From: Riana Tauro <riana.tauro@intel.com>
To: Karthik Poosa <karthik.poosa@intel.com>, <igt-dev@lists.freedesktop.org>
Cc: <anshuman.gupta@intel.com>, <badal.nilawar@intel.com>,
<rodrigo.vivi@intel.com>, <kamil.konieczny@linux.intel.com>
Subject: Re: [PATCH i-g-t v8 2/2] tests/intel/xe_pm_residency: Add subtest for ASPM Link state residency
Date: Tue, 16 Dec 2025 19:48:08 +0530 [thread overview]
Message-ID: <b49b01e7-935c-402e-93c9-eaf1f4ecaca6@intel.com> (raw)
In-Reply-To: <20251216080951.199893-3-karthik.poosa@intel.com>
Hi Karthik
On 12/16/2025 1:39 PM, Karthik Poosa wrote:
> Add subtest aspm_link_residency to verify PCIe ASPM.
> Active State Power Management (ASPM) is a power management mechanism
> for PCI Express (PCIe) devices that aims to save power while the devices
> are in a fully active state.
> This test uses link state counters from debugfs - dgfx_pcie_link_residencies
> to verify this.
>
> v2:
> - Add dedicated function to get pcie endpoint upstream port. (Badal)
> - Read residency counter as unsigned long long int instead of
> unsigned long int.
> - Print residency counter before sleep also.
> - Don't assert if sysfs not corresponding to aspm_link_state
> is not present. (Badal)
> - Run workload before validation of aspm link residency. (Anshuman)
>
> v3:
> - Move igt_device_get_pci_usp to separate patch. (Kamil)
> - Move reading of residency to separate function. (Badal)
>
> v4:
> - Add description about PCIe ASPM in commit message and code. (Kamil)
> - Add a NULL check for the return value of igt_device_get_pci_usp().
> - Resolve compilation warnings about using variable as format string
> to sscanf.
>
> v5:
> - Use igt_device_get_pci_upstream_port() which is the renamed version
> of igt_device_get_pci_usp().
>
> v6:
> - Refactor and enhance readability. (Badal)
> - Move save and restore of link states to separate functions. (Badal)
>
> v7:
> - Skip aspm_link_residency on integrated platforms as it not supported.
>
> Signed-off-by: Karthik Poosa <karthik.poosa@intel.com>
> Reviewed-by: Badal Nilawar <badal.nilawar@intel.com>
> ---
> tests/intel/xe_pm_residency.c | 179 ++++++++++++++++++++++++++++++++++
> 1 file changed, 179 insertions(+)
> mode change 100644 => 100755 tests/intel/xe_pm_residency.c
>
> diff --git a/tests/intel/xe_pm_residency.c b/tests/intel/xe_pm_residency.c
> old mode 100644
> new mode 100755
> index d33a87b13..04e560f9b
> --- a/tests/intel/xe_pm_residency.c
> +++ b/tests/intel/xe_pm_residency.c
> @@ -37,6 +37,27 @@ enum test_type {
> TEST_IDLE,
> };
>
> +enum link_state_index {
> + LINK_STATE_ASPM,
> + LINK_STATE_ASPM_L1_1,
> + LINK_STATE_ASPM_L1_2,
> + LINK_STATE_PCIPM_L1_1,
> + LINK_STATE_PCIPM_L1_2,
> + MAX_LINK_STATES,
> +};
> +
> +struct link_state_info {
> + const char *filename;
> + char state;
> + const char *parse_str;
> +} link_state_sysfs [] = {
Remove space before [
> + { "l1_aspm", 0, "PCIE LINK L1 RESIDENCY : "},
> + { "l1_1_aspm", 0, "NULL"},
> + { "l1_2_aspm", 0, "PCIE LINK L1.2 RESIDENCY : "},
> + { "l1_1_pcipm", 0, NULL},
> + { "l1_2_pcipm", 0, NULL},
> +};
What will be the value of state here?
> +
> /**
> * SUBTEST: gt-c6-on-idle
> * Description: Validate GT C6 state on idle
> @@ -64,6 +85,10 @@ enum test_type {
> * SUBTEST: cpg-gt-toggle
> * Description: Toggle GT coarse power gating states by acquiring/releasing
> * forcewake.
> + *
> + * SUBTEST: aspm_link_residency
> + * Description: Check for PCIe ASPM (Active State Power Management) link states
> + * entry while device is in D0.
> */
> IGT_TEST_DESCRIPTION("Tests for gtidle properties");
>
> @@ -255,6 +280,24 @@ static void idle_residency_on_exec(int fd, struct drm_xe_engine_class_instance *
> munmap(done, 4096);
> }
>
> +static void do_exec(int fd, struct drm_xe_engine_class_instance *hwe)
> +{
> + unsigned long *done;
> +
> + igt_info("Running on %s:%d\n",
> + xe_engine_class_string(hwe->engine_class), hwe->engine_instance);
> + done = mmap(0, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
> + igt_assert(done != MAP_FAILED);
> + memset(done, 0, 4096);
> +
> + igt_fork(child, 1)
> + exec_load(fd, hwe, done);
> +
> + *done = 1;
> + igt_waitchildren();
> + munmap(done, 4096);
> +}
Can't we just run a spinner. Is this necessary?
> +
> static void measure_power(struct igt_power *gpu, double *power)
> {
> struct power_sample power_sample[2];
> @@ -370,6 +413,127 @@ static void cpg_gt_toggle(int fd)
> powergate_status(fd, gt, "down");
> }
>
> +static uint64_t get_link_state_residency(int fd_xe, const char *parse_str)
> +{
> + int fd_debugfs_dir = 0;
> + int ret = 0;
> + char *ptr = NULL;
> + char path[256] = {0}, buf[1024] = {0};
> + uint64_t residency = 0;
> +
> + fd_debugfs_dir = igt_debugfs_dir(fd_xe);
> + igt_assert(fd_debugfs_dir >= 0);
> +
> + ret = igt_debugfs_simple_read(fd_debugfs_dir, "dgfx_pcie_link_residencies" , buf, sizeof(buf));
Extra space
Keep the maximum length to 100 lines throught the file
> + igt_assert_f(ret >= 0, "cannot read link residency file\n");
> + ptr = strstr(buf, parse_str);
> + igt_assert_f((ptr != NULL), "cannot find residency string %s\n", parse_str);
> + if (ptr != NULL) {
if you are asserting why check again?
> + sprintf(path, "%s%%llu", parse_str);
> + igt_debug("searching for-%s\n", parse_str);
Why do we need these debug messages ?
> + sscanf(ptr + strlen(parse_str), "%lu", &residency);
> + igt_info("Link residency %"PRIu64"\n", residency);
> + }
> + close(fd_debugfs_dir);
> + return residency;
> +}
> +
> +static void save_and_disable_link_states(int fd_pci_usp)
> +{
> + int i = 0;
> + int ret = 0;
> + char path[256] = {0};
> +
> + for (i = 0 ; i < MAX_LINK_STATES ; i++) {
> + sprintf(path, "%s", link_state_sysfs[i].filename);
> + if (faccessat(fd_pci_usp, path, R_OK, 0)) {
Use existing sysfs functions
> + igt_debug("%s not present to save\n", path);
unnecessary
> + continue;
> + }
> + ret = igt_sysfs_scanf(fd_pci_usp, path, "%c", &link_state_sysfs[i].state);
> + igt_assert_lt(0, ret);
> + igt_debug("saved %s = %c\n", link_state_sysfs[i].filename , link_state_sysfs[i].state);
extra space
> +
> + ret = igt_sysfs_printf(fd_pci_usp, path, "%c", '0');
> + igt_assert_lt(0, ret);
> + }
> +}
> +
> +static void restore_link_states(int fd_pci_usp)
> +{
> + int i = 0;
> + int ret = 0;
> + char path[256] = {0};
> +
> + /* Restore saved states of L1 sysfs entries. */
> + for (i = 0 ; i < MAX_LINK_STATES ; i++) {
> + sprintf(path, "%s", link_state_sysfs[i].filename);
> + if (faccessat(fd_pci_usp, path, R_OK, 0)) {
> + igt_debug("%s not present to restore\n", path);
> + continue;
> + }
> + ret = igt_sysfs_printf(fd_pci_usp, path, "%c", link_state_sysfs[i].state);
extra space
> + igt_assert_lt(0, ret);
> + igt_debug("restored %s to %c\n", link_state_sysfs[i].filename , link_state_sysfs[i].state);
extra space
> + }
> +}
> +
> +static void test_aspm_link_residency(int fd_xe, uint8_t aspm_link_state)
use the enum
> +{
> + struct pci_device *pci_dev;
> + int fd_pci_usp = 0;
> + char name[PATH_MAX];
> + int ret = 0;
> + char path[256] = {0};
> + uint64_t residency_pre = 0, residency_post = 0;
> +
> + igt_assert(aspm_link_state <= LINK_STATE_ASPM_L1_2);
> +
> + /* Get upstream port pci_dev */
> + pci_dev = igt_device_get_pci_upstream_port(fd_xe);
> + igt_assert_f(pci_dev != NULL, "Couldn't get pci device of upstream port\n");
> + igt_debug("Upstream port PCI device: %04x:%02x:%02x.%01x\n", pci_dev->domain,
> + pci_dev->bus, pci_dev->dev, pci_dev->func);
not necessary
> +
> + snprintf(name, sizeof(name), "/sys/bus/pci/devices/%04x:%02x:%02x.%01x/link",
> + pci_dev->domain, pci_dev->bus, pci_dev->dev, pci_dev->func);
use igt_device_get_pci_slot_name
> + fd_pci_usp = open(name, O_DIRECTORY);
> + igt_assert_f((fd_pci_usp >= 0), "Can't open link directory upstream port %s, ret %d\n", name, fd_pci_usp);
> +
> + /* Disable runtime PM as link ASPM entry happens during device is in D0 only. */
> + igt_assert(igt_setup_runtime_pm(fd_xe));
> + igt_disable_runtime_pm();
> +
> + /* Check if ASPM sysfs is present. */
> + sprintf(path, "%s", link_state_sysfs[aspm_link_state].filename);
> + igt_require_f(!faccessat(fd_pci_usp, path, R_OK, 0), "%s is not present\n", path);
Use igt_sysfs_has_attr
> + ret = igt_sysfs_scanf(fd_pci_usp, path, "%c", &link_state_sysfs[aspm_link_state].state);
> + igt_assert_f((ret > 0), "couldn't read residency for %s", path);
Start with Caps
> +
> + /* Save current state of all available link sysfs entries and disable all link states. */
> + save_and_disable_link_states(fd_pci_usp);
> +
> + /* Enable only the ASPM link state needed for test. */
> + igt_debug("Enabling %s\n", link_state_sysfs[aspm_link_state].filename);
> + sprintf(path, "%s", link_state_sysfs[aspm_link_state].filename);
> + ret = igt_sysfs_printf(fd_pci_usp, path, "%c", '1');
> +
> + /* Read link state residencies before and after idle wait time. */
> + residency_pre = get_link_state_residency(fd_xe, link_state_sysfs[aspm_link_state].parse_str);
Wrap at 100 lines
> + igt_info("Waiting for link to enter idle....\n");
> + sleep(5);
> + residency_post = get_link_state_residency(fd_xe, link_state_sysfs[aspm_link_state].parse_str);
Wrap at 100 lines
Thanks
Riana
> +
> + /* Restore saved link states. */
> + restore_link_states(fd_pci_usp);
> +
> + igt_restore_runtime_pm();
> + close(fd_pci_usp);
> + close(fd_xe);
> +
> + igt_assert_f(residency_post > residency_pre, "ASPM entry failed, pre %"PRIu64", post %"PRIu64"\n",
> + residency_pre, residency_post);
> +}
> int igt_main()
> {
> uint32_t d3cold_allowed;
> @@ -444,6 +608,21 @@ int igt_main()
> cpg_gt_toggle(fd);
> }
>
> + igt_describe("ASPM Link residency validation");
> + igt_subtest_with_dynamic("aspm_link_residency") {
> + igt_require(xe_has_vram(fd));
> + xe_for_each_gt(fd, gt) {
> + xe_for_each_engine(fd, hwe) {
> + if (gt == hwe->gt_id && !hwe->engine_instance) {
> + igt_dynamic_f("gt%u-engine-%s", gt,
> + xe_engine_class_string(hwe->engine_class))
> + do_exec(fd, hwe);
> + }
> + }
> + }
> + test_aspm_link_residency(fd, LINK_STATE_ASPM);
> + }
> +
> igt_fixture() {
> close(fd);
> }
next prev parent reply other threads:[~2025-12-16 14:18 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-12-16 8:09 [PATCH i-g-t v8 0/2] tests/intel/xe_pm_residency: Add ASPM Link residency test Karthik Poosa
2025-12-16 8:09 ` [PATCH i-g-t v8 1/2] lib/igt_device: Add API to get pci device upstream port Karthik Poosa
2025-12-16 11:21 ` Nilawar, Badal
2025-12-16 8:09 ` [PATCH i-g-t v8 2/2] tests/intel/xe_pm_residency: Add subtest for ASPM Link state residency Karthik Poosa
2025-12-16 14:18 ` Riana Tauro [this message]
2025-12-17 6:34 ` Poosa, Karthik
2025-12-16 14:27 ` ✓ Xe.CI.BAT: success for tests/intel/xe_pm_residency: Add ASPM Link residency test Patchwork
2025-12-16 14:27 ` ✓ i915.CI.BAT: " Patchwork
2025-12-16 16:48 ` ✓ i915.CI.Full: " Patchwork
2025-12-17 8:49 ` ✗ Xe.CI.Full: failure " 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=b49b01e7-935c-402e-93c9-eaf1f4ecaca6@intel.com \
--to=riana.tauro@intel.com \
--cc=anshuman.gupta@intel.com \
--cc=badal.nilawar@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=kamil.konieczny@linux.intel.com \
--cc=karthik.poosa@intel.com \
--cc=rodrigo.vivi@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;
as well as URLs for NNTP newsgroup(s).