Hi Jan,

On Sat, 28 Jul 2012, Jan Kratochvil wrote:

> 7.4.1 is too old.

Here is a patch against a fresh checkout from the git mirror.

> Primarily linux_nat_find_memory_regions is now reworked and renamed as
> linux_find_memory_regions as GDB can now gcore also from remotely run process
> via gdbserver.  The testcase should be verified it also works with gdbserver
> afterwards, see gdb/testsuite/boards/native-gdbserver.exp.

I only tested the added test case:

make -C gdb/testsuite/ check RUNTESTFLAGS="gdb.base/gcore-relro.exp"

make -C gdb/testsuite/ check RUNTESTFLAGS="--target_board=native-gdbserver 
gdb.base/gcore-relro.exp"

> Also there should be descriptive comments before each modified function.
> I can write those otherwise.

I'm not sure what kind you expect, so I'll let you write them. :-)

> Here is missing a check:
>     if (separate_debug_objfile_backlink != NULL)
>       continue;
> as separate debug info files (in /usr/lib/debug, sometimes with .debug
> extension) are irrelevant for gcore.

Should be good now.

> Therefore if "Anonymous:" entry is not present at all (=not supported by 
> Linux kernel on that host) we must write out very every area as we have 
> no way to verify for swap-cached pages (having Anonymous > 0).
> 
> Sure if "Anonymous:" is present and it is > 0 it also must be written 
> out.

Ditto.

> You can reproduce the "Anonymous:" problem with
> 	http://people.redhat.com/jkratoch/relroswap.c
> although it is not suitable for GDB testsuite as it needs to do huge memory
> allocations to swap out (+swap in) the testcase pages.
> (If you have a better idea how to make "Anonymous:" entry non-zero it is sure
> welcome.  I have tried madvise (MADV_DONTNEED) but it does not work.)

Starting the test makes my desktop unusable (no SSD for swap yet), so I 
cowardly gave up.

> > +set testfile "gcore-relro"
> 
> Is hould use 'standard_testfile' like other testcases recently modified 
> by Tom Tromey.

I hope that's ok now.


Cheers,
Jean-Marc

-- 
saffroy@gmail.com