From: "Poosa, Karthik" <karthik.poosa@intel.com>
To: Kamil Konieczny <kamil.konieczny@linux.intel.com>,
<igt-dev@lists.freedesktop.org>, <anshuman.gupta@intel.com>,
<badal.nilawar@intel.com>, <riana.tauro@intel.com>,
<rodrigo.vivi@intel.com>
Subject: Re: [PATCH i-g-t v10 2/2] tests/intel/xe_pm_residency: Add subtest for ASPM Link state residency
Date: Fri, 23 Jan 2026 18:05:01 +0530 [thread overview]
Message-ID: <126b57d8-9441-47e0-aa53-3e887a0e2210@intel.com> (raw)
In-Reply-To: <20260123120703.uraxy7njgzi47uxh@kamilkon-DESK.igk.intel.com>
[-- Attachment #1: Type: text/plain, Size: 12673 bytes --]
On 23-01-2026 17:37, Kamil Konieczny wrote:
> Hi,
> On 2026-01-23 at 16:04:11 +0530, Poosa, Karthik wrote:
>> On 22-01-2026 20:30, Kamil Konieczny wrote:
>>> Hi Karthik,
>>> On 2026-01-21 at 20:27:47 +0530, 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 the 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.
>>>>
>>>> v8:
>>>> - Address below review comments from Riana.
>>>> - Use igt_sysfs_has_attr() instead of faccess().
>>>> - Remove unnecessary spaces, debug logs, if checks.
>>>> - Wrap line length to 100 chars.
>>>> - Use spinner instead of mmap for workload.
>>>>
>>>> v9:
>>>> - Address review comments from Kamil.
>>>> - Simplify couple of igt_asserts.
>>>> - Remove extra spaces.
>>>>
>>>> v10:
>>>> - Add exit handler for aspm_link_residency test. (Riana)
>>>> - Remove unused and unnecessary variables.
>>>>
>>>> Signed-off-by: Karthik Poosa<karthik.poosa@intel.com>
>>>> Reviewed-by: Badal Nilawar<badal.nilawar@intel.com>
>>>> ---
>>>> tests/intel/xe_pm_residency.c | 196 ++++++++++++++++++++++++++++++++++
>>>> 1 file changed, 196 insertions(+)
>>>>
>>>> diff --git a/tests/intel/xe_pm_residency.c b/tests/intel/xe_pm_residency.c
>>>> index d33a87b13..2d965ef7d 100644
>>>> --- a/tests/intel/xe_pm_residency.c
>>>> +++ b/tests/intel/xe_pm_residency.c
>>>> @@ -31,12 +31,36 @@
>>>> const double tolerance = 0.1;
>>>> int fw_handle = -1;
>>>> +int fd_pci_usp;
>>> int fd_pci_usp = -1;
>>>
>>>> +bool rpm_disabled;
>>>> +bool link_states_saved;
>>>> enum test_type {
>>>> TEST_S2IDLE,
>>>> 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[] = {
>>>> + { "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}, +}; + /** * SUBTEST: gt-c6-on-idle * Description:
>>>> Validate GT C6 state on idle @@ -64,6 +88,10 @@ 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 +283,21 @@ static void idle_residency_on_exec(int fd, struct drm_xe_engine_class_instance *
>>>> munmap(done, 4096);
>>>> }
>>>> +static void do_spin(int fd, struct drm_xe_engine_class_instance *eci)
>>>> +{
>>>> + igt_spin_t *spin;
>>>> + uint64_t vm, ahnd;
>>>> +
>>>> + igt_info("Running spinner on %s:%d\n",
>>>> + xe_engine_class_string(eci->engine_class), eci->engine_instance);
>>>> + vm = xe_vm_create(fd, 0, 0);
>>>> + intel_allocator_init();
>>>> + ahnd = intel_allocator_open(fd, 0, INTEL_ALLOCATOR_RELOC);
>>>> + spin = igt_spin_new(fd, .ahnd = ahnd, .vm = vm, .hwe = eci);
>>>> + igt_measured_usleep(USEC_PER_SEC);
>>>> + igt_spin_free(fd, spin);
>>>> +}
>>>> +
>>>> static void measure_power(struct igt_power *gpu, double *power)
>>>> {
>>>> struct power_sample power_sample[2];
>>>> @@ -370,6 +413,143 @@ 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 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));
>>>> + igt_assert_f(ret >= 0, "Cannot read residency file dgfx_pcie_link_residencies, ret %d\n",
>>>> + ret);
>>>> +
>>>> + ptr = strstr(buf, parse_str);
>>>> + igt_assert_f(ptr, "Cannot find residency string %s\n", parse_str);
>>>> + ret = sscanf(ptr + strlen(parse_str), "%lu", &residency);
>>>> + igt_assert_f(ret > 0, "Couldn't read residency value, ret %d", ret);
>>>> + igt_info("Link residency %"PRIu64"\n", residency);
>>>> + close(fd_debugfs_dir);
>>>> +
>>>> + return residency;
>>>> +}
>>>> +
>>>> +static void save_and_disable_link_states(void)
>>>> +{
>>>> + 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 (!igt_sysfs_has_attr(fd_pci_usp, path))
>>>> + 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);
>>>> + ret = igt_sysfs_printf(fd_pci_usp, path, "%c", '0');
>>> Disable should be done in separate function, so there should be two:
>>>
>>> static void save_link_states(void)
>>> static void disable_link_states(void)
>> any reason for this Kamil, it will be unnecessary extra function.
> Because you mix here reading and setting. imho there could be
> a problem if setting first few succeeds but in middle fails.
ok
>
>
>>>> + igt_assert_lt(0, ret);
>>>> + }
>>>> + link_states_saved = true;
>>>> +}
>>>> +
>>>> +static void restore_link_states(void)
>>>> +{
>>>> + int i = 0;
>>>> + int ret = 0;
>>>> + char path[256] = {0};
>>>> +
>>>> + if (link_states_saved && fd_pci_usp >= 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 (!igt_sysfs_has_attr(fd_pci_usp, path))
>>>> + continue;
>>>> + ret = igt_sysfs_printf(fd_pci_usp, path, "%c", link_state_sysfs[i].state);
>>>> + igt_assert_lt(0, ret);
>>>> + igt_debug("restored %s to %c\n", link_state_sysfs[i].filename,
>>>> + link_state_sysfs[i].state);
>>>> + }
>>>> + link_states_saved = false;
>>>> + }
>>>> +}
>>>> +
>>>> +static void test_aspm_link_residency(int fd_xe, enum link_state_index aspm_link_state)
>>>> +{
>>>> + struct pci_device *pci_dev;
>>>> + char name[PATH_MAX];
>>>> + int ret = 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, "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);
>>>> +
>>>> + 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);
>>>> + 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();
>>>> + rpm_disabled = true;
>>>> +
>>>> + /* Check if ASPM sysfs is present. */
>>>> + igt_require_f(igt_sysfs_has_attr(fd_pci_usp, link_state_sysfs[aspm_link_state].filename),
>>>> + "%s is not present\n", link_state_sysfs[aspm_link_state].filename);
>>>> + ret = igt_sysfs_scanf(fd_pci_usp, link_state_sysfs[aspm_link_state].filename, "%c",
>>>> + &link_state_sysfs[aspm_link_state].state);
>>>> + igt_assert_f((ret > 0), "Couldn't read residency for %s",
>>>> + link_state_sysfs[aspm_link_state].filename);
>>>> +
>>>> + /* Save current state of all available link sysfs entries and disable all link states. */
>>>> + save_and_disable_link_states();
>>>> +
>>>> + /* Enable only the ASPM link state needed for test. */
>>>> + igt_debug("Enabling %s\n", link_state_sysfs[aspm_link_state].filename);
>>>> + ret = igt_sysfs_printf(fd_pci_usp, link_state_sysfs[aspm_link_state].filename, "%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);
>>>> + igt_info("Waiting for link to enter idle....\n");
>>>> + sleep(SLEEP_DURATION);
>>>> + residency_post = get_link_state_residency(fd_xe,
>>>> + link_state_sysfs[aspm_link_state].parse_str);
>>>> +
>>>> + restore_link_states();
>>>> + igt_restore_runtime_pm();
>>>> + rpm_disabled = false;
>>>> + 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);
>>>> +}
>>>> +
>>>> +static void aspm_residency_exit_handler(int sig)
>>>> +{
>>>> + restore_link_states();
>>>> + if (rpm_disabled) {
>>>> + igt_restore_runtime_pm();
>>>> + rpm_disabled = false;
>>>> + }
>>>> + if (fd_pci_usp)
>>> if (fd_pci_usp != -1)
>>>
>>>> + close(fd_pci_usp);
>>> Add:
>>> fd_pci_usp = -1;
>>>
>>>> +}
>>>> +
>>>> int igt_main()
>>>> {
>>>> uint32_t d3cold_allowed;
>>>> @@ -444,6 +624,22 @@ int igt_main()
>>>> cpg_gt_toggle(fd);
>>>> }
>>>> + igt_describe("ASPM Link residency validation");
>>>> + igt_subtest_with_dynamic("aspm_link_residency") {
>>> This should be igt_subtest():
>>>
>>> igt_subtest("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))
>>> What are you testing here? If you need to start and stop spinner
>>> please describe here why. If you really need it, it should be:
>> we want to run a basic workload before testing ASPM entry.
>>
>> So we are using spinner as workload here.
> ok, but why do you need any workload? It should also work without it?
Yes, it works even workload also.
This was suggested by Anshuman,
I believe this is intended to wake up the GPU and ensure the engines
execute some workload before entering ASPM.
>
>>> if (gt == hwe->gt_id && !hwe->engine_instance)
>>> do_spin(fd, hwe);
>>>
>>>> + do_spin(fd, hwe);
>>>> + }
>>>> + }
>>>> + }
>>> Add newline.
>>>
>>>> + igt_install_exit_handler(aspm_residency_exit_handler);
>>>> + test_aspm_link_residency(fd, LINK_STATE_ASPM);
>>> Btw any igt_assert outside of dynamic sub-subtest here will result
>>> in SIGABORT, not a proper test fail.
>>>
>>> Add cleanup here:
>>> aspm_residency_exit_handler(0);
>> but installing exit handler already does this !
> It could become a problem when someone adds new test after this one.
>
> Regards,
> Kamil
>
>>>> + }
>>>> +
>>>> igt_fixture() {
>>>> close(fd);
>>>> }
>>>> --
>>>> 2.25.1
>>>>
[-- Attachment #2: Type: text/html, Size: 15662 bytes --]
next prev parent reply other threads:[~2026-01-23 12:35 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-01-21 14:57 [PATCH i-g-t v10 0/2] tests/intel/xe_pm_residency: Add ASPM Link residency test Karthik Poosa
2026-01-21 14:57 ` [PATCH i-g-t v10 1/2] lib/igt_device: Add API to get pci device upstream port Karthik Poosa
2026-01-21 14:57 ` [PATCH i-g-t v10 2/2] tests/intel/xe_pm_residency: Add subtest for ASPM Link state residency Karthik Poosa
2026-01-22 15:00 ` Kamil Konieczny
2026-01-23 10:34 ` Poosa, Karthik
2026-01-23 12:07 ` Kamil Konieczny
2026-01-23 12:35 ` Poosa, Karthik [this message]
2026-01-23 14:06 ` Kamil Konieczny
2026-02-04 5:46 ` Poosa, Karthik
2026-01-21 16:55 ` ✗ Xe.CI.BAT: failure for tests/intel/xe_pm_residency: Add ASPM Link residency test Patchwork
2026-01-21 17:15 ` ✓ i915.CI.BAT: success " Patchwork
2026-01-22 2:05 ` ✗ Xe.CI.Full: failure " Patchwork
2026-01-22 3:52 ` ✗ i915.CI.Full: " Patchwork
-- strict thread matches above, loose matches on Subject: below --
2026-01-06 13:51 [PATCH i-g-t v10 0/2] " Karthik Poosa
2026-01-06 13:51 ` [PATCH i-g-t v10 2/2] tests/intel/xe_pm_residency: Add subtest for ASPM Link state residency Karthik Poosa
2026-01-07 6:24 ` Riana Tauro
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=126b57d8-9441-47e0-aa53-3e887a0e2210@intel.com \
--to=karthik.poosa@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=riana.tauro@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