linux-kernel - Re: [PATCH 03/25] dma-direct: take dma-ranges/offsets into account in resource mapping

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: <20220421173523.ig62jtvj7qbno6q7@mobilestation>
Date:   Thu, 21 Apr 2022 20:35:23 +0300
From:   Serge Semin <fancer.lancer@...il.com>
To:     Christoph Hellwig <hch@....de>
Cc:     Robin Murphy <robin.murphy@....com>,
        Serge Semin <Sergey.Semin@...kalelectronics.ru>,
        Gustavo Pimentel <gustavo.pimentel@...opsys.com>,
        Vinod Koul <vkoul@...nel.org>,
        Jingoo Han <jingoohan1@...il.com>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        Frank Li <Frank.Li@....com>,
        Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>,
        Marek Szyprowski <m.szyprowski@...sung.com>,
        Vladimir Murzin <vladimir.murzin@....com>,
        Alexey Malahov <Alexey.Malahov@...kalelectronics.ru>,
        Pavel Parkhomenko <Pavel.Parkhomenko@...kalelectronics.ru>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Rob Herring <robh@...nel.org>,
        Krzysztof Wilczyński <kw@...ux.com>,
        linux-pci@...r.kernel.org, dmaengine@...r.kernel.org,
        linux-kernel@...r.kernel.org, iommu@...ts.linux-foundation.org
Subject: Re: [PATCH 03/25] dma-direct: take dma-ranges/offsets into account
 in resource mapping

On Thu, Apr 21, 2022 at 04:45:36PM +0200, Christoph Hellwig wrote:
> On Wed, Apr 20, 2022 at 11:55:38AM +0300, Serge Semin wrote:
> > On Wed, Apr 20, 2022 at 10:47:46AM +0200, Christoph Hellwig wrote:
> > > I can't really comment on the dma-ranges exlcusion for P2P mappings,
> > > as that predates my involvedment, however:
> > 
> > My example wasn't specific to the PCIe P2P transfers, but about PCIe
> > devices reaching some platform devices over the system interconnect
> > bus.
> 
> So strike PCIe, but this our definition of Peer to Peer accesses.
> 
> > What if I get to have a physical address of a platform device and want
> > have that device being accessed by a PCIe peripheral device? The
> > dma_map_resource() seemed very much suitable for that. But considering
> > what you say it isn't.
> 

> dma_map_resource is the right thing for that.  But the physical address
> of MMIO ranges in the platform device should not have struct pages
> allocated for it, and thus the other dma_map_* APIs should not work on
> it to start with.

The problem is that the dma_map_resource() won't work for that, but
presumably the dma_map_sg()-like methods will (after some hacking with
the phys address, but anyway). Consider the system diagram in my
previous email. Here is what I would do to initialize a DMA
transaction between a platform device and a PCIe peripheral device:

1) struct resource *rsc = platform_get_resource(plat_dev, IORESOURCE_MEM, 0);

2) dma_addr_t dar = dma_map_resource(&pci_dev->dev, rsc->start, rsc->end - rsc->start + 1,
                                      DMA_FROM_DEVICE, 0);

3) dma_addr_t sar;
   void *tmp = dma_alloc_coherent(&pci_dev->dev, PAGE_SIZE, &sar, GFP_KERNEL);
   memset(tmp, 0xaa, PAGE_SIZE);

4) PCIe device: DMA.DAR=dar, DMA.SAR=sar. RUN.

If there is no dma-ranges specified in the PCIe Host controller
DT-node, the PCIe peripheral devices will see the rest of the system
memory as is (no offsets and remappings). But if there is dma-ranges
with some specific system settings it may affect the PCIe MRd/MWr TLPs
address translation including the addresses targeted to the MMIO
space. In that case the mapping performed on step 2) will return a
wrong DMA-address since the corresponding dma_direct_map_resource()
just returns the passed physical address missing the
'pci_dev->dma_range_map'-based mapping performed in
translate_phys_to_dma().

Note the mapping on step 3) works correctly because it calls the
translate_phys_to_dma() of the direct DMA interface thus taking the
PCie dma-ranges into account.

To sum up as I see it either restricting dma_map_resource() to map
just the intra-bus addresses was wrong or there must be some
additional mapping infrastructure for the denoted systems. Though I
don't see a way the dma_map_resource() could be fixed to be suitable
for each considered cases.

-Sergey

map the platforms