RE: Out of bounce buffers

I will be answering this one myself

Further debugging seems fortunately to vindicate the ATA drivers - I tried
to disable most other drivers - USB, if etc. And the problem goes away 

I now need to tie down where the problem occurs - but what seems to happen
is that a DMA read sanity check fails due to the segment has been
overwritten/changed/freed while the ATA driver was holding the same tag 


-----Original Message-----
From: owner-freebsd-current_at_freebsd.org
[mailto:owner-freebsd-current_at_freebsd.org] On Behalf Of Thomas Sparrevohn
Sent: 13 October 2008 22:08
To: 'Søren Schmidt'; Current_at_freebsd.org
Subject: Out of bounce buffers 



Hi 

I have had a weird issue since the change in April where the SATA Port
multipliers was introduced. The system started issuing "DATA load FAILURE"
in the ata-dmaload every time that disk subsystem got loaded. Because the
code prior from 10/04-2008 worked fine and Ignored the problem and simply
used the code prior to the SATA Port multiplier was introduced. The reason
was twofold - One I did not time to investigate and two I suspected it could
be a hw error - using cheap SATA disks.

Well finally I have had time to investigate and the problem occurs because
the system "runs out of bounce buffers" 

I have tested the entire disk for errors using a Hitachi tools and there
seems to no problems - down to the degree that I can get the system to work
with the current ATA subsystem as long as I only use one disk ;-) Regardless
of which of the 4 disks - 

however if I add a second disk to the subsystem it will eventually (after
17-40GB of restores) give me a "DATA load failures" - Naturally I suspected
the disk drives - however as I said - I can fully restore without the error
occurring as long as I only use one disk

Puzzled by this I started plastering the code with printf's to see what
happened - patch included just FYI - It turns out that the kernel runs out
of bounce buffers - which is very surprising - Can anybody give any advice
on this?