linux-kernel - Re: [PATCH v4 4/7] iommu: Add quirk for Intel graphic devices in map

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <281f22ed-8b36-fc20-4b08-27ef168f941d@arm.com>
Date:   Tue, 3 Nov 2020 12:00:33 +0000
From:   Robin Murphy <robin.murphy@....com>
To:     Lu Baolu <baolu.lu@...ux.intel.com>,
        Joerg Roedel <joro@...tes.org>, Tom Murphy <murphyt7@....ie>,
        David Woodhouse <dwmw2@...radead.org>,
        Christoph Hellwig <hch@...radead.org>
Cc:     Tvrtko Ursulin <tvrtko.ursulin@...ux.intel.com>,
        Ashok Raj <ashok.raj@...el.com>,
        Intel-gfx@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        iommu@...ts.linux-foundation.org
Subject: Re: [PATCH v4 4/7] iommu: Add quirk for Intel graphic devices in
 map_sg

On 2020-09-27 07:34, Lu Baolu wrote:
> Combining the sg segments exposes a bug in the Intel i915 driver which
> causes visual artifacts and the screen to freeze. This is most likely
> because of how the i915 handles the returned list. It probably doesn't
> respect the returned value specifying the number of elements in the list
> and instead depends on the previous behaviour of the Intel iommu driver
> which would return the same number of elements in the output list as in
> the input list.
> 
> Signed-off-by: Tom Murphy <murphyt7@....ie>
> Signed-off-by: Lu Baolu <baolu.lu@...ux.intel.com>
> ---
>   drivers/iommu/dma-iommu.c | 27 +++++++++++++++++++++++++++
>   1 file changed, 27 insertions(+)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 3526db774611..e7e4d758f51a 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -879,6 +879,33 @@ static int __finalise_sg(struct device *dev, struct scatterlist *sg, int nents,
>   	unsigned int cur_len = 0, max_len = dma_get_max_seg_size(dev);
>   	int i, count = 0;
>   
> +	/*
> +	 * The Intel graphic driver is used to assume that the returned
> +	 * sg list is not combound. This blocks the efforts of converting
> +	 * Intel IOMMU driver to dma-iommu api's. Add this quirk to make the
> +	 * device driver work and should be removed once it's fixed in i915
> +	 * driver.
> +	 */
> +	if (IS_ENABLED(CONFIG_DRM_I915) && dev_is_pci(dev) &&
> +	    to_pci_dev(dev)->vendor == PCI_VENDOR_ID_INTEL &&
> +	    (to_pci_dev(dev)->class >> 16) == PCI_BASE_CLASS_DISPLAY) {
> +		for_each_sg(sg, s, nents, i) {
> +			unsigned int s_iova_off = sg_dma_address(s);
> +			unsigned int s_length = sg_dma_len(s);
> +			unsigned int s_iova_len = s->length;
> +
> +			s->offset += s_iova_off;
> +			s->length = s_length;
> +			sg_dma_address(s) = dma_addr + s_iova_off;
> +			sg_dma_len(s) = s_length;
> +			dma_addr += s_iova_len;
> +
> +			pr_info_once("sg combining disabled due to i915 driver\n");
> +		}
> +
> +		return nents;
> +	}

BTW, a much less invasive workaround would be to simply override 
seg_mask to 0. That's enough to make sure that no segment looks eligible 
for merging.

Robin.

> +
>   	for_each_sg(sg, s, nents, i) {
>   		/* Restore this segment's original unaligned fields first */
>   		unsigned int s_iova_off = sg_dma_address(s);
>