Re: [PATCH v2] makedumpfile: add parameters to update_cyclic_region

[Baoquan He <bhe@redhat.com>]

On 11/26/13 at 05:50pm, HATAYAMA Daisuke wrote:
> (2013/11/26 16:57), Baoquan He wrote:
> >On 11/26/13 at 02:50pm, HATAYAMA Daisuke wrote:
> >>(2013/11/26 11:52), Baoquan He wrote:
> >>>On 11/25/13 at 01:33pm, HATAYAMA Daisuke wrote:
> >>>>(2013/11/25 11:31), Baoquan He wrote:
> >
> >>

> 
> For that, you need to pass a part with the currnet cycle to __exclude_unnecessary_pages(),
> not a whole (mmd->pfn_start, mmd->pfn_end). There might be similar part that needs fix,
> but sorry I don't have good memory...
> 
> int
> exclude_unnecessary_pages_cyclic(void)
> {
> <cut>
>                         if (mmd->pfn_end >= info->cyclic_start_pfn &&
>                             mmd->pfn_start <= info->cyclic_end_pfn) {
>                                 if (!__exclude_unnecessary_pages(mmd->mem_map,
>                                                                  mmd->pfn_start, mmd->pfn_end))
>                                         return FALSE;
>                         }
> 
> For ELF-to-ELF code, unfortunately, I gave up in the middle of source code reading.
> At lesst, if I remember correctly, I think the code relied on the current
> update_mmap_range() implementation. It might be hard to clean up there in a natural way.

Hi,

I reviewed code again, and understand what kind of pain you meant. I
think the hard part is write_elf_pages_cyclic(), because it is not kind
of like kdump format which loops from 0 to info->max_mapnr, then you can
use the loop MACRO.

struct cycle {
  uint64_t start_pfn;
  uint64_t end_pfn;
};

#define for_each_cycle(C, MAX_MAPNR) \
  for (first_cycle((C), (MAX_MAPNR)); !end_cycle(C); \
       update_cycle(C))

for_each_cycle(&cycle, info->max_mapnr) {
  if (!create_1st_bitmap_cyclic(&cycle))
    return FALSE;
  if (!exclude_unnecessary_pages_cyclic(&cycle))
    return FALSE;
  if (!write_kdump_bitmap1_cyclic(&cycle))
    return FALSE;
}


In fact, don't worry. Please check create_1st_bitmap_cyclic() again, its
handling is very similar to write_elf_pages_cyclic(), because both of
them traverse each PT_LOAD segment, and then do some handling. So you
can call create_1st_bitmap_cyclic() in the big loop which is from 0 to
info->max_mapnr, if rethink about write_elf_pages_cyclic(), you will
find it can also be called like this. Then the cycle loop MACRO can be
used either for elf format handling.

for_each_cycle(&cycle, info->max_mapnr) {
  if (!create_1st_bitmap_cyclic(&cycle))
    return FALSE;
  if (!exclude_unnecessary_pages_cyclic(&cycle))
    return FALSE;
  if (!write_elf_pages_cyclic(&cycle))
    return FALSE;
}

As for mmap, I don't think it will be affected by this change, anyway,
cycle has been passed in, not as a function parameter, but a global
variable (it's stored in global varible info, should be the same).

Baoquan
Thanks

> 
> -- 
> Thanks.
> HATAYAMA, Daisuke
> 
> 
> _______________________________________________
> kexec mailing list
> kexec@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kexec

_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec