Re: kern/54945 (nfs uvm_pagegetdirty panic)

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

I encountered a similar assertion failure, but in a different place:

panic: kernel diagnostic assertion "!UVM_OBJ_IS_CLEAN(pg->uobject) || uvm_pagegetdirty(pg) == UVM_PAGE_STATUS_CLEAN" failed: file "../../../../uvm/uvm_fault.c", line 1868
cpu0: Begin traceback...
0x00ce7cc0: at vpanic+0x130
0x00ce7cf0: at kern_assert+0x60
0x00ce7d30: at uvm_fault_internal+0x150c
0x00ce7e70: at trap+0x560
0x00ce7f20: user DTMISS trap by 0x181435c: srr1=0xc030
            r1=0xfffe9a50 cr=0x24022848 xer=0 ctr=0xfdd5ce5c esr=0 pid=0xf4
cpu0: End traceback...
Stopped in pid 858.1 (make) at  netbsd:vpanic+0x134:    or      r3, r29, r29

The machine is an evbppc/ibm4xx box with NFS root, and panic occurred when
compiling pkgsrc (working directory was also NFS). Kernel was as of Feb 18:

NetBSD obs266 9.99.47 NetBSD 9.99.47 (OBS266) #149: Tue Feb 18 11:44:40 JST 2020  rin@latipes:/build/src/sys/arch/evbppc/compile/OBS266 evbppc

with local changes, that work stable if applied to netbsd-9.

I tried a kernel built from the same source with UVMHIST enabled, but
it survived under heavy load over 36 hours, and I gave up testing ;-).

I found that some fixes for NFS were done in yamt-pagecache branch at
the same time as commit you merged to -HEAD on Jan 15:

http://www.nerv.org/netbsd/?q=id:20111102T215402Z.4a644bae705ba7daf75cfd19cf9f1d33bfb39f24

> Modified Files:
> ...
> 	src/sys/nfs [yamt-pagecache]: nfs_bio.c nfs_subs.c
> ...
> Log Message:
> page cache related changes
> ...
> - fix nfs commit range tracking.
> - fix nfs write clustering.  XXX hack
> ...

Neither changes in nfs_bio.c:

http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/nfs/nfs_bio.c.diff?r1=1.188&r2=1.188.2.1&only_with_tag=yamt-pagecache&f=H

nor nfs_subs.c:

http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/nfs/nfs_subs.c.diff?r1=1.221&r2=1.221.2.1&only_with_tag=yamt-pagecache&f=H

have been committed to -HEAD yet.

Shouldn't we merge these changes together?

Thanks,
rin