Re: [PATCH] PCI and PCI Hotplug update for 2.6.6-rc1

From: Greg KH (greg_at_kroah.com)
Date: 04/15/04

  • Next message: Greg KH: "Re: [PATCH] PCI and PCI Hotplug update for 2.6.6-rc1" 
    Date:	Thu, 15 Apr 2004 10:23:46 -0700
To: linux-kernel@vger.kernel.org

    ChangeSet 1.1692.3.10, 2004/03/31 14:57:52-08:00, dsaxena@plexity.net

    [PATCH] PCI: Allow arch-specific pci_set_dma_mask and friends

    The patch provides the ability for architectures to have custom
    implementations of pci_set_dma_mask() and friends (dac_set_dma_mask
    and set_consistent_dma_mask). The reason I need this is b/c I have
    a chipset (Intel ARM IXP425) that has a broken PCI interface that
    only allows PCI dma to/from the bottom 64MB of system memory. To get
    around this limitation, I trap a custom dma-mapping implementation that
    bounces buffers outside the 64MB window. At device discover time, my
    custom platform_notify() function gets called and it sets the dma_mask
    to (64MB-1) and in ARM's dma-mapping code, I check for dma_mask != 0xffffffff
    and if that is true, I call the special bounce helpers. This works great
    except that certain drivers (e100, ide-pci) call pci_set_dma_mask()
    with 0xffffffff and the generic implementation only allows for the
    architecture-defined pci_dma_supported() to return true or false. There
    is no method for the architecture to tell the PCI layer "I can't set
    the mask to 0xffffffff, but I can set it to this other value" and there
    is no way to pass that back to the driver. What this means is that if
    I have pci_set_dma_supported() return failure on full 32-bit DMA, the
    driver will not initialize the card; however, if I return true,
    pci_set_dma_mask() will set the dma mask to full 32-bits and I can no
    longer trap and will have buffers that are not dma-able and cause
    PCI master aborts. Both of those are not acceptable. IMHO, the
    driver shouldn't care if the architecture has to bounce DMA outside of
    64MB and since this is not something most architectures have to worry
    about, the easiest way to get around the issue is by allowing custom
    pci_set_dma_mask() for arches that need it but keeping the generic
    implementation for those that do not. In my case, it simply returns
    0 to the driver but keeps the device mask set to 64MB-1 so I can trap.

     drivers/pci/pci.c | 5 +++++
     1 files changed, 5 insertions(+)

    diff -Nru a/drivers/pci/pci.c b/drivers/pci/pci.c
    --- a/drivers/pci/pci.c Thu Apr 15 10:04:10 2004
    +++ b/drivers/pci/pci.c Thu Apr 15 10:04:10 2004
    @@ -700,6 +700,10 @@
             }
     }
     
    +#ifndef HAVE_ARCH_PCI_SET_DMA_MASK
    +/*
    + * These can be overridden by arch-specific implementations
    + */
     int
     pci_set_dma_mask(struct pci_dev *dev, u64 mask)
     {
    @@ -732,6 +736,7 @@
     
             return 0;
     }
    +#endif
          
     static int __devinit pci_init(void)
     {

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Greg KH: "Re: [PATCH] PCI and PCI Hotplug update for 2.6.6-rc1"

    Relevant Pages

    • Re: PCI card transfer with DMA/ busmastering
      ... I have a custom PCI card that I intend on using as a coprocessor to ... I've read a lot regarding DMA transfer, I assume this is the way to go. ... When I create my driver for my PCI card does the DMA stuff need to go ...
      (microsoft.public.development.device.drivers)
    • Call for testers: owners of ixgb(4) hardware
      ... I've came accross a bad mistake in the ixgbdriver that will prevent ... it from working on architectures where the DMA address is not the ... also test his card on sparc64 hardware, ...
      (freebsd-current)
    • Re: [PATCH 1/3 v3] dma: document dma_{un}map_{single|sg}_attrs() interface
      ... +It provides a mechanism for devices to explicitly order their DMA ... The same driver needs to compile on all architectures. ... we end with subtle driver bugs because the semantics aren't ...
      (Linux-Kernel)
    • Re: ALSA vs. non coherent DMA
      ... This is a mmap of the data record to be shared in realtime with apps. ... No using the appropriate page protection for a DMA coherent mapping ... architectures. ... an error or make it out via ifdef. ...
      (Linux-Kernel)
    • Re: [PATCH] asm-generic: add dma-mapping-linear.h
      ... the architectures I converted. ... int nents, enum dma_data_direction direction) ... whether there is any reason for the dma flush here, ... +dma_supported(struct device *dev, u64 mask) ...
      (Linux-Kernel)