x86/sev: improve efi runtime code support.
v4 cover x86/sev: improve efi runtime code support.
v4:
- address review comments from Ingo.
v3:
- pick up updates from Borislav
- add vmpl check to sev_es_efi_map_ghcbs_caas
v2 changes:
- rebase to latest master.
- update error message (Dionna).
- more details in the commit message (Borislav).
Gerd Hoffmann (3):
x86/sev/vc: fix efi runtime instruction emulation x86/sev: fix error handling in sev_es_efi_map_ghcbs_caas() x86/sev: Let sev_es_efi_map_ghcbs() map the caa pages too
arch/x86/include/asm/sev.h | 4 ++-- arch/x86/coco/sev/core.c | 23 +++++++++++++++++++---- arch/x86/coco/sev/vc-handle.c | 9 ++++++++- arch/x86/platform/efi/efi_64.c | 4 ++-- 4 files changed, 31 insertions(+), 9 deletions(-)
Signature · 1 line
2.50.0
PATCH v4 1/3 x86/sev/vc: fix efi runtime instruction emulation
In case efi_mm is active go use the userspace instruction decoder which supports fetching instructions from active_mm. This is needed to make instruction emulation work for EFI runtime code, so it can use cpuid and rdmsr.
EFI runtime code uses the cpuid instruction to gather information about the environment it is running in, such as SEV being enabled or not, and choose (if needed) the SEV code path for ioport access.
EFI runtime code uses the rdmsr instruction to get the location of the CAA page (see SVSM spec, section 4.2 - "Post Boot").
The big picture behind this is that the kernel needs to be able to properly handle #VC exceptions that come from EFI runtime services. Since EFI runtime services have a special page table mapping for the EFI virtual address space, the efi_mm context must be used when decoding instructions during #VC handling.
--- arch/x86/coco/sev/vc-handle.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
arch/x86/coco/sev/vc-handle.c +8 -1
@@ -17,6 +17,7 @@#include <linux/mm.h> #include <linux/io.h> #include <linux/psp-sev.h> +#include <linux/efi.h> #include <uapi/linux/sev-guest.h> #include <asm/init.h>
@@ -178,9 +179,15 @@ static enum es_result __vc_decode_kern_insn(struct es_em_ctxt *ctxt)return ES_OK ; } +/* +* User instruction decoding is also required for the EFI runtime. Even though +* the EFI runtime is running in kernel mode, it uses special EFI virtual +* address mappings that require the use of efi_mm to properly address and +* decode. +*/ static enum es_result vc_decode_insn (struct es_em_ctxt *ctxt) { -if (user_mode (ctxt->regs)) +if (user_mode (ctxt->regs) ||mm_is_efi (current->active_mm))return __vc_decode_user_insn (ctxt);else return __vc_decode_kern_insn (ctxt);
Signature · 1 line
2.50.0
Re: [PATCH v4 1/3] x86/sev/vc: fix efi runtime instruction emulation
Quoted 13 lines
In case
efi_mmis active go use the userspace instruction decoder which supports fetching instructions fromactive_mm. This is needed to make instruction emulation work for EFI runtime code, so it can use cpuid and rdmsr.EFI runtime code uses the cpuid instruction to gather information about the environment it is running in, such as SEV being enabled or not, and choose (if needed) the SEV code path for ioport access.
EFI runtime code uses the rdmsr instruction to get the location of the CAA page (see SVSM spec, section 4.2 - "Post Boot").
Quoted 7 lines
The big picture behind this is that the kernel needs to be able to properly handle #VC exceptions that come from EFI runtime services. Since EFI runtime services have a special page table mapping for the EFI virtual address space, the
efi_mmcontext must be used when decoding instructions during #VC handling.Signed-off-by Gerd Hoffmann <[email protected]>
Quoted 33 lines
--- arch/x86/coco/sev/vc-handle.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)arch/x86/coco/sev/vc-handle.c +8
-1@@ -17,6 +17,7 @@#include <linux/mm.h> #include <linux/io.h> #include <linux/psp-sev.h> +#include <linux/efi.h> #include <uapi/linux/sev-guest.h> #include <asm/init.h> @@ -178,9 +179,15 @@ static enum es_result __vc_decode_kern_insn(struct es_em_ctxt *ctxt)return ES_OK ; } +/* +* User instruction decoding is also required for the EFI runtime. Even though +* the EFI runtime is running in kernel mode, it uses special EFI virtual +* address mappings that require the use of efi_mm to properly address and +* decode. +*/ static enum es_result vc_decode_insn (struct es_em_ctxt *ctxt) { -if (user_mode (ctxt->regs)) +if (user_mode (ctxt->regs) ||mm_is_efi (current->active_mm))return __vc_decode_user_insn (ctxt);else return __vc_decode_kern_insn (ctxt);
[tip: x86/sev] x86/sev/vc: Fix EFI runtime instruction emulation
In reply to Gerd HoffmannThe following commit has been merged into the x86/sev branch of tip:
The following commit has been merged into the x86/sev branch of tip:
Commit-ID: 7b22e0432981c2fa230f1b493082b7e67112c4aa Gitweb: https://git.kernel.org/tip/7b22e0432981c2fa230f1b493082b7e67112c4aa Author: Gerd Hoffmann <[email protected]> AuthorDate: Thu, 26 Jun 2025 13:40:11 +02:00 Committer: Borislav Petkov (AMD) <[email protected]> CommitterDate: Fri, 27 Jun 2025 13:53:12 +02:00
x86/sev/vc: Fix EFI runtime instruction emulation
In case efi_mm is active go use the userspace instruction decoder which supports fetching instructions from active_mm. This is needed to make instruction emulation work for EFI runtime code, so it can use CPUID and RDMSR.
EFI runtime code uses the CPUID instruction to gather information about the environment it is running in, such as SEV being enabled or not, and choose (if needed) the SEV code path for ioport access.
EFI runtime code uses the RDMSR instruction to get the location of the CAA page (see SVSM spec, section 4.2 - "Post Boot").
The big picture behind this is that the kernel needs to be able to properly handle #VC exceptions that come from EFI runtime services. Since EFI runtime services have a special page table mapping for the EFI virtual address space, the efi_mm context must be used when decoding instructions during #VC handling.
[ bp: Massage. ]
--- arch/x86/coco/sev/vc-handle.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
arch/x86/coco/sev/vc-handle.c +8 -1
@@ -17,6 +17,7 @@#include <linux/mm.h> #include <linux/io.h> #include <linux/psp-sev.h> +#include <linux/efi.h> #include <uapi/linux/sev-guest.h> #include <asm/init.h>
@@ -178,9 +179,15 @@ static enum es_result __vc_decode_kern_insn(struct es_em_ctxt *ctxt)return ES_OK ; } +/* +* User instruction decoding is also required for the EFI runtime. Even though +* the EFI runtime is running in kernel mode, it uses special EFI virtual +* address mappings that require the use of efi_mm to properly address and +* decode. +*/ static enum es_result vc_decode_insn (struct es_em_ctxt *ctxt) { -if (user_mode (ctxt->regs)) +if (user_mode (ctxt->regs) ||mm_is_efi (current->active_mm))return __vc_decode_user_insn (ctxt);else return __vc_decode_kern_insn (ctxt);
PATCH v4 2/3 x86/sev: fix error handling in sev_es_efi_map_ghcbs_caas()
In reply to Gerd HoffmannPass up error codes from kernel_map_pages_in_pgd() instead of returning '1' on failure.
--- arch/x86/coco/sev/core.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/x86/coco/sev/core.c +4 -2
@@ -1050,6 +1050,7 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd) {struct sev_es_runtime_data *data;unsigned long address, pflags; +int retval;int cpu;u64 pfn;
@@ -1064,8 +1065,9 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd) address =__pa (&data->ghcb_page); pfn = address >>PAGE_SHIFT ; -if (kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags)) -return 1 ; + retval =kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags); +if (retval !=0 ) +return retval; }return 0 ;
Signature · 1 line
2.50.0
Borislav Petkov
maintainerRe: [PATCH v4 2/3] x86/sev: fix error handling in sev_es_efi_map_ghcbs_caas()
arch/x86/coco/sev/core.c int __init sev_es_efi_map_ghcbs(pgd_t *pgd) · line 1070
Quoted 5 lines · Gerd Hoffmann
arch/x86/coco/sev/core.c +3
-2-if (kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags)) -return 1 ; + retval =kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags); +if (retval !=0 ) +return retval;
Yeah, I'd understand if it made any sense to propagate the error upwards but this function is called exactly once by efi_setup_page_tables() and all it needs to return is success/failure which the caller uses on the spot.
So no point in doing any of that nonsense. I'll zap it from the set.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v4 3/3 x86/sev: Let sev_es_efi_map_ghcbs() map the caa pages too
In reply to Gerd HoffmannOVMF EFI firmware needs access to the CAA page to do SVSM protocol calls. For example, when the SVSM implements an EFI variable store, such calls will be necessary.
So add that to sev_es_efi_map_ghcbs() and also rename the function to reflect the additional job it is doing now.
--- arch/x86/include/asm/sev.h | 4 ++-- arch/x86/coco/sev/core.c | 17 +++++++++++++++-- arch/x86/platform/efi/efi_64.c | 4 ++-- 3 files changed, 19 insertions(+), 6 deletions(-)
arch/x86/include/asm/sev.h +2 -2
@@ -445,7 +445,7 @@ static __always_inline void sev_es_nmi_complete(void) cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT))__sev_es_nmi_complete (); } -extern int __init sev_es_efi_map_ghcbs (pgd_t *pgd); +extern int __init sev_es_efi_map_ghcbs_caas (pgd_t *pgd);extern void sev_enable (struct boot_params *bp);/*
@@ -556,7 +556,7 @@ static inline void sev_es_ist_enter(struct pt_regs *regs) { }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 int sev_es_efi_map_ghcbs_caas (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 ; }
arch/x86/coco/sev/core.c +15 -2
@@ -1045,11 +1045,13 @@ int __init sev_es_setup_ap_jump_table(struct real_mode_header *rmh)* This is needed by the OVMF UEFI firmware which will use whatever it finds in * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu * runtime GHCBs used by the kernel are also mapped in the EFI page-table. +* +* When running under SVSM the CCA page is needed too, so map it as well. */ -int __init sev_es_efi_map_ghcbs (pgd_t *pgd) +int __init sev_es_efi_map_ghcbs_caas (pgd_t *pgd) {struct sev_es_runtime_data *data; -unsigned long address, pflags; +unsigned long address, pflags, pflags_enc;int retval;int cpu;u64 pfn;
@@ -1058,6 +1060,7 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd)return 0 ; pflags = _PAGE_NX | _PAGE_RW; + pflags_enc =cc_mkenc (pflags);for_each_possible_cpu (cpu) { data =per_cpu (runtime_data, cpu);
@@ -1068,6 +1071,16 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd) retval =kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags);if (retval !=0 )return retval; + +if (snp_vmpl) { + address =per_cpu (svsm_caa_pa, cpu); +if (!address) +return 1 ; + + pfn = address >>PAGE_SHIFT ; +if (kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags_enc)) +return 1 ; + } }return 0 ;
arch/x86/platform/efi/efi_64.c +2 -2
@@ -216,8 +216,8 @@ int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)* When SEV-ES is active, the GHCB as set by the kernel will be used * by firmware. Create a 1:1 unencrypted mapping for each GHCB. */ -if (sev_es_efi_map_ghcbs (pgd)) { -pr_err ("Failed to create 1:1 mapping for the GHCBs!\n" ); +if (sev_es_efi_map_ghcbs_caas (pgd)) { +pr_err ("Failed to create 1:1 mapping for the GHCBs and CAAs!\n" );return 1 ; }
Signature · 1 line
2.50.0
[tip: x86/sev] x86/sev: Let sev_es_efi_map_ghcbs() map the CA pages too
The following commit has been merged into the x86/sev branch of tip:
The following commit has been merged into the x86/sev branch of tip:
Commit-ID: a7549636f67f973474ebe1ad262acc2aa4d1327d Gitweb: https://git.kernel.org/tip/a7549636f67f973474ebe1ad262acc2aa4d1327d Author: Gerd Hoffmann <[email protected]> AuthorDate: Thu, 26 Jun 2025 13:40:13 +02:00 Committer: Borislav Petkov (AMD) <[email protected]> CommitterDate: Fri, 27 Jun 2025 14:07:10 +02:00
x86/sev: Let sev_es_efi_map_ghcbs() map the CA pages too
OVMF EFI firmware needs access to the CA page to do SVSM protocol calls. For example, when the SVSM implements an EFI variable store, such calls will be necessary.
So add that to sev_es_efi_map_ghcbs() and also rename the function to reflect the additional job it is doing now.
[ bp: Massage. ]
--- arch/x86/coco/sev/core.c | 17 +++++++++++++++-- arch/x86/include/asm/sev.h | 4 ++-- arch/x86/platform/efi/efi_64.c | 4 ++-- 3 files changed, 19 insertions(+), 6 deletions(-)
arch/x86/coco/sev/core.c +15 -2
@@ -1045,11 +1045,13 @@ int __init sev_es_setup_ap_jump_table(struct real_mode_header *rmh)* This is needed by the OVMF UEFI firmware which will use whatever it finds in * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu * runtime GHCBs used by the kernel are also mapped in the EFI page-table. +* +* When running under SVSM the CA page is needed too, so map it as well. */ -int __init sev_es_efi_map_ghcbs (pgd_t *pgd) +int __init sev_es_efi_map_ghcbs_cas (pgd_t *pgd) { +unsigned long address, pflags, pflags_enc;struct sev_es_runtime_data *data; -unsigned long address, pflags;int cpu;u64 pfn;
@@ -1057,6 +1059,7 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd)return 0 ; pflags = _PAGE_NX | _PAGE_RW; + pflags_enc =cc_mkenc (pflags);for_each_possible_cpu (cpu) { data =per_cpu (runtime_data, cpu);
@@ -1066,6 +1069,16 @@ int __init sev_es_efi_map_ghcbs(pgd_t *pgd)if (kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags))return 1 ; + +if (snp_vmpl) { + address =per_cpu (svsm_caa_pa, cpu); +if (!address) +return 1 ; + + pfn = address >>PAGE_SHIFT ; +if (kernel_map_pages_in_pgd (pgd, pfn, address,1 , pflags_enc)) +return 1 ; + } }return 0 ;
arch/x86/include/asm/sev.h +2 -2
@@ -446,7 +446,7 @@ static __always_inline void sev_es_nmi_complete(void) cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT))__sev_es_nmi_complete (); } -extern int __init sev_es_efi_map_ghcbs (pgd_t *pgd); +extern int __init sev_es_efi_map_ghcbs_cas (pgd_t *pgd);extern void sev_enable (struct boot_params *bp);/*
@@ -554,7 +554,7 @@ static inline void sev_es_ist_enter(struct pt_regs *regs) { }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 int sev_es_efi_map_ghcbs_cas (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 ; }
arch/x86/platform/efi/efi_64.c +2 -2
@@ -216,8 +216,8 @@ int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)* When SEV-ES is active, the GHCB as set by the kernel will be used * by firmware. Create a 1:1 unencrypted mapping for each GHCB. */ -if (sev_es_efi_map_ghcbs (pgd)) { -pr_err ("Failed to create 1:1 mapping for the GHCBs!\n" ); +if (sev_es_efi_map_ghcbs_cas (pgd)) { +pr_err ("Failed to create 1:1 mapping for the GHCBs and CAs!\n" );return 1 ; }