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 08402C77B73 for ; Tue, 6 Jun 2023 15:25:08 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id CC54E10E375; Tue, 6 Jun 2023 15:25:07 +0000 (UTC) Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by gabe.freedesktop.org (Postfix) with ESMTPS id 0EBD310E372 for ; Tue, 6 Jun 2023 15:25:05 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1686065105; x=1717601105; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=4o626YxImc0QB1LcwRtp3P4sV5D+E8UcmFOTj0GKA/M=; b=Ayz/hlVy37U3OrFh5RwD26TQ/3GtvVfTxK2aVc/PELRDJQsvJxRu9R10 awiEN5h/A4v7e0Ur9ge8oTPcCMwnIH8/guM1dtm9KUlgH3+J3Cb+zlUnC 45ocyJs76txOSBe3sri+iUBQK/2Yxb5obw4OQkSgi0aYgYcp1+915HU43 T1COymOjIsw1+lLYjtc8P471yNMSShvaEpQrXzuXpnHqng/4x7fsVgIkJ Y1QybywX767ILPUioWFwNDBV1sYIeTeUef16HwnzK6W8j13C+4ImvtUkA CAE89FJWwnU03omXH4mcGeB5oFsULPI2exz3eKC37UcR6rCkLu5FQyhGV g==; X-IronPort-AV: E=McAfee;i="6600,9927,10733"; a="354205313" X-IronPort-AV: E=Sophos;i="6.00,221,1681196400"; d="scan'208";a="354205313" Received: from orsmga006.jf.intel.com ([10.7.209.51]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 06 Jun 2023 08:25:03 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10733"; a="686582044" X-IronPort-AV: E=Sophos;i="6.00,221,1681196400"; d="scan'208";a="686582044" Received: from akurkina-mobl1.ger.corp.intel.com (HELO mwauld-desk1.intel.com) ([10.252.4.184]) by orsmga006-auth.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 06 Jun 2023 08:25:01 -0700 From: Matthew Auld To: intel-xe@lists.freedesktop.org Date: Tue, 6 Jun 2023 16:24:18 +0100 Message-Id: <20230606152420.172591-4-matthew.auld@intel.com> X-Mailer: git-send-email 2.40.1 In-Reply-To: <20230606152420.172591-1-matthew.auld@intel.com> References: <20230606152420.172591-1-matthew.auld@intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Subject: [Intel-xe] [PATCH v5 3/5] drm/xe/bo: support tiered vram allocation for small-bar 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: , Cc: Lucas De Marchi Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" Add the new flag XE_BO_NEEDS_CPU_ACCESS, to force allocating in the mappable part of vram. If no flag is specified we do a topdown allocation, to limit the chances of stealing the precious mappable part, if we don't need it. If this is a full-bar system, then this all gets nooped. For kernel users, it looks like xe_bo_create_pin_map() is the central place which users should call if they want CPU access to the object, so add the flag there. We still need to plumb this through for userspace allocations. Also it looks like page-tables are using pin_map(), which is less than ideal. If we can already use the GPU to do page-table management, then maybe we should just force that for small-bar. Signed-off-by: Matthew Auld Cc: Gwan-gyeong Mun Cc: Thomas Hellström Cc: Lucas De Marchi Reviewed-by: Maarten Lankhorst Reviewed-by: Gwan-gyeong Mun --- drivers/gpu/drm/xe/tests/xe_migrate.c | 3 +- drivers/gpu/drm/xe/xe_bo.c | 48 ++++++++++++++++++--------- drivers/gpu/drm/xe/xe_bo.h | 1 + drivers/gpu/drm/xe/xe_ttm_vram_mgr.c | 4 +++ 4 files changed, 40 insertions(+), 16 deletions(-) diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c b/drivers/gpu/drm/xe/tests/xe_migrate.c index 60266fea7faa..d2d3cb58aa42 100644 --- a/drivers/gpu/drm/xe/tests/xe_migrate.c +++ b/drivers/gpu/drm/xe/tests/xe_migrate.c @@ -111,7 +111,8 @@ static void test_copy(struct xe_migrate *m, struct xe_bo *bo, struct xe_bo *sysmem = xe_bo_create_locked(xe, m->tile, NULL, bo->size, ttm_bo_type_kernel, - XE_BO_CREATE_SYSTEM_BIT); + XE_BO_CREATE_SYSTEM_BIT | + XE_BO_NEEDS_CPU_ACCESS); if (IS_ERR(sysmem)) { KUNIT_FAIL(test, "Failed to allocate sysmem bo for %s: %li\n", str, PTR_ERR(sysmem)); diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c index b94a80a32d86..35ffde8988ae 100644 --- a/drivers/gpu/drm/xe/xe_bo.c +++ b/drivers/gpu/drm/xe/xe_bo.c @@ -110,19 +110,28 @@ static void add_vram(struct xe_device *xe, struct xe_bo *bo, struct ttm_place *places, u32 bo_flags, u32 mem_type, u32 *c) { struct xe_tile *tile = mem_type_to_tile(xe, mem_type); + struct ttm_place place = { .mem_type = mem_type }; + u64 io_size = tile->mem.vram.io_size; XE_BUG_ON(!tile->mem.vram.size); - places[*c] = (struct ttm_place) { - .mem_type = mem_type, - /* - * For eviction / restore on suspend / resume objects - * pinned in VRAM must be contiguous - */ - .flags = bo_flags & (XE_BO_CREATE_PINNED_BIT | - XE_BO_CREATE_GGTT_BIT) ? - TTM_PL_FLAG_CONTIGUOUS : 0, - }; + /* + * For eviction / restore on suspend / resume objects + * pinned in VRAM must be contiguous + */ + if (bo_flags & (XE_BO_CREATE_PINNED_BIT | + XE_BO_CREATE_GGTT_BIT)) + place.flags |= TTM_PL_FLAG_CONTIGUOUS; + + if (io_size < tile->mem.vram.size) { + if (bo_flags & XE_BO_NEEDS_CPU_ACCESS) { + place.fpfn = 0; + place.lpfn = io_size >> PAGE_SHIFT; + } else { + place.flags |= TTM_PL_FLAG_TOPDOWN; + } + } + places[*c] = place; *c += 1; if (bo->props.preferred_mem_type == XE_BO_PROPS_INVALID) @@ -362,15 +371,20 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, struct ttm_resource *mem) { struct xe_device *xe = ttm_to_xe_device(bdev); - struct xe_tile *tile; switch (mem->mem_type) { case XE_PL_SYSTEM: case XE_PL_TT: return 0; case XE_PL_VRAM0: - case XE_PL_VRAM1: - tile = mem_type_to_tile(xe, mem->mem_type); + case XE_PL_VRAM1: { + struct xe_tile *tile = mem_type_to_tile(xe, mem->mem_type); + struct xe_ttm_vram_mgr_resource *vres = + to_xe_ttm_vram_mgr_resource(mem); + + if (vres->used_visible_size < mem->size) + return -EINVAL; + mem->bus.offset = mem->start << PAGE_SHIFT; if (tile->mem.vram.mapping && @@ -385,7 +399,7 @@ static int xe_ttm_io_mem_reserve(struct ttm_device *bdev, mem->bus.caching = ttm_write_combined; #endif return 0; - case XE_PL_STOLEN: + } case XE_PL_STOLEN: return xe_ttm_stolen_io_mem_reserve(xe, mem); default: return -EINVAL; @@ -1292,7 +1306,8 @@ struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile xe_ttm_stolen_cpu_access_needs_ggtt(xe)) flags |= XE_BO_CREATE_GGTT_BIT; - bo = xe_bo_create_locked_range(xe, tile, vm, size, start, end, type, flags); + bo = xe_bo_create_locked_range(xe, tile, vm, size, start, end, type, + flags | XE_BO_NEEDS_CPU_ACCESS); if (IS_ERR(bo)) return bo; @@ -1601,6 +1616,9 @@ int xe_bo_vmap(struct xe_bo *bo) xe_bo_assert_held(bo); + if (!(bo->flags & XE_BO_NEEDS_CPU_ACCESS)) + return -EINVAL; + if (!iosys_map_is_null(&bo->vmap)) return 0; diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h index 29eb7474f018..fa485e38aa69 100644 --- a/drivers/gpu/drm/xe/xe_bo.h +++ b/drivers/gpu/drm/xe/xe_bo.h @@ -31,6 +31,7 @@ #define XE_BO_DEFER_BACKING BIT(9) #define XE_BO_SCANOUT_BIT BIT(10) #define XE_BO_FIXED_PLACEMENT_BIT BIT(11) +#define XE_BO_NEEDS_CPU_ACCESS BIT(12) /* this one is trigger internally only */ #define XE_BO_INTERNAL_TEST BIT(30) #define XE_BO_INTERNAL_64K BIT(31) diff --git a/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c index 1a84abd35fcf..7cd8c42cc0ff 100644 --- a/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c +++ b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c @@ -372,11 +372,15 @@ int xe_ttm_vram_mgr_alloc_sgt(struct xe_device *xe, struct sg_table **sgt) { struct xe_tile *tile = &xe->tiles[res->mem_type - XE_PL_VRAM0]; + struct xe_ttm_vram_mgr_resource *vres = to_xe_ttm_vram_mgr_resource(res); struct xe_res_cursor cursor; struct scatterlist *sg; int num_entries = 0; int i, r; + if (vres->used_visible_size < res->size) + return -EOPNOTSUPP; + *sgt = kmalloc(sizeof(**sgt), GFP_KERNEL); if (!*sgt) return -ENOMEM; -- 2.40.1