[Xen-devel] [PATCH 0/2] x86: cleanup highmap after brk is concluded

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

Hi all,
a little while ago I sent a patch titled "x86/mm/init: respect memblock
reserved regions when destroying mappings"
(https://lkml.org/lkml/2011/1/31/232) to fix a serious boot crash
problem on Xen (full logs attached):

Pid: 0, comm: swapper Not tainted 2.6.38-rc6+ #1270 Hewlett-Packard HP xw8600 
Workstation/0A98h
RIP: e030:[<ffffffff81008314>]  [<ffffffff81008314>] 
get_phys_to_machine+0x44/0x50
RSP: e02b:ffffffff82001ca0  EFLAGS: 00010002
RAX: ffffffff824ce000 RBX: 0000000126004067 RCX: 0000000000000010
RDX: 0000000000000000 RSI: 00000001cfdc2000 RDI: 0000000000000004
RBP: ffffffff82001ca0 R08: 0000000000000020 R09: 0000000000000000
R10: 0000000000000007 R11: 00000000ffffffff R12: 0000000000126004
R13: 0000000000002004 R14: ffff880100000000 R15: ffff8801cfdc2000
FS:  0000000000000000(0000) GS:ffffffff82162000(0000) knlGS:0000000000000000
CS:  e033 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000002003000 CR4: 0000000000002660
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 0000000000000000 DR7: 0000000000000000
Process swapper (pid: 0, threadinfo ffffffff82000000, task ffffffff8200b020)
Stack:
 ffffffff82001cd0 ffffffff8100582c ffffffff81dce1bc ffffffff82001e10
 00000001cfdc2000 ffffffff82003880 ffffffff82001ce0 ffffffff8100587e
 ffffffff82001d98 ffffffff8100498f 00000000ffffffff 0000000000000007
Call Trace:
 [<ffffffff8100582c>] pte_mfn_to_pfn+0x8c/0xb0
 [<ffffffff8100587e>] xen_pgd_val+0xe/0x10
 [<ffffffff8100498f>] __raw_callee_save_xen_pgd_val+0x11/0x1e
 [<ffffffff813ba570>] ? xenboot_write_console+0x0/0xd0
 [<ffffffff821c24b8>] ? kernel_physical_mapping_init+0x83/0x1db
 [<ffffffff8195469f>] init_memory_mapping+0x31f/0x6d0
 [<ffffffff821989fd>] ? memblock_reserve+0x1b/0x21
 [<ffffffff8217de95>] setup_arch+0xa59/0xd89
 [<ffffffff819b9c90>] ? _raw_spin_unlock_irqrestore+0x20/0x30
 [<ffffffff810074bd>] ? __raw_callee_save_xen_irq_disable+0x11/0x1e
 [<ffffffff82177b35>] start_kernel+0xc6/0x4df
 [<ffffffff821772c5>] x86_64_start_reservations+0xa5/0xc9
 [<ffffffff8217b6fa>] xen_start_kernel+0x5d3/0x6a9


Even though a clear solution wasn't reached in the following discussion,
Yinghai Lu sent a patch to move cleanup_highmap() after reserve_brk() so
that we don't have to clear the initial mappings in two steps.
The patch is a nice cleanup and with few small changes to honour the
variable max_pfn_mapped can be used to fix the boot issue on Xen: all we
have to do is setting max_pfn_mapped to the last valid pfn mapped on Xen
that is the page baking _end.


The list of patches with diffstat follows, comments and suggestions are
very welcome:

Stefano Stabellini (1):
      xen: set max_pfn_mapped to the last pfn mapped

Yinghai Lu (1):
      x86: Cleanup highmap after brk is concluded

 arch/x86/kernel/head64.c |    3 ---
 arch/x86/kernel/setup.c  |    6 ++++++
 arch/x86/mm/init.c       |   19 -------------------
 arch/x86/mm/init_64.c    |   11 ++++++-----
 arch/x86/xen/mmu.c       |   13 +++++++------
 5 files changed, 19 insertions(+), 33 deletions(-)


A git branch based on 2.6.38-rc6 is available here:

git://xenbits.xen.org/people/sstabellini/linux-pvhvm.git 2.6.38-rc6-mm-fix

Cheers,

Stefano

Attachment: log
Description: Text document

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel