Re: Port Multiplier access with Sil 3124

[Linda Walsh <lkml@tlinx.org>]

Grant Grundler wrote:
> Linda,
> Please tell us which kernel version you have and include the dmesg
> output. Port multiplier support is generally working in 2.6.26 (and later
> releases) for several drivers.  PM support was added after 2.6.20.
---
    Sorry.  I'm a bit behind the edge at 2.6.27.3 --
but I tried a boot w/2.6.28.3 and didn't see a difference (had to return
to 27.3 as I accidently disabled a different driver and haven't regened it.

>
> Where did you find 350MB/s theoretical?
> Link is 8b10b encoded. So it's 10:1 of bits to bytes conversion of link rate.
> 3Gb/s --> 300MB/s in theory.
---
    Yep.  Forgot about that... was forgetting about the 2-bits of overhead,
so was using 3Gb/8/(1024*1024) -> 357.6, which I rounded down to 350... 
but forgot about the 10/8 encoding (just as I get rid of that conversion
habit from modem days, now I have to remember it as a special case for
SATA...:-)  ).

>
> For MB/s, it depends on the SATA controller. Don't expect more then
> 225-235 MB/s per port. At least one sata_sil3124 chip (3126?) is buggy
> and won't do more than 120MB/s read for all ports (170 MB/s write).
> See linux-ide archives for discussion on this.
---
    That's horrible!  I'd definitely call that a bug.  Right now,
the fastest I'm getting on my single ATA's is 70-80MB/s.  I could
get that per/ATA port.  So 120MB read for 4 SATA ports would be
a giant step backwards (170MB/s write?  write faster than read?)