From: Matthew Auld <matthew.auld@intel.com>
To: Sanjay Yadav <sanjay.kumar.yadav@intel.com>,
igt-dev@lists.freedesktop.org
Subject: Re: [PATCH v2] lib/intel_pat: Handle platforms without compressed PAT index
Date: Mon, 16 Mar 2026 11:45:54 +0000 [thread overview]
Message-ID: <96456cfc-9dea-4d12-9c91-edab96a6d9d8@intel.com> (raw)
In-Reply-To: <20260316065843.287818-2-sanjay.kumar.yadav@intel.com>
On 16/03/2026 06:58, Sanjay Yadav wrote:
> Previously intel_get_pat_idx_uc_comp() would silently return zero on
> platforms where no compressed PAT index exists (e.g. CRI),
> since uc_comp was zero-initialized by default. This could lead to
> incorrect behavior when callers assume the returned index is valid.
>
> Fix this by:
> - Adding XE_PAT_IDX_INVALID sentinel to detect unsupported platforms
> - Initializing uc_comp to XE_PAT_IDX_INVALID in xe_get_pat_sw_config()
> before parsing debugfs; only platforms whose kernel exposes
> IDX[XE_CACHE_NONE_COMPRESSION] in gt0/pat_sw_config will override it
> with a valid index
> - Adding igt_assert(HAS_FLATCCS()) and igt_assert_f() in
> intel_get_pat_idx_uc_comp() to fail with a clear error instead of
> returning an invalid index
>
> Also update the bo-comp-disable-bind test to distinguish between "CCS
> is disabled" and "CCS does not exist" using HAS_FLATCCS(). On platforms
> without CCS, the compressed PAT bind check is skipped since there is no
> valid compressed PAT index to test with.
>
> v2:
> - Rebase
> - Update commit message accordingly
>
> Suggested-by: Matthew Auld <matthew.auld@intel.com>
> Signed-off-by: Sanjay Yadav <sanjay.kumar.yadav@intel.com>
Assuming CI is still happy,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> ---
> lib/intel_pat.c | 5 +++++
> lib/intel_pat.h | 1 +
> tests/intel/xe_pat.c | 30 ++++++++++++++++++++++--------
> 3 files changed, 28 insertions(+), 8 deletions(-)
>
> diff --git a/lib/intel_pat.c b/lib/intel_pat.c
> index 9feeeb39d..525d9ad24 100644
> --- a/lib/intel_pat.c
> +++ b/lib/intel_pat.c
> @@ -37,6 +37,8 @@ int32_t xe_get_pat_sw_config(int drm_fd, struct intel_pat_cache *xe_pat_cache)
> }
>
> memset(xe_pat_cache, 0, sizeof(*xe_pat_cache));
> + xe_pat_cache->uc_comp = XE_PAT_IDX_INVALID;
> +
> while ((nread = getline(&line, &line_len, dbgfs_file)) != -1) {
> uint32_t value = 0;
> char *p = NULL;
> @@ -159,8 +161,11 @@ uint8_t intel_get_pat_idx_uc_comp(int fd)
> uint16_t dev_id = intel_get_drm_devid(fd);
>
> igt_assert(intel_gen(dev_id) >= 20);
> + igt_assert(HAS_FLATCCS(dev_id));
>
> intel_get_pat_idx(fd, &pat);
> + igt_assert_f(pat.uc_comp != XE_PAT_IDX_INVALID,
> + "No compressed PAT index available on this platform\n");
> return pat.uc_comp;
> }
>
> diff --git a/lib/intel_pat.h b/lib/intel_pat.h
> index e9ade2e2e..e5dd8a0af 100644
> --- a/lib/intel_pat.h
> +++ b/lib/intel_pat.h
> @@ -10,6 +10,7 @@
> #include <stdint.h>
>
> #define DEFAULT_PAT_INDEX ((uint8_t)-1) /* igt-core can pick 1way or better */
> +#define XE_PAT_IDX_INVALID ((uint8_t)-2) /* no such PAT index on this platform */
> #define XE_PAT_MAX_ENTRIES 32
>
> struct xe_pat_entry {
> diff --git a/tests/intel/xe_pat.c b/tests/intel/xe_pat.c
> index b92512164..e91b037a5 100644
> --- a/tests/intel/xe_pat.c
> +++ b/tests/intel/xe_pat.c
> @@ -934,15 +934,18 @@ static bool has_no_compression_hint(int fd)
> * Test category: functionality test
> * Description: Validates that binding a BO created with
> * the NO_COMPRESSION flag using a compressed PAT index fails
> - * with -EINVAL on Xe2+ platforms.
> + * with -EINVAL on Xe2+ platforms. On platforms where CCS
> + * does not exist, the test verifies uncompressed access works.
> */
>
> static void bo_comp_disable_bind(int fd)
> {
> size_t size = xe_get_default_alignment(fd);
> - uint8_t comp_pat_index, uncomp_pat_index;
> - bool supported;
> + uint16_t dev_id = intel_get_drm_devid(fd);
> + bool has_flatccs = HAS_FLATCCS(dev_id);
> + uint8_t uncomp_pat_index;
> uint32_t vm, bo;
> + bool supported;
> int ret;
>
> supported = has_no_compression_hint(fd);
> @@ -958,14 +961,25 @@ static void bo_comp_disable_bind(int fd)
> igt_assert_eq(ret, 0);
> vm = xe_vm_create(fd, 0, 0);
>
> - comp_pat_index = intel_get_pat_idx_uc_comp(fd);
> uncomp_pat_index = intel_get_pat_idx_uc(fd);
>
> - igt_assert_eq(__xe_vm_bind(fd, vm, 0, bo, 0, 0x100000,
> - size, 0, 0, NULL, 0,
> - 0, comp_pat_index, 0),
> - -EINVAL);
> + /*
> + * On platforms with CCS, binding a NO_COMPRESSION BO with a
> + * compressed PAT index must fail. On platforms without CCS,
> + * there is no valid compressed PAT index, so skip this check.
> + */
> + if (has_flatccs) {
> + uint8_t comp_pat_index = intel_get_pat_idx_uc_comp(fd);
> +
> + igt_assert_eq(__xe_vm_bind(fd, vm, 0, bo, 0, 0x100000,
> + size, 0, 0, NULL, 0,
> + 0, comp_pat_index, 0),
> + -EINVAL);
> + } else {
> + igt_debug("Platform has no CCS, skipping compressed PAT bind check\n");
> + }
>
> + /* Uncompressed bind must always succeed */
> igt_assert_eq(__xe_vm_bind(fd, vm, 0, bo, 0, 0x100000,
> size, 0, 0, NULL, 0,
> 0, uncomp_pat_index, 0),
next prev parent reply other threads:[~2026-03-16 11:46 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-16 6:58 [PATCH v2] lib/intel_pat: Handle platforms without compressed PAT index Sanjay Yadav
2026-03-16 11:45 ` Matthew Auld [this message]
2026-03-16 22:18 ` ✓ Xe.CI.BAT: success for lib/intel_pat: Handle platforms without compressed PAT index (rev2) Patchwork
2026-03-17 9:59 ` ✓ i915.CI.BAT: " Patchwork
2026-03-17 21:26 ` ✓ i915.CI.Full: " Patchwork
2026-03-17 22:01 ` ✗ 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=96456cfc-9dea-4d12-9c91-edab96a6d9d8@intel.com \
--to=matthew.auld@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=sanjay.kumar.yadav@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