Hi!
I'm working on a driver for my USB 2.0 high-speed webcam under Linux
and I'm using a tool
called usbsnoop, which was designed for Windows, to get an idea on the exchanged
information between the host and the webcam. By examining the produced
trace file, I found
that my PC is sending a bunch of isochronous URBs to the webcam where
each one contains
256 isochronous packets and each packet is 3072 bytes wide. This means
that every URB
points to a 256 * 3072 bytes sized buffer.
Here is an excerpt of the trace file:
[21303 ms] >>> URB 1640 going down >>>
-- URB_FUNCTION_ISOCH_TRANSFER:
PipeHandle = ff27dd8c [endpoint 0x00000081]
TransferFlags = 00000005 (USBD_TRANSFER_DIRECTION_IN,
~USBD_SHORT_TRANSFER_OK,
USBD_START_ISO_TRANSFER_ASAP
TransferBufferLength = 000c0000
TransferBuffer = fd319100
TransferBufferMDL = 00000000
StartFrame = 00000000
NumberOfPackets = 00000100 // in hex
IsoPacket[0].Offset = 0
IsoPacket[0].Length = 0
IsoPacket[1].Offset = 3072
IsoPacket[1].Length = 0
IsoPacket[2].Offset = 6144
IsoPacket[2].Length = 0
I tried to program this behavior in my custom, alpha stage, kernel
driver, however I was
disappointed since Linux fails to allocate, using kmalloc(), such a
huge buffer (which is
quite normal actually) to associated to the URB through the
transfer_buffer field. I also
tried __get_free_pages(GFP_KERNEL, 10) without any success whatsoever.
Splitting the transfer across multiple URB doesn't seem to work (I didn't really
investigate in depth this possibility).
Any ideas?
Thanks for your time!
P.S: Is it possible to CC me since I didn't subscribe to the mailing
list? Thanks!
Best regards,
Ilyes Gouta.
Hi Ilyes, you won't be able to allocate that much *contiguous* memory,
but you should be able to allocate enough non-contiguous memory (e.g.
by calling __get_free_page 256 times; not the same as calling
__get_free_pages(8) !). To use that memory, you can try using the usb
scatter/gather support (see usb.h); I don't know if it works with
isochronous urbs though. I've CC'd the usb development list - maybe
someone there can help.
Ciao,
Duncan.
On Saturday 25 November 2006 21:42, Ilyes Gouta wrote:
> Hi!
>
> I'm working on a driver for my USB 2.0 high-speed webcam under Linux
> and I'm using a tool
> called usbsnoop, which was designed for Windows, to get an idea on the exchanged
> information between the host and the webcam. By examining the produced
> trace file, I found
> that my PC is sending a bunch of isochronous URBs to the webcam where
> each one contains
> 256 isochronous packets and each packet is 3072 bytes wide. This means
> that every URB
> points to a 256 * 3072 bytes sized buffer.
>
> Here is an excerpt of the trace file:
>
> [21303 ms] >>> URB 1640 going down >>>
> -- URB_FUNCTION_ISOCH_TRANSFER:
> PipeHandle = ff27dd8c [endpoint 0x00000081]
> TransferFlags = 00000005 (USBD_TRANSFER_DIRECTION_IN,
> ~USBD_SHORT_TRANSFER_OK,
> USBD_START_ISO_TRANSFER_ASAP
> TransferBufferLength = 000c0000
> TransferBuffer = fd319100
> TransferBufferMDL = 00000000
> StartFrame = 00000000
> NumberOfPackets = 00000100 // in hex
> IsoPacket[0].Offset = 0
> IsoPacket[0].Length = 0
> IsoPacket[1].Offset = 3072
> IsoPacket[1].Length = 0
> IsoPacket[2].Offset = 6144
> IsoPacket[2].Length = 0
>
> I tried to program this behavior in my custom, alpha stage, kernel
> driver, however I was
> disappointed since Linux fails to allocate, using kmalloc(), such a
> huge buffer (which is
> quite normal actually) to associated to the URB through the
> transfer_buffer field. I also
> tried __get_free_pages(GFP_KERNEL, 10) without any success whatsoever.
>
> Splitting the transfer across multiple URB doesn't seem to work (I didn't really
> investigate in depth this possibility).
>
> Any ideas?
>
> Thanks for your time!
>
> P.S: Is it possible to CC me since I didn't subscribe to the mailing
> list? Thanks!
>
> Best regards,
> Ilyes Gouta.
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
On Mon, 27 Nov 2006, Duncan Sands wrote:
> Hi Ilyes, you won't be able to allocate that much *contiguous* memory,
> but you should be able to allocate enough non-contiguous memory (e.g.
> by calling __get_free_page 256 times; not the same as calling
> __get_free_pages(8) !). To use that memory, you can try using the usb
> scatter/gather support (see usb.h); I don't know if it works with
> isochronous urbs though. I've CC'd the usb development list - maybe
> someone there can help.
The scatter/gather library does not support iso.
> > Splitting the transfer across multiple URB doesn't seem to work (I didn't really
> > investigate in depth this possibility).
It should work. Go back and try it again more carefully.
Alan Stern