From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: intel-xe@lists.freedesktop.org
Cc: matthew.auld@intel.com
Subject: [Intel-xe] [PATCH v2 7/7] drm/xe/vm: Defer vm rebind until next exec if nothing to execute
Date: Wed, 15 Mar 2023 16:55:07 +0100 [thread overview]
Message-ID: <20230315155507.43933-8-thomas.hellstrom@linux.intel.com> (raw)
In-Reply-To: <20230315155507.43933-1-thomas.hellstrom@linux.intel.com>
If all compute engines of a vm in compute mode are idle,
defer a rebind to the next exec to avoid the VM unnecessarily trying
to make memory resident and compete with other VMs for available
memory space.
v2:
- Patch added to series.
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
drivers/gpu/drm/xe/xe_exec.c | 1 +
drivers/gpu/drm/xe/xe_vm.c | 18 ++++++++++++++++++
drivers/gpu/drm/xe/xe_vm.h | 17 +++++++++++++++++
drivers/gpu/drm/xe/xe_vm_types.h | 5 +++++
4 files changed, 41 insertions(+)
diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index 97fd1a311f2d..ea869f2452ef 100644
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -364,6 +364,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
&job->drm.s_fence->finished);
xe_sched_job_push(job);
+ xe_vm_reactivate_rebind(vm);
err_repin:
if (!xe_vm_no_dma_fences(vm))
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index e8e178922082..6c466c628395 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -226,6 +226,19 @@ static int wait_for_existing_preempt_fences(struct xe_vm *vm)
return 0;
}
+static bool xe_vm_is_idle(struct xe_vm *vm)
+{
+ struct xe_engine *e;
+
+ xe_vm_assert_held(vm);
+ list_for_each_entry(e, &vm->preempt.engines, compute.link) {
+ if (!xe_engine_is_idle(e))
+ return false;
+ }
+
+ return true;
+}
+
static void arm_preempt_fences(struct xe_vm *vm, struct list_head *list)
{
struct list_head *link;
@@ -548,6 +561,11 @@ static void preempt_rebind_work_func(struct work_struct *w)
if (err)
goto out_unlock_outer;
+ if (xe_vm_is_idle(vm)) {
+ vm->preempt.rebind_deactivated = true;
+ goto out_unlock;
+ }
+
/* Fresh preempt fences already installed. Everyting is running. */
if (!preempt_fences_waiting(vm))
goto out_unlock;
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index 3468ed9d0528..748dc16ebed9 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -98,6 +98,23 @@ extern struct ttm_device_funcs xe_ttm_funcs;
struct ttm_buffer_object *xe_vm_ttm_bo(struct xe_vm *vm);
+/**
+ * xe_vm_reactivate_rebind() - Reactivate the rebind functionality on compute
+ * vms.
+ * @vm: The vm.
+ *
+ * If the rebind functionality on a compute vm was disabled due
+ * to nothing to execute. Reactivate it and run the rebind worker.
+ * This function should be called after submitting a batch to a compute vm.
+ */
+static inline void xe_vm_reactivate_rebind(struct xe_vm *vm)
+{
+ if (xe_vm_in_compute_mode(vm) && vm->preempt.rebind_deactivated) {
+ vm->preempt.rebind_deactivated = false;
+ queue_work(system_unbound_wq, &vm->preempt.rebind_work);
+ }
+}
+
static inline bool xe_vma_is_userptr(struct xe_vma *vma)
{
return !vma->bo;
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index 2a3b911ab358..fada7896867f 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -293,6 +293,11 @@ struct xe_vm {
struct list_head engines;
/** @num_engines: number user engines attached to this VM */
int num_engines;
+ /**
+ * @rebind_deactivated: Whether rebind has been temporarily deactivated
+ * due to no work available. Protected by the vm resv.
+ */
+ bool rebind_deactivated;
/**
* @rebind_work: worker to rebind invalidated userptrs / evicted
* BOs
--
2.39.2
next prev parent reply other threads:[~2023-03-15 15:56 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-03-15 15:55 [Intel-xe] [PATCH v2 0/7] Cpu page-table updates and fixes Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 1/7] drm/xe: Use a define to set initial seqno for fences Thomas Hellström
2023-03-16 16:09 ` Matthew Brost
2023-03-16 16:45 ` Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 2/7] drm/xe/migrate: Update cpu page-table updates Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 3/7] drm/xe/tests: Support CPU page-table updates in the migrate test Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 4/7] drm/xe: Introduce xe_engine_is_idle() Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 5/7] drm/xe: Use a small negative initial seqno Thomas Hellström
2023-03-15 15:55 ` [Intel-xe] [PATCH v2 6/7] drm/xe/tests: Test both CPU- and GPU page-table updates with the migrate test Thomas Hellström
2023-03-15 15:55 ` Thomas Hellström [this message]
2023-03-15 23:23 ` [Intel-xe] [PATCH v2 7/7] drm/xe/vm: Defer vm rebind until next exec if nothing to execute Matthew Brost
2023-03-15 15:58 ` [Intel-xe] ✓ CI.Patch_applied: success for Cpu page-table updates and fixes (rev2) Patchwork
2023-03-15 15:59 ` [Intel-xe] ✓ CI.KUnit: " Patchwork
2023-03-15 16:03 ` [Intel-xe] ✓ CI.Build: " Patchwork
2023-03-15 16:14 ` [Intel-xe] ○ CI.BAT: info " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230315155507.43933-8-thomas.hellstrom@linux.intel.com \
--to=thomas.hellstrom@linux.intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=matthew.auld@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox