[Dri-devel] PPC Lockup (ati-pcigart-branch)

[Jeff Hartmann]

Dan Malek wrote:

> Jeff Hartmann wrote:
>
>> Okay let me try and explain things a little better.
>
>
> Got it.  That is what I was guessing, only surprised you are holding
> page structs, but that makes sense.
>
>> ......  Currently there is no kernel function to do
>> this explicitly
>
>
> I'm working on that.  The PowerPC port cheated by using BATs and
> trivial macros, but this doesn't work on some of the newer processors
> and more complex applications.  Other architectures did the same, and
> I am surprised there aren't generic kernel functions to track down this
> information.  In fact, these functions are already present for 4xx and
> 8xx processors, so don't write anything new.
>
>> Another thing that happens later is that we need the bus address of each
>> of these pages to program the card to do scatter gather dma from this
>> region.
>
>
> That's where this is going to fall apart on PowerPC.
>
>> .....  We do virt_to_bus(pagelist[i]->virtual) to accomplish this
>> translation.
>
>
> I have to write some code (or actually remove some #ifdefs) before
> this will work for you.
>
>> I know on the ia32 a pgd/pmd can actually point to 4MB pages rather then
>> a real pte.  Does the PowerPC have anything like this?
>
>
> Not yet.  It's on the way....
>
>> ....  I would doubt
>> that I would encounter anything like this from a vmalloc'ed area of
>> memory (since vmalloc is arch independent and it would call alloc_page
>> for each individual pte.)  Am I correct in this assumption?
>
>
> Yes.
>
>> Just FYI, the code I posted works fine on the ia32 platform (only tested
>> with the i386 classic 2-level page tables.)
>
>
> What you are doing so far should work too on PowerPC.
>
>> Another thing we might be running into here is that vmalloc does not
>> guarantee a virtually contiguous area of memory (or so I am told.)
>
>
> Ummm...of course it is virtually contiguous.  How could it be
> different?  You request a size, and it returns a base virtual address.
> If there were holes in it, how would you know?

Look at vread in vmalloc.c, I think it would handle holes in a
vmalloc'ed area (From a brief reading of the code.)  I've seen postings
about this on linux-kernel.  I don't see a vwrite implementation, but I
would assume you would have to do something similar for writes.

>
>
>> .....  I've
>> NEVER seen this in practice on an ia32 platform.
>
>
> It can't happen on any platform (or I don't understand something about
> the comment, which could very well be the case today).

I think it can, I've seen numerous people talk about it on
linux-kernel.  I've also been told that my /dev/agpgart isn't 'safe'
because it assumes vmalloc'ed memory is always virtually contiguous.  I
think this only happens when there isn't enough virtual address space in
the kernel and its fragmented (probably only happens on machines with
lots of memory.)

-Jeff


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/