Skip to contentSkip to conversation

x86/sev: Don't hang but terminate on failure to remap SVSM CA

PATCH In review Ard Biesheuvel 3 participants 3 messages Jan 6, 2025 x86

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/sev: Don't hang but terminate on failure to remap SVSM CA

From: Ard Biesheuvel <[email protected]>

Commit

09d35045cd0f ("x86/sev: Avoid WARN()s and panic()s in early boot code")

replaced a panic() that could potentially hit before the kernel is even mapped with a deadloop, to ensure that execution does not proceed when the condition in question hits.

As Tom suggests, it is better to terminate and return to the hypervisor in this case, using a newly invented failure code to describe the failure condition.

---
 arch/x86/coco/sev/core.c          | 4 ++--
 arch/x86/include/asm/sev-common.h | 1 +
 2 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/coco/sev/core.c +2 -2
index 499b41953e3c..86898547056e 100644
@@ -2356,8 +2356,8 @@ static __head void svsm_setup(struct cc_blob_sev_info *cc_info)
 	call.rax = SVSM_CORE_CALL(SVSM_CORE_REMAP_CA);
 	call.rcx = pa;
 	ret = svsm_perform_call_protocol(&call);
-	while (ret)
-		cpu_relax(); /* too early to panic */
+	if (ret)
+		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_CA_REMAP_FAIL);
 
 	RIP_REL_REF(boot_svsm_caa) = (struct svsm_ca *)pa;
 	RIP_REL_REF(boot_svsm_caa_pa) = pa;
arch/x86/include/asm/sev-common.h +1 -0
index 50f5666938c0..577b64dda8b4 100644
@@ -206,6 +206,7 @@ struct snp_psc_desc {
 #define GHCB_TERM_NO_SVSM		7	/* SVSM is not advertised in the secrets page */
 #define GHCB_TERM_SVSM_VMPL0		8	/* SVSM is present but has set VMPL to 0 */
 #define GHCB_TERM_SVSM_CAA		9	/* SVSM is present but CAA is not page aligned */
+#define GHCB_TERM_SVSM_CA_REMAP_FAIL	10	/* SVSM is present but CA could not be remapped */
 
 #define GHCB_RESP_CODE(v)		((v) & GHCB_MSR_INFO_MASK)
 
Signature · 1 line
2.47.1.613.gc27f4b7a9f-goog
Reviewed-by on v1 1/1
#1 lore raw
Quoted 10 lines · Ard Biesheuvel

From: Ard Biesheuvel <[email protected]>

Commit

09d35045cd0f ("x86/sev: Avoid WARN()s and panic()s in early boot code")

replaced a panic() that could potentially hit before the kernel is even mapped with a deadloop, to ensure that execution does not proceed when the condition in question hits.

Quoted 7 lines

As Tom suggests, it is better to terminate and return to the hypervisor in this case, using a newly invented failure code to describe the failure condition.

Sorry Ard, I hadn't realized that the series was already merged or I would have submitted the patch myself. Thanks for doing this!

Quoted 32 lines
---
 arch/x86/coco/sev/core.c          | 4 ++--
 arch/x86/include/asm/sev-common.h | 1 +
 2 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/coco/sev/core.c +2 -2
index 499b41953e3c..86898547056e 100644
@@ -2356,8 +2356,8 @@ static __head void svsm_setup(struct cc_blob_sev_info *cc_info)
 	call.rax = SVSM_CORE_CALL(SVSM_CORE_REMAP_CA);
 	call.rcx = pa;
 	ret = svsm_perform_call_protocol(&call);
-	while (ret)
-		cpu_relax(); /* too early to panic */
+	if (ret)
+		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_CA_REMAP_FAIL);
 
 	RIP_REL_REF(boot_svsm_caa) = (struct svsm_ca *)pa;
 	RIP_REL_REF(boot_svsm_caa_pa) = pa;
arch/x86/include/asm/sev-common.h +1 -0
index 50f5666938c0..577b64dda8b4 100644
@@ -206,6 +206,7 @@ struct snp_psc_desc {
 #define GHCB_TERM_NO_SVSM		7	/* SVSM is not advertised in the secrets page */
 #define GHCB_TERM_SVSM_VMPL0		8	/* SVSM is present but has set VMPL to 0 */
 #define GHCB_TERM_SVSM_CAA		9	/* SVSM is present but CAA is not page aligned */
+#define GHCB_TERM_SVSM_CA_REMAP_FAIL	10	/* SVSM is present but CA could not be remapped */
 
 #define GHCB_RESP_CODE(v)		((v) & GHCB_MSR_INFO_MASK)
 
#2 lore raw
Quoted 16 lines · Tom Lendacky
Quoted 14 lines · Ard Biesheuvel

From: Ard Biesheuvel <[email protected]>

Commit

09d35045cd0f ("x86/sev: Avoid WARN()s and panic()s in early boot code")

replaced a panic() that could potentially hit before the kernel is even mapped with a deadloop, to ensure that execution does not proceed when the condition in question hits.

As Tom suggests, it is better to terminate and return to the hypervisor in this case, using a newly invented failure code to describe the failure condition.

Quoted 7 lines
Quoted 3 lines

Sorry Ard, I hadn't realized that the series was already merged or I would have submitted the patch myself. Thanks for doing this!

No worries

Quoted 2 lines

Thanks