Re: [igt-dev] [PATCH i-g-t v3 3/3] test/xe_pm: Add vram_d3cold_threshold subtest

["Vivi, Rodrigo" <rodrigo.vivi@intel.com>]

On Fri, 2023-07-21 at 07:22 +0000, Gupta, Anshuman wrote:
> 
> 
> > -----Original Message-----
> > From: Vivi, Rodrigo <rodrigo.vivi@intel.com>
> > Sent: Friday, July 21, 2023 2:38 AM
> > To: Gupta, Anshuman <anshuman.gupta@intel.com>
> > Cc: igt-dev@lists.freedesktop.org; Dixit, Ashutosh
> > <ashutosh.dixit@intel.com>; Nilawar, Badal
> > <badal.nilawar@intel.com>;
> > Tauro, Riana <riana.tauro@intel.com>
> > Subject: Re: [PATCH i-g-t v3 3/3] test/xe_pm: Add
> > vram_d3cold_threshold
> > subtest
> > 
> > On Tue, Jul 18, 2023 at 04:38:52PM +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]
> > > 
> > > Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
> > > ---
> > >  tests/xe/xe_pm.c | 82
> > > ++++++++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 82 insertions(+)
> > > 
> > > diff --git a/tests/xe/xe_pm.c b/tests/xe/xe_pm.c index
> > > 11a4c3705..a8257b13c 100644
> > > --- a/tests/xe/xe_pm.c
> > > +++ b/tests/xe/xe_pm.c
> > > @@ -20,6 +20,7 @@
> > >  #include "lib/igt_device.h"
> > >  #include "lib/igt_kms.h"
> > >  #include "lib/igt_pm.h"
> > > +#include "lib/igt_sysfs.h"
> > >  #include "lib/igt_syncobj.h"
> > >  #include "lib/intel_reg.h"
> > > 
> > > @@ -31,6 +32,8 @@
> > >  #define NO_SUSPEND -1
> > >  #define NO_RPM -1
> > > 
> > > +#define SIZE (4096 * 1024)
> > > +
> > >  typedef struct {
> > >         int fd_xe;
> > >         struct pci_device *pci_xe;
> > > @@ -77,6 +80,22 @@ static void set_d3cold_allowed(struct
> > > pci_device
> > *pci,
> > >         close(fd);
> > >  }
> > > 
> > > +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_info("ret value %d", ret);
> > > +       igt_assert(ret > 0);
> > > +}
> > > +
> > >  static bool setup_d3(device_t device, enum igt_acpi_d_state
> > > state)  {
> > >         igt_dpms_turn_off_display(device.fd_xe);
> > > @@ -349,6 +368,64 @@ 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) {
> > > +       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;
> > > +       void *map;
> > > +       int i, sysfs_fd;
> > > +
> > > +       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;
> > > +       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);
> > > +       sysfs_fd = igt_sysfs_open(device.fd_xe);
> > > +       set_vram_d3cold_threshold(sysfs_fd, threshold);
> > > +       close(sysfs_fd);
> > > +
> > > +       /* Setup D3Cold but card should be in D3hot */
> > > +       igt_assert(setup_d3(device, IGT_ACPI_D3Cold));
> > 
> > should we add some wait here to ensure we gave the proper time for
> > everything to settle?
> > 
> > > +       igt_assert(in_d3(device, IGT_ACPI_D3Hot));
> > > +       igt_assert(out_of_d3(device, IGT_ACPI_D3Cold));
> > > +       gem_close(device.fd_xe, bo);
> > > +}
> > > +
> > >  igt_main
> > >  {
> > >         struct drm_xe_engine_class_instance *hwe; @@ -449,6
> > > +526,11 @@
> > > igt_main
> > >                 }
> > >         }
> > > 
> > > +       igt_describe("Validate whether card is limited to d3hot,
> > > if vram used
> > > vram threshold");
> > > +       igt_subtest("vram-d3cold-threshold") {
> > > +               test_vram_d3cold_threshold(device);
> > > +       }
> > 
> > we probably need the other test and confirming that we are really
> > getting
> > d3cold after touching this sysfs entry.
> > 
> > With the kernel series I just sent that should work.
> Shall we improvise this test to test d3cold again after freeing the
> xe bo ?

this sounds like a great idea!

> Thanks,
> Anshuman.
> > 
> > but let's start with this:
> > 
> > Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
> > 
> > > +
> > >         igt_fixture {
> > >                 set_d3cold_allowed(device.pci_xe,
> > > d3cold_allowed);
> > >                 igt_restore_runtime_pm();
> > > --
> > > 2.25.1
> > >