Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 4F867CD8CB9 for ; Wed, 10 Jun 2026 12:10:01 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id B318E10E913; Wed, 10 Jun 2026 12:10:00 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=intel.com header.i=@intel.com header.b="YPihqZ09"; dkim-atps=neutral Received: from mgamail.intel.com (mgamail.intel.com [192.198.163.8]) by gabe.freedesktop.org (Postfix) with ESMTPS id 92E3F10E913 for ; Wed, 10 Jun 2026 12:09:58 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1781093398; x=1812629398; h=from:to:cc:subject:date:message-id:mime-version: content-transfer-encoding; bh=pR/C1yEl4DlRcPES1dHB9D0cmRhL5siKxxYRc3cj6b0=; b=YPihqZ09P1Dx5jqJZtsJsPoJipxXlYju48maysgutM48Kqq2gXttUdXQ ThilrMAcgRNCOX7iXvNYaU4l+l8WP4HSr9wwWcobfUXSVR5Y5tCMXKLhA e2643QF1pHelSe4PnO62Twv9GOiOFlg9uTfMHTwnvtgUjbdBdkOH4A+KL NXuzPCat6lZUm8D9/CfAqMpFvHlf8oJ6KcOHTyP85R29uKx2anfpwLTb3 Y317a/UX8OZP8hhZZmZ0udX4/iegeTxJw241FvsPhpP0gl9jSxZ3KgQzF aVvZ+EuDaAZ7PMKskypW5OvQLGabYgT8W2BCeMOsxxUD0UX3V3THHanvg A==; X-CSE-ConnectionGUID: K9y9thFER+azoiz+SVyKMw== X-CSE-MsgGUID: oq5X1C+UT5yHe9ZJxuRtDQ== X-IronPort-AV: E=McAfee;i="6800,10657,11812"; a="99461494" X-IronPort-AV: E=Sophos;i="6.24,197,1774335600"; d="scan'208";a="99461494" Received: from fmviesa002.fm.intel.com ([10.60.135.142]) by fmvoesa102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 10 Jun 2026 05:09:58 -0700 X-CSE-ConnectionGUID: IlZzu1XKRxm2l2ry22G07A== X-CSE-MsgGUID: 0v4fXgXbQrmg1tuapxsebA== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.24,197,1774335600"; d="scan'208";a="269831883" Received: from varungup-desk.iind.intel.com ([10.190.238.71]) by fmviesa002-auth.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 10 Jun 2026 05:09:56 -0700 From: Arvind Yadav To: intel-xe@lists.freedesktop.org Cc: matthew.brost@intel.com, himal.prasad.ghimiray@intel.com, thomas.hellstrom@linux.intel.com, rodrigo.vivi@intel.com, tejas.upadhyay@intel.com Subject: [PATCH v2] drm/xe/guc: Avoid self-deadlock in queue destruction Date: Wed, 10 Jun 2026 17:39:45 +0530 Message-ID: <20260610120945.1726638-1-arvind.yadav@intel.com> X-Mailer: git-send-email 2.43.0 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: intel-xe@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel Xe graphics driver List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" GuC exec queue destruction can run on xe->destroy_wq. During queue cleanup, xe_exec_queue_fini() can drop the last reference to objects which eventually release the DRM device and run drmm cleanup actions. guc_submit_sw_fini(), registered as a drmm action, drains xe->destroy_wq. If the final drm_dev_put() happens from an xe->destroy_wq worker, teardown can try to drain the same workqueue the current worker is running on, causing a self-deadlock. Fix this by taking an extra drm_device reference across async exec queue cleanup and dropping it from system_dfl_wq. The deferred put work is preallocated during queue init, so the destroy path does not need a failing or nofail allocation. This keeps the final drm_dev_put() out of xe->destroy_wq while preserving exec queue teardown ordering. v2: - rebase Fixes: 2d2be279f1ca ("drm/xe: fix UAF around queue destruction") Cc: Matthew Brost Cc: Thomas Hellström Cc: Rodrigo Vivi Cc: Himal Prasad Ghimiray Reviewed-by: Tejas Upadhyay Signed-off-by: Arvind Yadav --- drivers/gpu/drm/xe/xe_guc_exec_queue_types.h | 3 + drivers/gpu/drm/xe/xe_guc_submit.c | 62 ++++++++++++++++++-- 2 files changed, 59 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h index e5e53b421f29..b5dfc9c9092d 100644 --- a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h +++ b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h @@ -13,6 +13,7 @@ struct dma_fence; struct xe_exec_queue; +struct xe_drm_dev_put_work; /** * struct xe_guc_exec_queue - GuC specific state for an xe_exec_queue @@ -35,6 +36,8 @@ struct xe_guc_exec_queue { struct xe_sched_msg static_msgs[MAX_STATIC_MSG_TYPE]; /** @destroy_async: do final destroy async from this worker */ struct work_struct destroy_async; + /** @dev_put_work: deferred drm_dev_put worker to avoid self-deadlock */ + struct xe_drm_dev_put_work *dev_put_work; /** @resume_time: time of last resume */ u64 resume_time; /** @state: GuC specific state for this xe_exec_queue */ diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c index b29cc08e6291..d1ae4339f375 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.c +++ b/drivers/gpu/drm/xe/xe_guc_submit.c @@ -10,6 +10,7 @@ #include #include +#include #include #include "abi/guc_actions_abi.h" @@ -47,6 +48,11 @@ #define XE_GUC_EXEC_QUEUE_CGP_CONTEXT_ERROR_LEN 6 +struct xe_drm_dev_put_work { + struct work_struct work; + struct drm_device *drm; +}; + static int guc_submit_reset_prepare(struct xe_guc *guc); static struct xe_guc * @@ -1674,6 +1680,8 @@ static void guc_exec_queue_fini(struct xe_exec_queue *q) xe_sched_entity_fini(&ge->entity); xe_sched_fini(&ge->sched); + kfree(ge->dev_put_work); + /* * RCU free due sched being exported via DRM scheduler fences * (timeline name). @@ -1681,20 +1689,55 @@ static void guc_exec_queue_fini(struct xe_exec_queue *q) kfree_rcu(ge, rcu); } -static void __guc_exec_queue_destroy_async(struct work_struct *w) +static void xe_drm_dev_put_work_func(struct work_struct *work) { - struct xe_guc_exec_queue *ge = - container_of(w, struct xe_guc_exec_queue, destroy_async); - struct xe_exec_queue *q = ge->q; + struct xe_drm_dev_put_work *put = + container_of(work, struct xe_drm_dev_put_work, work); + + drm_dev_put(put->drm); + kfree(put); +} + +static void guc_exec_queue_do_destroy(struct xe_exec_queue *q, bool defer_dev_put) +{ + struct xe_guc_exec_queue *ge = q->guc; struct xe_guc *guc = exec_queue_to_guc(q); + struct xe_device *xe = guc_to_xe(guc); + struct drm_device *drm = &xe->drm; + struct xe_drm_dev_put_work *put_work = NULL; - guard(xe_pm_runtime)(guc_to_xe(guc)); + if (defer_dev_put) { + put_work = ge->dev_put_work; + ge->dev_put_work = NULL; + + put_work->drm = drm; + INIT_WORK(&put_work->work, xe_drm_dev_put_work_func); + + /* + * Avoid running the final drm_dev_put() from xe->destroy_wq, + * since guc_submit_sw_fini() drains that queue. + */ + drm_dev_get(drm); + } + + guard(xe_pm_runtime)(xe); trace_xe_exec_queue_destroy(q); /* Confirm no work left behind accessing device structures */ cancel_delayed_work_sync(&ge->sched.base.work_tdr); xe_exec_queue_fini(q); + + if (put_work) + queue_work(system_dfl_wq, &put_work->work); +} + +static void __guc_exec_queue_destroy_async(struct work_struct *w) +{ + struct xe_guc_exec_queue *ge = + container_of(w, struct xe_guc_exec_queue, destroy_async); + + guc_exec_queue_do_destroy(ge->q, true); } static void guc_exec_queue_destroy_async(struct xe_exec_queue *q) @@ -1706,7 +1749,7 @@ static void guc_exec_queue_destroy_async(struct xe_exec_queue *q) /* We must block on kernel engines so slabs are empty on driver unload */ if (q->flags & EXEC_QUEUE_FLAG_PERMANENT || exec_queue_wedged(q)) - __guc_exec_queue_destroy_async(&q->guc->destroy_async); + guc_exec_queue_do_destroy(q, false); else queue_work(xe->destroy_wq, &q->guc->destroy_async); } @@ -1914,6 +1957,12 @@ static int guc_exec_queue_init(struct xe_exec_queue *q) if (!ge) return -ENOMEM; + ge->dev_put_work = kmalloc_obj(*ge->dev_put_work, GFP_KERNEL); + if (!ge->dev_put_work) { + kfree(ge); + return -ENOMEM; + } + q->guc = ge; ge->q = q; init_rcu_head(&ge->rcu); @@ -1989,6 +2038,7 @@ static int guc_exec_queue_init(struct xe_exec_queue *q) err_release_id: release_guc_id(guc, q); err_free: + kfree(ge->dev_put_work); kfree(ge); return err; -- 2.43.0