Igt-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Wang, X" <x.wang@intel.com>
To: Zongyao Bai <zongyao.bai@intel.com>, <igt-dev@lists.freedesktop.org>
Cc: <jia.yao@intel.com>
Subject: Re: [PATCH 3/3] tests/intel/xe_pat: Add dynamic PAT index selection for pat-index-auto
Date: Tue, 12 May 2026 15:44:48 -0700	[thread overview]
Message-ID: <4f68b11d-0dca-4b8c-bdb9-9dce30523aab@intel.com> (raw)
In-Reply-To: <20260511233622.2074382-4-zongyao.bai@intel.com>



On 5/11/2026 16:36, Zongyao Bai wrote:
> Introduce the `pat-index-auto` dynamic subtest to iteratively test
> PAT indices based on the hardware configuration, improving test
> coverage beyond hardcoded arrays.
>
> Build the test mode set by:
>      selecting a small common set first (WT, coherent, non-coherent)
>      adding randomized remaining valid entries
>
> Add pat-index-auto to xe.fast-feedback.testlist.
>
> Signed-off-by: Zongyao Bai <zongyao.bai@intel.com>
> ---
>   tests/intel-ci/xe.fast-feedback.testlist |   1 +
>   tests/intel/xe_pat.c                     | 158 +++++++++++++++++++++++
>   2 files changed, 159 insertions(+)
>
> diff --git a/tests/intel-ci/xe.fast-feedback.testlist b/tests/intel-ci/xe.fast-feedback.testlist
> index acff025fa..6851c1f49 100644
> --- a/tests/intel-ci/xe.fast-feedback.testlist
> +++ b/tests/intel-ci/xe.fast-feedback.testlist
> @@ -211,6 +211,7 @@ igt@xe_pat@pat-index-xelp
>   igt@xe_pat@pat-index-xehpc
>   igt@xe_pat@pat-index-xelpg
>   igt@xe_pat@pat-index-xe2
> +igt@xe_pat@pat-index-auto
>   igt@xe_waitfence@abstime
>   igt@xe_waitfence@engine
>   igt@xe_waitfence@reltime
> diff --git a/tests/intel/xe_pat.c b/tests/intel/xe_pat.c
> index 79f9a87aa..957f843ec 100644
> --- a/tests/intel/xe_pat.c
> +++ b/tests/intel/xe_pat.c
> @@ -35,6 +35,7 @@
>   
>   #define XE_COH_NONE          1
>   #define XE_COH_AT_LEAST_1WAY 2
> +#define PAT_NAME_MAX_LEN     64
>   
>   /*
>    * PAT index 18: XA (eXclusive Access) + UC (Uncached).
> @@ -113,6 +114,7 @@ static void userptr_coh_none(int fd)
>   
>   #define L3_CACHE_POLICY_WB	0
>   #define L3_CACHE_POLICY_XD	1
> +#define L3_CACHE_POLICY_XA	2
>   #define L3_CACHE_POLICY_UC	3
>   
>   #define L4_CACHE_POLICY_WB	0
> @@ -1345,6 +1347,147 @@ const struct pat_index_entry bmg_g21_pat_index_modes[] = {
>   	{ NULL, 27, false, "c2-2way",     XE_COH_AT_LEAST_1WAY       },
>   };
>   
> +static void add_pat_mode(struct pat_index_entry *modes_arr, int idx,
> +			 uint32_t pat, bool has_comp_en, int pat_idx, bool cpu_wc)
> +{
> +	modes_arr[idx].get_pat_index = NULL;
> +	modes_arr[idx].pat_index = pat_idx;
> +	modes_arr[idx].compressed = has_comp_en && !!(pat & XE2_COMP_EN);
> +	modes_arr[idx].name = NULL;
> +	modes_arr[idx].coh_mode = REG_FIELD_GET(XE2_COH_MODE, pat) >= COH_MODE_1WAY ?
> +				  XE_COH_AT_LEAST_1WAY : XE_COH_NONE;
> +	modes_arr[idx].force_cpu_wc = cpu_wc;
> +}
> +
> +/*
> + * Build pat_index_modes[] test table dynamically from hardware PAT table:
> + * 1. Read PAT entries from hardware
> + * 2. Pick common modes first (WT, Coherent, Non-coherent)
> + * 3. Pick random modes (shuffled remaining entries)
> + * 4. Return the modes array and count through output parameters
> + */
> +static int init_pat_index_modes(int fd, struct pat_index_entry **modes)
> +{
> +	const int TEST_MODE_CNT = 12;
 From the CI results, it seems the test is quit fast , why the 
TEST_MODE_CNT limited to 12 ?


      New IGT tests (4)

  *

    igt@xe_pat@pat-index-auto:

      o Statuses : 9 pass(s) 3 skip(s)
      o Exec time: [0.0, 12.56] s
  *

    igt@xe_pat@pat-index-auto@blt:

      o Statuses : 9 pass(s)
      o Exec time: [1.49, 4.86] s
  *

    igt@xe_pat@pat-index-auto@dw:

      o Statuses : 9 pass(s)
      o Exec time: [0.71, 1.98] s
  *

    igt@xe_pat@pat-index-auto@render:

      o Statuses : 9 pass(s)
      o Exec time: [1.91, 5.79] s
      o


> +	const int TEST_COMMON_MODE_MAX = 4;
> +
> +	/* Static storage for results - persists after function returns */
> +	static struct pat_index_entry pat_index_modes[XE_PAT_MAX_ENTRIES * 2];
> +	static struct intel_pat_cache pat_sw_config;
> +	static char name_storage[XE_PAT_MAX_ENTRIES * 2][PAT_NAME_MAX_LEN];
> +
> +	int n_entries;
> +	bool has_comp_en;
> +
> +	int test_cnt = igt_run_in_simulation() ? TEST_MODE_CNT / 2 : TEST_MODE_CNT;
> +	struct pat_index_entry common_modes[TEST_COMMON_MODE_MAX];
> +	struct pat_index_entry random_modes[XE_PAT_MAX_ENTRIES * 2];
> +	struct pat_index_entry tmp;
> +	int n_common = 0;
> +	int n_random = 0;
> +	int n_picked = 0;
> +	int i, j;
> +
> +	bool ensure_picked_wt = true;
> +	bool ensure_picked_coh = true;
> +	bool ensure_picked_noncoh = true;
> +
> +	/* Fetch PAT config from hardware */
> +	n_entries = xe_fetch_pat_sw_config(fd, &pat_sw_config);
> +	has_comp_en = HAS_FLATCCS(intel_get_drm_devid(fd));
> +
> +	igt_assert(n_entries > 0);
> +
> +	/* Step 1: Split entries into common_modes and random_modes */
> +	for (i = 0; i < n_entries; i++) {
> +		uint32_t pat = pat_sw_config.entries[i].pat;
> +		uint8_t pat_clos = REG_FIELD_GET(XE2_L3_CLOS, pat);
> +		uint8_t l3_policy = REG_FIELD_GET(XE2_L3_POLICY, pat);
> +		uint8_t l4_policy = REG_FIELD_GET(XE2_L4_POLICY, pat);
> +		uint8_t coh_mode = REG_FIELD_GET(XE2_COH_MODE, pat);
> +
> +		if (pat_sw_config.entries[i].rsvd)
> +			continue;
> +
> +		/* Common pat index type 1: WT mode (l3=xd, l4=wt, coh=0) */
> +		if (ensure_picked_wt &&
> +		    pat_clos == 0 &&
> +		    l3_policy == L3_CACHE_POLICY_XD &&
> +		    l4_policy == L4_CACHE_POLICY_WT &&
> +		    coh_mode == 0) {
> +			add_pat_mode(common_modes, n_common, pat, has_comp_en, i, false);
> +			n_common++;
> +			ensure_picked_wt = false;
> +			continue;
> +		}
> +		/* Common pat index type 2: Coherent mode (l3=wb, l4=uc, coh>=1way) */
> +		if (ensure_picked_coh &&
> +		    pat_clos == 0 &&
> +		    l3_policy == L3_CACHE_POLICY_WB &&
> +		    l4_policy == L4_CACHE_POLICY_UC &&
> +		    coh_mode >= COH_MODE_1WAY) {
> +			add_pat_mode(common_modes, n_common, pat, has_comp_en, i, false);
> +			n_common++;
> +			add_pat_mode(common_modes, n_common, pat, has_comp_en, i, true);
> +			n_common++;
> +			ensure_picked_coh = false;
> +			continue;
> +		}
> +		/* Common pat index type 3: Non-coherent mode (l3=uc, l4=uc, coh=0) */
> +		if (ensure_picked_noncoh &&
> +		    pat_clos == 0 &&
> +		    l3_policy == L3_CACHE_POLICY_UC &&
> +		    l4_policy == L4_CACHE_POLICY_UC &&
> +		    coh_mode == 0) {
> +			add_pat_mode(common_modes, n_common, pat, has_comp_en, i, false);
> +			n_common++;
> +			ensure_picked_noncoh = false;
> +			continue;
> +		}
> +
> +		/* Add to random_modes */
> +		add_pat_mode(random_modes, n_random, pat, has_comp_en, i, false);
> +		n_random++;
> +		if (coh_mode >= COH_MODE_1WAY) {
> +			add_pat_mode(random_modes, n_random, pat, has_comp_en, i, true);
> +			n_random++;
> +		}
> +	}
> +	igt_assert(n_common + n_random > 0);
> +
> +	/* Step 2: Fisher-Yates shuffle on random_modes */
> +	for (i = n_random - 1; i > 0; i--) {
> +		j = random() % (i + 1);
> +		tmp = random_modes[i];
> +		random_modes[i] = random_modes[j];
> +		random_modes[j] = tmp;
> +	}
> +
> +	/* Step 3: Merge common_modes and random_modes into pat_index_modes (up to test_cnt) */
> +	for (i = 0; i < n_common && n_picked < test_cnt; i++)
> +		pat_index_modes[n_picked++] = common_modes[i];
> +	for (i = 0; i < n_random && n_picked < test_cnt; i++)
> +		pat_index_modes[n_picked++] = random_modes[i];
> +
> +	/* Step 4: Assign simple numeric names and print debug info */
> +	for (i = 0; i < n_picked; i++) {
> +		snprintf(name_storage[i], PAT_NAME_MAX_LEN, "PAT_INDEX_%u%s",
> +			 pat_index_modes[i].pat_index,
> +			 pat_index_modes[i].force_cpu_wc ? "-force_cpu_wc" : "");
> +
> +		pat_index_modes[i].name = name_storage[i];
> +
> +		igt_debug("  [%d] idx=%2u comp=%d coh=%u cpu_wc=%d name=%s%s\n",
> +			  i, pat_index_modes[i].pat_index, pat_index_modes[i].compressed,
> +			  pat_index_modes[i].coh_mode, pat_index_modes[i].force_cpu_wc,
> +			  pat_index_modes[i].name,
> +			  (i < n_common) ? " (common)" : " (random)");
> +	}
> +
> +	*modes = pat_index_modes;
> +	return n_picked;
> +}
> +
>   const struct pat_index_entry xe3p_lpg_coherency_pat_index_modes[] = {
>   	{ NULL, 18, false, "xa-l3-uc",	 XE_COH_NONE          },
>   	{ NULL, 19, false, "xa-l3-1way", XE_COH_AT_LEAST_1WAY },
> @@ -1414,6 +1557,12 @@ static uint32_t create_object(int fd, int r, int size, uint16_t coh_mode,
>    * Description: Check some of the xe2 pat_index modes.
>    */
>   
> +/**
> + * SUBTEST: pat-index-auto
> + * Test category: functionality test
> + * Description: Dynamically generate and test PAT index modes from platform PAT configuration.
> + */
> +
>   /**
>    * SUBTEST: xa-app-transient-media-off
>    * Test category: functionality test
> @@ -2275,6 +2424,15 @@ int igt_main_args("V", NULL, help_str, opt_handler, NULL)
>   							     ARRAY_SIZE(xe2_pat_index_modes));
>   	}
>   
> +	igt_subtest_with_dynamic("pat-index-auto") {
> +		struct pat_index_entry *modes;
> +		int n_modes;
> +
> +		igt_require(intel_graphics_ver(dev_id) >= IP_VER(20, 0));
> +		n_modes = init_pat_index_modes(fd, &modes);
> +		subtest_pat_index_modes_with_regions(fd, modes, n_modes);
> +	}
> +
>   	igt_subtest("display-vs-wb-transient")
>   		display_vs_wb_transient(fd);
>   


  reply	other threads:[~2026-05-12 22:45 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-05-11 23:36 [PATCH 0/3] tests/intel/xe_pat: dynamic PAT index selection for pat-index-auto Zongyao Bai
2026-05-11 23:36 ` [PATCH 1/3] tests/intel/xe_pat: fix typo in r2 wc debug print Zongyao Bai
2026-05-11 23:36 ` [PATCH 2/3] tests/intel/xe_pat: drop redundant manual srand initialization Zongyao Bai
2026-05-11 23:36 ` [PATCH 3/3] tests/intel/xe_pat: Add dynamic PAT index selection for pat-index-auto Zongyao Bai
2026-05-12 22:44   ` Wang, X [this message]
2026-05-12  5:04 ` ✓ i915.CI.BAT: success for tests/intel/xe_pat: " Patchwork
2026-05-12  6:41 ` ✓ Xe.CI.BAT: " Patchwork
2026-05-12 13:47 ` ✗ Xe.CI.FULL: failure " Patchwork
2026-05-12 19:06 ` ✓ i915.CI.Full: success " 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=4f68b11d-0dca-4b8c-bdb9-9dce30523aab@intel.com \
    --to=x.wang@intel.com \
    --cc=igt-dev@lists.freedesktop.org \
    --cc=jia.yao@intel.com \
    --cc=zongyao.bai@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