On Mon, Mar 14, 2011 at 4:29 PM, Jacob Meuser
wrote:
>
> i386
>
> 32-bit mfi frames (current code)
> 32m23.04s 33m50.57s 10m49.27s
> 32m31.09s 33m48.24s 10m51.21s
>
> 64-bit mfi frames (with patch)
> 32m29.76s 33m36.95s 11m3.20s
> 32m41.15s
On Monday, March 14, Jacob Meuser wrote:
>
> so, the question is, is the impact on i386 enough to warrant using
> 32-bit frames on 32-bit platforms? if so, should this be decided
> at runtime or compile time?
>
> any other thoughts?
IMHO, simpler is better, and i386 may have PAE so more than 4G
the following diff make mfi(4) use 64-bit frames, and support 64-bit
dma addresses. these changes are based on freebsd's mfi(4). however,
freebsd only uses 64-bit frames 'if (sizeof(bus_addr_t)) == 8',
whereas this patch uses 64-bit frames unconditionally, for both 32-bit
and 64-bit platforms. I
