x86_64 add NX mask for PTE entry
PATCH 1/1 x86_64 add NX mask for PTE entry
Hi,
If function change_page_attr_addr calls revert_page to revert
to original pte value, mk_pte_phys does not mask NX bit. If NX bit is set on no NX hardware supported x86_64 machine, there is will be RSVD type page fault and system will crash. This patch adds NX mask bit for PTE entry.
include/asm-x86_64/pgtable.h +1 -0
@@ -366,6 +366,7 @@ static inline pte_t mk_pte_phys(unsigned {pte_t pte;pte_val (pte) = physpage |pgprot_val (pgprot); +pte_val (pte) &= __supported_pte_mask;return pte; }
Quoted 6 lines · bibo,mao
Hi,
If functionchange_page_attr_addrcallsrevert_pageto revert
to original pte value,mk_pte_physdoes not mask NX bit. If NX bit is set on no NX hardware supportedx86_64machine, there is will be RSVD type page fault and system will crash. This patch adds NX mask bit for PTE entry.
Hmm, weird. I wonder why that didn't trip up earlier. Did you actually see that happening?
-Andi
Quoted 10 lines · Andi Kleen
Quoted 6 lines · bibo,mao
Hi,
If functionchange_page_attr_addrcallsrevert_pageto revert
to original pte value,mk_pte_physdoes not mask NX bit. If NX bit is set on no NX hardware supportedx86_64machine, there is will be RSVD type page fault and system will crash. This patch adds NX mask bit for PTE entry.Hmm, weird. I wonder why that didn't trip up earlier. Did you actually see that happening?
I remember previous system worked well on 2.6.12 and 2.6.13 version, but recently even on older kernel(2.6.9) there will be many times of page fault error and system crashes at last when rebooting system.
I will paste the oops message tomorrow.
thanks
bibo,mao
Quoted 3 lines
-Andi
Also I think change_page_attr is buggy. if orginal page is not large page, when that page's previous attr is reverted revert_page() function will be called to make that page huge. Indeed I think there is no need to do this.
thanks
bibo,mao
Quoted 28 lines · bibo,mao
Quoted 10 lines · Andi Kleen
Quoted 6 lines · bibo,mao
Hi,
If functionchange_page_attr_addrcallsrevert_pageto revert
to original pte value,mk_pte_physdoes not mask NX bit. If NX bit is set on no NX hardware supportedx86_64machine, there is will be RSVD type page fault and system will crash. This patch adds NX mask bit for PTE entry.Hmm, weird. I wonder why that didn't trip up earlier. Did you actually see that happening?
I remember previous system worked well on 2.6.12 and 2.6.13 version, but recently even on older kernel(2.6.9) there will be many times of page fault error and system crashes at last when rebooting system.
I will paste the oops message tomorrow.
thanks
bibo,maoQuoted 3 lines
-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Quoted 4 lines · bibo,mao
Also I think
change_page_attris buggy. if orginal page is not large page, when that page's previous attr is revertedrevert_page()function will be called to make that page huge. Indeed I think there is no need to do this.
On x86-64 all pages in the direct mapping are initially large pages. On i386 not necessarily, but it deals with that.
-Andi
The following is oops message everytime when rebooting system, and oops place is random, my machine is Pentium Dual core with 1G memory, no NX protection supports.
PMD value should be 0x000000003e4001e3 but not 0x800000003e4001e3, possibly there should be finer dump_pagetable() function to point out detail error type by page fault error code.
thanks
bibo,mao
Kernel 2.6.19-rc2 on an x86_64
INIT: Sending processes the TERM signal
Unable to handle kernel paging request at ffff81003e5fcde0 RIP: [<ffffffff812b9918>] __lock_text_start+0x0/0xf PGD 8063 PUD 9063 PMD 800000003e4001e3 BAD Oops: 000b [1] SMP CPU 0 Modules linked in: i2c_dev i2c_core e1000 ext3 jbd Pid: 4629, comm: default.hotplug Not tainted 2.6.19-rc2 #2 RIP: 0010:[<ffffffff812b9918>] [<ffffffff812b9918>] __lock_text_start+0x0/0xf RSP: 0018:ffff81003354fe50 EFLAGS: 00010286 RAX: 00002b756d31c000 RBX: ffff81003e5fcda0 RCX: ffff81000000a400 RDX: ffff810033c02a90 RSI: 0000000000000005 RDI: ffff81003e5fcde0 RBP: ffff81003e8914c0 R08: 00002b756d0f7000 R09: ffff81000000a400 R10: 0000003d31831e88 R11: 0000000000000286 R12: ffff81003f4c6138 R13: 0000000000000000 R14: 0000000000000000 R15: ffff81003354feb8 FS: 0000000000000000(0000) GS:ffffffff81435000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b CR2: ffff81003e400fe0 CR3: 0000000034366000 CR4: 00000000000006e0 Process default.hotplug (pid: 4629, threadinfo ffff81003354e000, task ffff81003e ee5710) Stack: ffffffff810659de ffff81003f4c6138 ffff8100373ef8c8 00002b756d0f7000 ffffffff81061959 0000000000000008 ffff810001571860 ffff810033a44558 ffff81003e7a5940 0000000000000000 00000000006c4470 0000000000000000 Call Trace: [<ffffffff810659de>] unlink_file_vma+0x23/0x3d [<ffffffff81061959>] free_pgtables+0x66/0xb2 [<ffffffff81067a45>] exit_mmap+0x88/0xe8 [<ffffffff81030196>] mmput+0x31/0xa0 [<ffffffff810350ed>] do_exit+0x1ec/0x82e [<ffffffff8103dfd7>] sys_rt_sigaction+0x85/0xa4 [<ffffffff810357db>] sys_exit_group+0x0/0xe [<ffffffff81009b4e>] system_call+0x7e/0x83 Code: f0 ff 0f 79 09 f3 90 83 3f 00 7e f9 eb f2 c3 31 c0 87 07 85 RIP [<ffffffff812b9918>] __lock_text_start+0x0/0xf RSP <ffff81003354fe50>
CR2: ffff81003e5fcde0
Quoted 19 lines · Andi Kleen
Quoted 6 lines · bibo,mao
Hi,
If functionchange_page_attr_addrcallsrevert_pageto revert
to original pte value,mk_pte_physdoes not mask NX bit. If NX bit is set on no NX hardware supportedx86_64machine, there is will be RSVD type page fault and system will crash. This patch adds NX mask bit for PTE entry.Hmm, weird. I wonder why that didn't trip up earlier. Did you actually see that happening?
(quoted diff) +0
-2-Andi -To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/