| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20230127035616.508966-1-aik@amd.com>
Date: Fri, 27 Jan 2023 14:56:16 +1100
From: Alexey Kardashevskiy <aik@....com>
To: Alexey Kardashevskiy <aik@....com>
CC: <kvm@...r.kernel.org>, <x86@...nel.org>,
<linux-kernel@...r.kernel.org>,
Thomas Gleixner <tglx@...utronix.de>,
Sean Christopherson <seanjc@...gle.com>,
Jiri Kosina <jkosina@...e.cz>, Ingo Molnar <mingo@...hat.com>,
Dave Hansen <dave.hansen@...ux.intel.com>,
"Borislav Petkov" <bp@...en8.de>,
"Peter Zijlstra (Intel)" <peterz@...radead.org>,
"H. Peter Anvin" <hpa@...or.com>,
Tom Lendacky <thomas.lendacky@....com>
Subject: [Question PATCH kernel] x86/amd/sev/nmi+vc: Fix stack handling (why is this happening?)
AMD SEV-ES guest kernels compiled without CONFIG_PARAVIRT crash when
"perf" executes. "./perf record sleep 20" is an example.
Some debugging revealed this happens when CONFIG_PARAVIRT_XXL is not
defined. The problem seems to be that between DEFINE_IDTENTRY_RAW(exc_nmi)
and actual reading of DB7 (which in turn causes #VC) every function is
inlined and no stack frame is created (?). Replacing __always_inline with
noinline in local_db_save() or native_get_debugreg() fixes the problem.
The crash does not happen with CONFIG_PARAVIRT_XXL as in this case
paravirt_get_debugreg() is used and there is an indirect call via
PVOP_CALL1(). It has not been noticed as the most configs have CONFIG_XEN
enabled which enables CONFIG_PARAVIRT_XXL.
This happens with the recent tip/master, here is my test kernel
and the config:
https://github.com/aik/linux/commits/debug_dr7
Found this while testing DebugSwap (which also fixes the crash as
it eliminates DB7 interception == #VC):
https://lore.kernel.org/all/20230120031047.628097-1-aik@amd.com
Define local_db_save_exc_nmi() to demostrate that the problem better.
Why is this crash happening and how to fix that? I am still reading
the assembly but was hoping for a shortcut here :) Thanks,
aik-Standard-PC-i440FX-PIIX-1996 login: .[A[ 15.775303] BUG: NMI stack guard page was hit at 0000000003983d50 (stack is 000000007feb1fa4..00000000574369c2)
[ 15.775314] stack guard page: 0000 [#1] PREEMPT SMP NOPTI
[ 15.775316] CPU: 0 PID: 790 Comm: sleep Not tainted 6.2.0-rc4_aik-debugswap_ruby-954bhost #73
[ 15.775322] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS unknown unknown
[ 15.775323] RIP: 0010:error_entry+0x17/0x140
[ 15.775326] Code: f8 e9 98 fd ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 56 48 8b 74 24 08 48 89 7c 24 08 52 51 50 41 50 41 51 41 52 41 53 53 <55> 41 54 41 55 41 56 41 57 56 31 f6 31 d2 31 c9 45 31 c0 45 31 c9
[ 15.775328] RSP: 0000:fffffe2446b2b000 EFLAGS: 00010097
[ 15.775332] RAX: fffffe2446b2b0a8 RBX: 0000000000000000 RCX: ffffffffb3a00fed
[ 15.775333] RDX: 0000000000000000 RSI: ffffffffb3a00b69 RDI: fffffe2446b2b0a8
[ 15.775336] RBP: fffffe2446b2b0a8 R08: 0000000000000000 R09: 0000000000000000
[ 15.775337] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
[ 15.775338] R13: 0000000000000000 R14: 000000000002dd80 R15: 0000000000000000
[ 15.775339] FS: 0000000000000000(0000) GS:ffff94b17dc00000(0000) knlGS:ffff94b17dc00000
[ 15.775340] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 15.775341] CR2: fffffe2446b2aff8 CR3: 00080000167b8000 CR4: 00000000003506f0
[ 15.775342] Call Trace:
[ 15.775352] <NMI>
[ 15.775355] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775357] ? exc_page_fault+0x11/0x120
[ 15.775360] ? asm_exc_page_fault+0x22/0x30
[ 15.775364] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775365] ? exc_page_fault+0x11/0x120
[ 15.775367] ? asm_exc_page_fault+0x22/0x30
[ 15.775368] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775369] ? exc_page_fault+0x11/0x120
[ 15.775371] ? asm_exc_page_fault+0x22/0x30
[ 15.775372] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775373] ? exc_page_fault+0x11/0x120
[ 15.775374] ? asm_exc_page_fault+0x22/0x30
[ 15.775375] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775376] ? exc_page_fault+0x11/0x120
[ 15.775378] ? asm_exc_page_fault+0x22/0x30
[ 15.775379] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775380] ? exc_page_fault+0x11/0x120
[ 15.775381] ? asm_exc_page_fault+0x22/0x30
[ 15.775382] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775383] ? exc_page_fault+0x11/0x120
[ 15.775384] ? asm_exc_page_fault+0x22/0x30
[ 15.775385] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775386] ? exc_page_fault+0x11/0x120
[ 15.775388] ? asm_exc_page_fault+0x22/0x30
[ 15.775389] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775390] ? exc_page_fault+0x11/0x120
[ 15.775391] ? asm_exc_page_fault+0x22/0x30
[ 15.775392] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775393] ? exc_page_fault+0x11/0x120
[ 15.775395] ? asm_exc_page_fault+0x22/0x30
[ 15.775396] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775397] ? exc_page_fault+0x11/0x120
[ 15.775398] ? asm_exc_page_fault+0x22/0x30
[ 15.775399] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775400] ? exc_page_fault+0x11/0x120
[ 15.775401] ? asm_exc_page_fault+0x22/0x30
[ 15.775403] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775404] ? exc_page_fault+0x11/0x120
[ 15.775405] ? asm_exc_page_fault+0x22/0x30
[ 15.775406] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775407] ? exc_page_fault+0x11/0x120
[ 15.775408] ? asm_exc_page_fault+0x22/0x30
[ 15.775409] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775410] ? exc_page_fault+0x11/0x120
[ 15.775412] ? asm_exc_page_fault+0x22/0x30
[ 15.775413] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775414] ? exc_page_fault+0x11/0x120
[ 15.775415] ? asm_exc_page_fault+0x22/0x30
[ 15.775416] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775420] ? exc_page_fault+0x11/0x120
[ 15.775421] ? asm_exc_page_fault+0x22/0x30
[ 15.775422] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775423] ? exc_page_fault+0x11/0x120
[ 15.775425] ? asm_exc_page_fault+0x22/0x30
[ 15.775426] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775427] ? exc_page_fault+0x11/0x120
[ 15.775431] ? asm_exc_page_fault+0x22/0x30
[ 15.775432] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775433] ? exc_page_fault+0x11/0x120
[ 15.775435] ? asm_exc_page_fault+0x22/0x30
[ 15.775436] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775437] ? exc_page_fault+0x11/0x120
[ 15.775438] ? asm_exc_page_fault+0x22/0x30
[ 15.775439] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775440] ? exc_page_fault+0x11/0x120
[ 15.775441] ? asm_exc_page_fault+0x22/0x30
[ 15.775442] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775443] ? exc_page_fault+0x11/0x120
[ 15.775445] ? asm_exc_page_fault+0x22/0x30
[ 15.775446] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775447] ? exc_page_fault+0x11/0x120
[ 15.775448] ? asm_exc_page_fault+0x22/0x30
[ 15.775449] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775450] ? exc_page_fault+0x11/0x120
[ 15.775454] ? asm_exc_page_fault+0x22/0x30
[ 15.775455] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775456] ? exc_page_fault+0x11/0x120
[ 15.775458] ? asm_exc_page_fault+0x22/0x30
[ 15.775459] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775460] ? exc_page_fault+0x11/0x120
[ 15.775461] ? asm_exc_page_fault+0x22/0x30
[ 15.775462] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775463] ? exc_page_fault+0x11/0x120
[ 15.775465] ? asm_exc_page_fault+0x22/0x30
[ 15.775466] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775467] ? exc_page_fault+0x11/0x120
[ 15.775468] ? asm_exc_page_fault+0x22/0x30
[ 15.775469] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775470] ? exc_page_fault+0x11/0x120
[ 15.775471] ? asm_exc_page_fault+0x22/0x30
[ 15.775472] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775473] ? exc_page_fault+0x11/0x120
[ 15.775475] ? asm_exc_page_fault+0x22/0x30
[ 15.775476] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775477] ? exc_page_fault+0x11/0x120
[ 15.775478] ? asm_exc_page_fault+0x22/0x30
[ 15.775482] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775483] ? exc_page_fault+0x11/0x120
[ 15.775485] ? asm_exc_page_fault+0x22/0x30
[ 15.775486] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775487] ? exc_page_fault+0x11/0x120
[ 15.775488] ? asm_exc_page_fault+0x22/0x30
[ 15.775490] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775491] ? exc_page_fault+0x11/0x120
[ 15.775492] ? asm_exc_page_fault+0x22/0x30
[ 15.775493] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775494] ? exc_page_fault+0x11/0x120
[ 15.775495] ? asm_exc_page_fault+0x22/0x30
[ 15.775496] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775497] ? exc_page_fault+0x11/0x120
[ 15.775499] ? asm_exc_page_fault+0x22/0x30
[ 15.775500] ? check_preemption_disabled+0x8/0xe0
[ 15.775502] ? __sev_es_ist_enter+0x13/0x100
[ 15.775503] ? exc_nmi+0x10e/0x150
[ 15.775505] ? end_repeat_nmi+0x16/0x67
[ 15.775506] ? asm_exc_double_fault+0x30/0x30
[ 15.775507] ? asm_exc_double_fault+0x30/0x30
[ 15.775508] ? asm_exc_double_fault+0x30/0x30
[ 15.775509] </NMI>
[ 15.775509] <#VC>
[ 15.775510] ? __show_regs.cold+0x18e/0x23d
[ 15.775511] </#VC>
[ 15.775511] <#DF>
[ 15.775512] ? __die_body.cold+0x1a/0x1f
[ 15.775513] ? die+0x26/0x40
[ 15.775517] ? handle_stack_overflow+0x44/0x60
[ 15.775518] ? exc_double_fault+0x14b/0x180
[ 15.775519] ? asm_exc_double_fault+0x1f/0x30
[ 15.775520] ? restore_regs_and_return_to_kernel+0x22/0x22
[ 15.775521] ? asm_exc_page_fault+0x9/0x30
[ 15.775522] ? error_entry+0x17/0x140
[ 15.775523] </#DF>
[ 15.775523] WARNING: stack recursion on stack type 6
[ 15.775524] Modules linked in: msr efivarfs
[ 15.837935] ---[ end trace 0000000000000000 ]---
Signed-off-by: Alexey Kardashevskiy <aik@....com>
---
arch/x86/include/asm/debugreg.h | 29 ++++++++++++++++++++
arch/x86/kernel/nmi.c | 2 +-
2 files changed, 30 insertions(+), 1 deletion(-)
diff --git a/arch/x86/include/asm/debugreg.h b/arch/x86/include/asm/debugreg.h
index b049d950612f..396e2ea114dc 100644
--- a/arch/x86/include/asm/debugreg.h
+++ b/arch/x86/include/asm/debugreg.h
@@ -125,6 +125,35 @@ static __always_inline void local_db_restore(unsigned long dr7)
set_debugreg(dr7, 7);
}
+/* noinline here inline __always_inline'd native_get_debugreg(int regno) */
+static noinline unsigned long native_get_debugreg7(void)
+{
+ unsigned long dr7;
+ asm("mov %%db7, %0" :"=r" (dr7));
+ return dr7;
+}
+
+static __always_inline unsigned long local_db_save_exc_nmi(void)
+{
+ unsigned long dr7;
+
+ if (static_cpu_has(X86_FEATURE_HYPERVISOR) && !hw_breakpoint_active())
+ return 0;
+
+ dr7 = native_get_debugreg7();
+ dr7 &= ~0x400; /* architecturally set bit */
+ if (dr7)
+ set_debugreg(0, 7);
+ /*
+ * Ensure the compiler doesn't lower the above statements into
+ * the critical section; disabling breakpoints late would not
+ * be good.
+ */
+ barrier();
+
+ return dr7;
+}
+
#ifdef CONFIG_CPU_SUP_AMD
extern void set_dr_addr_mask(unsigned long mask, int dr);
#else
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index cec0bfa3bc04..400b5b6b74f6 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -503,7 +503,7 @@ DEFINE_IDTENTRY_RAW(exc_nmi)
*/
sev_es_ist_enter(regs);
- this_cpu_write(nmi_dr7, local_db_save());
+ this_cpu_write(nmi_dr7, local_db_save_exc_nmi());
irq_state = irqentry_nmi_enter(regs);
--
2.39.1
Powered by blists - more mailing lists