RE: kmalloc vs pci_alloc_consistent

RE: kmalloc vs pci_alloc_consistent

[Johnson, Stephen]

I found a nice thread of conversation that allowed me to see some of the
history behind pci_alloc_consistent().

I read one thing noted on a man page for kmalloc that should be highlighted.
"the GFP_DMA flag may be set to indicate the memory must be suitable for
DMA. ***This can mean different things on different platforms.*** For
example, on x86, it means that the memory must come from the first 16MB."

For ppc arch, how about
#define kmalloc  pci_alloc_unconsistent  /* just joking */




-----Original Message-----
From: Johnson, Stephen
Sent: Friday, July 18, 2003 8:34 AM
To: linuxppc-embedded@lists.linuxppc.org
Subject: kmalloc vs pci_alloc_consistent



I don't know the history behind pci_alloc and wanted to throw this out there
to get some clarification.

What's the benefit of pci_alloc over using kmalloc with flags for a
cache-safe buffer?


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

Re: kmalloc vs pci_alloc_consistent

[Chris Zimman]

On Fri, Jul 18, 2003 at 09:56:35AM -0500, Johnson, Stephen wrote:
> I read one thing noted on a man page for kmalloc that should be highlighted.
> "the GFP_DMA flag may be set to indicate the memory must be suitable for
> DMA. ***This can mean different things on different platforms.*** For
> example, on x86, it means that the memory must come from the first 16MB."

Just to clarify -- the 16MB lower memory limit is only true on x86 when doing
DMA via ISA.  PCI does not have this problem.

In addition to making sure you get pages that are usable for DMA,
pci_alloc_consistent() also takes care of cache coherency issues.

--Chris

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

kmalloc vs pci_alloc_consistent

[Johnson, Stephen]

I don't know the history behind pci_alloc and wanted to throw this out there
to get some clarification.

What's the benefit of pci_alloc over using kmalloc with flags for a
cache-safe buffer?


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

Re: kmalloc vs pci_alloc_consistent

[Matt Porter]

On Fri, Jul 18, 2003 at 08:34:24AM -0500, Johnson, Stephen wrote:
>
> I don't know the history behind pci_alloc and wanted to throw this out there
> to get some clarification.
>
> What's the benefit of pci_alloc over using kmalloc with flags for a
> cache-safe buffer?

Documentation/DMA-mapping.txt has a detailed explanation of the API.

Regards,
--
Matt Porter
mporter@kernel.crashing.org

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