Date:   Fri, 2 Jun 2017 12:49:34 -0600
From:   Logan Gunthorpe <logang@...tatee.com>
To:     "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
        Dan Williams <dan.j.williams@...el.com>,
        Linux MM <linux-mm@...ck.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Cc:     Stephen Bates <sbates@...thlin.com>
Subject: x86/mm/64: Possible bug with remove_pagetable in v4.12-rc1

Hi Friends,

I may have hit on a possible bug with remove_pagetables. However, so
far, I have only been able to reproduce it with my out-of-tree p2pmem
patchset (so I understand if you don't consider it a bug). At this time,
I have not able been able to reproduce it using device-dax.

Starting with v4.12-rc1 and including rc3, if I insert PCI bar memory
using devm_memremap_pages, then remove it, then insert it again I hit a
kernel BUG (see below).

I did a quick bisect to find the commit that causes this is:

e6ab9c4d4: x86/mm/64: Fix crash in remove_pagetable()

If I print the output of pgd_page_vaddr and p4d_offset, I get very
different addresses:

pgd_page_vaddr: ffff88026c04d000
p4d_offset: ffffffff81e0aea8

The version of p4d_offset my kernel is using is the one in
pgtable-nop4d.h which simply returns a casted version of pgd. This seems
slightly suspect but I can't say I understand this code all that well.

Thanks for your help,

Logan


> [  111.501162] ------------[ cut here ]------------
> [  111.506427] kernel BUG at arch/x86/mm/init_64.c:128!
> [  111.512080] invalid opcode: 0000 [#1] SMP
> [  111.516662] Modules linked in: mtr_p2pmem(O+) [last unloaded: mtr_p2pmem]
> [  111.524357] CPU: 4 PID: 2317 Comm: insmod Tainted: G           O    4.11.0-rc5.direct-00148-g052a6536f8db #435
> [  111.535667] Hardware name: Supermicro SYS-7047GR-TRF/X9DRG-QF, BIOS 3.0a 12/05/2013
> [  111.544352] task: ffff880274d54300 task.stack: ffffc90002edc000
> [  111.551079] RIP: 0010:sync_global_pgds+0x134/0x140
> [  111.556535] RSP: 0018:ffffc90002edfaa8 EFLAGS: 00010287
> [  111.562477] RAX: 00003ffffffff000 RBX: 0000000000000c00 RCX: ffffc08000000000
> [  111.570556] RDX: 0000000000000001 RSI: ffff88046d462000 RDI: ffff8802749ed000
> [  111.578628] RBP: ffff880000000000 R08: ffffffff81e0ac00 R09: 000000046d462067
> [  111.586710] R10: ffff88046d462000 R11: ffff880474bdd1e0 R12: ffffffff81e0ac00
> [  111.594789] R13: ffff8804775f3c00 R14: ffff8804775c1400 R15: ffffea0011dd7cc0
> [  111.602871] FS:  00007f3f39999700(0000) GS:ffff88047fc00000(0000) knlGS:0000000000000000
> [  111.612043] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  111.618569] CR2: 000055e04c19f1d8 CR3: 00000004771f0000 CR4: 00000000000406e0
> [  111.626651] Call Trace:
> [  111.629491]  ? kernel_physical_mapping_init+0x1c2/0x1e2
> [  111.635438]  ? init_memory_mapping+0x1c0/0x370
> [  111.640509]  ? walk_system_ram_range+0x6d/0xb0
> [  111.645578]  ? arch_add_memory+0x4b/0xe0
> [  111.650067]  ? devm_memremap_pages+0x276/0x3c0
> [  111.655141]  ? p2pmem_add_resource+0x25/0xc0
> [  111.660016]  ? mtramon_init+0x17a/0x1000 [mtr_p2pmem]
> [  111.665764]  ? 0xffffffffa000a000
> [  111.669570]  ? do_one_initcall+0x39/0x170
> [  111.674154]  ? do_init_module+0x55/0x1e5
> [  111.678643]  ? load_module+0x24ae/0x29a0
> [  111.683130]  ? SYSC_finit_module+0x91/0xc0
> [  111.687811]  ? SYSC_finit_module+0x91/0xc0
> [  111.692492]  ? entry_SYSCALL_64_fastpath+0x13/0x94
> [  111.697950] Code: 49 f7 45 00 9f ff ff ff 75 a3 49 89 75 00 eb 9d c6 05 a8 85 00 01 00 e9 06 ff ff ff 48 83 c4 10 5b 5d 41 5c 41 5d 41 5e 41 5f c3 <0f> 0b 66 2e 0f 1f 84 00 00 00 00 00 55 53 48 89 f3 48 89 d5 e8 
> [  111.719213] RIP: sync_global_pgds+0x134/0x140 RSP: ffffc90002edfaa8
> [  111.726340] ---[ end trace 34982079c1a73cd0 ]---

Powered by blists - more mailing lists