[lkml]   [2002]   [Jan]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: pci_alloc_consistent from interrupt == BAD

On Fri, 18 Jan 2002, Troy Benjegerdes wrote:

> Somehow the docs in DMA-mappings.txt say pci_alloc_consistent is allowed from
> interrupt, but this is a "bad thing" on at least arm and PPC non-cache
> coherent cpus.

Everything gets bad when it doesn't work for you. :-)

> On these cpus we have to allocate page tables for consistent_alloc.. I really
> dont' think we want to be doing this during interrupt context.
> This also causes the sym53c8xx_2 driver to not work on some embedded ppc 4xx
> boards, and in general, seems to be a 'bad thing' to allow.

I have noted that some ports may [ever] require pci_alloc_consistent not
to be called from interrupt context. Just I will look into this when time
will allow.

Btw, I thought long ago about using a thread in the driver to deal with
completions and some other events on which the driver may want to allocate
memory. But this would have added useless latency. Instead I preferred to
be careful about the driver to be as fast as possible when completing an
IO under interrupt.

Btw-bis, FreeBSD-5 wants to thread everything (as most of you know). We
will see the result... No need to say that I posted my disagreement when
this project started. As FreeBSD-5 should be ready when Linux-2.6 will be,
we will be able to compare interrupt threading cost against simple fast
interrupt on typical hardware (99% are UP in my guessing).

> For example, in the arch/arm/consistent.c:
> /*
> * This allocates one page of cache-coherent memory space and returns
> * both the virtual and a "dma" address to that space. It is not clear
> * whether this could be called from an interrupt context or not. For
> * now, we expressly forbid it, especially as some of the stuff we do
> * here is not interrupt context safe.
> *
> * Note that this does *not* zero the allocated area!
> */
> void *consistent_alloc(int gfp, size_t size, dma_addr_t *dma_handle)
> (arm's pci_alloc_consistent always calls consistent_alloc).

ARM arch is simplistic. Wanting to run modern O/Ses on this crap looks
utter stupidity to me. I don't care about ARM at all.

> The PPC version calls a similiar function when CONFIG_NOT_COHERENT_CACHE is
> defined. On 'regular' ppc machines, it's just a __get_free_pages, which is
> why no one from the pmac crowd has screamed.

I am not going to ever use not cache coherent hardware, even if I am ready
to make the sym driver work reliably on such brain-dead things. Just it is
not high priority stuff for now.


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:15    [W:0.113 / U:2.000 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site