 arbitration_command)
|
|
{
|
|
struct i915_address_space *vm = &ctx->ppgtt->vm;
|
|
struct i915_request *rq = NULL;
|
|
struct i915_vma *hws, *vma;
|
|
u32 *batch;
|
|
int err;
|
|
|
|
vma = i915_vma_instance(spin->obj, vm, NULL);
|
|
if (IS_ERR(vma))
|
|
return ERR_CAST(vma);
|
|
|
|
hws = i915_vma_instance(spin->hws, vm, NULL);
|
|
if (IS_ERR(hws))
|
|
return ERR_CAST(hws);
|
|
|
|
err = i915_vma_pin(vma, 0, 0, PIN_USER);
|
|
if (err)
|
|
return ERR_PTR(err);
|
|
|
|
err = i915_vma_pin(hws, 0, 0, PIN_USER);
|
|
if (err)
|
|
goto unpin_vma;
|
|
|
|
rq = i915_request_alloc(engine, ctx);
|
|
if (IS_ERR(rq)) {
|
|
err = PTR_ERR(rq);
|
|
goto unpin_hws;
|
|
}
|
|
|
|
err = move_to_active(vma, rq, 0);
|
|
if (err)
|
|
goto cancel_rq;
|
|
|
|
err = move_to_active(hws, rq, 0);
|
|
if (err)
|
|
goto cancel_rq;
|
|
|
|
batch = spin->batch;
|
|
|
|
*batch++ = MI_STORE_DWORD_IMM_GEN4;
|
|
*batch++ = lower_32_bits(hws_address(hws, rq));
|
|
*batch++ = upper_32_bits(hws_address(hws, rq));
|
|
*batch++ = rq->fence.seqno;
|
|
|
|
*batch++ = arbitration_command;
|
|
|
|
*batch++ = MI_BATCH_BUFFER_START | 1 << 8 | 1;
|
|
*batch++ = lower_32_bits(vma->node.start);
|
|
*batch++ = upper_32_bits(vma->node.start);
|
|
*batch++ = MI_BATCH_BUFFER_END; /* not reached */
|
|
|
|
i915_gem_chipset_flush(spin->i915);
|
|
|
|
err = engine->emit_bb_start(rq, vma->node.start, PAGE_SIZE, 0);
|
|
|
|
cancel_rq:
|
|
if (err) {
|
|
i915_request_skip(rq, err);
|
|
i915_request_add(rq);
|
|
}
|
|
unpin_hws:
|
|
i915_vma_unpin(hws);
|
|
unpin_vma:
|
|
i915_vma_unpin(vma);
|
|
return err ? ERR_PTR(err) : rq;
|
|
}
|
|
|
|
static u32
|
|
hws_seqno(const struct igt_spinner *spin, const struct i915_request *rq)
|
|
{
|
|
u32 *seqno = spin->seqno + seqno_offset(rq->fence.context);
|
|
|
|
return READ_ONCE(*seqno);
|
|
}
|
|
|
|
void igt_spinner_end(struct igt_spinner *spin)
|
|
{
|
|
*spin->batch = MI_BATCH_BUFFER_END;
|
|
i915_gem_chipset_flush(spin->i915);
|
|
}
|
|
|
|
void igt_spinner_fini(struct igt_spinner *spin)
|
|
{
|
|
igt_spinner_end(spin);
|
|
|
|
i915_gem_object_unpin_map(spin->obj);
|
|
i915_gem_object_put(spin->obj);
|
|
|
|
i915_gem_object_unpin_map(spin->hws);
|
|
i915_gem_object_put(spin->hws);
|
|
}
|
|
|
|
bool igt_wait_for_spinner(struct igt_spinner *spin, struct i915_request *rq)
|
|
{
|
|
return !(wait_for_us(i915_seqno_passed(hws_seqno(spin, rq),
|
|
rq->fence.seqno),
|
|
10) &&
|
|
wait_for(i915_seqno_passed(hws_seqno(spin, rq),
|
|
rq->fence.seqno),
|
|
1000));
|
|
}
|