Skip to contentSkip to conversation

x86/efistub: Don't bother enabling SEV in the EFI stub

PATCH In review Ard Biesheuvel 3 participants 4 messages Apr 14, 2025

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/efistub: Don't bother enabling SEV in the EFI stub

From: Ard Biesheuvel <[email protected]>

One of the last things the EFI stub does before handing over to the core kernel when booting as a SEV guest is enabling SEV, even though this is mostly redundant: one of the first things the core kernel does is calling sme_enable(), after setting up the early GDT and IDT but before even setting up the kernel page tables.

So let's just drop this call to sev_enable(), and rely on the core kernel to initiaize SEV correctly.

Cc Tom Lendacky <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Dionna Amalie Glaze <[email protected]>
Cc Kevin Loughlin <[email protected]>
---
 arch/x86/include/asm/sev.h              | 2 --
 drivers/firmware/efi/libstub/x86-stub.c | 6 ------
 2 files changed, 8 deletions(-)

I'm still waiting for my SEV-capable hardware to turn up so this was build tested only.

arch/x86/include/asm/sev.h +0 -2
index 7427eb233990..1db906d6924b 100644
@@ -442,7 +442,6 @@ static __always_inline void sev_es_nmi_complete(void)
 		__sev_es_nmi_complete();
 }
 extern int __init sev_es_efi_map_ghcbs(pgd_t *pgd);
-extern void sev_enable(struct boot_params *bp);
 
 /*
  * RMPADJUST modifies the RMP permissions of a page of a lesser-
@@ -533,7 +532,6 @@ static inline void sev_es_ist_exit(void) { }
 static inline int sev_es_setup_ap_jump_table(struct real_mode_header *rmh) { return 0; }
 static inline void sev_es_nmi_complete(void) { }
 static inline int sev_es_efi_map_ghcbs(pgd_t *pgd) { return 0; }
-static inline void sev_enable(struct boot_params *bp) { }
 static inline int pvalidate(unsigned long vaddr, bool rmp_psize, bool validate) { return 0; }
 static inline int rmpadjust(unsigned long vaddr, bool rmp_psize, unsigned long attrs) { return 0; }
 static inline void setup_ghcb(void) { }
drivers/firmware/efi/libstub/x86-stub.c +0 -6
index cafc90d4caaf..52ed13ec4fe9 100644
@@ -925,12 +925,6 @@ void __noreturn efi_stub_entry(efi_handle_t handle,
 		goto fail;
 	}
 
-	/*
-	 * Call the SEV init code while still running with the firmware's
-	 * GDT/IDT, so #VC exceptions will be handled by EFI.
-	 */
-	sev_enable(boot_params);
-
 	efi_5level_switch();
 
 	enter_kernel(kernel_entry, boot_params);
Signature · 1 line
2.49.0.604.gff1f9ca942-goog
#1 lore raw
Quoted 10 lines · Ard Biesheuvel

From: Ard Biesheuvel <[email protected]>

One of the last things the EFI stub does before handing over to the core kernel when booting as a SEV guest is enabling SEV, even though this is mostly redundant: one of the first things the core kernel does is calling sme_enable(), after setting up the early GDT and IDT but before even setting up the kernel page tables.

So let's just drop this call to sev_enable(), and rely on the core kernel to initiaize SEV correctly.

SEV support does some things in sev_enable() that aren't done later in the kernel proper. For example, we check certain hypervisor features and save the CC blob into the boot parameters. And as I look closer, I see that we don't call initialize_identity_maps() from the EFI stub so we aren't calling snp_check_features() from the EFI stub, which we should have been doing.

Just removing the call does cause an SNP boot to crash. My testing shows just ensuring that the cc_blob_address in the boot_params structure is set results in a successful boot under SNP. So some of sev_enable() needs to remain and some things need to be moved into the kernel proper.

Thanks,
Tom

Quoted 51 lines
Cc Tom Lendacky <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Dionna Amalie Glaze <[email protected]>
Cc Kevin Loughlin <[email protected]>
---
 arch/x86/include/asm/sev.h              | 2 --
 drivers/firmware/efi/libstub/x86-stub.c | 6 ------
 2 files changed, 8 deletions(-)

I'm still waiting for my SEV-capable hardware to turn up so this was build tested only.

arch/x86/include/asm/sev.h +0 -2
index 7427eb233990..1db906d6924b 100644
@@ -442,7 +442,6 @@ static __always_inline void sev_es_nmi_complete(void)
 		__sev_es_nmi_complete();
 }
 extern int __init sev_es_efi_map_ghcbs(pgd_t *pgd);
-extern void sev_enable(struct boot_params *bp);
 
 /*
  * RMPADJUST modifies the RMP permissions of a page of a lesser-
@@ -533,7 +532,6 @@ static inline void sev_es_ist_exit(void) { }
 static inline int sev_es_setup_ap_jump_table(struct real_mode_header *rmh) { return 0; }
 static inline void sev_es_nmi_complete(void) { }
 static inline int sev_es_efi_map_ghcbs(pgd_t *pgd) { return 0; }
-static inline void sev_enable(struct boot_params *bp) { }
 static inline int pvalidate(unsigned long vaddr, bool rmp_psize, bool validate) { return 0; }
 static inline int rmpadjust(unsigned long vaddr, bool rmp_psize, unsigned long attrs) { return 0; }
 static inline void setup_ghcb(void) { }
drivers/firmware/efi/libstub/x86-stub.c +0 -6
index cafc90d4caaf..52ed13ec4fe9 100644
@@ -925,12 +925,6 @@ void __noreturn efi_stub_entry(efi_handle_t handle,
 		goto fail;
 	}
 
-	/*
-	 * Call the SEV init code while still running with the firmware's
-	 * GDT/IDT, so #VC exceptions will be handled by EFI.
-	 */
-	sev_enable(boot_params);
-
 	efi_5level_switch();
 
 	enter_kernel(kernel_entry, boot_params);
maintainer
#2 lore raw

Hi Tom,

Thanks for taking a look.

Quoted 10 lines · Tom Lendacky
Quoted 8 lines · Ard Biesheuvel

From: Ard Biesheuvel <[email protected]>

One of the last things the EFI stub does before handing over to the core kernel when booting as a SEV guest is enabling SEV, even though this is mostly redundant: one of the first things the core kernel does is calling sme_enable(), after setting up the early GDT and IDT but before even setting up the kernel page tables.

Quoted 6 lines
Quoted 2 lines

So let's just drop this call to sev_enable(), and rely on the core kernel to initiaize SEV correctly.

SEV support does some things in sev_enable() that aren't done later in the kernel proper. For example, we check certain hypervisor features and save the CC blob into the boot parameters.

I misread the last part: the kernel proper checks struct boot_params and setup_data, whereas the decompressor checks both of those as well as the EFI config table.

It would make sense for the EFI stub to populate the field in struct boot_params directly.

Quoted 5 lines

And as I look closer, I see
that we don't call initialize_identity_maps() from the EFI stub so we aren't calling snp_check_features() from the EFI stub, which we should have been doing.

The EFI stub has its own check for this, based on the same underlying logic (have_unsupported_snp_features()). But it checks this much earlier so it can fail gracefully and return to the firmware, rather than terminate.

Quoted 5 lines

Just removing the call does cause an SNP boot to crash. My testing shows just ensuring that the cc_blob_address in the boot_params structure is set results in a successful boot under SNP. So some of sev_enable() needs to remain and some things need to be moved into the kernel proper.

Thanks, this is very helpful. I'll dig a bit deeper and try to address the points you raised.

#3 lore raw
Quoted 36 lines · Ard Biesheuvel

Hi Tom,

Thanks for taking a look.

Quoted 10 lines · Tom Lendacky
Quoted 8 lines · Ard Biesheuvel

From: Ard Biesheuvel <[email protected]>

One of the last things the EFI stub does before handing over to the core kernel when booting as a SEV guest is enabling SEV, even though this is mostly redundant: one of the first things the core kernel does is calling sme_enable(), after setting up the early GDT and IDT but before even setting up the kernel page tables.

Quoted 6 lines
Quoted 2 lines

So let's just drop this call to sev_enable(), and rely on the core kernel to initiaize SEV correctly.

SEV support does some things in sev_enable() that aren't done later in the kernel proper. For example, we check certain hypervisor features and save the CC blob into the boot parameters.

I misread the last part: the kernel proper checks struct boot_params and setup_data, whereas the decompressor checks both of those as well as the EFI config table.

It would make sense for the EFI stub to populate the field in struct boot_params directly.

Quoted 5 lines

And as I look closer, I see
that we don't call initialize_identity_maps() from the EFI stub so we aren't calling snp_check_features() from the EFI stub, which we should have been doing.

Quoted 4 lines

The EFI stub has its own check for this, based on the same underlying logic (have_unsupported_snp_features()). But it checks this much earlier so it can fail gracefully and return to the firmware, rather than terminate.

Ah, whew! I forgot we have a slightly different call for that in the EFI stub, thanks for reminding me.

Thanks,
Tom

Quoted 9 lines
Quoted 6 lines

Just removing the call does cause an SNP boot to crash. My testing shows just ensuring that the cc_blob_address in the boot_params structure is set results in a successful boot under SNP. So some of sev_enable() needs to remain and some things need to be moved into the kernel proper.

Thanks, this is very helpful. I'll dig a bit deeper and try to address the points you raised.