lkml.org 
[lkml]   [2011]   [Jan]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH 2/2] USB: ehci: tegra: Align DMA transfers to 32 bytes
    On Mon, Jan 24, 2011 at 08:36:55AM -0800, Alan Stern wrote:
    > On Thu, 20 Jan 2011, Robert Morell wrote:
    >
    > > The Tegra2 USB controller doesn't properly deal with misaligned DMA
    > > buffers, causing corruption. This is especially prevalent with USB
    > > network adapters, where skbuff alignment is often in the middle of a
    > > 4-byte dword.
    > >
    > > To avoid this, allocate a temporary buffer for the DMA if the provided
    > > buffer isn't sufficiently aligned.
    > >
    > > Signed-off-by: Robert Morell <rmorell@nvidia.com>
    > > ---
    > > drivers/usb/host/ehci-tegra.c | 92 +++++++++++++++++++++++++++++++++++++++++
    > > 1 files changed, 92 insertions(+), 0 deletions(-)
    > >
    > > diff --git a/drivers/usb/host/ehci-tegra.c b/drivers/usb/host/ehci-tegra.c
    > > index 2341904..7cdfc65 100644
    > > --- a/drivers/usb/host/ehci-tegra.c
    > > +++ b/drivers/usb/host/ehci-tegra.c
    > > @@ -32,6 +32,8 @@
    > > #define TEGRA_USB_USBMODE_HOST (3 << 0)
    > > #define TEGRA_USB_PORTSC1_PTC(x) (((x) & 0xf) << 16)
    > >
    > > +#define TEGRA_USB_DMA_ALIGN 32
    > > +
    > > struct tegra_ehci_context {
    > > bool valid;
    > > u32 command;
    > > @@ -461,6 +463,94 @@ static int tegra_ehci_bus_resume(struct usb_hcd *hcd)
    > > }
    > > #endif
    > >
    > > +struct temp_buffer {
    > > + void *kmalloc_ptr;
    > > + void *old_xfer_buffer;
    > > + u8 data[0];
    > > +};
    > > +
    > > +static void free_temp_buffer(struct urb *urb, int status)
    > > +{
    > > + enum dma_data_direction dir;
    > > + struct temp_buffer *temp;
    > > +
    > > + if (!(urb->transfer_flags & URB_ALIGNED_TEMP_BUFFER))
    > > + return;
    > > +
    > > + dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
    > > +
    > > + temp = container_of(urb->transfer_buffer, struct temp_buffer,
    > > + data);
    > > +
    > > + if (dir == DMA_FROM_DEVICE && !status)
    > > + memcpy(temp->old_xfer_buffer, temp->data,
    > > + urb->transfer_buffer_length);
    >
    > Even if status is nonzero, there may be valid data in the buffer. You
    > should skip that test.

    Thanks for looking, Alan. I added that test based on earlier feedback.
    I think the big concern here is security: if the URB fails in such a way
    that the buffer is not overwritten, then we may copy out freed kernel
    data to userspace.

    Are there specific status codes that I can check for here? I guess the
    only other option is to remove the direction check from the alloc path
    or alloc with GFP_ZERO.

    Thanks,
    Robert

    > No other problems that I can see.
    >
    > Alan Stern
    >
    >


    \
     
     \ /
      Last update: 2011-01-24 23:55    [W:4.081 / U:0.068 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site