[Crash-utility] Fix fill_mem_map_cache() to read page-size-or-less size correctly

[Dave Anderson]

Hi Kazu,

This seems to be an extrememly rare condition, but in any case, your
patch is the correct thing to do.  

However, I do have two dumpfiles where the change causes the warning message
to be displayed unnecessarily, which was only meant to be displayed if the
mem_map cache is *not* virtually-mapped.  But the single check for 
!(vt->flags & V_MEM_MAP) is not always sufficient, so it should also
check for !(machdep->flags & VMEMMAP) like so:

--- a/memory.c	2019-09-17 14:20:24.586069004 -0400
+++ b/memory.c	2019-09-17 14:36:36.765286602 -0400
@@ -6348,13 +6348,13 @@ fill_mem_map_cache(ulong pp, ulong ppend
                 if (cnt > size)
                         cnt = size;
 
-		if (!readmem(addr, KVADDR, bufptr, size,
+		if (!readmem(addr, KVADDR, bufptr, cnt,
                     "virtual page struct cache", RETURN_ON_ERROR|QUIET)) {
-			BZERO(bufptr, size);
-			if (!(vt->flags & V_MEM_MAP) && ((addr+size) < ppend)) 
+			BZERO(bufptr, cnt);
+			if (!((vt->flags & V_MEM_MAP) || (machdep->flags & VMEMMAP)) && ((addr+cnt) < ppend))
 				error(WARNING, 
 		                   "mem_map[] from %lx to %lx not accessible\n",
-					addr, addr+size);
+					addr, addr+cnt);
 		}
 
 		addr += cnt;

You OK with that?

Thanks,
  Dave




 

----- Original Message -----
> Hi Kazu,
> 
> I will be out of the office until Monday September 16th.  I will check
> out your patch when I get back.
> 
> Thanks,
>   Dave
> 
> 
> ----- Original Message -----
> > 
> > fill_mem_map_cache() intends to read page-size-or-less size if it cannot
> > read the whole cache size, but it seems it doesn't correctly, and shows
> > just zero for existing data.
> > 
> >   crash> kmem -p 1000
> >         PAGE       PHYSICAL      MAPPING       INDEX CNT FLAGS
> >   ffffea0000000040     1000                0        0  0 0
> >   crash> rd ffffea0000000040
> >   ffffea0000000040:  000fffff00000400                    ........
> > 
> > I think the size below should be cnt, and confirmed that the patch works
> > well with a dump by makedumpfile -e option.
> > 
> > Signed-off-by: Kazuhito Hagio <k-hagio at ab.jp.nec.com>
> > ---
> >  memory.c | 8 ++++----
> >  1 file changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/memory.c b/memory.c
> > index ac7e73679405..4584673ed0ae 100644
> > --- a/memory.c
> > +++ b/memory.c
> > @@ -6348,13 +6348,13 @@ fill_mem_map_cache(ulong pp, ulong ppend, char
> > *page_cache)
> >                  if (cnt > size)
> >                          cnt = size;
> >  
> > -                if (!readmem(addr, KVADDR, bufptr, size,
> > +                if (!readmem(addr, KVADDR, bufptr, cnt,
> >                      "virtual page struct cache", RETURN_ON_ERROR|QUIET)) {
> > -                        BZERO(bufptr, size);
> > -                        if (!(vt->flags & V_MEM_MAP) && ((addr+size) <
> > ppend))
> > +                        BZERO(bufptr, cnt);
> > +                        if (!(vt->flags & V_MEM_MAP) && ((addr+cnt) <
> > ppend))
> >                                  error(WARNING,
> >                                     "mem_map[] from %lx to %lx not
> >                                     accessible\n",
> > -                                        addr, addr+size);
> > +                                        addr, addr+cnt);
> >                  }
> >  
> >                  addr += cnt;
> > --
> > 2.18.1
> > 
> 
> --
> Crash-utility mailing list
> Crash-utility at redhat.com
> https://www.redhat.com/mailman/listinfo/crash-utility
>