(2012/07/23 19:00), Dave Young wrote: > On 07/17/2012 11:15 AM, Takao Indoh wrote: > >> Hi Cong, >> >> When I tested kdump with 3.5.0-rc6 kernel, I found a problem of kdump >> kernel's panic in find_early_table_space(). >> >> init_memory_mapping: [mem 0x00000000-0x36ffafff] >> Kernel panic - not syncing: Cannot find space for the kernel page tables >> Pid: 0, comm: swapper Not tainted 3.5.0-rc6 #17 >> Call Trace: >> [<ffffffff8158549b>] panic+0xb8/0x1c8 >> [<ffffffff8158565d>] ? printk+0x48/0x4a >> [<ffffffff8157304c>] init_memory_mapping+0x46c/0x530 >> [<ffffffff818a73c7>] setup_arch+0x669/0xb0e >> [<ffffffff8158565d>] ? printk+0x48/0x4a >> [<ffffffff818a3a1f>] start_kernel+0x9b/0x34a >> [<ffffffff818a332d>] x86_64_start_reservations+0x131/0x136 >> [<ffffffff818a341f>] x86_64_start_kernel+0xed/0xf4 >> >> In find_early_table_space(), a kernel tries to find free area below 512M >> for pgtable using memblock_find_in_range, but it fails because kdump >> kernel does not have enough free space below 512M due to the memmap >> restriction. This is the memmap option specified against kdump kernel >> when crashkernel=128M. >> >> memmap=560K at 64K memmap=130492K at 770608K >> >> Only 560KB area is available and it is not sufficient for pgtable (it >> seems that about 1.8MB area is needed for pgtable). This problem is >> fixed by your revert patch. I hope this patch gets merged. > > > I can reproduce this issue as well, probably related to some x86 mm init > commits, this alloc failure does not happen with reverting below commits: > > bd2753b2dda7bb43c7468826de75f49c6a7e8965 > 722bc6b16771ed80871e1fd81c86d3627dda2ac8 Yeah, my result of bisect is as follows and at first I thought the commit 722bc6 caused this regression. 722bc6b16771ed80871e1fd81c86d3627dda2ac8 is the first bad commit commit 722bc6b16771ed80871e1fd81c86d3627dda2ac8 Author: WANG Cong <xiyou.wangcong at gmail.com> Date: Mon Mar 5 15:05:13 2012 -0800 IIUC, this commit just fixes a bug of counting pgtable entries. As the result, another problem came up to the surface. In the case of my machine(16GB memory), before applying 722bc6, find_early_table_space() requests about 12KB free area and it can be got from 560K at 64K area luckily. I think the size find_early_table_space() requests was a bug. After the bug is fixed by the commit 722bc6, find_early_table_space() requires 1.8MB area and it fails as I wrote. Thanks, Takao Indoh > >> >> Thanks, >> Takao Indoh >> >> (2012/06/12 14:21), Cong Wang wrote: >>> From: Cong Wang <xiyou.wangcong at gmail.com> >>> >>> This reverts the following commit: >>> >>> commit 8548c84da2f47e71bbbe300f55edb768492575f7 >>> Author: Takashi Iwai <tiwai at suse.de> >>> Date: Sun Oct 23 23:19:12 2011 +0200 >>> >>> x86: Fix S4 regression >>> >>> Commit 4b239f458 ("x86-64, mm: Put early page table high") causes a S4 >>> regression since 2.6.39, namely the machine reboots occasionally at S4 >>> resume. It doesn't happen always, overall rate is about 1/20. But, >>> like other bugs, once when this happens, it continues to happen. >>> >>> This patch fixes the problem by essentially reverting the memory >>> assignment in the older way. >>> >>> According to the previous discussion: >>> http://marc.info/?l=linux-kernel&m=133161674120253&w=2 >>> it seems that so far the best solution is just reverting it. >>> >>> Takashi, could you help to test if the S4 regression is still >>> there after this patch? >>> >>> Reported-by: CAI Qian <caiqian at redhat.com> >>> Cc: Dave Young <dyoung at redhat.com> >>> Cc: "H. Peter Anvin" <hpa at zytor.com> >>> Cc: Rafael J. Wysocki <rjw at sisk.pl> >>> Cc: Yinghai Lu <yinghai.lu at oracle.com> >>> Cc: Takashi Iwai <tiwai at suse.de> >>> Signed-off-by: Cong Wang <xiyou.wangcong at gmail.com> >>> >>> --- >>> arch/x86/mm/init.c | 3 ++- >>> 1 files changed, 2 insertions(+), 1 deletions(-) >>> >>> diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c >>> index bc4e9d8..7ab7975 100644 >>> --- a/arch/x86/mm/init.c >>> +++ b/arch/x86/mm/init.c >>> @@ -74,8 +74,9 @@ static void __init find_early_table_space(struct map_range *mr, unsigned long en >>> #ifdef CONFIG_X86_32 >>> /* for fixmap */ >>> tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE); >>> -#endif >>> + >>> good_end = max_pfn_mapped << PAGE_SHIFT; >>> +#endif >>> >>> base = memblock_find_in_range(start, good_end, tables, PAGE_SIZE); >>> if (!base) >>> > > >