From mboxrd@z Thu Jan 1 00:00:00 1970 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 A6FACCDD552 for ; Wed, 18 Sep 2024 19:53:36 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 3E09B10E63A; Wed, 18 Sep 2024 19:53:36 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=intel.com header.i=@intel.com header.b="YeFyNcBw"; dkim-atps=neutral Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.9]) by gabe.freedesktop.org (Postfix) with ESMTPS id 28D0210E63D for ; Wed, 18 Sep 2024 19:53:35 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1726689215; x=1758225215; h=from:to:subject:date:message-id:in-reply-to:references: mime-version:content-transfer-encoding; bh=yNJNgtIETtJ3ys8iyzVZaCuCYVIVNTiL3sDsWJ7yu/0=; b=YeFyNcBwxscnu1U51o6N3c/0rWgL8PjWse5tTRGEV/dD4yDPnBaC8np4 wIuFtYRa9/wLYevypmNM0QMFXHCl+37yjDQFHHjfSJTfSWo/JfO78yleu GsWG1fyg+csikC+DH21MUQG+4uvzIX6DiyCcNVYS2wtwlE2OaRHpKrJ2t FfUuqEIXEFq8X3Wy0o2iK1hM+WON9ELQTFhNh/TGMkVvw3lWpNqByiyfG AGOb/NyrmoZTdUszov1rKlQZVqONeKx+ylbbHMglsv8Xf4LV2PgmMWNCF /DeT1vjp80GWG6ijkD4PXEBxEuSjBNwrE9D0fVnZdXzUJgy8ldp9o1kC8 A==; X-CSE-ConnectionGUID: KfmUsAHWRn24rPFAg1E1Dg== X-CSE-MsgGUID: wp4/Fa8FSOe21bP68VmrsQ== X-IronPort-AV: E=McAfee;i="6700,10204,11199"; a="48152931" X-IronPort-AV: E=Sophos;i="6.10,239,1719903600"; d="scan'208";a="48152931" Received: from orviesa002.jf.intel.com ([10.64.159.142]) by orvoesa101.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 18 Sep 2024 12:53:34 -0700 X-CSE-ConnectionGUID: MKEmh6EcSpWDnbcM6UeLIw== X-CSE-MsgGUID: OeI+T7+aRyKzSp3DUZ+bfw== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.10,239,1719903600"; d="scan'208";a="100406151" Received: from orsosgc001.jf.intel.com ([10.165.21.138]) by orviesa002-auth.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 18 Sep 2024 12:53:35 -0700 From: Ashutosh Dixit To: intel-xe@lists.freedesktop.org Subject: [PATCH 4/7] drm/xe/oa: Signal output fences Date: Wed, 18 Sep 2024 12:53:26 -0700 Message-ID: <20240918195330.3167756-5-ashutosh.dixit@intel.com> X-Mailer: git-send-email 2.41.0 In-Reply-To: <20240918195330.3167756-1-ashutosh.dixit@intel.com> References: <20240918195330.3167756-1-ashutosh.dixit@intel.com> MIME-Version: 1.0 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" Introduce 'struct xe_oa_fence' which includes the dma_fence used to signal output fences in the xe_sync array. The fences are signaled asynchronously. When there are no output fences to signal, the OA configuration wait is synchronously re-introduced into the ioctl. v2: Don't wait in the work, use callback + delayed work (Matt B) Use a single, not a per-fence spinlock (Matt Brost) v3: Move ofence alloc before job submission (Matt) Assert, don't fail, from dma_fence_add_callback (Matt) Additional dma_fence_get for dma_fence_wait (Matt) Change dma_fence_wait to non-interruptible (Matt) v4: Introduce last_fence to prevent uaf if stream is closed with pending OA config jobs v5: Remove oa_fence_lock, move spinlock back into xe_oa_fence to prevent uaf Suggested-by: Matthew Brost Reviewed-by: Jonathan Cavitt Signed-off-by: Ashutosh Dixit Reviewed-by: Matthew Brost --- drivers/gpu/drm/xe/xe_oa.c | 119 ++++++++++++++++++++++++++----- drivers/gpu/drm/xe/xe_oa_types.h | 3 + 2 files changed, 105 insertions(+), 17 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c index c72cc7b0451a2..6aee2ba454ad1 100644 --- a/drivers/gpu/drm/xe/xe_oa.c +++ b/drivers/gpu/drm/xe/xe_oa.c @@ -100,6 +100,17 @@ struct xe_oa_config_bo { struct xe_bb *bb; }; +struct xe_oa_fence { + /* @base: dma fence base */ + struct dma_fence base; + /* @lock: lock for the fence */ + spinlock_t lock; + /* @work: work to signal @base */ + struct delayed_work work; + /* @cb: callback to schedule @work */ + struct dma_fence_cb cb; +}; + #define DRM_FMT(x) DRM_XE_OA_FMT_TYPE_##x static const struct xe_oa_format oa_formats[] = { @@ -172,10 +183,10 @@ static struct xe_oa_config *xe_oa_get_oa_config(struct xe_oa *oa, int metrics_se return oa_config; } -static void free_oa_config_bo(struct xe_oa_config_bo *oa_bo) +static void free_oa_config_bo(struct xe_oa_config_bo *oa_bo, struct dma_fence *last_fence) { xe_oa_config_put(oa_bo->oa_config); - xe_bb_free(oa_bo->bb, NULL); + xe_bb_free(oa_bo->bb, last_fence); kfree(oa_bo); } @@ -655,7 +666,8 @@ static void xe_oa_free_configs(struct xe_oa_stream *stream) xe_oa_config_put(stream->oa_config); llist_for_each_entry_safe(oa_bo, tmp, stream->oa_config_bos.first, node) - free_oa_config_bo(oa_bo); + free_oa_config_bo(oa_bo, stream->last_fence); + dma_fence_put(stream->last_fence); } static void xe_oa_store_flex(struct xe_oa_stream *stream, struct xe_lrc *lrc, @@ -954,40 +966,113 @@ xe_oa_alloc_config_buffer(struct xe_oa_stream *stream, struct xe_oa_config *oa_c return oa_bo; } +static void xe_oa_update_last_fence(struct xe_oa_stream *stream, struct dma_fence *fence) +{ + dma_fence_put(stream->last_fence); + stream->last_fence = dma_fence_get(fence); +} + +static void xe_oa_fence_work_fn(struct work_struct *w) +{ + struct xe_oa_fence *ofence = container_of(w, typeof(*ofence), work.work); + + /* Signal fence to indicate new OA configuration is active */ + dma_fence_signal(&ofence->base); + dma_fence_put(&ofence->base); +} + +static void xe_oa_config_cb(struct dma_fence *fence, struct dma_fence_cb *cb) +{ + /* Additional empirical delay needed for NOA programming after registers are written */ +#define NOA_PROGRAM_ADDITIONAL_DELAY_US 500 + + struct xe_oa_fence *ofence = container_of(cb, typeof(*ofence), cb); + + INIT_DELAYED_WORK(&ofence->work, xe_oa_fence_work_fn); + queue_delayed_work(system_unbound_wq, &ofence->work, + usecs_to_jiffies(NOA_PROGRAM_ADDITIONAL_DELAY_US)); + dma_fence_put(fence); +} + +static const char *xe_oa_get_driver_name(struct dma_fence *fence) +{ + return "xe_oa"; +} + +static const char *xe_oa_get_timeline_name(struct dma_fence *fence) +{ + return "unbound"; +} + +static const struct dma_fence_ops xe_oa_fence_ops = { + .get_driver_name = xe_oa_get_driver_name, + .get_timeline_name = xe_oa_get_timeline_name, +}; + static int xe_oa_emit_oa_config(struct xe_oa_stream *stream, struct xe_oa_config *config) { #define NOA_PROGRAM_ADDITIONAL_DELAY_US 500 struct xe_oa_config_bo *oa_bo; - int err = 0, us = NOA_PROGRAM_ADDITIONAL_DELAY_US; + struct xe_oa_fence *ofence; + int i, err, num_signal = 0; struct dma_fence *fence; - long timeout; - /* Emit OA configuration batch */ + ofence = kzalloc(sizeof(*ofence), GFP_KERNEL); + if (!ofence) { + err = -ENOMEM; + goto exit; + } + oa_bo = xe_oa_alloc_config_buffer(stream, config); if (IS_ERR(oa_bo)) { err = PTR_ERR(oa_bo); goto exit; } + /* Emit OA configuration batch */ fence = xe_oa_submit_bb(stream, XE_OA_SUBMIT_ADD_DEPS, oa_bo->bb); if (IS_ERR(fence)) { err = PTR_ERR(fence); goto exit; } - /* Wait till all previous batches have executed */ - timeout = dma_fence_wait_timeout(fence, false, 5 * HZ); - dma_fence_put(fence); - if (timeout < 0) - err = timeout; - else if (!timeout) - err = -ETIME; - if (err) - drm_dbg(&stream->oa->xe->drm, "dma_fence_wait_timeout err %d\n", err); + /* Point of no return: initialize and set fence to signal */ + spin_lock_init(&ofence->lock); + dma_fence_init(&ofence->base, &xe_oa_fence_ops, &ofence->lock, 0, 0); - /* Additional empirical delay needed for NOA programming after registers are written */ - usleep_range(us, 2 * us); + for (i = 0; i < stream->num_syncs; i++) { + if (stream->syncs[i].flags & DRM_XE_SYNC_FLAG_SIGNAL) + num_signal++; + xe_sync_entry_signal(&stream->syncs[i], &ofence->base); + } + + /* Additional dma_fence_get in case we dma_fence_wait */ + if (!num_signal) + dma_fence_get(&ofence->base); + + /* Update last fence too before adding callback */ + xe_oa_update_last_fence(stream, fence); + + /* Add job fence callback to schedule work to signal ofence->base */ + err = dma_fence_add_callback(fence, &ofence->cb, xe_oa_config_cb); + xe_gt_assert(stream->gt, !err || err == -ENOENT); + if (err == -ENOENT) + xe_oa_config_cb(fence, &ofence->cb); + + /* If nothing needs to be signaled we wait synchronously */ + if (!num_signal) { + dma_fence_wait(&ofence->base, false); + dma_fence_put(&ofence->base); + } + + /* Done with syncs */ + for (i = 0; i < stream->num_syncs; i++) + xe_sync_entry_cleanup(&stream->syncs[i]); + kfree(stream->syncs); + + return 0; exit: + kfree(ofence); return err; } diff --git a/drivers/gpu/drm/xe/xe_oa_types.h b/drivers/gpu/drm/xe/xe_oa_types.h index 99f4b2d4bdcf6..c8e0df13faf83 100644 --- a/drivers/gpu/drm/xe/xe_oa_types.h +++ b/drivers/gpu/drm/xe/xe_oa_types.h @@ -239,6 +239,9 @@ struct xe_oa_stream { /** @no_preempt: Whether preemption and timeslicing is disabled for stream exec_q */ u32 no_preempt; + /** @last_fence: fence to use in stream destroy when needed */ + struct dma_fence *last_fence; + /** @num_syncs: size of @syncs array */ u32 num_syncs; -- 2.41.0