From patchwork Fri Dec 22 04:31:59 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Dave Airlie X-Patchwork-Id: 13502895 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 620F7C4706F for ; Fri, 22 Dec 2023 04:33:41 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 399BB10E73E; Fri, 22 Dec 2023 04:33:35 +0000 (UTC) Received: from us-smtp-delivery-44.mimecast.com (us-smtp-delivery-44.mimecast.com [205.139.111.44]) by gabe.freedesktop.org (Postfix) with ESMTPS id 589D110E741 for ; Fri, 22 Dec 2023 04:33:32 +0000 (UTC) Received: from mimecast-mx02.redhat.com (mimecast-mx02.redhat.com [66.187.233.88]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.3, cipher=TLS_AES_256_GCM_SHA384) id us-mta-628-GjuJcvw7Pi2NugIX1ZVTMg-1; Thu, 21 Dec 2023 23:33:27 -0500 X-MC-Unique: GjuJcvw7Pi2NugIX1ZVTMg-1 Received: from smtp.corp.redhat.com (int-mx04.intmail.prod.int.rdu2.redhat.com [10.11.54.4]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by mimecast-mx02.redhat.com (Postfix) with ESMTPS id 3DB31185A780; Fri, 22 Dec 2023 04:33:27 +0000 (UTC) Received: from dreadlord.redhat.com (unknown [10.64.136.90]) by smtp.corp.redhat.com (Postfix) with ESMTP id 3E8912026D66; Fri, 22 Dec 2023 04:33:26 +0000 (UTC) From: Dave Airlie To: dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org Subject: [PATCH 10/11] nouveau: push event block/allowing out of the fence context Date: Fri, 22 Dec 2023 14:31:59 +1000 Message-ID: <20231222043308.3090089-11-airlied@gmail.com> In-Reply-To: <20231222043308.3090089-1-airlied@gmail.com> References: <20231222043308.3090089-1-airlied@gmail.com> MIME-Version: 1.0 X-Scanned-By: MIMEDefang 3.4.1 on 10.11.54.4 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: gmail.com X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" There is a deadlock between the irq and fctx locks, the irq handling takes irq then fctx lock the fence signalling takes fctx then irq lock This splits the fence signalling path so the code that hits the irq lock is done in a separate work queue. This seems to fix crashes/hangs when using nouveau gsp with i915 primary GPU. Signed-off-by: Dave Airlie --- drivers/gpu/drm/nouveau/nouveau_fence.c | 28 ++++++++++++++++++++----- drivers/gpu/drm/nouveau/nouveau_fence.h | 5 ++++- 2 files changed, 27 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c index ca762ea55413..5057d976fa57 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fence.c +++ b/drivers/gpu/drm/nouveau/nouveau_fence.c @@ -62,7 +62,7 @@ nouveau_fence_signal(struct nouveau_fence *fence) if (test_bit(DMA_FENCE_FLAG_USER_BITS, &fence->base.flags)) { struct nouveau_fence_chan *fctx = nouveau_fctx(fence); - if (!--fctx->notify_ref) + if (atomic_dec_and_test(&fctx->notify_ref)) drop = 1; } @@ -103,6 +103,7 @@ nouveau_fence_context_kill(struct nouveau_fence_chan *fctx, int error) void nouveau_fence_context_del(struct nouveau_fence_chan *fctx) { + cancel_work_sync(&fctx->allow_block_work); nouveau_fence_context_kill(fctx, 0); nvif_event_dtor(&fctx->event); fctx->dead = 1; @@ -167,6 +168,18 @@ nouveau_fence_wait_uevent_handler(struct nvif_event *event, void *repv, u32 repc return ret; } +static void +nouveau_fence_work_allow_block(struct work_struct *work) +{ + struct nouveau_fence_chan *fctx = container_of(work, struct nouveau_fence_chan, + allow_block_work); + + if (atomic_read(&fctx->notify_ref) == 0) + nvif_event_block(&fctx->event); + else + nvif_event_allow(&fctx->event); +} + void nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_chan *fctx) { @@ -178,6 +191,7 @@ nouveau_fence_context_new(struct nouveau_channel *chan, struct nouveau_fence_cha } args; int ret; + INIT_WORK(&fctx->allow_block_work, nouveau_fence_work_allow_block); INIT_LIST_HEAD(&fctx->flip); INIT_LIST_HEAD(&fctx->pending); spin_lock_init(&fctx->lock); @@ -521,15 +535,19 @@ static bool nouveau_fence_enable_signaling(struct dma_fence *f) struct nouveau_fence *fence = from_fence(f); struct nouveau_fence_chan *fctx = nouveau_fctx(fence); bool ret; + bool do_work; - if (!fctx->notify_ref++) - nvif_event_allow(&fctx->event); + if (atomic_inc_return(&fctx->notify_ref) == 0) + do_work = true; ret = nouveau_fence_no_signaling(f); if (ret) set_bit(DMA_FENCE_FLAG_USER_BITS, &fence->base.flags); - else if (!--fctx->notify_ref) - nvif_event_block(&fctx->event); + else if (atomic_dec_and_test(&fctx->notify_ref)) + do_work = true; + + if (do_work) + schedule_work(&fctx->allow_block_work); return ret; } diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.h b/drivers/gpu/drm/nouveau/nouveau_fence.h index 64d33ae7f356..28f5cf013b89 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fence.h +++ b/drivers/gpu/drm/nouveau/nouveau_fence.h @@ -3,6 +3,7 @@ #define __NOUVEAU_FENCE_H__ #include +#include #include struct nouveau_drm; @@ -45,7 +46,9 @@ struct nouveau_fence_chan { char name[32]; struct nvif_event event; - int notify_ref, dead, killed; + struct work_struct allow_block_work; + atomic_t notify_ref; + int dead, killed; }; struct nouveau_fence_priv {