On Sun, Oct 17, 2010 at 10:52 PM, Rajat Sharma <fs.rajat@xxxxxxxxx> wrote:
Hi,
I was trying to attempt DIRECT I/O on 4K aligned kernel buffer, but it looks like filesystems are not happy with kernel buffers. I tried doing IO on ext3, ext4 and NFS. Also I tried to allocate kernel buffer with kmalloc, vmalloc and static kernel buffer, but nothing help.
My guess is get_user_pages called to pin pages in memory is not happy with kernel pages and it is returning -EFAULT.
As a workaround, I tried to map virtual memory in process address space using do_mmap and it worked fine. Although it helped, it looks like a hack and causes performance penalty.
Please let me know if there is another cleaner way of doing this?
Thanks,
Rajat
