Re: [PATCH] amr(4) testers needed...

Nikolay Pavlov wrote:

> On Thursday, 13 July 2006 at 12:51:20 -0700, Doug Ambrisko wrote:
> 
>>Scott Long writes:
>>| Joao Barros wrote:
>>| > On 7/11/06, John Baldwin <john_at_baldwin.cx> wrote:
>>| > 
>>| >> I have a patch for amr(4) that is a forward-port of a Scott Long patch 
>>| >> for 4.x
>>| >> that fixes some severe data corruption with amr(4) + PAE on 4.x with > 
>>| >> 4GB of
>>| >> RAM.  I need the patch tested on current though so I can get it into 
>>| >> HEAD and
>>| >> eventually into 4.x.  The patch for head is at
>>| >> http://www.freebsd.org/~jhb/patches/amr_head.patch  It shouldn't break
>>| >> anything and should basically be a nop.  I think the patch will apply 
>>| >> to 6.x
>>| >> (and possibly 5.x) as well.  Thanks!
>>| >>
>>| > 
>>| > I can test it on i386 without PAE. Is that test enough?
>>| 
>>| Note that this problem is only present when you're using a management
>>| app at the same time as heavy disk activity is going on. If you're
>>| not using a management app (and few people are, the lack of apps is why
>>| I didn't catch this in the first place), then you're completely safe.
>>| John's email might be taken as a little alarmist in this respect.
>>
>>Hmm, I wonder why we don't see it or maybe we are :-(  We don't usually
>>hit the disk hard but have a management app. running.  I've seen
>>some strangeness.
> 
> 
> I am using this utility every hour in cron, now i have 4 processes in D
> state 
> 
> 24429  ??  D      0:00.01 /usr/local/sbin/megarc -ldInfo -LAll -a0
> 35394  ??  D      0:00.02 /usr/local/sbin/megarc -ldInfo -LAll -a0
> 37976  ??  D      0:00.02 /usr/local/sbin/megarc -ldInfo -LAll -a0
> 40526  ??  D      0:00.02 /usr/local/sbin/megarc -ldInfo -LAll -a0
> 
> I understand Scott position, but worried that megarc not killing by
> SIGKILL signal, it's not normal. I want to try preceding version of amr
> driver. Also i have 5.5 box as load balance pair for this one and never
> see such hangs on it.
> 
> 

Instead of rolling the driver back, please add the patch that John
provided.  It is safe and correct.

Scott