On Wed, Jun 9, 2010 at 10:37 AM, Dave Young <hidave.darkstar@xxxxxxxxx> wrote: > On Tue, Jun 8, 2010 at 12:43 AM, Catalin Marinas > <catalin.marinas@xxxxxxx> wrote: >> On Mon, 2010-06-07 at 11:00 +0100, Dave Young wrote: >>> On Mon, Jun 7, 2010 at 5:19 PM, Catalin Marinas <catalin.marinas@xxxxxxx> wrote: >>> > On Mon, 2010-06-07 at 06:20 +0100, Dave Young wrote: >>> >> On Fri, Jun 4, 2010 at 9:55 PM, Dave Young <hidave.darkstar@xxxxxxxxx> wrote: >>> >> > On Fri, Jun 4, 2010 at 6:50 PM, Catalin Marinas <catalin.marinas@xxxxxxx> wrote: >>> >> >> Dave Young <hidave.darkstar@xxxxxxxxx> wrote: >>> >> >>> With mmotm 2010-06-03-16-36, I gots tuns of kmemleaks >>> >> >> >>> >> >> Do you have CONFIG_NO_BOOTMEM enabled? I posted a patch for this but >>> >> >> hasn't been reviewed yet (I'll probably need to repost, so if it fixes >>> >> >> the problem for you a Tested-by would be nice): >>> >> >> >>> >> >> http://lkml.org/lkml/2010/5/4/175 >>> >> > >>> >> > >>> >> > I'd like to test, but I can not access the test pc during weekend. So >>> >> > I will test it next monday. >>> >> >>> >> Bad news, the patch does not fix this issue. >>> > >>> > Thanks for trying. Could you please just disable CONFIG_NO_BOOTMEM and >>> > post the kmemleak reported leaks again? >>> >>> Still too many suspected leaks, results similar with >>> (CONFIG_NO_BOOTMEM = y && apply your patch), looks like a little >>> different from original ones? I just copy some of them here: >>> >>> unreferenced object 0xde3c7420 (size 44): >>> comm "bash", pid 1631, jiffies 4294897023 (age 223.573s) >>> hex dump (first 32 bytes): >>> 05 05 00 00 ad 4e ad de ff ff ff ff ff ff ff ff .....N.......... >>> 98 42 d9 c1 00 00 00 00 50 fe 63 c1 10 32 8f dd .B......P.c..2.. >>> backtrace: >>> [<c1498ad2>] kmemleak_alloc+0x4a/0x83 >>> [<c10c1ace>] kmem_cache_alloc+0xde/0x12a >>> [<c10b421b>] anon_vma_fork+0x31/0x88 >>> [<c102c71d>] dup_mm+0x1d3/0x38f >>> [<c102d20d>] copy_process+0x8ce/0xf39 >>> [<c102d990>] do_fork+0x118/0x295 >>> [<c1007fe0>] sys_clone+0x1f/0x24 >>> [<c10029b1>] ptregs_clone+0x15/0x24 >>> [<ffffffff>] 0xffffffff >> >> I'll try to test the mmotm kernel as well. I don't get any kmemleak >> reports with the 2.6.35-rc1 kernel. > > Manually bisected mm patches, the memleak caused by following patch: > > mm-extend-ksm-refcounts-to-the-anon_vma-root.patch Add following debug code: void drop_anon_vma(struct anon_vma *anon_vma) { + int a, b; + a = anonvma_external_refcount(anon_vma); + b = anonvma_external_refcount(anon_vma->root); + if (!a || !b) { + printk("drop_anon_vma: ref %d ", a); + printk("root ref %d\n", b); + } result in below debug output: [ 52.948614] drop_anon_vma: ref 0 root ref 0 [ 52.949770] Pid: 1403, comm: ps Not tainted 2.6.35-rc1-mm1 #29 [ 52.951386] Call Trace: [ 52.952062] [<c14b1128>] ? printk+0x20/0x24 [ 52.953210] [<c10b409c>] drop_anon_vma+0x37/0xb3 [ 52.954503] [<c10b418c>] unlink_anon_vmas+0x74/0xc4 [ 52.955854] [<c10aeaa0>] free_pgtables+0x45/0x95 [ 52.957142] [<c10b00fd>] exit_mmap+0xab/0xfe [ 52.958325] [<c102fafa>] ? exit_mm+0xdd/0xec [ 52.959497] [<c102c25d>] mmput+0x49/0xcf [ 52.960605] [<c102fb01>] exit_mm+0xe4/0xec [ 52.961750] [<c103137c>] do_exit+0x1b4/0x64b [ 52.962921] [<c1031875>] do_group_exit+0x62/0x85 [ 52.964212] [<c10318ab>] sys_exit_group+0x13/0x17 [ 52.965523] [<c14b344d>] syscall_call+0x7/0xb So I guess the refcount break, either drop-without-get or over-drop > > cc Rik van Riel > >> >> Can you send me your .config file? Do you have CONFIG_HUGETLBFS enabled? >> >> Thanks. >> >> -- >> Catalin >> >> > > > > -- > Regards > dave > -- Regards dave -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href