[Crash-utility] Need to update page->flags

[Dave Anderson]

David Wilder wrote:

> Dave Anderson wrote:
>
> > David Wilder wrote:
> >
> >> The definitions for page->flags in crash are no longer correct in crash
> >> resulting in incorrect output for the kmem -i command.   The page flags
> >> are currently defined for V22 V24 and one flag for V26.    The flag
> >> definitions are changing in the 2.6 kernel often, crash will need to
> >> update these definitions for each 2.6.x.  Yuck!    Can we limit the flag
> >> definitions to just a few kernel versions?  Any suggestions on what
> >> versions we should have definitions for?
> >>
> >
> > I've long since been keeping up with these, because most
> > of the hardwired page flag definitions that are #define'd
> > in memory.c are not used anymore.  For example, there's
> > this construct in dump_mem_map() which just displays the
> > hex value:
> >
> >   if (v22) {
> >        ...translate hardwired flags...
> >   } else if (THIS_KERNEL_VERSION > LINUX(2,4,9)) {
> >        fprintf(fp, "%lx\n", flags);
> >   } else {
> >        ...translate old hardwired 2.4 flags...
> >   }
> >
> > But there are a few other places that need to know about
> > specific flags, like PG_reserved and PG_slab for example,
> > whose bits get determined in vm_init().  Which one has
> > changed that upsets "kmem -i"?
> >
> > Dave
> >
>
> I was first looking at PG_slab.  But I now see this bit in vm_init()
>
> if (VALID_MEMBER(page_pte)) {
>                 if (THIS_KERNEL_VERSION < LINUX(2,6,0))
>                         vt->PG_slab = 10;
>                 else if (THIS_KERNEL_VERSION >= LINUX(2,6,0))
>                         vt->PG_slab = 7;
>         }
>
> This would set the correct value for vt->PG_slab = 7 except that  pte is
> not a member of the struct page so vt->PG_slab=0.
>

Seems fixable...

>
> vt->PG_reserved is not getting set in PG_reserved_flag_init().  But that
> looks like a problem resolving the symbol sys_read.   So in
> dump_mem_map() it is falling back to v24_PG_reserved....
>

I wouldn't think it would have anything to do with the sys_read resolution
because kt->stext would be used.  It has to be due to some other problem
in that function.

Dave


>
> case GET_TOTALRAM_PAGES:
>                                 if (vt->PG_reserved)
>                                         PG_reserved_flag = vt->PG_reserved;
>                                 else
>                                         PG_reserved_flag = v22 ?
>                                                 1 << v22_PG_reserved :
>                                                 1 << v24_PG_reserved;
> ....
> That should mess up the calculation for total memory for kmem -i.  But
> in my case the number is still correct..
>
> --
> David Wilder
> IBM Linux Technology Center
> Beaverton, Oregon, USA
> dwilder at us.ibm.com
> (503)578-3789
>
> --
> Crash-utility mailing list
> Crash-utility at redhat.com
> https://www.redhat.com/mailman/listinfo/crash-utility