Greetings,

On 12-Sep-01 Oliver Neukum wrote:
>> The second bug report said the module works fine when the person limits
>> his
>> memory to 960 MB (how, I dont know). With full memory, he gets artefacts
> 
> With a 'mem=' kernel command line I presume.
> 
>> which to me suggests the second frame buffer is allocated, but somehow in
>> memory which isnt accessible (every other frame is bright pink --> filled
>> with 0xFF before YUV->RGB translation?)
> 
> Not accessible by whom ?
> The USB controller or the CPU ? 1GB should be accessible to the USB 
> controller.

The CPU; there are 3 buffers involved; a DMA-able buffer used for the ISO
transfers (kmalloc()ed), a ring buffer for whole frames (vmalloc()ed), and
finally the image buffer that is used for decompression (rvmalloc()ed, and
mmap()able). The frame buffers seem to have trouble during vmalloc.

I�m wondering... I�m vmallocing 3 buffers in rapid succession during open();
is there possibly a limit to the number of vmalloc()ed areas a process may
have?

Also I noticed my driver is the only USB driver allocating memory with
vmalloc(); others use rvmalloc() or vmalloc_32() directly.

>> Unfortunately Im not rich enough to put 1GB of RAM in my machine to test
>> this... ;-(
> 
> If it is indeed a problem of accessability by the CPU you might use the
> patch 
> Andrea Arcangeli has written which allows you the use highmen with 128MB.

You mean his patch simulates the highmem stuff on machines with less than 1
GB? That might be worth a try (mine has 320 MB RAM; plenty to play around
with). But no, it doesn�t seem like a CPU problem (the data is copied from
the ISO buffers to the frame buffers with a simple memcpy()).

 - Nemosoft

-----------------------------------------------------------------------------
Try SorceryNet!   One of the best IRC-networks around!   irc.sorcery.net:9000
URL: never        IRC: nemosoft      IscaBBS (bbs.isca.uiowa.edu): Nemosoft
                        >> Never mind the daylight << 

_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to