From patchwork Fri Aug 2 19:51:49 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Rob Herring X-Patchwork-Id: 11074101 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 059DA14E5 for ; Fri, 2 Aug 2019 19:52:11 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id EAD1F28898 for ; Fri, 2 Aug 2019 19:52:10 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id DF334288AF; Fri, 2 Aug 2019 19:52:10 +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 2256028898 for ; Fri, 2 Aug 2019 19:52:10 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id EE2BB6EDD5; Fri, 2 Aug 2019 19:52:06 +0000 (UTC) X-Original-To: dri-devel@lists.freedesktop.org Delivered-To: dri-devel@lists.freedesktop.org Received: from mail-io1-f66.google.com (mail-io1-f66.google.com [209.85.166.66]) by gabe.freedesktop.org (Postfix) with ESMTPS id 80E746EDD5 for ; Fri, 2 Aug 2019 19:52:02 +0000 (UTC) Received: by mail-io1-f66.google.com with SMTP id j6so34879714ioa.5 for ; Fri, 02 Aug 2019 12:52:02 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=pwIhp2y8f0D+4kEqwpU/HoJmB8e622d9feuXt2Phmd8=; b=A5vpVuk6mkDwvr/NfzVtKztZ38YS9reGZJNoUrc+kh6AhE4QT3R62zYwH3lzdAGP6v 5YIrj0sI9+UXzv1EtzObUjiAbMgjMq+QxKxISqptpE2vnw5NMRC+K/dUasIJLDHZQAsJ 28nW4tKfnENNO6W0TOee5hoAsupMw8JgOcHKRyxbSkqn9ld5Hot4eYifU1D9AjqT6bIV bAIU/3lWpezsQL39rIDXT1B9gA0cITWRf6pdRMgNvJ5abMd0VvgBbjagdqSt+TCAibxs CircdeE4JFReypjEwIGgH6ME4+IMd8ZyMDSkSbDtTCRzU6irKGOCMFBzHYNkYBXK6tg6 m2cg== X-Gm-Message-State: APjAAAVM72/mSE+l7/j+8PEV+zGFAalVaJF0XsfXghgHrPn+aV66RskT 6xEDVUx7vNYgiCoClXwkpigzGkI= X-Google-Smtp-Source: APXvYqwFCknLnDg9IXjcuifaj05vXndhvCrxtVzWGcOgTPebkotfUg9vDjjH50JjaUp8uix8/eZe0w== X-Received: by 2002:a5d:8747:: with SMTP id k7mr6621802iol.20.1564775521339; Fri, 02 Aug 2019 12:52:01 -0700 (PDT) Received: from xps15.herring.priv ([64.188.179.254]) by smtp.googlemail.com with ESMTPSA id s4sm98419934iop.25.2019.08.02.12.52.00 (version=TLS1_3 cipher=AEAD-AES256-GCM-SHA384 bits=256/256); Fri, 02 Aug 2019 12:52:00 -0700 (PDT) From: Rob Herring To: dri-devel@lists.freedesktop.org Subject: [PATCH v3 7/8] drm/panfrost: Add support for GPU heap allocations Date: Fri, 2 Aug 2019 13:51:49 -0600 Message-Id: <20190802195150.23207-8-robh@kernel.org> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20190802195150.23207-1-robh@kernel.org> References: <20190802195150.23207-1-robh@kernel.org> 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: Tomeu Vizoso , Maxime Ripard , Sean Paul , Steven Price , David Airlie , Boris Brezillon , Alyssa Rosenzweig , Robin Murphy Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" X-Virus-Scanned: ClamAV using ClamSMTP The midgard/bifrost GPUs need to allocate GPU heap memory which is allocated on GPU page faults and not pinned in memory. The vendor driver calls this functionality GROW_ON_GPF. This implementation assumes that BOs allocated with the PANFROST_BO_NOEXEC flag are never mmapped or exported. Both of those may actually work, but I'm unsure if there's some interaction there. It would cause the whole object to be pinned in memory which would defeat the point of this. On faults, we map in 2MB at a time in order to utilize huge pages (if enabled). Currently, once we've mapped pages in, they are only unmapped if the BO is freed. Once we add shrinker support, we can unmap pages with the shrinker. Cc: Tomeu Vizoso Cc: Boris Brezillon Cc: Robin Murphy Cc: Steven Price Acked-by: Alyssa Rosenzweig Signed-off-by: Rob Herring Reviewed-by: Steven Price --- drivers/gpu/drm/panfrost/TODO | 2 - drivers/gpu/drm/panfrost/panfrost_drv.c | 11 +- drivers/gpu/drm/panfrost/panfrost_gem.c | 36 ++++++- drivers/gpu/drm/panfrost/panfrost_gem.h | 8 ++ drivers/gpu/drm/panfrost/panfrost_mmu.c | 129 ++++++++++++++++++++++-- include/uapi/drm/panfrost_drm.h | 1 + 6 files changed, 172 insertions(+), 15 deletions(-) diff --git a/drivers/gpu/drm/panfrost/TODO b/drivers/gpu/drm/panfrost/TODO index c2e44add37d8..64129bf73933 100644 --- a/drivers/gpu/drm/panfrost/TODO +++ b/drivers/gpu/drm/panfrost/TODO @@ -14,8 +14,6 @@ The hard part is handling when more address spaces are needed than what the h/w provides. -- Support pinning pages on demand (GPU page faults). - - Support userspace controlled GPU virtual addresses. Needed for Vulkan. (Tomeu) - Support for madvise and a shrinker. diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c index 7ebd82d8d570..a7126b5f8e5d 100644 --- a/drivers/gpu/drm/panfrost/panfrost_drv.c +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c @@ -50,7 +50,12 @@ static int panfrost_ioctl_create_bo(struct drm_device *dev, void *data, struct drm_panfrost_create_bo *args = data; if (!args->size || args->pad || - (args->flags & ~PANFROST_BO_NOEXEC)) + (args->flags & ~(PANFROST_BO_NOEXEC | PANFROST_BO_HEAP))) + return -EINVAL; + + /* Heaps should never be executable */ + if ((args->flags & PANFROST_BO_HEAP) && + !(args->flags & PANFROST_BO_NOEXEC)) return -EINVAL; bo = panfrost_gem_create_with_handle(file, dev, args->size, args->flags, @@ -265,6 +270,10 @@ static int panfrost_ioctl_mmap_bo(struct drm_device *dev, void *data, return -ENOENT; } + /* Don't allow mmapping of heap objects as pages are not pinned. */ + if (to_panfrost_bo(gem_obj)->is_heap) + return -EINVAL; + ret = drm_gem_create_mmap_offset(gem_obj); if (ret == 0) args->offset = drm_vma_node_offset_addr(&gem_obj->vma_node); diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.c b/drivers/gpu/drm/panfrost/panfrost_gem.c index 63731f6c5223..f3d5f61714ae 100644 --- a/drivers/gpu/drm/panfrost/panfrost_gem.c +++ b/drivers/gpu/drm/panfrost/panfrost_gem.c @@ -27,13 +27,35 @@ static void panfrost_gem_free_object(struct drm_gem_object *obj) drm_mm_remove_node(&bo->node); spin_unlock(&pfdev->mm_lock); + if (bo->sgts) { + int i; + int n_sgt = bo->base.base.size / SZ_2M; + + for (i = 0; i < n_sgt; i++) { + if (bo->sgts[i].sgl) { + dma_unmap_sg(pfdev->dev, bo->sgts[i].sgl, + bo->sgts[i].nents, DMA_BIDIRECTIONAL); + sg_free_table(&bo->sgts[i]); + } + } + kfree(bo->sgts); + } + drm_gem_shmem_free_object(obj); } +static int panfrost_gem_pin(struct drm_gem_object *obj) +{ + if (to_panfrost_bo(obj)->is_heap) + return -EINVAL; + + return drm_gem_shmem_pin(obj); +} + static const struct drm_gem_object_funcs panfrost_gem_funcs = { .free = panfrost_gem_free_object, .print_info = drm_gem_shmem_print_info, - .pin = drm_gem_shmem_pin, + .pin = panfrost_gem_pin, .unpin = drm_gem_shmem_unpin, .get_sg_table = drm_gem_shmem_get_sg_table, .vmap = drm_gem_shmem_vmap, @@ -87,7 +109,10 @@ static int panfrost_gem_map(struct panfrost_device *pfdev, struct panfrost_gem_o if (ret) return ret; - return panfrost_mmu_map(bo); + if (!bo->is_heap) + ret = panfrost_mmu_map(bo); + + return ret; } struct panfrost_gem_object * @@ -101,7 +126,11 @@ panfrost_gem_create_with_handle(struct drm_file *file_priv, struct drm_gem_shmem_object *shmem; struct panfrost_gem_object *bo; - size = roundup(size, PAGE_SIZE); + /* Round up heap allocations to 2MB to keep fault handling simple */ + if (flags & PANFROST_BO_HEAP) + size = roundup(size, SZ_2M); + else + size = roundup(size, PAGE_SIZE); shmem = drm_gem_shmem_create_with_handle(file_priv, dev, size, handle); if (IS_ERR(shmem)) @@ -109,6 +138,7 @@ panfrost_gem_create_with_handle(struct drm_file *file_priv, bo = to_panfrost_bo(&shmem->base); bo->noexec = !!(flags & PANFROST_BO_NOEXEC); + bo->is_heap = !!(flags & PANFROST_BO_HEAP); ret = panfrost_gem_map(pfdev, bo); if (ret) diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.h b/drivers/gpu/drm/panfrost/panfrost_gem.h index 132f02399b7b..b628c9b67784 100644 --- a/drivers/gpu/drm/panfrost/panfrost_gem.h +++ b/drivers/gpu/drm/panfrost/panfrost_gem.h @@ -9,10 +9,12 @@ struct panfrost_gem_object { struct drm_gem_shmem_object base; + struct sg_table *sgts; struct drm_mm_node node; bool is_mapped :1; bool noexec :1; + bool is_heap :1; }; static inline @@ -21,6 +23,12 @@ struct panfrost_gem_object *to_panfrost_bo(struct drm_gem_object *obj) return container_of(to_drm_gem_shmem_obj(obj), struct panfrost_gem_object, base); } +static inline +struct panfrost_gem_object *drm_mm_node_to_panfrost_bo(struct drm_mm_node *node) +{ + return container_of(node, struct panfrost_gem_object, node); +} + struct drm_gem_object *panfrost_gem_create_object(struct drm_device *dev, size_t size); struct panfrost_gem_object * diff --git a/drivers/gpu/drm/panfrost/panfrost_mmu.c b/drivers/gpu/drm/panfrost/panfrost_mmu.c index 7d44328b280f..aa83bf46360b 100644 --- a/drivers/gpu/drm/panfrost/panfrost_mmu.c +++ b/drivers/gpu/drm/panfrost/panfrost_mmu.c @@ -2,6 +2,7 @@ /* Copyright 2019 Linaro, Ltd, Rob Herring */ #include #include +#include #include #include #include @@ -9,6 +10,7 @@ #include #include #include +#include #include #include "panfrost_device.h" @@ -235,12 +237,12 @@ void panfrost_mmu_unmap(struct panfrost_gem_object *bo) size_t unmapped_page; size_t pgsize = get_pgsize(iova, len - unmapped_len); - unmapped_page = ops->unmap(ops, iova, pgsize); - if (!unmapped_page) - break; - - iova += unmapped_page; - unmapped_len += unmapped_page; + if (ops->iova_to_phys(ops, iova)) { + unmapped_page = ops->unmap(ops, iova, pgsize); + WARN_ON(unmapped_page != pgsize); + } + iova += pgsize; + unmapped_len += pgsize; } mmu_hw_do_operation(pfdev, 0, bo->node.start << PAGE_SHIFT, @@ -276,6 +278,105 @@ static const struct iommu_gather_ops mmu_tlb_ops = { .tlb_sync = mmu_tlb_sync_context, }; +static struct drm_mm_node *addr_to_drm_mm_node(struct panfrost_device *pfdev, int as, u64 addr) +{ + struct drm_mm_node *node; + u64 offset = addr >> PAGE_SHIFT; + + drm_mm_for_each_node(node, &pfdev->mm) { + if (offset >= node->start && offset < (node->start + node->size)) + return node; + } + return NULL; +} + +#define NUM_FAULT_PAGES (SZ_2M / PAGE_SIZE) + +int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as, u64 addr) +{ + int ret, i; + struct drm_mm_node *node; + struct panfrost_gem_object *bo; + struct address_space *mapping; + pgoff_t page_offset; + struct sg_table *sgt; + struct page **pages; + + node = addr_to_drm_mm_node(pfdev, as, addr); + if (!node) + return -ENOENT; + + bo = drm_mm_node_to_panfrost_bo(node); + if (!bo->is_heap) { + dev_WARN(pfdev->dev, "matching BO is not heap type (GPU VA = %llx)", + node->start << PAGE_SHIFT); + return -EINVAL; + } + /* Assume 2MB alignment and size multiple */ + addr &= ~((u64)SZ_2M - 1); + page_offset = addr >> PAGE_SHIFT; + page_offset -= node->start; + + mutex_lock(&bo->base.pages_lock); + + if (!bo->base.pages) { + bo->sgts = kvmalloc_array(bo->base.base.size / SZ_2M, + sizeof(struct sg_table), GFP_KERNEL | __GFP_ZERO); + if (!bo->sgts) + return -ENOMEM; + + pages = kvmalloc_array(bo->base.base.size >> PAGE_SHIFT, + sizeof(struct page *), GFP_KERNEL | __GFP_ZERO); + if (!pages) { + kfree(bo->sgts); + bo->sgts = NULL; + return -ENOMEM; + } + bo->base.pages = pages; + bo->base.pages_use_count = 1; + } else + pages = bo->base.pages; + + mapping = bo->base.base.filp->f_mapping; + mapping_set_unevictable(mapping); + + for (i = page_offset; i < page_offset + NUM_FAULT_PAGES; i++) { + pages[i] = shmem_read_mapping_page(mapping, i); + if (IS_ERR(pages[i])) { + mutex_unlock(&bo->base.pages_lock); + ret = PTR_ERR(pages[i]); + goto err_pages; + } + } + + mutex_unlock(&bo->base.pages_lock); + + sgt = &bo->sgts[page_offset / (SZ_2M / PAGE_SIZE)]; + ret = sg_alloc_table_from_pages(sgt, pages + page_offset, + NUM_FAULT_PAGES, 0, SZ_2M, GFP_KERNEL); + if (ret) + goto err_pages; + + if (!dma_map_sg(pfdev->dev, sgt->sgl, sgt->nents, DMA_BIDIRECTIONAL)) { + ret = -EINVAL; + goto err_map; + } + + mmu_map_sg(pfdev, addr, IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, sgt); + + bo->is_mapped = true; + + dev_dbg(pfdev->dev, "mapped page fault @ %llx", addr); + + return 0; + +err_map: + sg_free_table(sgt); +err_pages: + drm_gem_shmem_put_pages(&bo->base); + return ret; +} + static const char *access_type_name(struct panfrost_device *pfdev, u32 fault_status) { @@ -312,9 +413,7 @@ static irqreturn_t panfrost_mmu_irq_handler_thread(int irq, void *data) { struct panfrost_device *pfdev = data; u32 status = mmu_read(pfdev, MMU_INT_RAWSTAT); - int i; - - dev_err(pfdev->dev, "mmu irq status=%x\n", status); + int i, ret; for (i = 0; status; i++) { u32 mask = BIT(i) | BIT(i + 16); @@ -336,6 +435,18 @@ static irqreturn_t panfrost_mmu_irq_handler_thread(int irq, void *data) access_type = (fault_status >> 8) & 0x3; source_id = (fault_status >> 16); + /* Page fault only */ + if ((status & mask) == BIT(i)) { + WARN_ON(exception_type < 0xC1 || exception_type > 0xC4); + + ret = panfrost_mmu_map_fault_addr(pfdev, i, addr); + if (!ret) { + mmu_write(pfdev, MMU_INT_CLEAR, BIT(i)); + status &= ~mask; + continue; + } + } + /* terminal fault, print info about the fault */ dev_err(pfdev->dev, "Unhandled Page fault in AS%d at VA 0x%016llX\n" diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h index 17fb5d200f7a..9150dd75aad8 100644 --- a/include/uapi/drm/panfrost_drm.h +++ b/include/uapi/drm/panfrost_drm.h @@ -83,6 +83,7 @@ struct drm_panfrost_wait_bo { }; #define PANFROST_BO_NOEXEC 1 +#define PANFROST_BO_HEAP 2 /** * struct drm_panfrost_create_bo - ioctl argument for creating Panfrost BOs.