RE: [Patch] Move swiotlb_init early on X86_64

[Pallipadi, Venkatesh]

>-----Original Message-----
>From: Andi Kleen [mailto:ak@xxxxxxx] 
>Sent: Thursday, March 02, 2006 5:32 PM
>To: Zou, Nanhai
>Cc: Zhang, Yanmin; Luck, Tony; LKML; Andrew Morton; Pallipadi, 
>Venkatesh
>Subject: Re: [Patch] Move swiotlb_init early on X86_64
>
>On Friday 03 March 2006 00:35, Zou Nan hai wrote:
>
>> This patch modify the bootmem allocator,
>> let normal bootmem allocation on 64 bit system first go above 4G
>> address.
>
>That's very ugly and likely to break some architectures. Sorry
>but #ifdefs is the wrong way to do this.
>
>Passing a limit parameter is better and use that in the swiotlb
>allocation. If you're worried about changing too many callers
>you could add a new entry point.
>

Another potential issue with this approach:
On a 64 bit system with less than 4G phys memory, we will fail
to get any memory above 4G and fall back to start from '0'.
This is different from original behaviour, where goal was 
MAX_DMA_ADDRESS (16M) and we would allocate memory starting 
from 16M. As a result, we will now eat up memory in 0-16M range 
and may break some legacy drivers as they will not get any memory.

If we go this way, then we should fallback to original goal if we 
are not able to get greater than 4G memory.

Thanks,
Venki
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/