From: Francois Dugast <francois.dugast@intel.com>
To: igt-dev@lists.freedesktop.org
Cc: "Francois Dugast" <francois.dugast@intel.com>,
"Matthew Brost" <matthew.brost@intel.com>,
"Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Subject: [PATCH i-g-t] tests/intel/xe_exec_system_allocator: Test madvise(MADV_DONTNEED)
Date: Fri, 24 Apr 2026 10:16:47 +0200 [thread overview]
Message-ID: <20260424081647.516473-1-francois.dugast@intel.com> (raw)
Extend the compute test to make use of madvise(MADV_DONTNEED) on the input
array before reading from it after computation, so that its effects can be
checked by looking at GT stats or tracing for example with:
sudo bpftrace -e 'kprobe:xe_migrate_from_vram { printf("%llu,%d,VRAM->SYSTEM\n", nsecs, tid); }'
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Francois Dugast <francois.dugast@intel.com>
---
tests/intel/xe_exec_system_allocator.c | 32 +++++++++++++++++++++++++-
1 file changed, 31 insertions(+), 1 deletion(-)
diff --git a/tests/intel/xe_exec_system_allocator.c b/tests/intel/xe_exec_system_allocator.c
index 5580099f7..5a6334b73 100644
--- a/tests/intel/xe_exec_system_allocator.c
+++ b/tests/intel/xe_exec_system_allocator.c
@@ -2267,6 +2267,7 @@ static void read_gt_stats_snapshot(int fd,
/* compute flags */
#define TOUCH_ONCE (0x1 << 0)
#define ACCESS_DEVICE_HOST (0x1 << 1)
+#define USE_MADV_DONTNEED (0x1 << 2)
/**
* SUBTEST: compute
@@ -2285,6 +2286,18 @@ static void read_gt_stats_snapshot(int fd,
* Description: Run a simple compute kernel %arg[1] on a 2MB malloc'ed buffer
* Test category: performance test
*
+ * SUBTEST: eu-fault-4k-range-device-host-madvise-dontneed
+ * Description: Run a simple compute kernel %arg[1] on a 4KB malloc'ed buffer for which madvise(MADV_DONTNEED) is used
+ * Test category: performance test
+ *
+ * SUBTEST: eu-fault-64k-range-device-host-madvise-dontneed
+ * Description: Run a simple compute kernel %arg[1] on a 64KB malloc'ed buffer for which madvise(MADV_DONTNEED) is used
+ * Test category: performance test
+ *
+ * SUBTEST: eu-fault-2m-range-device-host-madvise-dontneed
+ * Description: Run a simple compute kernel %arg[1] on a 2MB malloc'ed buffer for which madvise(MADV_DONTNEED) is used
+ * Test category: performance test
+ *
* arg[1]:
*
* @once-device: touch the buffer only once, from the device
@@ -2321,7 +2334,7 @@ test_compute(int fd, struct drm_xe_engine_class_instance *eci, size_t size,
xe_wait_ufence(fd, &bo_sync->sync, USER_FENCE_VALUE, 0, FIVE_SEC);
env.loop_count = (flags & TOUCH_ONCE) ? 1 : env.array_size;
- env.skip_results_check = !(flags & ACCESS_DEVICE_HOST);
+ env.skip_results_check = true;
env.vm = vm;
read_gt_stats_snapshot(fd, eci, &stats_before);
@@ -2343,6 +2356,22 @@ test_compute(int fd, struct drm_xe_engine_class_instance *eci, size_t size,
xe_run_intel_compute_kernel_on_engine(fd, eci, &env, EXECENV_PREF_SYSTEM);
+ if (flags & ACCESS_DEVICE_HOST) {
+ float total = 0;
+
+ if (flags & USE_MADV_DONTNEED)
+ madvise(compute_input, size, MADV_DONTNEED);
+
+ /*
+ * No functional purpose, just force a read on host
+ */
+ for (int j = 0; j < env.array_size; j++)
+ total += compute_input[i];
+
+ if (!total)
+ igt_debug("compute_input is zeros\n");
+ }
+
free(compute_input);
}
@@ -2524,6 +2553,7 @@ int igt_main()
{ "once-device-host", TOUCH_ONCE | ACCESS_DEVICE_HOST },
{ "range-device", 0 },
{ "range-device-host", ACCESS_DEVICE_HOST },
+ { "range-device-host-madvise-dontneed", ACCESS_DEVICE_HOST | USE_MADV_DONTNEED },
{ NULL },
};
--
2.43.0
next reply other threads:[~2026-04-24 8:17 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-04-24 8:16 Francois Dugast [this message]
2026-04-24 15:30 ` ✓ Xe.CI.BAT: success for tests/intel/xe_exec_system_allocator: Test madvise(MADV_DONTNEED) Patchwork
2026-04-24 16:07 ` ✓ i915.CI.BAT: " Patchwork
2026-04-24 16:40 ` ✗ Xe.CI.FULL: failure " Patchwork
2026-04-24 20:07 ` ✗ i915.CI.Full: " Patchwork
2026-05-07 7:21 ` [PATCH i-g-t] " Zbigniew Kempczyński
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=20260424081647.516473-1-francois.dugast@intel.com \
--to=francois.dugast@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=matthew.brost@intel.com \
--cc=thomas.hellstrom@linux.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