[Bug corefiles/34555] gdb memory exhaustion on gdb.base/many-headers.exp

https://sourceware.org/bugzilla/show_bug.cgi?id=34555

--- Comment #2 from Rainer Orth <ro at CeBiTec dot Uni-Bielefeld.DE> ---
Some more insight: while there is the size difference between Linux and
Solaris, that's not the issue here.  Comparing the coredumps with
elfdump, one sees a crucial difference:

While the Solaris core includes the actual data (though without disk
space consumption) of the anon segments

* elfdump:

Program Header[9]:
    p_vaddr:      0x7fffbe6d0000      p_flags:    0
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0                   p_memsz:    0x1000
    p_offset:     0                   p_align:    0
    pr_mflags:    [ CORE_NODATA ]

Program Header[10]:
    p_vaddr:      0x7fffbe6d2000      p_flags:    [ PF_W PF_R ]
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0x1000              p_memsz:    0x1000
    p_offset:     0x387cec            p_align:    0

* pmap:

00007FFFBE6D0000     4K -------* [ unknown ]
00007FFFBE6D2000     4K rw-----  [ anon ]

Linux does not:

* elfdump:

Program Header[6]:
    p_vaddr:      0x7ac94c08b000      p_flags:    0
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0                   p_memsz:    0x1000
    p_offset:     0xab3000            p_align:    0x1000
    pr_mflags:    [ CORE_NODATA ]

Program Header[7]:
    p_vaddr:      0x7ac94c08c000      p_flags:    [ PF_W PF_R ]
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0                   p_memsz:    0x1000
    p_offset:     0xab3000            p_align:    0x1000
    pr_mflags:    [ CORE_NODATA ]

* pmap:

7AC94C08B000      4K -------* [ unknown ]
7AC94C08C000      4K rw-----* [ unknown ]

On Linux, the anon segment is marked CORE_NODATA, while on Solaris it's
not.  However, that is easily changed: with coreadm -P default-anon, one
can exclude the anon segments from the coredumps.  This changes the
corefile size from the original

-rw------- 1 ro gcc 797M Aug 21 08:48 many-headers.core

to just

-rw------- 1 ro gcc  16M Aug 21 14:59 many-headers.core

elfdump/pmap confirm that both segments are now marked CORE_NODATA, as
on Linux:

* elfdump:

Program Header[9]:
    p_vaddr:      0x7fffbe6d0000      p_flags:    0
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0                   p_memsz:    0x1000
    p_offset:     0                   p_align:    0
    pr_mflags:    [ CORE_NODATA ]

Program Header[10]:
    p_vaddr:      0x7fffbe6d2000      p_flags:    [ PF_W PF_R ]
    p_paddr:      0                   p_type:     [ PT_LOAD ]
    p_filesz:     0                   p_memsz:    0x1000
    p_offset:     0                   p_align:    0
    pr_mflags:    [ CORE_NODATA ]

* pmap:

00007FFFBE6D0000     4K -------* [ unknown ]
00007FFFBE6D2000     4K rw-----* [ unknown ]

However, the gdb memory consumption is still as insanely large as before.

So we are clearly dealing with a gdb issue alone here.

-- 
You are receiving this mail because:
You are on the CC list for the bug.