Re: TSTILE lockups on disk dump/backups

[Don Lee <MacPPC2%c.icompute.com@localhost>]

I installed the 10.1 kernel on my recently upgraded NetBSD 9.4 system. Until the 10.1 kernel, my umps were running, with the occasional TSTILE lockup. All seemed to be working with the 10.1 kernel, with no TSTILE lockups so far, until the weekly backups ran.

First symptom was that the “verify” pass of my dumps was generating thousands of errors. I broke the verify apart from the dump, and ran the verify pass on the dump files already generated. I discovered that the dumps were the problem, not the verify. I tried running the gzipped dump files to disk, and verifying those dump files. Dump files written to disk, and then read from disk ran fine. Dumps run to NFS, and then read from NFS generate errors.

As an added check, I copied the dumps that I ran to disk over to NFS (using cp(1)), and ran the verify on the NFS copy. Verify ops from NFS of copied dump files came through without error. This strongly suggests that the "dump | gzip” operation was generating bad data.

The pipe for my dump operations is:


The dumps run with:

	LABEL=YYMMDD # time/date label
	STAGE=target.gz
	FS=	/usr	# filesystem to be dumped - / /usr /var /home
	dump -t -0  -f - -L ${LABEL} -a ${FS} | gzip -c > ${STAGE}

Evidence suggests strongly that data piped into gzip and run to NFS get corrupted.

uname -a:

	NetBSD mercy 10.1 NetBSD 10.1 (GENERIC) #0: Mon Dec 16 13:08:11 UTC 2024 mkrepro%mkrepro.NetBSD.org@localhost:/usr/src/sys/arch/macppc/compile/GENERIC macppc 

Am I doing something wrong, or do we have an NFS bug in 10.1??

-dgl-


>> On Jan 1, 2025, at 9:48 AM, Rin Okuyama <rokuyama.rk%gmail.com@localhost> wrote:
>> 
>> Hi,
>> 
>> Can you please try kernel from 10.1 release (or -current)?
>> 
>> https://ftp.netbsd.org/pub/NetBSD/NetBSD-10.1/macppc/binary/sets/
>> 
>> It should be working with 9.x userland.
>> 
>> For netbsd-9, pmap for powerpc/oea is known to be broken.
>> But, unfortunately, it exceeds our capacity to pull up
>> fixes made to -current/netbsd-10 at the moment...
>> 
>> Thanks,
>> rin
>> 
>> On 2024/12/31 14:34, Don Lee wrote:
>>> Another instance of the tstile hang with dump/gzip/nfs
> [snip]
>