linux-kernel - [PATCH kernel v2] x86/compressed/64: reduce #VC nesting for intercepted CPUID for SEV-SNP guest

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20230926040526.957240-1-aik@amd.com>
Date:   Tue, 26 Sep 2023 14:05:26 +1000
From:   Alexey Kardashevskiy <aik@....com>
To:     <x86@...nel.org>
CC:     <linux-kernel@...r.kernel.org>, Ingo Molnar <mingo@...hat.com>,
        "Borislav Petkov" <bp@...en8.de>,
        Tom Lendacky <thomas.lendacky@....com>,
        "Alexey Kardashevskiy" <aik@....com>
Subject: [PATCH kernel v2] x86/compressed/64: reduce #VC nesting for intercepted CPUID for SEV-SNP guest

For certain intercepts an SNP guest uses the GHCB protocol to talk to
the hypervisor from the #VC handler. The protocol requires a shared page so
there is one per vCPU. In case NMI arrives in a middle of #VC or the NMI
handler triggers a #VC, there is another "backup" GHCB page which stores
the content of the first one while SVM_VMGEXIT_NMI_COMPLETE is sent.
The vc_raw_handle_exception() handler manages main and backup GHCB pages
via __sev_get_ghcb/__sev_put_ghcb.

This works fine for #VC and occasional NMIs. This does not work so fine if
the #VC handler causes intercept + another #VC, if NMI arrives during
the second #VC, there are no more pages for SVM_VMGEXIT_NMI_COMPLETE.
The problem place is the #VC CPUID handler. Running perf in the SNP guest
crashes with:

Kernel panic - not syncing: Unable to handle #VC exception! GHCB and Backup GHCB are already in use

vc_raw_handle_exception #1: exit_code 72 (CPUID) eax d ecx 1
We lock the main GHCB and while it is locked we get to
snp_cpuid_postprocess() which executes "rdmsr" of MSR_IA32_XSS==0xda0 which
triggers:

vc_raw_handle_exception #2: exit_code 7c (MSR) ecx da0
Here we lock the backup ghcb.

And then PMC NMI comes which cannot complete as there is no GHCB page left
to use:

CPU: 5 PID: 566 Comm: touch Not tainted 6.5.0-rc2-aik-ad9c-g7413e71d3dcf-dirty #27
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS unknown unknown
Call Trace:
 <NMI>
 dump_stack_lvl+0x44/0x60
 panic+0x222/0x310
 ____sev_get_ghcb+0x21e/0x220
 __sev_es_nmi_complete+0x28/0xf0
 exc_nmi+0x1ac/0x1c0
 end_repeat_nmi+0x16/0x67
...
 </NMI>
 <TASK>
 vc_raw_handle_exception+0x9e/0x2c0
 kernel_exc_vmm_communication+0x4d/0xa0
 asm_exc_vmm_communication+0x31/0x60
RIP: 0010:snp_cpuid+0x2ad/0x420

Add a helper similar to rdmsr_safe() for making a direct hypercall in the SEV-ES
environment. Use the new helper instead of the raw "rdmsr" to avoid the extra
#VC event.

Fixes: ee0bfa08a345 ("x86/compressed/64: Add support for SEV-SNP CPUID table in #VC handlers")
Signed-off-by: Alexey Kardashevskiy <aik@....com>
---

This is made on top of (which has the "efi/unaccepted: Make sure unaccepted table is mapped"
fix for booting SNP):
b996cbe1203c (tip/master) 15 hours ago Ingo Molnar Merge branch into tip/master: 'x86/tdx'

plus:
https://lore.kernel.org/lkml/a5856fa1ebe3879de91a8f6298b6bbd901c61881.1690578565.git.thomas.lendacky@amd.com/


rdmsr_safe_GHCB() is in arch/x86/include/asm/svm.h as this is where
the ghcb struct is defined.


---
Changes:
v2:
* de-uglify by defining rdmsr_safe_GHCB()
---
 arch/x86/include/asm/svm.h   | 14 ++++++++++++++
 arch/x86/kernel/sev-shared.c |  5 +++--
 2 files changed, 17 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/svm.h b/arch/x86/include/asm/svm.h
index fb8366af59da..866ef7d6b7a9 100644
--- a/arch/x86/include/asm/svm.h
+++ b/arch/x86/include/asm/svm.h
@@ -673,4 +673,18 @@ DEFINE_GHCB_ACCESSORS(sw_exit_info_2)
 DEFINE_GHCB_ACCESSORS(sw_scratch)
 DEFINE_GHCB_ACCESSORS(xcr0)
 
+/* Paravirt SEV-ES rdmsr which avoids extra #VC event */
+#define rdmsr_safe_GHCB(msr, low, high, ghcb, ctxt) ({				\
+	int __ret;								\
+										\
+	ghcb_set_rcx((ghcb), (msr));						\
+	__ret = sev_es_ghcb_hv_call((ghcb), (ctxt), SVM_EXIT_MSR, 0, 0);	\
+	if (__ret == ES_OK) {							\
+		low  = (ghcb)->save.rax;					\
+		high = (ghcb)->save.rdx;					\
+		/* Invalidate qwords for likely another following GHCB call */	\
+		vc_ghcb_invalidate(ghcb);					\
+	}									\
+	__ret; })
+
 #endif
diff --git a/arch/x86/kernel/sev-shared.c b/arch/x86/kernel/sev-shared.c
index e73c90c9cc5b..b3fb9d0a07c6 100644
--- a/arch/x86/kernel/sev-shared.c
+++ b/arch/x86/kernel/sev-shared.c
@@ -479,8 +479,9 @@ static int snp_cpuid_postprocess(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
 			if (leaf->eax & BIT(3)) {
 				unsigned long lo, hi;
 
-				asm volatile("rdmsr" : "=a" (lo), "=d" (hi)
-						     : "c" (MSR_IA32_XSS));
+				if (rdmsr_safe_GHCB(MSR_IA32_XSS, lo, hi, ghcb, ctxt) != ES_OK)
+					return -EINVAL;
+
 				xss = (hi << 32) | lo;
 			}
 
-- 
2.41.0