From: Rodrigo Vivi <rodrigo.vivi@intel.com>
To: Anshuman Gupta <anshuman.gupta@intel.com>
Cc: igt-dev@lists.freedesktop.org, badal.nilawar@intel.com
Subject: Re: [igt-dev] [PATCH i-g-t v2 2/2] test/xe_pm: Add vram_d3cold_threshold subtest
Date: Fri, 11 Aug 2023 17:09:29 -0400 [thread overview]
Message-ID: <ZNajiVS45lA404W7@intel.com> (raw)
In-Reply-To: <20230811111026.1327273-3-anshuman.gupta@intel.com>
On Fri, Aug 11, 2023 at 04:40:26PM +0530, Anshuman Gupta wrote:
> Adding a vram_d3cold_threshold subtest, which creates a Xe bo and
> set the vram_d3cold_threshold according to vram used and bo size.
> Test setups the d3cold and expect card to be limited to d3hot.
>
> v2:
> - Add subtest doc.
> v3:
> - skip the test on igfx. [Riana]
> - Test doc enhancement. [Riana]
> - Create the bo before vram query. [Riana]
> - Use xe_bo_map insead of xe_bo_mmap_offset and mmap. [Riana]
> - Close the bo handle. [Riana]
> v3:
> - Restore the vram_d3cold_threshold value. [Badal]
> - Don't fail the test if there is no vram_d3cold_threshold.
> - Test d3cold after closing the xe BO.
> v4:
> - When d3cold is not allowed, don't use out_of_d3() as device will
> be runtime suspended.
> - Set vram_d3cold_threshold to vram_used + Xe BO size such that
> it can enter to d3cold after freeing up the Xe BO.
> - Use forcewake handle to wake the device.
>
> Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
> ---
> tests/xe/xe_pm.c | 128 +++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 128 insertions(+)
>
> diff --git a/tests/xe/xe_pm.c b/tests/xe/xe_pm.c
> index 23b8246ed..480382f9e 100644
> --- a/tests/xe/xe_pm.c
> +++ b/tests/xe/xe_pm.c
> @@ -19,6 +19,7 @@
> #include "igt.h"
> #include "lib/igt_device.h"
> #include "lib/igt_pm.h"
> +#include "lib/igt_sysfs.h"
> #include "lib/igt_syncobj.h"
> #include "lib/intel_reg.h"
>
> @@ -30,12 +31,16 @@
> #define NO_SUSPEND -1
> #define NO_RPM -1
>
> +#define SIZE (4096 * 1024)
> +
> typedef struct {
> int fd_xe;
> struct pci_device *pci_xe;
> struct pci_device *pci_root;
> } device_t;
>
> +uint64_t orig_threshold;
> +
> /* runtime_usage is only available if kernel build CONFIG_PM_ADVANCED_DEBUG */
> static bool runtime_usage_available(struct pci_device *pci)
> {
> @@ -76,6 +81,49 @@ static void set_d3cold_allowed(struct pci_device *pci,
> close(fd);
> }
>
> +static uint64_t get_vram_d3cold_threshold(int sysfs)
> +{
> + uint64_t threshold;
> + char path[64];
> + int ret;
> +
> + sprintf(path, "device/vram_d3cold_threshold");
> + igt_require_f(!faccessat(sysfs, path, R_OK, 0), "vram_d3cold_threshold is not present\n");
> +
> + ret = igt_sysfs_scanf(sysfs, path, "%lu", &threshold);
> + igt_assert(ret > 0);
> +
> + return threshold;
> +}
> +
> +static void set_vram_d3cold_threshold(int sysfs, uint64_t threshold)
> +{
> + char path[64];
> + int ret;
> +
> + sprintf(path, "device/vram_d3cold_threshold");
> +
> + if (!faccessat(sysfs, path, R_OK | W_OK, 0))
> + ret = igt_sysfs_printf(sysfs, path, "%lu", threshold);
> + else
> + igt_warn("vram_d3cold_threshold is not present\n");
> +
> + igt_assert(ret > 0);
> +}
> +
> +static void vram_d3cold_threshold_restore(int sig)
> +{
> + int fd, sysfs_fd;
> +
> + fd = drm_open_driver(DRIVER_XE);
> + sysfs_fd = igt_sysfs_open(fd);
> +
> + set_vram_d3cold_threshold(sysfs_fd, orig_threshold);
> +
> + close(sysfs_fd);
> + close(fd);
> +}
> +
> static bool setup_d3(device_t device, enum igt_acpi_d_state state)
> {
> switch (state) {
> @@ -341,11 +389,82 @@ NULL));
> igt_assert(in_d3(device, d_state));
> }
>
> +/**
> + * SUBTEST: vram-d3cold-threshold
> + * Description:
> + * Validate whether card is limited to d3hot while vram used
> + * is greater than vram_d3cold_threshold.
> + * Run type: FULL
> + */
> +static void test_vram_d3cold_threshold(device_t device, int sysfs_fd)
> +{
> + struct drm_xe_query_mem_usage *mem_usage;
> + struct drm_xe_device_query query = {
> + .extensions = 0,
> + .query = DRM_XE_DEVICE_QUERY_MEM_USAGE,
> + .size = 0,
> + .data = 0,
> + };
> + uint64_t vram_used_mb = 0, vram_total_mb = 0, threshold;
> + uint32_t bo, flags;
> + int handle, i;
> + bool active;
> + void *map;
> +
> + igt_require(xe_has_vram(device.fd_xe));
> +
> + flags = vram_memory(device.fd_xe, 0);
> + igt_require_f(flags, "Device doesn't support vram memory region\n");
> +
> + igt_assert_eq(igt_ioctl(device.fd_xe, DRM_IOCTL_XE_DEVICE_QUERY, &query), 0);
> + igt_assert_neq(query.size, 0);
> +
> + mem_usage = malloc(query.size);
> + igt_assert(mem_usage);
> +
> + query.data = to_user_pointer(mem_usage);
> + igt_assert_eq(igt_ioctl(device.fd_xe, DRM_IOCTL_XE_DEVICE_QUERY, &query), 0);
> +
> + for (i = 0; i < mem_usage->num_regions; i++) {
> + if (mem_usage->regions[i].mem_class == XE_MEM_REGION_CLASS_VRAM) {
> + vram_used_mb += (mem_usage->regions[i].used / (1024 * 1024));
> + vram_total_mb += (mem_usage->regions[i].total_size / (1024 * 1024));
> + }
> + }
> +
> + threshold = vram_used_mb + (SIZE / 1024 /1024);
> + igt_require(threshold < vram_total_mb);
> +
> + bo = xe_bo_create_flags(device.fd_xe, 0, SIZE, flags);
> + map = xe_bo_map(device.fd_xe, bo, SIZE);
> + memset(map, 0, SIZE);
> + munmap(map, SIZE);
> + set_vram_d3cold_threshold(sysfs_fd, threshold);
> +
> + /* Setup D3Cold but card should be in D3hot */
> + igt_assert(setup_d3(device, IGT_ACPI_D3Cold));
> + sleep(1);
> + igt_assert(in_d3(device, IGT_ACPI_D3Hot));
> + igt_assert(igt_pm_get_acpi_real_d_state(device.pci_root) == IGT_ACPI_D0);
> + gem_close(device.fd_xe, bo);
> +
> + /* open and close fw handle to wake the device */
hmmm I would expect that we should wake the device with mem_access when cleaning
up the stuff...
Maybe I'm wrong, but I'm in the process of moving the mem_access to xe_pm,
documenting it and taking care of other suspicious cases like xe_exec without VM.
https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/551 assigned for myself.
But anyway, while I don't get to that, let's move with this patch as is.
Maybe adding an "XXX" along with the comment line above? up to you.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
> + handle = igt_debugfs_open(device.fd_xe, "forcewake_all", O_RDONLY);
> + igt_assert(handle >= 0);
> + active = igt_get_runtime_pm_status() == IGT_RUNTIME_PM_STATUS_ACTIVE;
> + close(handle);
> + igt_assert(active);
> +
> + /* Test D3Cold again after freeing up the Xe BO */
> + igt_assert(in_d3(device, IGT_ACPI_D3Cold));
> +}
> +
> igt_main
> {
> struct drm_xe_engine_class_instance *hwe;
> device_t device;
> char d3cold_allowed[2];
> + int sysfs_fd;
> const struct s_state {
> const char *name;
> enum igt_suspend_state state;
> @@ -378,6 +497,7 @@ igt_main
>
> get_d3cold_allowed(device.pci_xe, d3cold_allowed);
> igt_assert(igt_setup_runtime_pm(device.fd_xe));
> + sysfs_fd = igt_sysfs_open(device.fd_xe);
> }
>
> for (const struct s_state *s = s_states; s->name; s++) {
> @@ -437,7 +557,15 @@ igt_main
> }
> }
>
> + igt_describe("Validate whether card is limited to d3hot, if vram used > vram threshold");
> + igt_subtest("vram-d3cold-threshold") {
> + orig_threshold = get_vram_d3cold_threshold(sysfs_fd);
> + igt_install_exit_handler(vram_d3cold_threshold_restore);
> + test_vram_d3cold_threshold(device, sysfs_fd);
> + }
> +
> igt_fixture {
> + close(sysfs_fd);
> set_d3cold_allowed(device.pci_xe, d3cold_allowed);
> igt_restore_runtime_pm();
> xe_device_put(device.fd_xe);
> --
> 2.25.1
>
next prev parent reply other threads:[~2023-08-11 21:09 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-08-11 11:10 [igt-dev] [PATCH i-g-t v2 0/2] Add vram_d3cold_threshold test Anshuman Gupta
2023-08-11 11:10 ` [igt-dev] [PATCH i-g-t v2 1/2] lib/igt_pm: Remove pci_system_{init, cleanup} Anshuman Gupta
2023-08-11 11:10 ` [igt-dev] [PATCH i-g-t v2 2/2] test/xe_pm: Add vram_d3cold_threshold subtest Anshuman Gupta
2023-08-11 21:09 ` Rodrigo Vivi [this message]
2023-08-14 3:51 ` Gupta, Anshuman
2023-08-14 6:21 ` Anshuman Gupta
2023-08-11 11:20 ` [igt-dev] ✗ GitLab.Pipeline: warning for Add vram_d3cold_threshold test (rev3) Patchwork
2023-08-11 11:47 ` [igt-dev] ○ CI.xeBAT: info " Patchwork
2023-08-11 11:56 ` [igt-dev] ✓ Fi.CI.BAT: success " Patchwork
2023-08-12 15:04 ` [igt-dev] ✗ Fi.CI.IGT: failure " Patchwork
-- strict thread matches above, loose matches on Subject: below --
2023-06-28 11:26 [igt-dev] [PATCH i-g-t v2 0/2] vram d3cold threshold test Anshuman Gupta
2023-06-28 11:26 ` [igt-dev] [PATCH i-g-t v2 2/2] test/xe_pm: Add vram_d3cold_threshold subtest Anshuman Gupta
2023-07-05 9:19 ` 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=ZNajiVS45lA404W7@intel.com \
--to=rodrigo.vivi@intel.com \
--cc=anshuman.gupta@intel.com \
--cc=badal.nilawar@intel.com \
--cc=igt-dev@lists.freedesktop.org \
/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