From patchwork Tue Aug 7 16:08:41 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 10558849 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 9702D14E5 for ; Tue, 7 Aug 2018 16:08:58 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 8402D2A587 for ; Tue, 7 Aug 2018 16:08:58 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 76C232A58F; Tue, 7 Aug 2018 16:08:58 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-5.2 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_MED autolearn=ham version=3.3.1 Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher DHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.wl.linuxfoundation.org (Postfix) with ESMTPS id EEF422A534 for ; Tue, 7 Aug 2018 16:08:57 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id F15D76E186; Tue, 7 Aug 2018 16:08:54 +0000 (UTC) X-Original-To: dri-devel@lists.freedesktop.org Delivered-To: dri-devel@lists.freedesktop.org Received: from fireflyinternet.com (mail.fireflyinternet.com [109.228.58.192]) by gabe.freedesktop.org (Postfix) with ESMTPS id 63CB289109; Tue, 7 Aug 2018 16:08:53 +0000 (UTC) X-Default-Received-SPF: pass (skip=forwardok (res=PASS)) x-ip-name=78.156.65.138; Received: from haswell.alporthouse.com (unverified [78.156.65.138]) by fireflyinternet.com (Firefly Internet (M1)) with ESMTP id 12685727-1500050 for multiple; Tue, 07 Aug 2018 17:08:43 +0100 From: Chris Wilson To: intel-gfx@lists.freedesktop.org, amd-gfx@lists.freedesktop.org Subject: [PATCH v4] drm/amdgpu: Transfer fences to dmabuf importer Date: Tue, 7 Aug 2018 17:08:41 +0100 Message-Id: <20180807160841.31028-1-chris@chris-wilson.co.uk> X-Mailer: git-send-email 2.18.0 In-Reply-To: <20180807104500.31264-1-chris@chris-wilson.co.uk> References: <20180807104500.31264-1-chris@chris-wilson.co.uk> MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Alex Deucher , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" X-Virus-Scanned: ClamAV using ClamSMTP amdgpu only uses shared-fences internally, but dmabuf importers rely on implicit write hazard tracking via the reservation_object.fence_excl. For example, the importer use the write hazard for timing a page flip to only occur after the exporter has finished flushing its write into the surface. As such, on exporting a dmabuf, we must either flush all outstanding fences (for we do not know which are writes and should have been exclusive) or alternatively create a new exclusive fence that is the composite of all the existing shared fences, and so will only be signaled when all earlier fences are signaled (ensuring that we can not be signaled before the completion of any earlier write). v2: reservation_object is already locked by amdgpu_bo_reserve() Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107341 Testcase: igt/amd_prime/amd-to-i915 References: 8e94a46c1770 ("drm/amdgpu: Attach exclusive fence to prime exported bo's. (v5)") Signed-off-by: Chris Wilson Cc: Alex Deucher Cc: "Christian König" --- This time, hopefully proofread and references complete. -Chris --- drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c | 68 ++++++++++++++++++++--- 1 file changed, 60 insertions(+), 8 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c index 1c5d97f4b4dd..dff2b01a3d89 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -37,6 +37,7 @@ #include "amdgpu_display.h" #include #include +#include static const struct dma_buf_ops amdgpu_dmabuf_ops; @@ -188,6 +189,57 @@ amdgpu_gem_prime_import_sg_table(struct drm_device *dev, return ERR_PTR(ret); } +static int +__reservation_object_make_exclusive(struct reservation_object *obj) +{ + struct reservation_object_list *fobj; + struct dma_fence_array *array; + struct dma_fence **fences; + unsigned int count, i; + + fobj = reservation_object_get_list(obj); + if (!fobj) + return 0; + + count = !!rcu_access_pointer(obj->fence_excl); + count += fobj->shared_count; + + fences = kmalloc_array(count, sizeof(*fences), GFP_KERNEL); + if (!fences) + return -ENOMEM; + + for (i = 0; i < fobj->shared_count; i++) { + struct dma_fence *f = + rcu_dereference_protected(fobj->shared[i], + reservation_object_held(obj)); + + fences[i] = dma_fence_get(f); + } + + if (rcu_access_pointer(obj->fence_excl)) { + struct dma_fence *f = + rcu_dereference_protected(obj->fence_excl, + reservation_object_held(obj)); + + fences[i] = dma_fence_get(f); + } + + array = dma_fence_array_create(count, fences, + dma_fence_context_alloc(1), 0, + false); + if (!array) + goto err_fences_put; + + reservation_object_add_excl_fence(obj, &array->base); + return 0; + +err_fences_put: + for (i = 0; i < count; i++) + dma_fence_put(fences[i]); + kfree(fences); + return -ENOMEM; +} + /** * amdgpu_gem_map_attach - &dma_buf_ops.attach implementation * @dma_buf: shared DMA buffer @@ -219,16 +271,16 @@ static int amdgpu_gem_map_attach(struct dma_buf *dma_buf, if (attach->dev->driver != adev->dev->driver) { /* - * Wait for all shared fences to complete before we switch to future - * use of exclusive fence on this prime shared bo. + * We only create shared fences for internal use, but importers + * of the dmabuf rely on exclusive fences for implicitly + * tracking write hazards. As any of the current fences may + * correspond to a write, we need to convert all existing + * fences on the reservation object into a single exclusive + * fence. */ - r = reservation_object_wait_timeout_rcu(bo->tbo.resv, - true, false, - MAX_SCHEDULE_TIMEOUT); - if (unlikely(r < 0)) { - DRM_DEBUG_PRIME("Fence wait failed: %li\n", r); + r = __reservation_object_make_exclusive(bo->tbo.resv); + if (r) goto error_unreserve; - } } /* pin buffer into GTT */