Re: linux-next: boot failure after merge of the memblock tree

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Su Hua <suhua.tanke@xxxxxxxxx> 于2024年10月25日周五 16:19写道:
>
> Appreciate everyone.
>
> Mike Rapoport <rppt@xxxxxxxxxx> 于2024年10月25日周五 14:57写道:
> >
> > Hi Stephen,
> >
> > On Tue, Oct 22, 2024 at 05:39:21PM +1100, Stephen Rothwell wrote:
> > > Hi all,
> > >
> > > After merging the memblock tree, today's linux-next build
> > > (powerpc_pseries_le_defconfig) failed my qemu boot test like this:
> > >
> > > mem auto-init: stack:all(zero), heap alloc:off, heap free:off
> > > BUG: Unable to handle kernel data access on read at 0x00001878
> > > Faulting instruction address: 0xc0000000004f00e4
> > > Oops: Kernel access of bad area, sig: 7 [#1]
> > > LE PAGE_SIZE=4K MMU=Radix SMP NR_CPUS=2048 NUMA pSeries
> > > Modules linked in:
> > > CPU: 0 UID: 0 PID: 0 Comm: swapper Not tainted 6.12.0-rc4-06078-g367eaba2691a #1
> > > Hardware name: IBM pSeries (emulated by qemu) POWER10 (architected) 0x801200 0xf000006 of:SLOF,HEAD pSeries
> > > NIP:  c0000000004f00e4 LR: c000000000489df8 CTR: 0000000000000000
> > > REGS: c0000000028cfae0 TRAP: 0300   Not tainted  (6.12.0-rc4-06078-g367eaba2691a)
> > > MSR:  8000000002001033 <SF,VEC,ME,IR,DR,RI,LE>  CR: 84000240  XER: 00000000
> > > CFAR: c0000000004f2c48 DAR: 0000000000001878 DSISR: 00080000 IRQMASK: 3
> > > GPR00: c00000000204994c c0000000028cfd80 c0000000016a4300 c00c000000040000
> > > GPR04: 0000000000000001 0000000000001000 0000000000000007 c000000002a11178
> > > GPR08: 0000000000000000 0000000000001800 c00000007fffe720 0000000000002001
> > > GPR12: 0000000000000000 c000000002a6a000 0000000000000000 00000000018855c0
> > > GPR16: c000000002940270 c00c000000000000 0000000000040000 0000000000000000
> > > GPR20: 0000000000000000 ffffffffffffffff 0000000000000001 ffffffffffffffff
> > > GPR24: 00c0000000000000 0000000000000000 0000000000000000 0000000008000000
> > > GPR28: 0000000000000000 0000000000002a6b 0000000000000000 0000000000001000
> > > NIP [c0000000004f00e4] set_pfnblock_flags_mask+0x74/0x140
> > > LR [c000000000489df8] reserve_bootmem_region+0x2a8/0x2c0
> > > Call Trace:
> > > c0000000028cfd80] [c0000000028cfdd0] 0xc0000000028cfdd0 (unreliable)
> > > c0000000028cfe20] [c00000000204994c] memblock_free_all+0x144/0x2d0
> > > c0000000028cfea0] [c000000002016354] mem_init+0x5c/0x70
> > > c0000000028cfec0] [c00000000204547c] mm_core_init+0x158/0x1dc
> > > c0000000028cff30] [c000000002004350] start_kernel+0x608/0x944
> > > c0000000028cffe0] [c00000000000e99c] start_here_common+0x1c/0x20
> > > Code: 4182000c 79082d28 7d4a4214 e9230000 3d020137 38e8ce78 79284620 792957a0 79081f24 79295d24 7d07402a 7d284a14 <e9090078> 7c254040 41800094 e9290088
> > > ---[ end trace 0000000000000000 ]---
> > >
> > > Kernel panic - not syncing: Attempted to kill the idle task!
> > >
> > > Caused by commit
> > >
> > >   ad48825232a9 ("memblock: uniformly initialize all reserved pages to MIGRATE_MOVABLE")
> > >
> > > I bisected the failure to this commit and have reverted it for today.
> >
> > Apparently set_pfnblock_flags_mask() is unhappy when called for
> > uninitialized struct page. With the patch below
> >
> > qemu-system-ppc64el -M pseries -cpu power10 -smp 16 -m 32G -vga none -nographic -kernel $KERNEL
> >
> > boots up to mounting root filesystem.
> >
> > diff --git a/mm/mm_init.c b/mm/mm_init.c
> > index 49dbd30e71ad..2395970314e7 100644
> > --- a/mm/mm_init.c
> > +++ b/mm/mm_init.c
> > @@ -723,10 +723,10 @@ static void __meminit init_reserved_page(unsigned long pfn, int nid)
> >                         break;
> >         }
> >
> > +       __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
> > +
> >         if (pageblock_aligned(pfn))
> >                 set_pageblock_migratetype(pfn_to_page(pfn), MIGRATE_MOVABLE);
> > -
> > -       __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
>
> Indeed, when #ifdef NODE_NOT_IN_PAGE_FLAGS is defined, there is no
> problem, and this is why my
> test environment did not reveal any issues. However, when
> NODE_NOT_IN_PAGE_FLAGS is not defined,
> page_to_nid needs to use page->flags to get the node ID, which depends
> on __init_single_page for initialization.
>
> >  }
> >  #else
> >  static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
> >
> > > --
> > > Cheers,
> > > Stephen Rothwell
> >
> >
> >
> > --
> > Sincerely yours,
> > Mike.
>
> Sincerely yours,
> Su

Hi Mike
Could you please advise whether the fix for this issue should be
submitted by you or me
as a new patch, or should I submit a patch that adjusts the code
position, just like this:

diff --git a/mm/mm_init.c b/mm/mm_init.c
index 4ba5607aaf19..5a8114fb02ae 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -723,6 +723,9 @@ static void __meminit init_reserved_page(unsigned
long pfn, int nid)
                        break;
        }
        __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
+
+       if (pageblock_aligned(pfn))
+               set_pageblock_migratetype(pfn_to_page(pfn), MIGRATE_MOVABLE);
 }
 #else

Sincerely yours,
Su





[Index of Archives]     [Linux Kernel]     [Linux USB Development]     [Yosemite News]     [Linux SCSI]

  Powered by Linux