Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752313AbcCHN7G (ORCPT ); Tue, 8 Mar 2016 08:59:06 -0500 Received: from foss.arm.com ([217.140.101.70]:33885 "EHLO foss.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751412AbcCHN6z (ORCPT ); Tue, 8 Mar 2016 08:58:55 -0500 Subject: Re: [PATCH 1/4] ARM64: dma: Add support for NO_KERNEL_MAPPING attribute To: Stephen Boyd , linux-kernel@vger.kernel.org References: <1457428939-26659-1-git-send-email-stephen.boyd@linaro.org> <1457428939-26659-2-git-send-email-stephen.boyd@linaro.org> Cc: linux-arm@lists.infradead.org, Greg Kroah-Hartman , Mimi Zohar , Andrew Morton , Mark Brown , Laura Abbott , Laura Abbott , Arnd Bergmann , Marek Szyprowski , Catalin Marinas , Will Deacon From: Robin Murphy Message-ID: <56DEDA9B.8000900@arm.com> Date: Tue, 8 Mar 2016 13:58:51 +0000 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.5.1 MIME-Version: 1.0 In-Reply-To: <1457428939-26659-2-git-send-email-stephen.boyd@linaro.org> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4730 Lines: 129 On 08/03/16 09:22, Stephen Boyd wrote: > Both the IOMMU and non-IOMMU allocations don't respect the > NO_KERNEL_MAPPING attribute, therefore drivers can't save virtual > address space and time spent mapping large buffers that are > intended only for userspace. Plumb this attribute into the code > for both types of DMA ops. I have to say I'm a little dubious about how much time we save by not creating one mapping once, and instead going on to repeatedly create and tear down lots of mappings - my initial hunch would place that number somewhere below 0. Similarly for the degree of pressure on our hundreds of gigabytes of vmalloc space, especially on "memory constrained" systems. Some real data would be a lot more convincing ;) Robin. > Cc: Robin Murphy > Cc: Laura Abbott > Cc: Arnd Bergmann > Cc: Marek Szyprowski > Cc: Catalin Marinas > Cc: Will Deacon > Signed-off-by: Stephen Boyd > --- > arch/arm64/mm/dma-mapping.c | 39 ++++++++++++++++++++++++++++++--------- > 1 file changed, 30 insertions(+), 9 deletions(-) > > diff --git a/arch/arm64/mm/dma-mapping.c b/arch/arm64/mm/dma-mapping.c > index 331c4ca6205c..06a593653f23 100644 > --- a/arch/arm64/mm/dma-mapping.c > +++ b/arch/arm64/mm/dma-mapping.c > @@ -169,6 +169,9 @@ static void *__dma_alloc(struct device *dev, size_t size, > > /* create a coherent mapping */ > page = virt_to_page(ptr); > + if (dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) > + return page; > + > coherent_ptr = dma_common_contiguous_remap(page, size, VM_USERMAP, > prot, NULL); > if (!coherent_ptr) > @@ -194,7 +197,8 @@ static void __dma_free(struct device *dev, size_t size, > if (!is_device_dma_coherent(dev)) { > if (__free_from_pool(vaddr, size)) > return; > - vunmap(vaddr); > + if (!dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) > + vunmap(vaddr); > } > __dma_free_coherent(dev, size, swiotlb_addr, dma_handle, attrs); > } > @@ -567,6 +571,9 @@ static void *__iommu_alloc_attrs(struct device *dev, size_t size, > if (!pages) > return NULL; > > + if (dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) > + return pages; > + > addr = dma_common_pages_remap(pages, size, VM_USERMAP, prot, > __builtin_return_address(0)); > if (!addr) > @@ -624,18 +631,32 @@ static void __iommu_free_attrs(struct device *dev, size_t size, void *cpu_addr, > if (WARN_ON(!area || !area->pages)) > return; > iommu_dma_free(dev, area->pages, iosize, &handle); > - dma_common_free_remap(cpu_addr, size, VM_USERMAP); > + if (!dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) > + dma_common_free_remap(cpu_addr, size, VM_USERMAP); > } else { > iommu_dma_unmap_page(dev, handle, iosize, 0, NULL); > __free_pages(virt_to_page(cpu_addr), get_order(size)); > } > } > > +static struct page **__iommu_get_pages(void *cpu_addr, struct dma_attrs *attrs) > +{ > + struct vm_struct *area; > + > + if (dma_get_attr(DMA_ATTR_NO_KERNEL_MAPPING, attrs)) > + return cpu_addr; > + > + area = find_vm_area(cpu_addr); > + if (area) > + return area->pages; > + return NULL; > +} > + > static int __iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma, > void *cpu_addr, dma_addr_t dma_addr, size_t size, > struct dma_attrs *attrs) > { > - struct vm_struct *area; > + struct page **pages; > int ret; > > vma->vm_page_prot = __get_dma_pgprot(attrs, vma->vm_page_prot, > @@ -644,11 +665,11 @@ static int __iommu_mmap_attrs(struct device *dev, struct vm_area_struct *vma, > if (dma_mmap_from_coherent(dev, vma, cpu_addr, size, &ret)) > return ret; > > - area = find_vm_area(cpu_addr); > - if (WARN_ON(!area || !area->pages)) > + pages = __iommu_get_pages(cpu_addr, attrs); > + if (WARN_ON(!pages)) > return -ENXIO; > > - return iommu_dma_mmap(area->pages, size, vma); > + return iommu_dma_mmap(pages, size, vma); > } > > static int __iommu_get_sgtable(struct device *dev, struct sg_table *sgt, > @@ -656,12 +677,12 @@ static int __iommu_get_sgtable(struct device *dev, struct sg_table *sgt, > size_t size, struct dma_attrs *attrs) > { > unsigned int count = PAGE_ALIGN(size) >> PAGE_SHIFT; > - struct vm_struct *area = find_vm_area(cpu_addr); > + struct page **pages = __iommu_get_pages(cpu_addr, attrs); > > - if (WARN_ON(!area || !area->pages)) > + if (WARN_ON(!pages)) > return -ENXIO; > > - return sg_alloc_table_from_pages(sgt, area->pages, count, 0, size, > + return sg_alloc_table_from_pages(sgt, pages, count, 0, size, > GFP_KERNEL); > } > >