lkml.org 
[lkml]   [2005]   [Apr]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Subjectmmap + dma_alloc_coherent
Date
Hi,

I would like to mmap a kernel buffer, allocated with pci_alloc_consistent()
for DMA, to userspace and came up with the following. Since there seem to be
some (unresolved) issues (see below) with this and I would like to do the
RightThing(TM), I would appreciate your comments about my stuff.

As for the unresolved issues, I found the following LKML threads to be very
helpful in understanding the problem:
1. (DMA API issues)
http://marc.theaimsgroup.com/?l=linux-kernel&m=108757847518687&w=2
2. (can device drivers return non-ram via vm_ops->nopage?)
http://marc.theaimsgroup.com/?l=linux-kernel&m=107978968703503&w=2

What I did (with comments on problematic / not fully clear to me parts):


pci_probe():
my_buffer = pci_alloc_consistent() (size: PAGE_SIZE << 4)

------------------------------------------------------------------------
my_mmap():
vma->vm_ops = &my_vm_ops;
vma->vm_flags |= (VM_RESERVED | VM_IO);
my_vma_open(vma);

my_vm_ops = {
.open = my_vm_open,
.close = my_vm_close,
.nopage = my_vm_nopage,
}

Q: Is VM_IO needed here? I took it from the sg.c driver.
Q: I choosed nopage because remap_page_range does not work on RAM pages.
Correct?
------------------------------------------------------------------------

my_vm_open():
increment vma usage count

my_vm_close():
decrement vma usage count

my_vm_nopage():
offset = (address - vma->vm_start) + (vma->vm_pgoff << PAGE_SHIFT);
pageptr = my_buffer + offset;
page = virt_to_page(pageptr);

/* got it, now increment the count */
get_page(page);

if (type)
{
*type = VM_FAULT_MINOR;
}
return page;

Q: As seen in the threads mentioned above I should not use virt_to_page() on
addresses I got from pci_alloc_consistent/dma_alloc_coherent. What is the
right way to handle this?
Q: get_page() increments the page refcount. Where is the correspondig
put_page() operation?
--------------------------------------------------------------------------

pci_remove():
pci_free_consistent(my_buffer, ...)

--------------------------------------------------------------------------


I first tried the above and failed. Somehow my driver seemed to screw up the
page tables. I noticed a function sg_correct4mmap() in the sg.c driver which
"fixes" refcount handling on pages allocated using __get_free_pages() with
order > 0. After implementing this things improved. Here are the changes:

my_mmap():
if (!mmap_called)
{
correct4mmap(my_buffer, 1);
mmap_called = 1;
}

release():
if (mmap_called)
{
if (vma_usage_count == 0)
{
correct4mmap(my_buffer, 0)
mmap_called = 0;
}
}

Q: Can someone please briefly explain, why (if at all) this is needed?
-----------------------------------------------------------------------------

Somehow the whole thing does not "feel" right so I would really like some
comments on this. I think this could be helpful to other, too.

Quoting Jeff Garzik from one of the older threads:
"My suggestion/request to the VM wizards would be to directly provide mmap
helpers for dma/mmio/pio, that Does The Right Thing. And require their
use in every driver. Don't give driver writers the opportunity to think
about this stuff and/or screw it up."

There are such helper functions on ARM and Russel tried to push them into
the generic DMA API (the last time in June 2004, I think). Can any progress
be expected regarding these helper functions?


Thanks for your time.

-Rolf

--
Rolf Offermanns <roffermanns@sysgo.com>
SYSGO AG Tel.: +49-6136-9948-0
Am Pfaffenstein 14 Fax: +49-6136-9948-10
55270 Klein-Winternheim http://www.sysgo.com

-
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/

\
 
 \ /
  Last update: 2005-04-13 12:46    [W:0.033 / U:0.148 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site