EXT: RE: crash: read error on type: "memory section root table"

[Agrain Patrick]

-----Message d'origine-----
De : HAGIO KAZUHITO(萩尾　一仁) <k-hagio-ab at nec.com> 
Envoyé : mercredi 6 avril 2022 10:06
À : Agrain Patrick <patrick.agrain at al-enterprise.com>
Cc : kexec at lists.infradead.org; Discussion list for crash utility usage, maintenance and development <crash-utility at redhat.com>
Objet : RE: EXT: RE: crash: read error on type: "memory section root table"

-----Original Message-----
> -----Original Message-----
> > Hello,
> >
> > Suggested trace above gives following information after a crash -d 8 command:
> > <...>
> > kernel NR_CPUS: 2
> > <readmem: ffffffffa4925820, KVADDR, "high_memory", 8, (FOE), 
> > 56017b542648>
> > <read_diskdump: addr: ffffffffa4925820 paddr: 12925820 cnt: 8>
> > read_diskdump: paddr/pfn: 12925820/12925 -> cache physical page: 
> > 12925000
> > GETBUF(328 -> 0)
> > FREEBUF(0)
> > GETBUF(328 -> 0)
> > FREEBUF(0)
> > PAGESIZE=4096
> > mem_section_size = 16384
> > NR_SECTION_ROOTS = 2048
> > NR_MEM_SECTIONS = 524288
> > SECTIONS_PER_ROOT = 256
> > SECTION_ROOT_MASK = 0xff
> > PAGES_PER_SECTION = 32768
> > <readmem: ffffffffa4926db0, KVADDR, "mem_section", 8, (FOE), 
> > 7ffd1b6bb000>
> > <read_diskdump: addr: ffffffffa4926db0 paddr: 12926db0 cnt: 8>
> > read_diskdump: paddr/pfn: 12926db0/12926 -> cache physical page: 
> > 12926000
> > <readmem: ffff904c7f7fc000, KVADDR, "memory section root table", 
> > 16384, (FOE), 56017da26fd0>
> > <read_diskdump: addr: ffff904c7f7fc000 paddr: 3f7fc000 cnt: 4096>
> > read_diskdump: paddr/pfn: 3f7fc000/3f7fc -> cache physical page: 
> > 3f7fc000
> > crash: PAG3 - errno=2 r=0 pd.size=49
> > read_diskdump: READ_ERROR: cannot cache page: 3f7fc000
> > crash: read error: kernel virtual address: ffff904c7f7fc000  type: "memory section root table"
> 
> hmm, r=0 means end of file, can you check again whether pd.offset 
> exceeds the dumpfile size?  If so, somehow the dumpfile is shorter than expected.
> 
> I think a RHEL-based kexec-tools does "sync" after makedumpfile, but 
> can you check?

> > Note 2: The debug message of makedumpfile report 'Write bytes     : 17364943', but the file is ~5MB for '-d 31' opton.

This also looks the same situation.

Does cp command always work on your machine to capture /proc/vmcore?
e.g. with a RHEL-based kexec-tools:

  core_collector cp

The size of a vmcore should become almost same as memory size.

Yes, if I'm using the /proc/vmcore instead, crash has no problem to read it... except that in a very few number of cases, the /proc/vmcore has less bytes than expected (1031389184 bytes on a 1GB RAM system) and then crash also fails.

Thanks,
Kazu