x86: SEV-ES Guest Support
Subsystems (8)
v2 cover x86: SEV-ES Guest Support
Hi,
here is an updated version of the patch-set to enable Linux to run as a guest in an SEV-ES enabled Hypervisor. The first version can be found here:
https://lore.kernel.org/lkml/[email protected]/
The first post also includes a more elaborate description of the implementation requirements and details. A branch containing these patches is here:
https://git.kernel.org/pub/scm/linux/kernel/git/joro/linux.git/log/?h=sev-es-client-v5.6-rc6
There are lots of small changes since the first version, here is a list of the major ones, which address most of the valuable review comments I received, thanks for that!
Changes since v1:
- Rebased to v5.6-rc6
- Factored out instruction decoding part of the UMIP handler and re-used it in the SEV-ES code.
- Several enhancements of the instruction decoder as needed by SEV-ES
- The instruction fetch and memory access code for instruction emulation now handles different user execution modes as well as segment bases.
- Added emulation of (REP) MOVS instructions
- Added handling for nesting #VC handlers - which fixed the NMI issues.
- Pass error_code as a parameter to the #VC exception handlers
- Reworked early exception dispatch function
- Moved the GHCB pages out of the per-cpu areas and only allocate them when they are actually needed. The per-cpu areas only store a pointer now.
- Removed emulation for INVD, now it will just cause an error if used.
- Added prefixes to the function names.
- Fixed a bug which broke bare-metal boot with mem_encrypt=on
The last missing change I have on my list is to rework the NMI handling patch. I decided to postpone this until Thomas' Gleixners rework of the x86 entry code is ready and merged, because the NMI handling will conflict with these changes.
Please review.
Thanks,
Joerg
Doug Covelli (1):
x86/vmware: Add VMware specific handling for VMMCALL under SEV-ES
Joerg Roedel (51):
KVM: SVM: Add GHCB Accessor functions
x86/traps: Move some definitions to <asm/trap_defs.h> x86/insn: Make inat-tables.c suitable for pre-decompression code x86/umip: Factor out instruction fetch
x86/umip: Factor out instruction decoding
x86/insn: Add insn_get_modrm_reg_off()
x86/insn: Add insn_rep_prefix() helper x86/boot/compressed: Fix debug_puthex() parameter type x86/boot/compressed/64: Disable red-zone usage x86/boot/compressed/64: Add IDT Infrastructure x86/boot/compressed/64: Rename kaslr_64.c to ident_map_64.c x86/boot/compressed/64: Add page-fault handler x86/boot/compressed/64: Always switch to own page-table x86/boot/compressed/64: Don't pre-map memory in KASLR code x86/boot/compressed/64: Change add_identity_map() to take start and
end
x86/boot/compressed/64: Add stage1 #VC handler x86/boot/compressed/64: Call set_sev_encryption_mask earlier x86/boot/compressed/64: Check return value of
kernel_ident_mapping_init()
x86/boot/compressed/64: Add function to map a page unencrypted x86/boot/compressed/64: Setup GHCB Based VC Exception handler x86/fpu: Move xgetbv()/xsetbv() into separate header x86/idt: Move IDT to data segment
x86/idt: Split idt_data setup out of set_intr_gate() x86/idt: Move two function from k/idt.c to i/a/desc.h x86/head/64: Install boot GDT
x86/head/64: Reload GDT after switch to virtual addresses x86/head/64: Load segment registers earlier
x86/head/64: Switch to initial stack earlier
x86/head/64: Build k/head64.c with -fno-stack-protector x86/head/64: Load IDT earlier
x86/head/64: Move early exception dispatch to C code x86/sev-es: Add SEV-ES Feature Detection
x86/sev-es: Compile early handler code into kernel image x86/sev-es: Setup early #VC handler
x86/sev-es: Setup GHCB based boot #VC handler
x86/sev-es: Support nested #VC exceptions
x86/sev-es: Wire up existing #VC exit-code handlers x86/sev-es: Handle instruction fetches from user-space x86/sev-es: Harden runtime #VC handler for exceptions from user-space x86/sev-es: Filter exceptions not supported from user-space x86/sev-es: Handle MMIO String Instructions
x86/sev-es: Handle RDTSCP Events
x86/sev-es: Handle #AC Events
x86/sev-es: Handle #DB Events
x86/paravirt: Allow hypervisor specific VMMCALL handling under SEV-ES x86/realmode: Add SEV-ES specific trampoline entry point x86/head/64: Don't call verify_cpu() on starting APs x86/head/64: Rename start_cpu0
x86/sev-es: Support CPU offline/online
x86/cpufeature: Add SEV_ES_GUEST CPU Feature
x86/sev-es: Add NMI state tracking
Tom Lendacky (18):
KVM: SVM: Add GHCB definitions x86/cpufeatures: Add SEV-ES CPU feature x86/sev-es: Add support for handling IOIO exceptions x86/sev-es: Add CPUID handling to #VC handler x86/sev-es: Setup per-cpu GHCBs for the runtime handler x86/sev-es: Add Runtime #VC Exception Handler x86/sev-es: Handle MMIO events
x86/sev-es: Handle MSR events
x86/sev-es: Handle DR7 read/write events x86/sev-es: Handle WBINVD Events
x86/sev-es: Handle RDTSC Events
x86/sev-es: Handle RDPMC Events
x86/sev-es: Handle INVD Events
x86/sev-es: Handle MONITOR/MONITORX Events x86/sev-es: Handle MWAIT/MWAITX Events x86/sev-es: Handle VMMCALL Events
x86/kvm: Add KVM specific VMMCALL handling under SEV-ES x86/realmode: Setup AP jump table
arch/x86/Kconfig | 1 + arch/x86/boot/Makefile | 2 +- arch/x86/boot/compressed/Makefile | 8 +- arch/x86/boot/compressed/head_64.S | 41 + arch/x86/boot/compressed/ident_map_64.c | 320 ++++++ arch/x86/boot/compressed/idt_64.c | 53 + arch/x86/boot/compressed/idt_handlers_64.S | 82 ++ arch/x86/boot/compressed/kaslr.c | 36 +- arch/x86/boot/compressed/kaslr_64.c | 153 --- arch/x86/boot/compressed/misc.h | 34 +- arch/x86/boot/compressed/sev-es.c | 177 +++ arch/x86/entry/entry_64.S | 52 + arch/x86/include/asm/cpu.h | 2 +- arch/x86/include/asm/cpufeatures.h | 2 + arch/x86/include/asm/desc.h | 28 + arch/x86/include/asm/desc_defs.h | 10 + arch/x86/include/asm/fpu/internal.h | 29 +- arch/x86/include/asm/fpu/xcr.h | 32 + arch/x86/include/asm/insn-eval.h | 6 + arch/x86/include/asm/mem_encrypt.h | 5 + arch/x86/include/asm/msr-index.h | 3 + arch/x86/include/asm/pgtable.h | 2 +- arch/x86/include/asm/processor.h | 1 + arch/x86/include/asm/realmode.h | 4 + arch/x86/include/asm/segment.h | 2 +- arch/x86/include/asm/setup.h | 1 - arch/x86/include/asm/sev-es.h | 119 ++ arch/x86/include/asm/svm.h | 103 ++ arch/x86/include/asm/trap_defs.h | 50 + arch/x86/include/asm/traps.h | 51 +- arch/x86/include/asm/x86_init.h | 16 +- arch/x86/include/uapi/asm/svm.h | 11 + arch/x86/kernel/Makefile | 5 + arch/x86/kernel/cpu/amd.c | 9 +- arch/x86/kernel/cpu/scattered.c | 1 + arch/x86/kernel/cpu/vmware.c | 50 +- arch/x86/kernel/head64.c | 57 +- arch/x86/kernel/head_32.S | 4 +- arch/x86/kernel/head_64.S | 169 ++- arch/x86/kernel/idt.c | 52 +- arch/x86/kernel/kvm.c | 35 +- arch/x86/kernel/nmi.c | 8 + arch/x86/kernel/sev-es-shared.c | 444 ++++++++ arch/x86/kernel/sev-es.c | 1165 ++++++++++++++++++++ arch/x86/kernel/smpboot.c | 4 +- arch/x86/kernel/traps.c | 3 + arch/x86/kernel/umip.c | 49 +- arch/x86/lib/insn-eval.c | 130 +++ arch/x86/mm/extable.c | 1 + arch/x86/mm/mem_encrypt.c | 11 +- arch/x86/mm/mem_encrypt_identity.c | 3 + arch/x86/realmode/init.c | 12 + arch/x86/realmode/rm/header.S | 3 + arch/x86/realmode/rm/trampoline_64.S | 20 + arch/x86/tools/gen-insn-attr-x86.awk | 50 +- tools/arch/x86/tools/gen-insn-attr-x86.awk | 50 +- 56 files changed, 3352 insertions(+), 419 deletions(-) create mode 100644 arch/x86/boot/compressed/ident_map_64.c create mode 100644 arch/x86/boot/compressed/idt_64.c create mode 100644 arch/x86/boot/compressed/idt_handlers_64.S delete mode 100644 arch/x86/boot/compressed/kaslr_64.c create mode 100644 arch/x86/boot/compressed/sev-es.c create mode 100644 arch/x86/include/asm/fpu/xcr.h create mode 100644 arch/x86/include/asm/sev-es.h create mode 100644 arch/x86/include/asm/trap_defs.h create mode 100644 arch/x86/kernel/sev-es-shared.c create mode 100644 arch/x86/kernel/sev-es.c
Signature · 1 line
2.17.1
PATCH v2 1/70 KVM: SVM: Add GHCB definitions
From: Tom Lendacky <[email protected]>
Extend the vmcb_safe_area with SEV-ES fields and add a new 'struct ghcb' which will be used for guest-hypervisor communication.
--- arch/x86/include/asm/svm.h | 42 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 42 insertions(+)
arch/x86/include/asm/svm.h +42 -0
@@ -201,6 +201,48 @@ struct __attribute__ ((__packed__)) vmcb_save_area {u64 br_to;u64 last_excp_from;u64 last_excp_to; + +/* +* The following part of the save area is valid only for +* SEV-ES guests when referenced through the GHCB. +*/ +u8 reserved_7[104 ]; +u64 reserved_8;/* rax already available at 0x01f8 */ +u64 rcx; +u64 rdx; +u64 rbx; +u64 reserved_9;/* rsp already available at 0x01d8 */ +u64 rbp; +u64 rsi; +u64 rdi; +u64 r8; +u64 r9; +u64 r10; +u64 r11; +u64 r12; +u64 r13; +u64 r14; +u64 r15; +u8 reserved_10[16 ]; +u64 sw_exit_code; +u64 sw_exit_info_1; +u64 sw_exit_info_2; +u64 sw_scratch; +u8 reserved_11[56 ]; +u64 xcr0; +u8 valid_bitmap[16 ]; +u64 x87_state_gpa; +u8 reserved_12[1016 ]; +}; + +struct __attribute__ ((__packed__ ))ghcb { +struct vmcb_save_area save; + +u8 shared_buffer[2032 ]; + +u8 reserved_1[10 ]; +u16 protocol_version;/* negotiated SEV-ES/GHCB protocol version */ +u32 ghcb_usage; };struct __attribute__ ((__packed__ ))vmcb {
Signature · 1 line
2.17.1
Borislav Petkov
maintainerPATCH 1/1 KVM: SVM: Use __packed shorthard
I guess we can do that ontop.
---
From: Borislav Petkov <[email protected]>
Date: Mon, 23 Mar 2020 14:20:08 +0100
... to make it more readable.
No functional changes.
--- arch/x86/include/asm/svm.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
arch/x86/include/asm/svm.h +8 -8
@@ -151,14 +151,14 @@ struct __attribute__ ((__packed__)) vmcb_control_area {#define SVM_NESTED_CTL_NP_ENABLE BIT(0)#define SVM_NESTED_CTL_SEV_ENABLE BIT(1) -struct __attribute__ ((__packed__ ))vmcb_seg { +struct vmcb_seg {u16 selector;u16 attrib;u32 limit;u64 base; -}; +} __packed; -struct __attribute__ ((__packed__ ))vmcb_save_area { +struct vmcb_save_area {struct vmcb_seg es;struct vmcb_seg cs;struct vmcb_seg ss;
@@ -233,9 +233,9 @@ struct __attribute__ ((__packed__)) vmcb_save_area {u8 valid_bitmap[16 ];u64 x87_state_gpa;u8 reserved_12[1016 ]; -}; +} __packed; -struct __attribute__ ((__packed__ ))ghcb { +struct ghcb {struct vmcb_save_area save;u8 shared_buffer[2032 ];
@@ -243,12 +243,12 @@ struct __attribute__ ((__packed__)) ghcb {u8 reserved_1[10 ];u16 protocol_version;/* negotiated SEV-ES/GHCB protocol version */ u32 ghcb_usage; -}; +} __packed; -struct __attribute__ ((__packed__ ))vmcb { +struct vmcb {struct vmcb_control_area control;struct vmcb_save_area save; -}; +} __packed;#define SVM_CPUID_FUNC 0x8000000a
Signature · 7 lines
2.21.0
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [PATCH] KVM: SVM: Use __packed shorthard
Quoted 6 lines · Borislav Petkov
I guess we can do that ontop.
---
From: Borislav Petkov <[email protected]>
Date: Mon, 23 Mar 2020 14:20:08 +0100
Quoted 8 lines
... to make it more readable.
No functional changes.
Signed-off-by Borislav Petkov <[email protected]>--- arch/x86/include/asm/svm.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-)
Added it to the patch-set, thanks Boris.
PATCH v2 2/70 KVM: SVM: Add GHCB Accessor functions
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Building a correct GHCB for the hypervisor requires setting valid bits in the GHCB. Simplify that process by providing accessor functions to set values and to update the valid bitmap.
--- arch/x86/include/asm/svm.h | 61 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 61 insertions(+)
arch/x86/include/asm/svm.h +61 -0
@@ -333,4 +333,65 @@ struct __attribute__ ((__packed__)) vmcb {#define SVM_CR0_SELECTIVE_MASK (X86_CR0_TS | X86_CR0_MP) +/* GHCB Accessor functions */ + +#define DEFINE_GHCB_INDICES (field) \ + u16 idx = offsetof(struct vmcb_save_area, field) / 8; \ + u16 byte_idx = idx / 8; \ + u16 bit_idx = idx % 8; \ + BUILD_BUG_ON(byte_idx > ARRAY_SIZE(ghcb->save.valid_bitmap)); + +#define GHCB_SET_VALID (ghcb, field) \ + { \ + DEFINE_GHCB_INDICES(field) \ + (ghcb)->save.valid_bitmap[byte_idx] |= BIT(bit_idx); \ + } + +#define DEFINE_GHCB_SETTER (field) \ + static inline void \ + ghcb_set_##field(struct ghcb *ghcb, u64 value) \ + { \ + GHCB_SET_VALID(ghcb, field) \ + (ghcb)->save.field = value; \ + } + +#define DEFINE_GHCB_ACCESSORS (field) \ + static inline bool ghcb_is_valid_##field(const struct ghcb *ghcb) \ + { \ + DEFINE_GHCB_INDICES(field) \ + return !!((ghcb)->save.valid_bitmap[byte_idx] \ + & BIT(bit_idx)); \ + } \ + \ + static inline void \ + ghcb_set_##field(struct ghcb *ghcb, u64 value) \ + { \ + GHCB_SET_VALID(ghcb, field) \ + (ghcb)->save.field = value; \ + } + +DEFINE_GHCB_ACCESSORS(cpl) +DEFINE_GHCB_ACCESSORS(rip) +DEFINE_GHCB_ACCESSORS(rsp) +DEFINE_GHCB_ACCESSORS(rax) +DEFINE_GHCB_ACCESSORS(rcx) +DEFINE_GHCB_ACCESSORS(rdx) +DEFINE_GHCB_ACCESSORS(rbx) +DEFINE_GHCB_ACCESSORS(rbp) +DEFINE_GHCB_ACCESSORS(rsi) +DEFINE_GHCB_ACCESSORS(rdi) +DEFINE_GHCB_ACCESSORS(r8) +DEFINE_GHCB_ACCESSORS(r9) +DEFINE_GHCB_ACCESSORS(r10) +DEFINE_GHCB_ACCESSORS(r11) +DEFINE_GHCB_ACCESSORS(r12) +DEFINE_GHCB_ACCESSORS(r13) +DEFINE_GHCB_ACCESSORS(r14) +DEFINE_GHCB_ACCESSORS(r15) +DEFINE_GHCB_ACCESSORS(sw_exit_code) +DEFINE_GHCB_ACCESSORS(sw_exit_info_1) +DEFINE_GHCB_ACCESSORS(sw_exit_info_2) +DEFINE_GHCB_ACCESSORS(sw_scratch) +DEFINE_GHCB_ACCESSORS(xcr0) +#endif
Signature · 1 line
2.17.1
PATCH v2 3/70 x86/cpufeatures: Add SEV-ES CPU feature
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Add CPU feature detection for Secure Encrypted Virtualization with Encrypted State. This feature enhances SEV by also encrypting the guest register state, making it in-accessible to the hypervisor.
--- arch/x86/include/asm/cpufeatures.h | 1 + arch/x86/kernel/cpu/amd.c | 3 ++- arch/x86/kernel/cpu/scattered.c | 1 + 3 files changed, 4 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
@@ -234,6 +234,7 @@#define X86_FEATURE_EPT_AD ( 8*32+17)/* Intel Extended Page Table access-dirty bit */ #define X86_FEATURE_VMCALL ( 8*32+18)/* "" Hypervisor supports the VMCALL instruction */ #define X86_FEATURE_VMW_VMMCALL ( 8*32+19)/* "" VMware prefers VMMCALL hypercall instruction */ +#define X86_FEATURE_SEV_ES ( 8*32+20)/* AMD Secure Encrypted Virtualization - Encrypted State */ /* Intel-defined CPU features, CPUID level 0x00000007:0 (EBX), word 9 */ #define X86_FEATURE_FSGSBASE ( 9*32+ 0)/* RDFSBASE, WRFSBASE, RDGSBASE, WRGSBASE instructions*/
arch/x86/kernel/cpu/amd.c +2 -1
@@ -581,7 +581,7 @@ static void early_detect_mem_encrypt(struct cpuinfo_x86 *c)* If BIOS has not enabled SME then don't advertise the * SME feature (set in scattered.c). * For SEV: If BIOS has not enabled SEV then don't advertise the -* SEV feature (set in scattered.c). +* SEV and SEV_ES feature (set in scattered.c). * * In all cases, since support for SME and SEV requires long mode, * don't advertise the feature under CONFIG_X86_32.
@@ -612,6 +612,7 @@ static void early_detect_mem_encrypt(struct cpuinfo_x86 *c)setup_clear_cpu_cap (X86_FEATURE_SME ); clear_sev:setup_clear_cpu_cap (X86_FEATURE_SEV ); +setup_clear_cpu_cap (X86_FEATURE_SEV_ES ); } }
arch/x86/kernel/cpu/scattered.c +1 -0
@@ -41,6 +41,7 @@ static const struct cpuid_bit cpuid_bits[] = { { X86_FEATURE_MBA, CPUID_EBX, 6, 0x80000008, 0 }, { X86_FEATURE_SME, CPUID_EAX, 0, 0x8000001f, 0 }, { X86_FEATURE_SEV, CPUID_EAX, 1, 0x8000001f, 0 }, + { X86_FEATURE_SEV_ES, CPUID_EAX, 3, 0x8000001f, 0 }, { 0, 0, 0, 0, 0 } };
Signature · 1 line
2.17.1
PATCH v2 4/70 x86/traps: Move some definitions to <asm/trap_defs.h>
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Move the definition of x86 trap vector numbers and the page-fault error code bits to the new header file asm/trap_defs.h. This makes it easier to include them into pre-decompression boot code. No functional changes.
--- arch/x86/include/asm/trap_defs.h | 49 ++++++++++++++++++++++++++++++++ arch/x86/include/asm/traps.h | 44 +--------------------------- 2 files changed, 50 insertions(+), 43 deletions(-) create mode 100644 arch/x86/include/asm/trap_defs.h
arch/x86/include/asm/trap_defs.h new file +49 -0
@@ -0,0 +1,49 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _ASM_X86_TRAP_DEFS_H +#define _ASM_X86_TRAP_DEFS_H + +/* Interrupts/Exceptions */ +enum { +X86_TRAP_DE =0 ,/* 0, Divide-by-zero */ +X86_TRAP_DB ,/* 1, Debug */ +X86_TRAP_NMI ,/* 2, Non-maskable Interrupt */ +X86_TRAP_BP ,/* 3, Breakpoint */ +X86_TRAP_OF ,/* 4, Overflow */ +X86_TRAP_BR ,/* 5, Bound Range Exceeded */ +X86_TRAP_UD ,/* 6, Invalid Opcode */ +X86_TRAP_NM ,/* 7, Device Not Available */ +X86_TRAP_DF ,/* 8, Double Fault */ +X86_TRAP_OLD_MF ,/* 9, Coprocessor Segment Overrun */ +X86_TRAP_TS ,/* 10, Invalid TSS */ +X86_TRAP_NP ,/* 11, Segment Not Present */ +X86_TRAP_SS ,/* 12, Stack Segment Fault */ +X86_TRAP_GP ,/* 13, General Protection Fault */ +X86_TRAP_PF ,/* 14, Page Fault */ +X86_TRAP_SPURIOUS ,/* 15, Spurious Interrupt */ +X86_TRAP_MF ,/* 16, x87 Floating-Point Exception */ +X86_TRAP_AC ,/* 17, Alignment Check */ +X86_TRAP_MC ,/* 18, Machine Check */ +X86_TRAP_XF ,/* 19, SIMD Floating-Point Exception */ +X86_TRAP_IRET =32 ,/* 32, IRET Exception */ +}; + +/* +* Page fault error code bits: +* +* bit 0 == 0: no page found 1: protection fault +* bit 1 == 0: read access 1: write access +* bit 2 == 0: kernel-mode access 1: user-mode access +* bit 3 == 1: use of reserved bit detected +* bit 4 == 1: fault was an instruction fetch +* bit 5 == 1: protection keys block access +*/ +enum x86_pf_error_code { +X86_PF_PROT =1 <<0 , +X86_PF_WRITE =1 <<1 , +X86_PF_USER =1 <<2 , +X86_PF_RSVD =1 <<3 , +X86_PF_INSTR =1 <<4 , +X86_PF_PK =1 <<5 , +}; + +#endif /* _ASM_X86_TRAP_DEFS_H */
arch/x86/include/asm/traps.h +1 -43
54 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 5/70 x86/insn: Make inat-tables.c suitable for pre-decompression code
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The inat-tables.c file has some arrays in it that contain pointers to other arrays. These pointers need to be relocated when the kernel image is moved to a different location.
The pre-decompression boot-code has no support for applying ELF relocations, so initialize these arrays at runtime in the pre-decompression code to make sure all pointers are correctly initialized.
--- arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- tools/arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- 2 files changed, 98 insertions(+), 2 deletions(-)
arch/x86/tools/gen-insn-attr-x86.awk +49 -1
@@ -362,6 +362,9 @@ function convert_operands(count,opnd, i,j,imm,mod) END { if (awkchecked != "") exit 1 + + print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \
@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]"," - print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n" + + print "/* Escape opcode map array */" + print "static const insn_attr_t *inat_escape_tables[INAT_ESC_MAX + 1]" \ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* Group opcode map array */" + print "static const insn_attr_t *inat_group_tables[INAT_GRP_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* AVX opcode map array */" + print "static const insn_attr_t *inat_avx_tables[X86_VEX_M_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "static void inat_init_tables(void)" + print "{" + + # print escape opcode map's array + print "\t/* Print Escape opcode map array */" + for (i = 0; i < geid; i++) + for (j = 0; j < max_lprefix; j++) + if (etable[i,j]) + print "\tinat_escape_tables["i"]["j"] = "etable[i,j]";" + print "" + + # print group opcode map's array + print "\t/* Print Group opcode map array */" + for (i = 0; i < ggid; i++) + for (j = 0; j < max_lprefix; j++) + if (gtable[i,j]) + print "\tinat_group_tables["i"]["j"] = "gtable[i,j]";" + print "" + # print AVX opcode map's array + print "\t/* Print AVX opcode map array */" + for (i = 0; i < gaid; i++) + for (j = 0; j < max_lprefix; j++) + if (atable[i,j]) + print "\tinat_avx_tables["i"]["j"] = "atable[i,j]";" + + print "}" + print "#endif" }
tools/arch/x86/tools/gen-insn-attr-x86.awk +49 -1
61 lines not on this page · raw message
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 05/70] x86/insn: Make inat-tables.c suitable for pre-decompression code
+ Masami.
Quoted 154 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
The
inat-tables.cfile has some arrays in it that contain pointers to other arrays. These pointers need to be relocated when the kernel image is moved to a different location.The pre-decompression boot-code has no support for applying ELF relocations, so initialize these arrays at runtime in the pre-decompression code to make sure all pointers are correctly initialized.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- tools/arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- 2 files changed, 98 insertions(+), 2 deletions(-)arch/x86/tools/gen-insn-attr-x86.awk +49
-1@@ -362,6 +362,9 @@ function convert_operands(count,opnd, i,j,imm,mod) END { if (awkchecked != "") exit 1 + + print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]"," - print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n" + + print "/* Escape opcode map array */" + print "static const insn_attr_t *inat_escape_tables[INAT_ESC_MAX + 1]" \ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* Group opcode map array */" + print "static const insn_attr_t *inat_group_tables[INAT_GRP_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* AVX opcode map array */" + print "static const insn_attr_t *inat_avx_tables[X86_VEX_M_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "static void inat_init_tables(void)" + print "{" + + # print escape opcode map's array + print "\t/* Print Escape opcode map array */" + for (i = 0; i < geid; i++) + for (j = 0; j < max_lprefix; j++) + if (etable[i,j]) + print "\tinat_escape_tables["i"]["j"] = "etable[i,j]";" + print "" + + # print group opcode map's array + print "\t/* Print Group opcode map array */" + for (i = 0; i < ggid; i++) + for (j = 0; j < max_lprefix; j++) + if (gtable[i,j]) + print "\tinat_group_tables["i"]["j"] = "gtable[i,j]";" + print "" + # print AVX opcode map's array + print "\t/* Print AVX opcode map array */" + for (i = 0; i < gaid; i++) + for (j = 0; j < max_lprefix; j++) + if (atable[i,j]) + print "\tinat_avx_tables["i"]["j"] = "atable[i,j]";" + + print "}" + print "#endif" }tools/arch/x86/tools/gen-insn-attr-x86.awk +49
-1@@ -362,6 +362,9 @@ function convert_operands(count,opnd, i,j,imm,mod) END { if (awkchecked != "") exit 1 + + print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]"," - print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n" + + print "/* Escape opcode map array */" + print "static const insn_attr_t *inat_escape_tables[INAT_ESC_MAX + 1]" \ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* Group opcode map array */" + print "static const insn_attr_t *inat_group_tables[INAT_GRP_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* AVX opcode map array */" + print "static const insn_attr_t *inat_avx_tables[X86_VEX_M_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "static void inat_init_tables(void)" + print "{" + + # print escape opcode map's array + print "\t/* Print Escape opcode map array */" + for (i = 0; i < geid; i++) + for (j = 0; j < max_lprefix; j++) + if (etable[i,j]) + print "\tinat_escape_tables["i"]["j"] = "etable[i,j]";" + print "" + + # print group opcode map's array + print "\t/* Print Group opcode map array */" + for (i = 0; i < ggid; i++) + for (j = 0; j < max_lprefix; j++) + if (gtable[i,j]) + print "\tinat_group_tables["i"]["j"] = "gtable[i,j]";" + print "" + # print AVX opcode map's array + print "\t/* Print AVX opcode map array */" + for (i = 0; i < gaid; i++) + for (j = 0; j < max_lprefix; j++) + if (atable[i,j]) + print "\tinat_avx_tables["i"]["j"] = "atable[i,j]";" + + print "}" + print "#endif" }Signature · 1 line
2.17.1
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [PATCH 05/70] x86/insn: Make inat-tables.c suitable for pre-decompression code
Hi,
Quoted 13 lines · On Wed, 25 Mar 2020 16:39:45 +0100 Borislav Petkov <[email protected]> wrote:
+ Masami.
Quoted 10 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
The
inat-tables.cfile has some arrays in it that contain pointers to other arrays. These pointers need to be relocated when the kernel image is moved to a different location.The pre-decompression boot-code has no support for applying ELF relocations, so initialize these arrays at runtime in the pre-decompression code to make sure all pointers are correctly initialized.
I need to check the whole series, but as far as I can understand from this patch, this seems not allowing to store the address value in static pointers. It may break more things, for example _kprobe_blacklist records the NOKPROBE_SYMBOL() symbol addresses at the build time.
I have some comments here.
Quoted 20 lines
Quoted 20 lines
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- tools/arch/x86/tools/gen-insn-attr-x86.awk | 50 +++++++++++++++++++++- 2 files changed, 98 insertions(+), 2 deletions(-)arch/x86/tools/gen-insn-attr-x86.awk +3
-0@@ -362,6 +362,9 @@ function convert_operands(count,opnd, i,j,imm,mod) END { if (awkchecked != "") exit 1 + + print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk END { · line 396
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk +3
-1@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]"," - print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n"
I think the definitions of inat_*_tables can be shared in both case. If __BOOT_COMPRESSED is set, we can define inat_init_tables() as a initialize function, and if not, it will be just a dummy "do {} while (0)".
BTW, where is the __BOOT_COMPRESSED defined?
Quoted 9 lines
Quoted 9 lines
arch/x86/tools/gen-insn-attr-x86.awk +9
-0+ + print "/* Escape opcode map array */" + print "static const insn_attr_t *inat_escape_tables[INAT_ESC_MAX + 1]" \ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "/* Group opcode map array */" + print "static const insn_attr_t *inat_group_tables[INAT_GRP_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];"
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk END { · line 413
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk +8
-0+ print "" + + print "/* AVX opcode map array */" + print "static const insn_attr_t *inat_avx_tables[X86_VEX_M_MAX + 1]"\ + "[INAT_LSTPFX_MAX + 1];" + print "" + + print "static void inat_init_tables(void)"
This functions should be "inline".
And I can not see the call-site of inat_init_tables() in this patch.
If possible, please include call-site with definition (especially new init function) so that I can check the init call timing too.
Quoted 19 lines
Quoted 19 lines
arch/x86/tools/gen-insn-attr-x86.awk +19
-0+ print "{" + + # print escape opcode map's array + print "\t/* Print Escape opcode map array */" + for (i = 0; i < geid; i++) + for (j = 0; j < max_lprefix; j++) + if (etable[i,j]) + print "\tinat_escape_tables["i"]["j"] = "etable[i,j]";" + print "" + + # print group opcode map's array + print "\t/* Print Group opcode map array */" + for (i = 0; i < ggid; i++) + for (j = 0; j < max_lprefix; j++) + if (gtable[i,j]) + print "\tinat_group_tables["i"]["j"] = "gtable[i,j]";" + print "" + # print AVX opcode map's array + print "\t/* Print AVX opcode map array */"
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk END { · line 440
Quoted 8 lines
arch/x86/tools/gen-insn-attr-x86.awk +7
-0+ for (i = 0; i < gaid; i++) + for (j = 0; j < max_lprefix; j++) + if (atable[i,j]) + print "\tinat_avx_tables["i"]["j"] = "atable[i,j]";" + + print "}" + print "#endif" }
The code itself looks good to me.
Thank you,
Signature · 1 line
Masami Hiramatsu <[email protected]>
Re: [PATCH 05/70] x86/insn: Make inat-tables.c suitable for pre-decompression code
Hi Masami,
Quoted 17 lines · Masami Hiramatsu
Quoted 13 lines · On Wed, 25 Mar 2020 16:39:45 +0100 Borislav Petkov <[email protected]> wrote:
+ Masami.
Quoted 10 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
The
inat-tables.cfile has some arrays in it that contain pointers to other arrays. These pointers need to be relocated when the kernel image is moved to a different location.The pre-decompression boot-code has no support for applying ELF relocations, so initialize these arrays at runtime in the pre-decompression code to make sure all pointers are correctly initialized.
Quoted 4 lines
I need to check the whole series, but as far as I can understand from this patch, this seems not allowing to store the address value in static pointers. It may break more things, for example
_kprobe_blacklistrecords theNOKPROBE_SYMBOL()symbol addresses at the build time.
The runtime-initialization function is only used in the pre-decompression boot code (arch/x86/boot/compressed/) which is not part of the running kernel image. At that stage of booting there is no support for kprobe or tracing or any other neat features that might break things here.
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
arch/x86/tools/gen-insn-attr-x86.awk +2
-0+ print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]","
Quoted 8 lines
Quoted 4 lines
arch/x86/tools/gen-insn-attr-x86.awk END { · line 396
Quoted 4 lines
arch/x86/tools/gen-insn-attr-x86.awk +3
-1- print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n"I think the definitions of inat_*_tables can be shared in both case. If
__BOOT_COMPRESSEDis set, we can defineinat_init_tables()as a initialize function, and if not, it will be just a dummy "do {} while (0)".
The inat_*_tables are all declared const, so this way it is not possible to change them at runtime. For the running kernel image this is fine, as there are ELF relocations which fix things up, but at the pre-decompression boot stage there are no ELF relocations which can fix the tables, so the pointers in there need to be initialized at runtime.
Quoted 1 line
BTW, where is the
__BOOT_COMPRESSEDdefined?
It is defined in arch/x86/boot/compressed/sev-es.c by patch
x86/boot/compressed/64: Setup GHCB Based VC Exception handler
which also includes parts of the instruction decoder into the pre-decompression boot code and adds the only call-site for inat_init_tables().
Quoted 4 lines
Quoted 1 line
Review END { · line 397
Quoted 1 line
(quoted diff) +1
-0+ print "static void inat_init_tables(void)"This functions should be "inline".
And I can not see the call-site ofinat_init_tables()in this patch.
The call-site is added with the patch that includes the instruction decoder into the pre-decompression code. If possible I'd like to keep those things separate, as both patches are already pretty big by themselfes (and they do different things, in different parts of the code).
Quoted 2 lines
If possible, please include call-site with definition (especially new init function) so that I can check the init call timing too.
The function is called at the first #VC exception after a GHCB has been set up. Call-path is: boot_vc_handler -> sev_es_setup_ghcb -> inat_init_tables.
See
https://git.kernel.org/pub/scm/linux/kernel/git/joro/linux.git/tree/arch/x86/boot/compressed/sev-es.c?h=sev-es-client-v5.6-rc6
for the full code there.
Thanks,
Joerg
Re: [PATCH 05/70] x86/insn: Make inat-tables.c suitable for pre-decompression code
Quoted 25 lines · On Thu, 16 Apr 2020 17:24:06 +0200 Joerg Roedel <[email protected]> wrote:
Hi Masami,
Quoted 17 lines · Masami Hiramatsu
Quoted 13 lines · On Wed, 25 Mar 2020 16:39:45 +0100 Borislav Petkov <[email protected]> wrote:
+ Masami.
Quoted 10 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
The
inat-tables.cfile has some arrays in it that contain pointers to other arrays. These pointers need to be relocated when the kernel image is moved to a different location.The pre-decompression boot-code has no support for applying ELF relocations, so initialize these arrays at runtime in the pre-decompression code to make sure all pointers are correctly initialized.
Quoted 4 lines
I need to check the whole series, but as far as I can understand from this patch, this seems not allowing to store the address value in static pointers. It may break more things, for example
_kprobe_blacklistrecords theNOKPROBE_SYMBOL()symbol addresses at the build time.
Quoted 5 lines
The runtime-initialization function is only used in the pre-decompression boot code (
arch/x86/boot/compressed/) which is not part of the running kernel image. At that stage of booting there is no support for kprobe or tracing or any other neat features that might break things here.
Ah, I got it. So you intended to port the instruction decoder to pre-decompression boot code, correct?
Quoted 18 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
arch/x86/tools/gen-insn-attr-x86.awk +2
-0+ print "#ifndef __BOOT_COMPRESSED\n" + # print escape opcode map's array print "/* Escape opcode map array */" print "const insn_attr_t * const inat_escape_tables[INAT_ESC_MAX + 1]" \@@ -388,6 +391,51 @@ END { for (j = 0; j < max_lprefix; j++) if (atable[i,j]) print " ["i"]["j"] = "atable[i,j]","Quoted 8 lines
Quoted 4 lines
arch/x86/tools/gen-insn-attr-x86.awk END { · line 396
Quoted 4 lines
arch/x86/tools/gen-insn-attr-x86.awk +3
-1- print "};" + print "};\n" + + print "#else /* !__BOOT_COMPRESSED */\n"I think the definitions of inat_*_tables can be shared in both case. If
__BOOT_COMPRESSEDis set, we can defineinat_init_tables()as a initialize function, and if not, it will be just a dummy "do {} while (0)".
Quoted 2 lines
The inat_*_tables are all declared const, so this way it is not possible to change them at runtime.
Indeed.
Quoted 4 lines
For the running kernel image this is fine, as
there are ELF relocations which fix things up, but at the pre-decompression boot stage there are no ELF relocations which can fix the tables, so the pointers in there need to be initialized at runtime.
OK.
Quoted 10 lines
Quoted 2 lines
BTW, where is the
__BOOT_COMPRESSEDdefined?It is defined in
arch/x86/boot/compressed/sev-es.cby patchx86/boot/compressed/64: Setup GHCB Based VC Exception handler
which also includes parts of the instruction decoder into the pre-decompression boot code and adds the only call-site for
inat_init_tables().
Thanks, I understand it.
Quoted 11 lines
Quoted 5 lines
Quoted 2 lines
Review END { · line 397
Quoted 2 lines
(quoted diff) +1
-0+ print "static void inat_init_tables(void)"This functions should be "inline".
And I can not see the call-site ofinat_init_tables()in this patch.The call-site is added with the patch that includes the instruction decoder into the pre-decompression code. If possible I'd like to keep those things separate, as both patches are already pretty big by themselfes (and they do different things, in different parts of the code).
OK, if you will send v2, please CC both to me.
Quoted 7 lines
Quoted 3 lines
If possible, please include call-site with definition (especially new init function) so that I can check the init call timing too.
The function is called at the first #VC exception after a GHCB has been set up. Call-path is:
boot_vc_handler->sev_es_setup_ghcb->inat_init_tables.
sound good to me.
Thank you,
Quoted 10 lines
See
https://git.kernel.org/pub/scm/linux/kernel/git/joro/linux.git/tree/arch/x86/boot/compressed/sev-es.c?h=sev-es-client-v5.6-rc6for the full code there.
Thanks,
Joerg
Signature · 1 line
Masami Hiramatsu <[email protected]>
Re: [PATCH 05/70] x86/insn: Make inat-tables.c suitable for pre-decompression code
Quoted 5 lines · Masami Hiramatsu
On Thu, 16 Apr 2020 17:24:06 +0200
Joerg Roedel <[email protected]> wrote:Ah, I got it. So you intended to port the instruction decoder to pre-decompression boot code, correct?
Right, it is needed there to decode instructions which cause #VC exceptions when running as an SEV-ES guest.
Quoted 7 lines
Quoted 5 lines
The call-site is added with the patch that includes the instruction decoder into the pre-decompression code. If possible I'd like to keep those things separate, as both patches are already pretty big by themselfes (and they do different things, in different parts of the code).
OK, if you will send v2, please CC both to me.
Will do, I added you to the cc-list for future posts of this series.
Regards,
Joerg
PATCH v2 6/70 x86/umip: Factor out instruction fetch
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Factor out the code to fetch the instruction from user-space to a helper function.
--- arch/x86/include/asm/insn-eval.h | 2 ++ arch/x86/kernel/umip.c | 26 +++++----------------- arch/x86/lib/insn-eval.c | 38 ++++++++++++++++++++++++++++++++ 3 files changed, 46 insertions(+), 20 deletions(-)
arch/x86/include/asm/insn-eval.h +2 -0
@@ -19,5 +19,7 @@ void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs);int insn_get_modrm_rm_off (struct insn *insn,struct pt_regs *regs);unsigned long insn_get_seg_base (struct pt_regs *regs,int seg_reg_idx);int insn_get_code_seg_params (struct pt_regs *regs); +int insn_fetch_from_user (struct pt_regs *regs, +unsigned char buf[MAX_INSN_SIZE ]);#endif /* _ASM_X86_INSN_EVAL_H */
arch/x86/kernel/umip.c +6 -20
@@ -317,11 +317,11 @@ static void force_sig_info_umip_fault(void __user *addr, struct pt_regs *regs)*/ bool fixup_umip_exception (struct pt_regs *regs) { -int not_copied, nr_copied, reg_offset, dummy_data_size, umip_inst; -unsigned long seg_base =0 , *reg_addr; +int nr_copied, reg_offset, dummy_data_size, umip_inst;/* 10 bytes is the maximum size of the result of UMIP instructions */ unsigned char dummy_data[10 ] = {0 };unsigned char buf[MAX_INSN_SIZE ]; +unsigned long *reg_addr;void __user *uaddr;struct insn insn;int seg_defs;
@@ -329,26 +329,12 @@ bool fixup_umip_exception(struct pt_regs *regs)if (!regs)return false; -/* -* If not in user-space long mode, a custom code segment could be in -* use. This is true in protected mode (if the process defined a local -* descriptor table), or virtual-8086 mode. In most of the cases -* seg_base will be zero as in USER_CS. -*/ -if (!user_64bit_mode (regs)) - seg_base =insn_get_seg_base (regs,INAT_SEG_REG_CS ); - -if (seg_base ==-1L ) -return false; - - not_copied =copy_from_user (buf, (void __user *)(seg_base + regs->ip), -sizeof (buf)); - nr_copied =sizeof (buf) - not_copied; + nr_copied =insn_fetch_from_user (regs, buf);/* -* The copy_from_user above could have failed if user code is protected -* by a memory protection key. Give up on emulation in such a case. -* Should we issue a page fault? +* The insn_fetch_from_user above could have failed if user code +* is protected by a memory protection key. Give up on emulation +* in such a case. Should we issue a page fault? */ if (!nr_copied)return false;
arch/x86/lib/insn-eval.c +38 -0
@@ -1369,3 +1369,41 @@ void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs)return (void __user *)-1L ; } } + +/** +* insn_fetch_from_user() - Copy instruction bytes from user-space memory +* @regs: Structure with register values as seen when entering kernel mode +* @buf: Array to store the fetched instruction +* +* Gets the linear address of the instruction and copies the instruction bytes +* to the buf. +* +* Returns: +* +* Number of instruction bytes copied. +* +* 0 if nothing was copied. +*/ +int insn_fetch_from_user (struct pt_regs *regs, +unsigned char buf[MAX_INSN_SIZE ]) +{ +unsigned long seg_base =0 ; +int not_copied; + +/* +* If not in user-space long mode, a custom code segment could be in +* use. This is true in protected mode (if the process defined a local +* descriptor table), or virtual-8086 mode. In most of the cases +* seg_base will be zero as in USER_CS. +*/ +if (!user_64bit_mode (regs)) + seg_base =insn_get_seg_base (regs,INAT_SEG_REG_CS ); + +if (seg_base ==-1L ) +return 0 ; + + not_copied =copy_from_user (buf, (void __user *)(seg_base + regs->ip), +MAX_INSN_SIZE ); + +return MAX_INSN_SIZE - not_copied; +}
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 06/70] x86/umip: Factor out instruction fetch
Quoted 4 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Factor out the code to fetch the instruction from user-space to a helper function.
Add "No functional changes." here.
Quoted 6 lines
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/include/asm/insn-eval.h | 2 ++ arch/x86/kernel/umip.c | 26 +++++----------------- arch/x86/lib/insn-eval.c | 38 ++++++++++++++++++++++++++++++++ 3 files changed, 46 insertions(+), 20 deletions(-)
...
arch/x86/lib/insn-eval.c void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs) · line 1388
Quoted 2 lines
arch/x86/lib/insn-eval.c +2
-0+int insn_fetch_from_user (struct pt_regs *regs, +unsigned char buf[MAX_INSN_SIZE ])
No need for that linebreak - fits in 80 cols.
Quoted 5 lines
arch/x86/lib/insn-eval.c +5
-0+{ +unsigned long seg_base =0 ; +int not_copied; + +/*
arch/x86/lib/insn-eval.c void __user *insn_get_addr_ref(struct insn *insn, struct pt_regs *regs) · line 1403
Quoted 10 lines
arch/x86/lib/insn-eval.c +10
-0+* If not in user-space long mode, a custom code segment could be in +* use. This is true in protected mode (if the process defined a local +* descriptor table), or virtual-8086 mode. In most of the cases +* seg_base will be zero as in USER_CS. +*/ +if (!user_64bit_mode (regs)) + seg_base =insn_get_seg_base (regs,INAT_SEG_REG_CS ); + +if (seg_base ==-1L ) +return 0 ;
This reads strange: seg_base is changed only inside that if test so I guess we could test it there too:
if (!user_64bit_mode(regs)) {
seg_base = insn_get_seg_base(regs, INAT_SEG_REG_CS);
if (seg_base == -1L)
return 0;
}which is a small enough change to not require a separate patch.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 7/70 x86/umip: Factor out instruction decoding
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Factor out the code used to decode an instruction with the correct address and operand sizes to a helper function.
--- arch/x86/include/asm/insn-eval.h | 2 ++ arch/x86/kernel/umip.c | 23 +--------------- arch/x86/lib/insn-eval.c | 45 ++++++++++++++++++++++++++++++++ 3 files changed, 48 insertions(+), 22 deletions(-)
arch/x86/include/asm/insn-eval.h +2 -0
@@ -21,5 +21,7 @@ unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx);int insn_get_code_seg_params (struct pt_regs *regs);int insn_fetch_from_user (struct pt_regs *regs,unsigned char buf[MAX_INSN_SIZE ]); +bool insn_decode (struct pt_regs *regs,struct insn *insn, +unsigned char buf[MAX_INSN_SIZE ],int buf_size);#endif /* _ASM_X86_INSN_EVAL_H */
arch/x86/kernel/umip.c +1 -22
@@ -324,7 +324,6 @@ bool fixup_umip_exception(struct pt_regs *regs)unsigned long *reg_addr;void __user *uaddr;struct insn insn; -int seg_defs;if (!regs)return false;
@@ -339,27 +338,7 @@ bool fixup_umip_exception(struct pt_regs *regs)if (!nr_copied)return false; -insn_init (&insn, buf, nr_copied,user_64bit_mode (regs)); - -/* -* Override the default operand and address sizes with what is specified -* in the code segment descriptor. The instruction decoder only sets -* the address size it to either 4 or 8 address bytes and does nothing -* for the operand bytes. This OK for most of the cases, but we could -* have special cases where, for instance, a 16-bit code segment -* descriptor is used. -* If there is an address override prefix, the instruction decoder -* correctly updates these values, even for 16-bit defaults. -*/ - seg_defs =insn_get_code_seg_params (regs); -if (seg_defs == -EINVAL ) -return false; - - insn.addr_bytes =INSN_CODE_SEG_ADDR_SZ (seg_defs); - insn.opnd_bytes =INSN_CODE_SEG_OPND_SZ (seg_defs); - -insn_get_length (&insn); -if (nr_copied < insn.length) +if (!insn_decode (regs, &insn, buf, nr_copied))return false; umip_inst =identify_insn (&insn);
arch/x86/lib/insn-eval.c +45 -0
48 lines not on this page · raw message
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 07/70] x86/umip: Factor out instruction decoding
Quoted 4 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Factor out the code used to decode an instruction with the correct address and operand sizes to a helper function.
As for the previous one: "No functional changes."
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 8/70 x86/insn: Add insn_get_modrm_reg_off()
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add a function to the instruction decoder which returns the pt_regs offset of the register specified in the reg field of the modrm byte.
--- arch/x86/include/asm/insn-eval.h | 1 + arch/x86/lib/insn-eval.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
arch/x86/include/asm/insn-eval.h +1 -0
@@ -17,6 +17,7 @@void __user *insn_get_addr_ref (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_rm_off (struct insn *insn,struct pt_regs *regs); +int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs);unsigned long insn_get_seg_base (struct pt_regs *regs,int seg_reg_idx);int insn_get_code_seg_params (struct pt_regs *regs); int insn_fetch_from_user(struct pt_regs *regs,
arch/x86/lib/insn-eval.c +23 -0
@@ -20,6 +20,7 @@enum reg_type {REG_TYPE_RM =0 , +REG_TYPE_REG ,REG_TYPE_INDEX ,REG_TYPE_BASE , };
@@ -441,6 +442,13 @@ static int get_reg_offset(struct insn *insn, struct pt_regs *regs, regno +=8 ;break ; +case REG_TYPE_REG : + regno =X86_MODRM_REG (insn->modrm.value); + +if (X86_REX_R (insn->rex_prefix.value)) + regno +=8 ; +break ; +case REG_TYPE_INDEX : regno =X86_SIB_INDEX (insn->sib.value);if (X86_REX_X (insn->rex_prefix.value))
@@ -809,6 +817,21 @@ int insn_get_modrm_rm_off(struct insn *insn, struct pt_regs *regs)return get_reg_offset (insn, regs,REG_TYPE_RM ); } +/** +* insn_get_modrm_reg_off() - Obtain register in reg part of the ModRM byte +* @insn: Instruction containing the ModRM byte +* @regs: Register values as seen when entering kernel mode +* +* Returns: +* +* The register indicated by the reg part of the ModRM byte. The +* register is obtained as an offset from the base of pt_regs. +*/ +int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs) +{ +return get_reg_offset (insn, regs,REG_TYPE_REG ); +} +/** * get_seg_base_limit() - obtain base address and limit of a segment * @insn: Instruction. Must be valid.
Signature · 1 line
2.17.1
Re: [PATCH 08/70] x86/insn: Add insn_get_modrm_reg_off()
Quoted 5 lines · On Thu, 19 Mar 2020 10:13:05 +0100 Joerg Roedel <[email protected]> wrote:
From: Joerg Roedel <[email protected]>
Add a function to the instruction decoder which returns the
pt_regsoffset of the register specified in the reg field of the modrm byte.
This looks good to me.
Thank you,
Quoted 69 lines
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/include/asm/insn-eval.h | 1 + arch/x86/lib/insn-eval.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)arch/x86/include/asm/insn-eval.h +1
-0@@ -17,6 +17,7 @@void __user *insn_get_addr_ref (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_rm_off (struct insn *insn,struct pt_regs *regs); +int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs);unsigned long insn_get_seg_base (struct pt_regs *regs,int seg_reg_idx);int insn_get_code_seg_params (struct pt_regs *regs); int insn_fetch_from_user(struct pt_regs *regs,arch/x86/lib/insn-eval.c +23
-0@@ -20,6 +20,7 @@enum reg_type {REG_TYPE_RM =0 , +REG_TYPE_REG ,REG_TYPE_INDEX ,REG_TYPE_BASE , };@@ -441,6 +442,13 @@ static int get_reg_offset(struct insn *insn, struct pt_regs *regs, regno +=8 ;break ; +case REG_TYPE_REG : + regno =X86_MODRM_REG (insn->modrm.value); + +if (X86_REX_R (insn->rex_prefix.value)) + regno +=8 ; +break ; +case REG_TYPE_INDEX : regno =X86_SIB_INDEX (insn->sib.value);if (X86_REX_X (insn->rex_prefix.value))@@ -809,6 +817,21 @@ int insn_get_modrm_rm_off(struct insn *insn, struct pt_regs *regs)return get_reg_offset (insn, regs,REG_TYPE_RM ); } +/** +* insn_get_modrm_reg_off() - Obtain register in reg part of the ModRM byte +* @insn: Instruction containing the ModRM byte +* @regs: Register values as seen when entering kernel mode +* +* Returns: +* +* The register indicated by the reg part of the ModRM byte. The +* register is obtained as an offset from the base of pt_regs. +*/ +int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs) +{ +return get_reg_offset (insn, regs,REG_TYPE_REG ); +} +/** * get_seg_base_limit() - obtain base address and limit of a segment * @insn: Instruction. Must be valid. Signature · 1 line
2.17.1
Signature · 1 line
Masami Hiramatsu <[email protected]>
PATCH v2 9/70 x86/insn: Add insn_rep_prefix() helper
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add a function to check whether an instruction has a REP prefix.
--- arch/x86/include/asm/insn-eval.h | 1 + arch/x86/lib/insn-eval.c | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+)
arch/x86/include/asm/insn-eval.h +1 -0
@@ -15,6 +15,7 @@#define INSN_CODE_SEG_OPND_SZ (params) (params & 0xf)#define INSN_CODE_SEG_PARAMS (oper_sz, addr_sz) (oper_sz | (addr_sz << 4)) +bool insn_rep_prefix (struct insn *insn);void __user *insn_get_addr_ref (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_rm_off (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs);
arch/x86/lib/insn-eval.c +24 -0
@@ -53,6 +53,30 @@ static bool is_string_insn(struct insn *insn) } } +/** +* insn_rep_prefix() - Determine if instruction has a REP prefix +* @insn: Instruction containing the prefix to inspect +* +* Returns: +* +* true if the instruction has a REP prefix, false if not. +*/ +bool insn_rep_prefix (struct insn *insn) +{ +int i; + +insn_get_prefixes (insn); + +for (i =0 ; i < insn->prefixes.nbytes; i++) { +insn_byte_t p = insn->prefixes.bytes[i]; + +if (p ==0xf2 || p ==0xf3 ) +return true; + } + +return false; +} +/** * get_seg_reg_override_idx() - obtain segment register override index * @insn: Valid instruction with segment override prefixes
Signature · 1 line
2.17.1
Re: [PATCH 09/70] x86/insn: Add insn_rep_prefix() helper
Quoted 15 lines · On Thu, 19 Mar 2020 10:13:06 +0100 Joerg Roedel <[email protected]> wrote:
From: Joerg Roedel <[email protected]>
Add a function to check whether an instruction has a REP prefix.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/include/asm/insn-eval.h | 1 + arch/x86/lib/insn-eval.c | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+)arch/x86/include/asm/insn-eval.h +0
-0
arch/x86/include/asm/insn-eval.h line 18
Quoted 4 lines
arch/x86/include/asm/insn-eval.h +1
-0@@ -15,6 +15,7 @@#define INSN_CODE_SEG_OPND_SZ (params) (params & 0xf)#define INSN_CODE_SEG_PARAMS (oper_sz, addr_sz) (oper_sz | (addr_sz << 4)) +bool insn_rep_prefix (struct insn *insn);
Can you make it "insn_has_rep_prefix()"?
Thank you,
Quoted 41 lines
arch/x86/include/asm/insn-eval.h +0
-0void __user *insn_get_addr_ref (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_rm_off (struct insn *insn,struct pt_regs *regs);int insn_get_modrm_reg_off (struct insn *insn,struct pt_regs *regs);arch/x86/lib/insn-eval.c +24
-0@@ -53,6 +53,30 @@ static bool is_string_insn(struct insn *insn) } } +/** +* insn_rep_prefix() - Determine if instruction has a REP prefix +* @insn: Instruction containing the prefix to inspect +* +* Returns: +* +* true if the instruction has a REP prefix, false if not. +*/ +bool insn_rep_prefix (struct insn *insn) +{ +int i; + +insn_get_prefixes (insn); + +for (i =0 ; i < insn->prefixes.nbytes; i++) { +insn_byte_t p = insn->prefixes.bytes[i]; + +if (p ==0xf2 || p ==0xf3 ) +return true; + } + +return false; +} +/** * get_seg_reg_override_idx() - obtain segment register override index * @insn: Valid instruction with segment override prefixes Signature · 1 line
2.17.1
Signature · 1 line
Masami Hiramatsu <[email protected]>
PATCH v2 10/70 x86/boot/compressed: Fix debug_puthex() parameter type
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
In the CONFIG_X86_VERBOSE_BOOTUP=Y case the debug_puthex() macro just turns into __puthex, which takes 'unsigned long' as parameter. But in the CONFIG_X86_VERBOSE_BOOTUP=N case it is a function which takes 'unsigned char *', causing compile warnings when the function is used. Fix the parameter type to get rid of the warnings.
--- arch/x86/boot/compressed/misc.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/boot/compressed/misc.h +1 -1
@@ -59,7 +59,7 @@ void __puthex(unsigned long value);static inline void debug_putstr (const char *s) { } -static inline void debug_puthex (const char *s) +static inline void debug_puthex (unsigned long value) { }#define debug_putaddr (x)/* */
Signature · 1 line
2.17.1
[tip: x86/boot] x86/boot/compressed: Fix debug_puthex() parameter type
The following commit has been merged into the x86/boot branch of tip:
The following commit has been merged into the x86/boot branch of tip:
Commit-ID: c90beea22a2bece4b0bbb39789bf835504421594 Gitweb: https://git.kernel.org/tip/c90beea22a2bece4b0bbb39789bf835504421594 Author: Joerg Roedel <[email protected]> AuthorDate: Thu, 19 Mar 2020 10:13:07 +01:00 Committer: Borislav Petkov <[email protected]> CommitterDate: Sat, 28 Mar 2020 12:14:26 +01:00
x86/boot/compressed: Fix debug_puthex() parameter type
In the CONFIG_X86_VERBOSE_BOOTUP=Y case, the debug_puthex() macro just turns into __puthex(), which takes 'unsigned long' as parameter.
But in the CONFIG_X86_VERBOSE_BOOTUP=N case, it is a function which takes 'unsigned char *', causing compile warnings when the function is used. Fix the parameter type to get rid of the warnings.
--- arch/x86/boot/compressed/misc.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/boot/compressed/misc.h +1 -1
@@ -59,7 +59,7 @@ void __puthex(unsigned long value);static inline void debug_putstr (const char *s) { } -static inline void debug_puthex (const char *s) +static inline void debug_puthex (unsigned long value) { }#define debug_putaddr (x)/* */
PATCH v2 11/70 x86/boot/compressed/64: Disable red-zone usage
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The x86-64 ABI defines a red-zone on the stack:
The 128-byte area beyond the location pointed to by %rsp is considered to be reserved and shall not be modified by signal or interrupt handlers. 10 Therefore, functions may use this area for temporary data that is not needed across function calls. In particular, leaf functions may use this area for their entire stack frame, rather than adjusting the stack pointer in the prologue and epilogue. This area is known as the red zone.
This is not compatible with exception handling, so disable it for the pre-decompression boot code.
--- arch/x86/boot/Makefile | 2 +- arch/x86/boot/compressed/Makefile | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
arch/x86/boot/Makefile +1 -1
@@ -65,7 +65,7 @@ clean-files += cpustr.h # --------------------------------------------------------------------------- -KBUILD_CFLAGS := $(REALMODE_CFLAGS) -D_SETUP +KBUILD_CFLAGS := $(REALMODE_CFLAGS) -D_SETUP -mno-red-zone KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) GCOV_PROFILE := n
arch/x86/boot/compressed/Makefile +2 -2
@@ -30,7 +30,7 @@ KBUILD_CFLAGS := -m$(BITS) -O2 KBUILD_CFLAGS += -fno-strict-aliasing $(call cc-option, -fPIE, -fPIC) KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING cflags-$(CONFIG_X86_32) := -march=i386 -cflags-$(CONFIG_X86_64) := -mcmodel=small +cflags-$(CONFIG_X86_64) := -mcmodel=small -mno-red-zone KBUILD_CFLAGS += $(cflags-y) KBUILD_CFLAGS += -mno-mmx -mno-sse KBUILD_CFLAGS += $(call cc-option,-ffreestanding)
@@ -87,7 +87,7 @@ endif vmlinux-objs-$(CONFIG_ACPI) += $(obj)/acpi.o -$(obj)/eboot.o: KBUILD_CFLAGS += -fshort-wchar -mno-red-zone +$(obj)/eboot.o: KBUILD_CFLAGS += -fshort-wchar vmlinux-objs-$(CONFIG_EFI_STUB) += $(obj)/eboot.o \ $(objtree)/drivers/firmware/efi/libstub/lib.a
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 11/70] x86/boot/compressed/64: Disable red-zone usage
Quoted 7 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
The x86-64 ABI defines a red-zone on the stack:
The 128-byte area beyond the location pointed to by %rsp is considered to be reserved and shall not be modified by signal or interrupt handlers. 10 Therefore, functions may use this area for
^^
That 10 is the footnote number from the pdf. :)
Quoted 6 lines
temporary data that is not needed across function calls. In particular, leaf functions may use this area for their entire stack frame, rather than adjusting the stack pointer in the prologue and epilogue. This area is known as the red zone.
This is not compatible with exception handling, so disable it for the
I could use some blurb as to what the problem is, for future reference.
Quoted 35 lines
pre-decompression boot code.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/boot/Makefile | 2 +- arch/x86/boot/compressed/Makefile | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)arch/x86/boot/Makefile +1
-1@@ -65,7 +65,7 @@ clean-files += cpustr.h # --------------------------------------------------------------------------- -KBUILD_CFLAGS := $(REALMODE_CFLAGS) -D_SETUP +KBUILD_CFLAGS := $(REALMODE_CFLAGS) -D_SETUP -mno-red-zone KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) GCOV_PROFILE := narch/x86/boot/compressed/Makefile +1
-1@@ -30,7 +30,7 @@ KBUILD_CFLAGS := -m$(BITS) -O2 KBUILD_CFLAGS += -fno-strict-aliasing $(call cc-option, -fPIE, -fPIC) KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING cflags-$(CONFIG_X86_32) := -march=i386 -cflags-$(CONFIG_X86_64) := -mcmodel=small +cflags-$(CONFIG_X86_64) := -mcmodel=small -mno-red-zone KBUILD_CFLAGS += $(cflags-y) KBUILD_CFLAGS += -mno-mmx -mno-sse KBUILD_CFLAGS += $(call cc-option,-ffreestanding)
arch/x86/boot/compressed/Makefile endif · line 93
Quoted 8 lines
arch/x86/boot/compressed/Makefile +1
-1@@ -87,7 +87,7 @@ endif vmlinux-objs-$(CONFIG_ACPI) += $(obj)/acpi.o -$(obj)/eboot.o: KBUILD_CFLAGS += -fshort-wchar -mno-red-zone +$(obj)/eboot.o: KBUILD_CFLAGS += -fshort-wchar vmlinux-objs-$(CONFIG_EFI_STUB) += $(obj)/eboot.o \ $(objtree)/drivers/firmware/efi/libstub/lib.a
That last chunk is not needed anymore after
c2d0b470154c ("efi/libstub/x86: Incorporate eboot.c into libstub")
AFAICT.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 12/70 x86/boot/compressed/64: Add IDT Infrastructure
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add code needed to setup an IDT in the early pre-decompression boot-code. The IDT is loaded first in startup_64, which is after EfiExitBootServices() has been called, and later reloaded when the kernel image has been relocated to the end of the decompression area.
This allows to setup different IDT handlers before and after the relocation.
--- arch/x86/boot/compressed/Makefile | 1 + arch/x86/boot/compressed/head_64.S | 34 ++++++++++ arch/x86/boot/compressed/idt_64.c | 43 +++++++++++++ arch/x86/boot/compressed/idt_handlers_64.S | 75 ++++++++++++++++++++++ arch/x86/boot/compressed/misc.h | 5 ++ arch/x86/include/asm/desc_defs.h | 3 + 6 files changed, 161 insertions(+) create mode 100644 arch/x86/boot/compressed/idt_64.c create mode 100644 arch/x86/boot/compressed/idt_handlers_64.S
arch/x86/boot/compressed/Makefile +1 -0
@@ -81,6 +81,7 @@ vmlinux-objs-$(CONFIG_EARLY_PRINTK) += $(obj)/early_serial_console.o vmlinux-objs-$(CONFIG_RANDOMIZE_BASE) += $(obj)/kaslr.o ifdef CONFIG_X86_64 vmlinux-objs-$(CONFIG_RANDOMIZE_BASE) += $(obj)/kaslr_64.o + vmlinux-objs-y += $(obj)/idt_64.o $(obj)/idt_handlers_64.o vmlinux-objs-y += $(obj)/mem_encrypt.o vmlinux-objs-y += $(obj)/pgtable_64.o endif
arch/x86/boot/compressed/head_64.S +34 -0
@@ -33,6 +33,7 @@ #include <asm/processor-flags.h> #include <asm/asm-offsets.h> #include <asm/bootparam.h> +#include <asm/desc_defs.h> #include "pgtable.h" /*
@@ -358,6 +359,10 @@ SYM_CODE_START(startup_64) movq %rax, gdt64+2(%rip) lgdt gdt64(%rip) + pushq %rsi + call load_stage1_idt + popq %rsi + /* * paging_prepare() sets up the trampoline and checks if we need to * enable 5-level paging.
@@ -465,6 +470,16 @@ SYM_FUNC_END_ALIAS(efi_stub_entry) .text SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated) +/* + * Reload GDT after relocation - The GDT at the non-relocated position + * might be overwritten soon by the in-place decompression, so reload + * GDT at the relocated address. The GDT is referenced by exception + * handling and needs to be set up correctly. + */ + leaq gdt(%rip), %rax + movq %rax, gdt64+2(%rip) + lgdt gdt64(%rip) + /* * Clear BSS (stack is currently empty) */
@@ -475,6 +490,13 @@ SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated) shrq $3, %rcx rep stosq +/* + * Load stage2 IDT + */ + pushq %rsi + call load_stage2_idt + popq %rsi + /* * Do the extraction, and jump to the new kernel.. */
@@ -628,6 +650,18 @@ SYM_DATA_START_LOCAL(gdt) .quad 0x0000000000000000 /* TS continued */ SYM_DATA_END_LABEL(gdt, SYM_L_LOCAL, gdt_end) +SYM_DATA_START(boot_idt_desc) + .word boot_idt_end - boot_idt + .quad 0 +SYM_DATA_END(boot_idt_desc) + .balign 8 +SYM_DATA_START(boot_idt) + .rept BOOT_IDT_ENTRIES + .quad 0 + .quad 0 + .endr +SYM_DATA_END_LABEL(boot_idt, SYM_L_GLOBAL, boot_idt_end) + #ifdef CONFIG_EFI_MIXED SYM_DATA_LOCAL(efi32_boot_args, .long 0, 0) SYM_DATA(efi_is64, .byte 1)
arch/x86/boot/compressed/idt_64.c new file +43 -0
43 lines not on this page · raw message
arch/x86/boot/compressed/idt_handlers_64.S new file +75 -0
75 lines not on this page · raw message
arch/x86/boot/compressed/misc.h +5 -0
@@ -23,6 +23,7 @@#include <asm/page.h> #include <asm/boot.h> #include <asm/bootparam.h> +#include <asm/desc_defs.h> #define BOOT_CTYPE_H #include <linux/acpi.h>
@@ -133,4 +134,8 @@ int count_immovable_mem_regions(void);static inline int count_immovable_mem_regions (void ) {return 0 ; }#endif +/* idt_64.c */ +extern gate_desc boot_idt[BOOT_IDT_ENTRIES ]; +extern struct desc_ptr boot_idt_desc; +#endif /* BOOT_COMPRESSED_MISC_H */
arch/x86/include/asm/desc_defs.h +3 -0
@@ -109,6 +109,9 @@ struct desc_ptr {#endif /* !__ASSEMBLY__ */ +/* Boot IDT definitions */ +#define BOOT_IDT_ENTRIES 32 +/* Access rights as returned by LAR */ #define AR_TYPE_RODATA (0 * (1 << 9))#define AR_TYPE_RWDATA (1 * (1 << 9))
Signature · 1 line
2.17.1
Re: [PATCH 12/70] x86/boot/compressed/64: Add IDT Infrastructure
Quoted 23 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Add code needed to setup an IDT in the early pre-decompression boot-code. The IDT is loaded first in
startup_64, which is afterEfiExitBootServices()has been called, and later reloaded when the kernel image has been relocated to the end of the decompression area.This allows to setup different IDT handlers before and after the relocation.
Signed-off-by Joerg Roedel <[email protected]>---
arch/x86/boot/compressed/head_64.S +3
-0@@ -465,6 +470,16 @@ SYM_FUNC_END_ALIAS(efi_stub_entry) .text SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated) +/* + * Reload GDT after relocation - The GDT at the non-relocated position + * might be overwritten soon by the in-place decompression, so reload
arch/x86/boot/compressed/head_64.S SYM_FUNC_END_ALIAS(efi_stub_entry) · line 485
Quoted 10 lines
arch/x86/boot/compressed/head_64.S +7
-0+ * GDT at the relocated address. The GDT is referenced by exception + * handling and needs to be set up correctly. + */ + leaq gdt(%rip), %rax + movq %rax, gdt64+2(%rip) + lgdt gdt64(%rip) + /* * Clear BSS (stack is currently empty) */
Note that this is now done in mainline as of commit c98a76eabbb6e, just prior to jumping to .Lrelocated, so this can be dropped on the next rebase.
Thanks.
Re: [PATCH 12/70] x86/boot/compressed/64: Add IDT Infrastructure
Hi Arvind,
Quoted 11 lines · Arvind Sankar
Quoted 10 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
arch/x86/boot/compressed/head_64.S +9
-0+/* + * Reload GDT after relocation - The GDT at the non-relocated position + * might be overwritten soon by the in-place decompression, so reload + * GDT at the relocated address. The GDT is referenced by exception + * handling and needs to be set up correctly. + */ + leaq gdt(%rip), %rax + movq %rax, gdt64+2(%rip) + lgdt gdt64(%rip)
Quoted 8 lines
Quoted 4 lines
+
/*
* Clear BSS (stack is currently empty)
*/Note that this is now done in mainline as of commit
c98a76eabbb6e, just prior to jumping to .Lrelocated, so this can be dropped on the next rebase.
Thanks for the heads-up, I removed this hunk.
Regards,
Joerg
PATCH v2 13/70 x86/boot/compressed/64: Rename kaslr_64.c to ident_map_64.c
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The file contains only code related to identity mapped page-tables. Rename the file and compile it always in.
---
arch/x86/boot/compressed/Makefile | 2 +-
.../boot/compressed/{kaslr_64.c => ident_map_64.c} | 12 ++++++++++++
arch/x86/boot/compressed/kaslr.c | 9 ---------
arch/x86/boot/compressed/misc.h | 8 ++++++++
4 files changed, 21 insertions(+), 10 deletions(-)
rename arch/x86/boot/compressed/{kaslr_64.c => ident_map_64.c} (93%)arch/x86/boot/compressed/Makefile +1 -1
@@ -80,7 +80,7 @@ vmlinux-objs-y := $(obj)/vmlinux.lds $(obj)/kernel_info.o $(obj)/head_$(BITS).o vmlinux-objs-$(CONFIG_EARLY_PRINTK) += $(obj)/early_serial_console.o vmlinux-objs-$(CONFIG_RANDOMIZE_BASE) += $(obj)/kaslr.o ifdef CONFIG_X86_64 - vmlinux-objs-$(CONFIG_RANDOMIZE_BASE) += $(obj)/kaslr_64.o + vmlinux-objs-y += $(obj)/ident_map_64.o vmlinux-objs-y += $(obj)/idt_64.o $(obj)/idt_handlers_64.o vmlinux-objs-y += $(obj)/mem_encrypt.o vmlinux-objs-y += $(obj)/pgtable_64.o
arch/x86/boot/compressed/ident_map_64.c renamed from arch/x86/boot/compressed/kaslr_64.c +12 -0
@@ -29,6 +29,18 @@#define __PAGE_OFFSET __PAGE_OFFSET_BASE#include "../../mm/ident_map.c" +#ifdef CONFIG_X86_5LEVEL +unsigned int __pgtable_l5_enabled; +unsigned int pgdir_shift =39 ; +unsigned int ptrs_per_p4d =1 ; +#endif + +/* Used by PAGE_KERN* macros: */ +pteval_t __default_kernel_pte_mask__read_mostly = ~0 ; + +/* Used by pgtable.h asm code to force instruction serialization. */ +unsigned long __force_order; +/* Used to track our page table allocation area. */ struct alloc_pgt_data {unsigned char *pgt_buf;
arch/x86/boot/compressed/kaslr.c +0 -9
@@ -43,17 +43,8 @@#define STATIC #include <linux/decompress/mm.h> -#ifdef CONFIG_X86_5LEVEL -unsigned int __pgtable_l5_enabled; -unsigned int pgdir_shift__ro_after_init =39 ; -unsigned int ptrs_per_p4d__ro_after_init =1 ; -#endif -extern unsigned long get_cmd_line_ptr (void ); -/* Used by PAGE_KERN* macros: */ -pteval_t __default_kernel_pte_mask__read_mostly = ~0 ; -/* Simplified build-specific string for starting entropy. */ static const char build_str[] =UTS_RELEASE " (" LINUX_COMPILE_BY "@" LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION ;
arch/x86/boot/compressed/misc.h +8 -0
@@ -134,6 +134,14 @@ int count_immovable_mem_regions(void);static inline int count_immovable_mem_regions (void ) {return 0 ; }#endif +/* ident_map_64.c */ +#ifdef CONFIG_X86_5LEVEL +extern unsigned int __pgtable_l5_enabled, pgdir_shift, ptrs_per_p4d; +#endif + +/* Used by PAGE_KERN* macros: */ +extern pteval_t __default_kernel_pte_mask; +/* idt_64.c */ extern gate_desc boot_idt[BOOT_IDT_ENTRIES ];extern struct desc_ptr boot_idt_desc;
Signature · 1 line
2.17.1
PATCH v2 14/70 x86/boot/compressed/64: Add page-fault handler
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Install a page-fault handler to add an identity mapping to addresses not yet mapped. Also do some checking whether the error code is sane.
This makes non SEV-ES machines use the exception handling infrastructure in the pre-decompressions boot code too, making it less likely to break in the future.
--- arch/x86/boot/compressed/ident_map_64.c | 38 ++++++++++++++++++++++ arch/x86/boot/compressed/idt_64.c | 2 ++ arch/x86/boot/compressed/idt_handlers_64.S | 2 ++ arch/x86/boot/compressed/misc.h | 6 ++++ 4 files changed, 48 insertions(+)
arch/x86/boot/compressed/ident_map_64.c +38 -0
@@ -19,11 +19,13 @@/* No PAGE_TABLE_ISOLATION support needed either: */ #undef CONFIG_PAGE_TABLE_ISOLATION +#include "error.h" #include "misc.h" /* These actually do the work of building the kernel identity maps. */ #include <asm/init.h> #include <asm/pgtable.h> +#include <asm/trap_defs.h> /* Use the static base for this part of the boot process */ #undef __PAGE_OFFSET#define __PAGE_OFFSET __PAGE_OFFSET_BASE
@@ -163,3 +165,39 @@ void finalize_identity_maps(void) {write_cr3 (top_level_pgt); } + +static void pf_error (unsigned long error_code,unsigned long address, +struct pt_regs *regs) +{ +error_putstr ("Unexpected page-fault:" ); +error_putstr ("\nError Code: " ); +error_puthex (error_code); +error_putstr ("\nCR2: 0x" ); +error_puthex (address); +error_putstr ("\nRIP relative to _head: 0x" ); +error_puthex (regs->ip - (unsigned long )_head); +error_putstr ("\n" ); + +error ("Stopping.\n" ); +} + +void do_boot_page_fault (struct pt_regs *regs) +{ +unsigned long address =native_read_cr2 (); +unsigned long error_code = regs->orig_ax; + +/* +* Check for unexpected error codes. Unexpected are: +* - Faults on present pages +* - User faults +* - Reserved bits set +*/ +if (error_code & (X86_PF_PROT |X86_PF_USER |X86_PF_RSVD )) +pf_error (error_code, address, regs); + +/* +* Error code is sane - now identity map the 2M region around +* the faulting address. +*/ +add_identity_map (address &PMD_MASK ,PMD_SIZE ); +}
arch/x86/boot/compressed/idt_64.c +2 -0
@@ -39,5 +39,7 @@ void load_stage2_idt(void) { boot_idt_desc.address = (unsigned long )boot_idt; +set_idt_entry (X86_TRAP_PF , boot_pf_handler); +load_boot_idt (&boot_idt_desc); }
arch/x86/boot/compressed/idt_handlers_64.S +2 -0
@@ -73,3 +73,5 @@ SYM_FUNC_END(\name) .text .code64 + +EXCEPTION_HANDLER boot_pf_handler do_boot_page_fault error_code=1
arch/x86/boot/compressed/misc.h +6 -0
@@ -37,6 +37,9 @@#define memptr unsigned#endif +/* boot/compressed/vmlinux start and end markers */ +extern char _head[], _end[]; +/* misc.c */ extern memptr free_mem_ptr;extern memptr free_mem_end_ptr;
@@ -146,4 +149,7 @@ extern pteval_t __default_kernel_pte_mask;extern gate_desc boot_idt[BOOT_IDT_ENTRIES ];extern struct desc_ptr boot_idt_desc; +/* IDT Entry Points */ +void boot_pf_handler (void ); +#endif /* BOOT_COMPRESSED_MISC_H */
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 14/70] x86/boot/compressed/64: Add page-fault handler
Quoted 36 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Install a page-fault handler to add an identity mapping to addresses not yet mapped. Also do some checking whether the error code is sane.
This makes non SEV-ES machines use the exception handling infrastructure in the pre-decompressions boot code too, making it less likely to break in the future.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/boot/compressed/ident_map_64.c | 38 ++++++++++++++++++++++ arch/x86/boot/compressed/idt_64.c | 2 ++ arch/x86/boot/compressed/idt_handlers_64.S | 2 ++ arch/x86/boot/compressed/misc.h | 6 ++++ 4 files changed, 48 insertions(+)arch/x86/boot/compressed/ident_map_64.c +2
-0@@ -19,11 +19,13 @@/* No PAGE_TABLE_ISOLATION support needed either: */ #undef CONFIG_PAGE_TABLE_ISOLATION +#include "error.h" #include "misc.h" /* These actually do the work of building the kernel identity maps. */ #include <asm/init.h> #include <asm/pgtable.h> +#include <asm/trap_defs.h> /* Use the static base for this part of the boot process */ #undef __PAGE_OFFSET#define __PAGE_OFFSET __PAGE_OFFSET_BASE
arch/x86/boot/compressed/ident_map_64.c void finalize_identity_maps(void) · line 170
Quoted 6 lines
arch/x86/boot/compressed/ident_map_64.c +3
-0@@ -163,3 +165,39 @@ void finalize_identity_maps(void) {write_cr3 (top_level_pgt); } + +static void pf_error (unsigned long error_code,unsigned long address, +struct pt_regs *regs)
AFAICT, that function is called below only so just merge its body into the call site instead...
Quoted 50 lines
arch/x86/boot/compressed/ident_map_64.c +33
-0+{ +error_putstr ("Unexpected page-fault:" ); +error_putstr ("\nError Code: " ); +error_puthex (error_code); +error_putstr ("\nCR2: 0x" ); +error_puthex (address); +error_putstr ("\nRIP relative to _head: 0x" ); +error_puthex (regs->ip - (unsigned long )_head); +error_putstr ("\n" ); + +error ("Stopping.\n" ); +} + +void do_boot_page_fault (struct pt_regs *regs) +{ +unsigned long address =native_read_cr2 (); +unsigned long error_code = regs->orig_ax; + +/* +* Check for unexpected error codes. Unexpected are: +* - Faults on present pages +* - User faults +* - Reserved bits set +*/ +if (error_code & (X86_PF_PROT |X86_PF_USER |X86_PF_RSVD )) +pf_error (error_code, address, regs); + +/* +* Error code is sane - now identity map the 2M region around +* the faulting address. +*/ +add_identity_map (address &PMD_MASK ,PMD_SIZE ); +}arch/x86/boot/compressed/idt_64.c +2
-0@@ -39,5 +39,7 @@ void load_stage2_idt(void) { boot_idt_desc.address = (unsigned long )boot_idt; +set_idt_entry (X86_TRAP_PF , boot_pf_handler); +load_boot_idt (&boot_idt_desc); }arch/x86/boot/compressed/idt_handlers_64.S +0
-0
arch/x86/boot/compressed/idt_handlers_64.S SYM_FUNC_END(\name) · line 77
Quoted 5 lines
arch/x86/boot/compressed/idt_handlers_64.S +2
-0@@ -73,3 +73,5 @@ SYM_FUNC_END(\name) .text .code64 + +EXCEPTION_HANDLER boot_pf_handler do_boot_page_fault error_code=1
boot_page_fault do_boot_page_fault
equivalent to the PF handler proper naming pls. Grepping "page_fault" would give you all then.
Quoted 25 lines
arch/x86/boot/compressed/misc.h +6
-0@@ -37,6 +37,9 @@#define memptr unsigned#endif +/* boot/compressed/vmlinux start and end markers */ +extern char _head[], _end[]; +/* misc.c */ extern memptr free_mem_ptr;extern memptr free_mem_end_ptr;@@ -146,4 +149,7 @@ extern pteval_t __default_kernel_pte_mask;extern gate_desc boot_idt[BOOT_IDT_ENTRIES ];extern struct desc_ptr boot_idt_desc; +/* IDT Entry Points */ +void boot_pf_handler (void ); +#endif /* BOOT_COMPRESSED_MISC_H */ Signature · 1 line
2.17.1
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 15/70 x86/boot/compressed/64: Always switch to own page-table
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
When booted through startup_64 the kernel keeps running on the EFI page-table until the KASLR code sets up its own page-table. Without KASLR the pre-decompression boot code never switches off the EFI page-table. Change that by unconditionally switching to our own page-table once the kernel is relocated.
This makes sure we can make changes to the mapping when necessary, for example map pages unencrypted in SEV and SEV-ES guests.
Also remove the debug_putstr() calls in initialize_identity_maps() because the function now runs before console_init() is called.
--- arch/x86/boot/compressed/head_64.S | 3 +- arch/x86/boot/compressed/ident_map_64.c | 51 +++++++++++++++---------- arch/x86/boot/compressed/kaslr.c | 3 -- 3 files changed, 32 insertions(+), 25 deletions(-)
arch/x86/boot/compressed/head_64.S +2 -1
@@ -491,10 +491,11 @@ SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated) rep stosq /* - * Load stage2 IDT + * Load stage2 IDT and switch to our own page-table */ pushq %rsi call load_stage2_idt + call initialize_identity_maps popq %rsi /*
arch/x86/boot/compressed/ident_map_64.c +30 -21
76 lines not on this page · raw message
arch/x86/boot/compressed/kaslr.c +0 -3
@@ -903,9 +903,6 @@ void choose_random_location(unsigned long input, boot_params->hdr.loadflags |=KASLR_FLAG ; -/* Prepare to add new identity pagetables on demand. */ -initialize_identity_maps (); -/* Record the various known unsafe memory ranges. */ mem_avoid_init (input, input_size, *output);
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 15/70] x86/boot/compressed/64: Always switch to own page-table
Quoted 9 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
When booted through
startup_64the kernel keeps running on the EFI page-table until the KASLR code sets up its own page-table. Without KASLR the pre-decompression boot code never switches off the EFI page-table. Change that by unconditionally switching to our own page-table once the kernel is relocated.This makes sure we can make changes to the mapping when necessary, for
Pls use passive voice in your commit message: no "we" or "I", etc, and describe your changes in imperative mood.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 16/70 x86/boot/compressed/64: Don't pre-map memory in KASLR code
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
With the page-fault handler in place the identity mapping can be built on-demand. So remove the code which manually creates the mappings and unexport/remove the functions used for it.
--- arch/x86/boot/compressed/ident_map_64.c | 16 ++-------------- arch/x86/boot/compressed/kaslr.c | 24 +----------------------- arch/x86/boot/compressed/misc.h | 10 ---------- 3 files changed, 3 insertions(+), 47 deletions(-)
arch/x86/boot/compressed/ident_map_64.c +2 -14
@@ -89,11 +89,9 @@ phys_addr_t physical_mask = (1ULL << __PHYSICAL_MASK_SHIFT) - 1;static struct x86_mapping_info mapping_info;/* -* Adds the specified range to what will become the new identity mappings. -* Once all ranges have been added, the new mapping is activated by calling -* finalize_identity_maps() below. +* Adds the specified range to the identity mappings. */ -void add_identity_map (unsigned long start,unsigned long size) +static void add_identity_map (unsigned long start,unsigned long size) {unsigned long end = start + size;
@@ -165,16 +163,6 @@ void initialize_identity_maps(void)write_cr3 (top_level_pgt); } -/* -* This switches the page tables to the new level4 that has been built -* via calls to add_identity_map() above. If booted via startup_32(), -* this is effectively a no-op. -*/ -void finalize_identity_maps (void ) -{ -write_cr3 (top_level_pgt); -} -static void pf_error (unsigned long error_code,unsigned long address,struct pt_regs *regs) {
arch/x86/boot/compressed/kaslr.c +1 -23
@@ -399,8 +399,6 @@ static void mem_avoid_init(unsigned long input, unsigned long input_size,*/ mem_avoid[MEM_AVOID_ZO_RANGE ].start = input; mem_avoid[MEM_AVOID_ZO_RANGE ].size = (output + init_size) - input; -add_identity_map (mem_avoid[MEM_AVOID_ZO_RANGE ].start, - mem_avoid[MEM_AVOID_ZO_RANGE ].size);/* Avoid initrd. */ initrd_start = (u64 )boot_params->ext_ramdisk_image <<32 ;
@@ -420,14 +418,10 @@ static void mem_avoid_init(unsigned long input, unsigned long input_size, ; mem_avoid[MEM_AVOID_CMDLINE ].start = cmd_line; mem_avoid[MEM_AVOID_CMDLINE ].size = cmd_line_size; -add_identity_map (mem_avoid[MEM_AVOID_CMDLINE ].start, - mem_avoid[MEM_AVOID_CMDLINE ].size);/* Avoid boot parameters. */ mem_avoid[MEM_AVOID_BOOTPARAMS ].start = (unsigned long )boot_params; mem_avoid[MEM_AVOID_BOOTPARAMS ].size =sizeof (*boot_params); -add_identity_map (mem_avoid[MEM_AVOID_BOOTPARAMS ].start, - mem_avoid[MEM_AVOID_BOOTPARAMS ].size);/* We don't need to set a mapping for setup_data. */
@@ -436,11 +430,6 @@ static void mem_avoid_init(unsigned long input, unsigned long input_size,/* Enumerate the immovable memory regions */ num_immovable_mem =count_immovable_mem_regions (); - -#ifdef CONFIG_X86_VERBOSE_BOOTUP -/* Make sure video RAM can be used. */ -add_identity_map (0 ,PMD_SIZE ); -#endif }/*
@@ -919,19 +908,8 @@ void choose_random_location(unsigned long input,warn ("Physical KASLR disabled: no suitable memory region!" ); }else {/* Update the new physical address location. */ -if (*output != random_addr) { -add_identity_map (random_addr, output_size); +if (*output != random_addr) *output = random_addr; - } - -/* -* This loads the identity mapping page table. -* This should only be done if a new physical address -* is found for the kernel, otherwise we should keep -* the old page table to make it be like the "nokaslr" -* case. -*/ -finalize_identity_maps (); }
arch/x86/boot/compressed/misc.h +0 -10
@@ -98,17 +98,7 @@ static inline void choose_random_location(unsigned long input,#endif #ifdef CONFIG_X86_64 -void initialize_identity_maps (void ); -void add_identity_map (unsigned long start,unsigned long size); -void finalize_identity_maps (void );extern unsigned char _pgtable[]; -#else -static inline void initialize_identity_maps (void ) -{ } -static inline void add_identity_map (unsigned long start,unsigned long size) -{ } -static inline void finalize_identity_maps (void ) -{ }#endif #ifdef CONFIG_EARLY_PRINTK
Signature · 1 line
2.17.1
PATCH v2 17/70 x86/boot/compressed/64: Change add_identity_map() to take start and end
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Changing the function to take start and end as parameters instead of start and size simplifies the callers, which don't need to calculate the size if they already have start and end.
--- arch/x86/boot/compressed/ident_map_64.c | 15 +++++---------- 1 file changed, 5 insertions(+), 10 deletions(-)
arch/x86/boot/compressed/ident_map_64.c +5 -10
@@ -91,10 +91,8 @@ static struct x86_mapping_info mapping_info;/* * Adds the specified range to the identity mappings. */ -static void add_identity_map (unsigned long start,unsigned long size) +static void add_identity_map (unsigned long start,unsigned long end) { -unsigned long end = start + size; -/* Align boundary to 2M. */ start =round_down (start,PMD_SIZE ); end =round_up (end,PMD_SIZE );
@@ -109,8 +107,6 @@ static void add_identity_map(unsigned long start, unsigned long size)/* Locates and clears a region for a new top level page table. */ void initialize_identity_maps (void ) { -unsigned long start, size; -/* If running as an SEV guest, the encryption mask is required. */ set_sev_encryption_mask ();
@@ -157,9 +153,7 @@ void initialize_identity_maps(void)* New page-table is set up - map the kernel image and load it * into cr3. */ - start = (unsigned long )_head; - size = _end - _head; -add_identity_map (start, size); +add_identity_map ((unsigned long )_head, (unsigned long )_end);write_cr3 (top_level_pgt); }
@@ -180,7 +174,8 @@ static void pf_error(unsigned long error_code, unsigned long address,void do_boot_page_fault (struct pt_regs *regs) { -unsigned long address =native_read_cr2 (); +unsigned long address =native_read_cr2 () &PMD_MASK ; +unsigned long end = address +PMD_SIZE ;unsigned long error_code = regs->orig_ax;/*
@@ -196,5 +191,5 @@ void do_boot_page_fault(struct pt_regs *regs)* Error code is sane - now identity map the 2M region around * the faulting address. */ -add_identity_map (address &PMD_MASK ,PMD_SIZE ); +add_identity_map (address, end); }
Signature · 1 line
2.17.1
PATCH v2 18/70 x86/boot/compressed/64: Add stage1 #VC handler
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add the first handler for #VC exceptions. At stage 1 there is no GHCB yet becaue we might still be on the EFI page table and thus can't map memory unencrypted.
The stage 1 handler is limited to the MSR based protocol to talk to the hypervisor and can only support CPUID exit-codes, but that is enough to get to stage 2.
--- arch/x86/boot/compressed/Makefile | 1 + arch/x86/boot/compressed/idt_64.c | 4 ++ arch/x86/boot/compressed/idt_handlers_64.S | 4 ++ arch/x86/boot/compressed/misc.h | 1 + arch/x86/boot/compressed/sev-es.c | 42 ++++++++++++++ arch/x86/include/asm/msr-index.h | 1 + arch/x86/include/asm/sev-es.h | 45 +++++++++++++++ arch/x86/include/asm/trap_defs.h | 1 + arch/x86/kernel/sev-es-shared.c | 65 ++++++++++++++++++++++ 9 files changed, 164 insertions(+) create mode 100644 arch/x86/boot/compressed/sev-es.c create mode 100644 arch/x86/include/asm/sev-es.h create mode 100644 arch/x86/kernel/sev-es-shared.c
arch/x86/boot/compressed/Makefile +1 -0
@@ -84,6 +84,7 @@ ifdef CONFIG_X86_64 vmlinux-objs-y += $(obj)/idt_64.o $(obj)/idt_handlers_64.o vmlinux-objs-y += $(obj)/mem_encrypt.o vmlinux-objs-y += $(obj)/pgtable_64.o + vmlinux-objs-$(CONFIG_AMD_MEM_ENCRYPT) += $(obj)/sev-es.o endif vmlinux-objs-$(CONFIG_ACPI) += $(obj)/acpi.o
arch/x86/boot/compressed/idt_64.c +4 -0
@@ -31,6 +31,10 @@ void load_stage1_idt(void) { boot_idt_desc.address = (unsigned long )boot_idt; +#ifdef CONFIG_AMD_MEM_ENCRYPT +set_idt_entry (X86_TRAP_VC , boot_stage1_vc_handler); +#endif +load_boot_idt (&boot_idt_desc); }
arch/x86/boot/compressed/idt_handlers_64.S +4 -0
@@ -75,3 +75,7 @@ SYM_FUNC_END(\name) .code64 EXCEPTION_HANDLER boot_pf_handler do_boot_page_fault error_code=1 + +#ifdef CONFIG_AMD_MEM_ENCRYPT +EXCEPTION_HANDLER boot_stage1_vc_handler vc_no_ghcb_handler error_code=1 +#endif
arch/x86/boot/compressed/misc.h +1 -0
@@ -141,5 +141,6 @@ extern struct desc_ptr boot_idt_desc;/* IDT Entry Points */ void boot_pf_handler (void ); +void boot_stage1_vc_handler (void );#endif /* BOOT_COMPRESSED_MISC_H */
arch/x86/boot/compressed/sev-es.c new file +42 -0
42 lines not on this page · raw message
arch/x86/include/asm/msr-index.h +1 -0
@@ -432,6 +432,7 @@#define MSR_AMD64_IBSBRTARGET 0xc001103b#define MSR_AMD64_IBSOPDATA4 0xc001103d#define MSR_AMD64_IBS_REG_COUNT_MAX 8/* includes MSR_AMD64_IBSBRTARGET */ +#define MSR_AMD64_SEV_ES_GHCB 0xc0010130#define MSR_AMD64_SEV 0xc0010131#define MSR_AMD64_SEV_ENABLED_BIT 0#define MSR_AMD64_SEV_ENABLED BIT_ULL(MSR_AMD64_SEV_ENABLED_BIT)
arch/x86/include/asm/sev-es.h new file +45 -0
45 lines not on this page · raw message
arch/x86/include/asm/trap_defs.h +1 -0
@@ -24,6 +24,7 @@ enum {X86_TRAP_AC ,/* 17, Alignment Check */ X86_TRAP_MC ,/* 18, Machine Check */ X86_TRAP_XF ,/* 19, SIMD Floating-Point Exception */ +X86_TRAP_VC =29 ,/* 29, VMM Communication Exception */ X86_TRAP_IRET = 32, /* 32, IRET Exception */ };
arch/x86/kernel/sev-es-shared.c new file +65 -0
65 lines not on this page · raw message
Signature · 1 line
2.17.1
Re: [PATCH 18/70] x86/boot/compressed/64: Add stage1 #VC handler
Quoted 22 lines · Joerg Roedel
arch/x86/include/asm/sev-es.h new file +16
-0@@ -0,0 +1,45 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* AMD Encrypted Register State Support +* +* Author: Joerg Roedel <[email protected]> +*/ + +#ifndef __ASM_ENCRYPTED_STATE_H +#define __ASM_ENCRYPTED_STATE_H + +#include <linux/types.h> + +#define GHCB_SEV_CPUID_REQ 0x004UL +#define GHCB_CPUID_REQ_EAX 0 +#define GHCB_CPUID_REQ_EBX 1 +#define GHCB_CPUID_REQ_ECX 2
arch/x86/include/asm/sev-es.h line 26
Quoted 10 lines
arch/x86/include/asm/sev-es.h new file +10
-0+#define GHCB_CPUID_REQ_EDX 3 +#define GHCB_CPUID_REQ (fn, reg) (GHCB_SEV_CPUID_REQ | \ + (((unsigned long)reg & 3) << 30) | \ + (((unsigned long)fn) << 32)) + +#define GHCB_SEV_CPUID_RESP 0x005UL +#define GHCB_SEV_TERMINATE 0x100UL + +#define GHCB_SEV_GHCB_RESP_CODE (v) ((v) & 0xfff) +#define VMGEXIT () { asm volatile("rep; vmmcall\n\r"); }
Since preemption and irqs should be disabled before updating the GHCB and its MSR and until the contents have been accessed following VMGEXIT, should there be checks in place to ensure that's always the case?
Quoted 106 lines
arch/x86/include/asm/sev-es.h new file +19
-0+ +static inline u64 lower_bits (u64 val,unsigned int bits) +{ +u64 mask = (1ULL << bits) -1 ; + +return (val & mask); +} + +static inline u64 copy_lower_bits (u64 out,u64 in,unsigned int bits) +{ +u64 mask = (1ULL << bits) -1 ; + + out &= ~mask; + out |=lower_bits (in, bits); + +return out; +} + +#endif arch/x86/include/asm/trap_defs.h +1
-0@@ -24,6 +24,7 @@ enum {X86_TRAP_AC ,/* 17, Alignment Check */ X86_TRAP_MC ,/* 18, Machine Check */ X86_TRAP_XF ,/* 19, SIMD Floating-Point Exception */ +X86_TRAP_VC =29 ,/* 29, VMM Communication Exception */ X86_TRAP_IRET = 32, /* 32, IRET Exception */ };arch/x86/kernel/sev-es-shared.c new file +65
-0@@ -0,0 +1,65 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* AMD Encrypted Register State Support +* +* Author: Joerg Roedel <[email protected]> +* +* This file is not compiled stand-alone. It contains code shared +* between the pre-decompression boot code and the running Linux kernel +* and is included directly into both code-bases. +*/ + +/* +* Boot VC Handler - This is the first VC handler during boot, there is no GHCB +* page yet, so it only supports the MSR based communication with the +* hypervisor and only the CPUID exit-code. +*/ +void __init vc_no_ghcb_handler (struct pt_regs *regs,unsigned long exit_code) +{ +unsigned int fn =lower_bits (regs->ax,32 ); +unsigned long val; + +/* Only CPUID is supported via MSR protocol */ +if (exit_code !=SVM_EXIT_CPUID ) + goto fail; + +sev_es_wr_ghcb_msr (GHCB_CPUID_REQ (fn,GHCB_CPUID_REQ_EAX )); +VMGEXIT (); + val =sev_es_rd_ghcb_msr (); +if (GHCB_SEV_GHCB_RESP_CODE (val) !=GHCB_SEV_CPUID_RESP ) + goto fail; + regs->ax = val >>32 ; + +sev_es_wr_ghcb_msr (GHCB_CPUID_REQ (fn,GHCB_CPUID_REQ_EBX )); +VMGEXIT (); + val =sev_es_rd_ghcb_msr (); +if (GHCB_SEV_GHCB_RESP_CODE (val) !=GHCB_SEV_CPUID_RESP ) + goto fail; + regs->bx = val >>32 ; + +sev_es_wr_ghcb_msr (GHCB_CPUID_REQ (fn,GHCB_CPUID_REQ_ECX )); +VMGEXIT (); + val =sev_es_rd_ghcb_msr (); +if (GHCB_SEV_GHCB_RESP_CODE (val) !=GHCB_SEV_CPUID_RESP ) + goto fail; + regs->cx = val >>32 ; + +sev_es_wr_ghcb_msr (GHCB_CPUID_REQ (fn,GHCB_CPUID_REQ_EDX )); +VMGEXIT (); + val =sev_es_rd_ghcb_msr (); +if (GHCB_SEV_GHCB_RESP_CODE (val) !=GHCB_SEV_CPUID_RESP ) + goto fail; + regs->dx = val >>32 ; + + regs->ip +=2 ; + +return ; + +fail: +sev_es_wr_ghcb_msr (GHCB_SEV_TERMINATE ); +VMGEXIT (); + +/* Shouldn't get here - if we do halt the machine */ +while (true) + asmvolatile ("hlt\n" ); +}Signature · 1 line
2.17.1
Re: [PATCH 18/70] x86/boot/compressed/64: Add stage1 #VC handler
Quoted 7 lines · David Rientjes
arch/x86/include/asm/sev-es.h line 26
Quoted 2 lines · Joerg Roedel
arch/x86/include/asm/sev-es.h new file +2
-0+#define GHCB_SEV_GHCB_RESP_CODE (v) ((v) & 0xfff) +#define VMGEXIT () { asm volatile("rep; vmmcall\n\r"); }Since preemption and irqs should be disabled before updating the GHCB and its MSR and until the contents have been accessed following VMGEXIT, should there be checks in place to ensure that's always the case?
Good point, some checking is certainly helpful. Currently it is the case, because the GHCB is accessed and used only:
1) At boot when only the boot CPU is running
2) In the #VC handler, which does not enable interrupts
3) In the NMI handler, which is also not preemptible
I can also add code to sev_es_get/put_ghcb to make sure these conditions are met. All this does not prevent the preemption by NMIs, which could cause another nested #VC exception, but that is handled separatly.
Regards,
Joerg
Borislav Petkov
maintainerRe: [PATCH 18/70] x86/boot/compressed/64: Add stage1 #VC handler
In reply to Joerg Roedelarch/x86/boot/compressed/idt_handlers_64.S SYM_FUNC_END(\name) · line 80
Quoted 11 lines · Joerg Roedel
arch/x86/boot/compressed/idt_handlers_64.S +3
-0@@ -75,3 +75,7 @@ SYM_FUNC_END(\name) .code64 EXCEPTION_HANDLER boot_pf_handler do_boot_page_fault error_code=1 + +#ifdef CONFIG_AMD_MEM_ENCRYPT +EXCEPTION_HANDLER boot_stage1_vc_handler vc_no_ghcb_handler error_code=1
Like the others
boot_stage1_vc do_boot_stage1_vc ...
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 19/70 x86/boot/compressed/64: Call set_sev_encryption_mask earlier
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Call set_sev_encryption_mask() while still on the stage 1 #VC-handler, because the stage 2 handler needs our own page-tables to be set up, to which calling set_sev_encryption_mask() is a prerequisite.
--- arch/x86/boot/compressed/head_64.S | 8 +++++++- arch/x86/boot/compressed/ident_map_64.c | 3 --- 2 files changed, 7 insertions(+), 4 deletions(-)
arch/x86/boot/compressed/head_64.S +7 -1
@@ -491,9 +491,15 @@ SYM_FUNC_START_LOCAL_NOALIGN(.Lrelocated) rep stosq /* - * Load stage2 IDT and switch to our own page-table + * If running as an SEV guest, the encryption mask is required in the + * page-table setup code below. When the guest also has SEV-ES enabled + * set_sev_encryption_mask() will cause #VC exceptions, but the stage2 + * handler can't map its GHCB because the page-table is not set up yet. + * So set up the encryption mask here while still on the stage1 #VC + * handler. Then load stage2 IDT and switch to our own page-table. */ pushq %rsi + call set_sev_encryption_mask call load_stage2_idt call initialize_identity_maps popq %rsi
arch/x86/boot/compressed/ident_map_64.c +0 -3
@@ -107,9 +107,6 @@ static void add_identity_map(unsigned long start, unsigned long end)/* Locates and clears a region for a new top level page table. */ void initialize_identity_maps (void ) { -/* If running as an SEV guest, the encryption mask is required. */ -set_sev_encryption_mask (); -/* Exclude the encryption mask from __PHYSICAL_MASK */ physical_mask &= ~sme_me_mask;
Signature · 1 line
2.17.1
PATCH v2 20/70 x86/boot/compressed/64: Check return value of kernel_ident_mapping_init()
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The function can fail to create an identity mapping, check for that and bail out if it happens.
--- arch/x86/boot/compressed/ident_map_64.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/boot/compressed/ident_map_64.c +5 -2
@@ -93,6 +93,8 @@ static struct x86_mapping_info mapping_info;*/ static void add_identity_map (unsigned long start,unsigned long end) { +int ret; +/* Align boundary to 2M. */ start =round_down (start,PMD_SIZE ); end =round_up (end,PMD_SIZE );
@@ -100,8 +102,9 @@ static void add_identity_map(unsigned long start, unsigned long end)return ;/* Build the mapping. */ -kernel_ident_mapping_init (&mapping_info, (pgd_t *)top_level_pgt, - start, end); + ret =kernel_ident_mapping_init (&mapping_info, (pgd_t *)top_level_pgt, start, end); +if (ret) +error ("Error: kernel_ident_mapping_init() failed\n" ); }/* Locates and clears a region for a new top level page table. */
Signature · 1 line
2.17.1
PATCH v2 21/70 x86/boot/compressed/64: Add function to map a page unencrypted
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
This function is needed to map the GHCB for SEV-ES guests. The GHCB is used for communication with the hypervisor, so its content must not be encrypted.
--- arch/x86/boot/compressed/ident_map_64.c | 125 ++++++++++++++++++++++++ arch/x86/boot/compressed/misc.h | 1 + 2 files changed, 126 insertions(+)
arch/x86/boot/compressed/ident_map_64.c +125 -0
@@ -26,6 +26,7 @@#include <asm/init.h> #include <asm/pgtable.h> #include <asm/trap_defs.h> +#include <asm/cmpxchg.h> /* Use the static base for this part of the boot process */ #undef __PAGE_OFFSET#define __PAGE_OFFSET __PAGE_OFFSET_BASE
@@ -157,6 +158,130 @@ void initialize_identity_maps(void)write_cr3 (top_level_pgt); } +static pte_t *split_large_pmd (struct x86_mapping_info *info, +pmd_t *pmdp,unsigned long __address) +{ +unsigned long page_flags; +unsigned long address; +pte_t *pte; +pmd_t pmd; +int i; + + pte = (pte_t *)info->alloc_pgt_page (info->context); +if (!pte) +return NULL ; + + address = __address &PMD_MASK ; +/* No large page - clear PSE flag */ + page_flags = info->page_flag & ~_PAGE_PSE; + +/* Populate the PTEs */ +for (i =0 ; i <PTRS_PER_PMD ; i++) { +set_pte (&pte[i],__pte (address | page_flags)); + address +=PAGE_SIZE ; + } + +/* +* Ideally we need to clear the large PMD first and do a TLB +* flush before we write the new PMD. But the 2M range of the +* PMD might contain the code we execute and/or the stack +* we are on, so we can't do that. But that should be safe here +* because we are going from large to small mappings and we are +* also the only user of the page-table, so there is no chance +* of a TLB multihit. +*/ + pmd =__pmd ((unsigned long )pte | info->kernpg_flag); +set_pmd (pmdp, pmd); +/* Flush TLB to establish the new PMD */ +write_cr3 (top_level_pgt); + +return pte +pte_index (__address); +} + +static void clflush_page (unsigned long address) +{ +unsigned int flush_size; +char *cl, *start, *end; + +/* +* Hardcode cl-size to 64 - CPUID can't be used here because that might +* cause another #VC exception and the GHCB is not ready to use yet. +*/ + flush_size =64 ; + start = (char *)(address &PAGE_MASK ); + end = start +PAGE_SIZE ; + +/* +* First make sure there are no pending writes on the cache-lines to +* flush. +*/ + asmvolatile ("mfence" : : :"memory" ); + +for (cl = start; cl != end; cl += flush_size) +clflush (cl); +} + +static int __set_page_decrypted (struct x86_mapping_info *info, +unsigned long address) +{ +unsigned long scratch, *target; +pgd_t *pgdp = (pgd_t *)top_level_pgt; +p4d_t *p4dp; +pud_t *pudp; +pmd_t *pmdp; +pte_t *ptep, pte; + +/* +* First make sure there is a PMD mapping for 'address'. +* It should already exist, but keep things generic. +* +* To map the page just read from it and fault it in if there is no +* mapping yet. add_identity_map() can't be called here because that +* would unconditionally map the address on PMD level, destroying any +* PTE-level mappings that might already exist. Also do something +* useless with 'scratch' so the access won't be optimized away. +*/ + target = (unsigned long *)address; + scratch = *target; +arch_cmpxchg (target, scratch, scratch); + +/* +* The page is mapped at least with PMD size - so skip checks and walk +* directly to the PMD. +*/ + p4dp =p4d_offset (pgdp, address); + pudp =pud_offset (p4dp, address); + pmdp =pmd_offset (pudp, address); + +if (pmd_large (*pmdp)) + ptep =split_large_pmd (info, pmdp, address); +else + ptep =pte_offset_kernel (pmdp, address); + +if (!ptep) +return -ENOMEM ; + +/* Clear encryption flag and write new pte */ + pte =pte_clear_flags (*ptep, _PAGE_ENC); +set_pte (ptep, pte); + +/* Flush TLB to map the page unencrypted */ +write_cr3 (top_level_pgt); + +/* +* Changing encryption attributes of a page requires to flush it from +* the caches. +*/ +clflush_page (address); + +return 0 ; +} + +int set_page_decrypted (unsigned long address) +{ +return __set_page_decrypted (&mapping_info, address); +} +static void pf_error (unsigned long error_code,unsigned long address,struct pt_regs *regs) {
arch/x86/boot/compressed/misc.h +1 -0
@@ -98,6 +98,7 @@ static inline void choose_random_location(unsigned long input,#endif #ifdef CONFIG_X86_64 +extern int set_page_decrypted (unsigned long address);extern unsigned char _pgtable[];#endif
Signature · 1 line
2.17.1
Re: [PATCH 21/70] x86/boot/compressed/64: Add function to map a page unencrypted
Quoted 128 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
This function is needed to map the GHCB for SEV-ES guests. The GHCB is used for communication with the hypervisor, so its content must not be encrypted.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/boot/compressed/ident_map_64.c | 125 ++++++++++++++++++++++++ arch/x86/boot/compressed/misc.h | 1 + 2 files changed, 126 insertions(+)arch/x86/boot/compressed/ident_map_64.c +101
-0@@ -26,6 +26,7 @@#include <asm/init.h> #include <asm/pgtable.h> #include <asm/trap_defs.h> +#include <asm/cmpxchg.h> /* Use the static base for this part of the boot process */ #undef __PAGE_OFFSET#define __PAGE_OFFSET __PAGE_OFFSET_BASE@@ -157,6 +158,130 @@ void initialize_identity_maps(void)write_cr3 (top_level_pgt); } +static pte_t *split_large_pmd (struct x86_mapping_info *info, +pmd_t *pmdp,unsigned long __address) +{ +unsigned long page_flags; +unsigned long address; +pte_t *pte; +pmd_t pmd; +int i; + + pte = (pte_t *)info->alloc_pgt_page (info->context); +if (!pte) +return NULL ; + + address = __address &PMD_MASK ; +/* No large page - clear PSE flag */ + page_flags = info->page_flag & ~_PAGE_PSE; + +/* Populate the PTEs */ +for (i =0 ; i <PTRS_PER_PMD ; i++) { +set_pte (&pte[i],__pte (address | page_flags)); + address +=PAGE_SIZE ; + } + +/* +* Ideally we need to clear the large PMD first and do a TLB +* flush before we write the new PMD. But the 2M range of the +* PMD might contain the code we execute and/or the stack +* we are on, so we can't do that. But that should be safe here +* because we are going from large to small mappings and we are +* also the only user of the page-table, so there is no chance +* of a TLB multihit. +*/ + pmd =__pmd ((unsigned long )pte | info->kernpg_flag); +set_pmd (pmdp, pmd); +/* Flush TLB to establish the new PMD */ +write_cr3 (top_level_pgt); + +return pte +pte_index (__address); +} + +static void clflush_page (unsigned long address) +{ +unsigned int flush_size; +char *cl, *start, *end; + +/* +* Hardcode cl-size to 64 - CPUID can't be used here because that might +* cause another #VC exception and the GHCB is not ready to use yet. +*/ + flush_size =64 ; + start = (char *)(address &PAGE_MASK ); + end = start +PAGE_SIZE ; + +/* +* First make sure there are no pending writes on the cache-lines to +* flush. +*/ + asmvolatile ("mfence" : : :"memory" ); + +for (cl = start; cl != end; cl += flush_size) +clflush (cl); +} + +static int __set_page_decrypted (struct x86_mapping_info *info, +unsigned long address) +{ +unsigned long scratch, *target; +pgd_t *pgdp = (pgd_t *)top_level_pgt; +p4d_t *p4dp; +pud_t *pudp; +pmd_t *pmdp; +pte_t *ptep, pte; + +/* +* First make sure there is a PMD mapping for 'address'. +* It should already exist, but keep things generic. +* +* To map the page just read from it and fault it in if there is no +* mapping yet. add_identity_map() can't be called here because that +* would unconditionally map the address on PMD level, destroying any +* PTE-level mappings that might already exist. Also do something +* useless with 'scratch' so the access won't be optimized away. +*/ + target = (unsigned long *)address; + scratch = *target; +arch_cmpxchg (target, scratch, scratch); + +/* +* The page is mapped at least with PMD size - so skip checks and walk +* directly to the PMD. +*/ + p4dp =p4d_offset (pgdp, address); + pudp =pud_offset (p4dp, address); + pmdp =pmd_offset (pudp, address); + +if (pmd_large (*pmdp)) + ptep =split_large_pmd (info, pmdp, address); +else + ptep =pte_offset_kernel (pmdp, address); +
arch/x86/boot/compressed/ident_map_64.c void initialize_identity_maps(void) · line 270
Quoted 10 lines
arch/x86/boot/compressed/ident_map_64.c +10
-0+if (!ptep) +return -ENOMEM ; + +/* Clear encryption flag and write new pte */ + pte =pte_clear_flags (*ptep, _PAGE_ENC); +set_pte (ptep, pte); + +/* Flush TLB to map the page unencrypted */ +write_cr3 (top_level_pgt); +
Is there a guarantee that this flushes the tlb if cr3 == top_level_pgt alrady without an invlpg?
Quoted 29 lines
arch/x86/boot/compressed/ident_map_64.c +14
-0+/* +* Changing encryption attributes of a page requires to flush it from +* the caches. +*/ +clflush_page (address); + +return 0 ; +} + +int set_page_decrypted (unsigned long address) +{ +return __set_page_decrypted (&mapping_info, address); +} +static void pf_error (unsigned long error_code,unsigned long address,struct pt_regs *regs) {arch/x86/boot/compressed/misc.h +1
-0@@ -98,6 +98,7 @@ static inline void choose_random_location(unsigned long input,#endif #ifdef CONFIG_X86_64 +extern int set_page_decrypted (unsigned long address);extern unsigned char _pgtable[];#endif
Re: [PATCH 21/70] x86/boot/compressed/64: Add function to map a page unencrypted
Quoted 10 lines · David Rientjes
arch/x86/boot/compressed/ident_map_64.c void initialize_identity_maps(void) · line 270
Quoted 8 lines
arch/x86/boot/compressed/ident_map_64.c +8
-0+ +/* Clear encryption flag and write new pte */ + pte =pte_clear_flags (*ptep, _PAGE_ENC); +set_pte (ptep, pte); + +/* Flush TLB to map the page unencrypted */ +write_cr3 (top_level_pgt); +Is there a guarantee that this flushes the tlb if cr3 ==
top_level_pgtalrady without an invlpg?
Ahh, good catch.
It *never* flushes global pages. For a generic function like this, that seems pretty dangerous because the PTEs it goes after could quite easily be Global. It's also not _obviously_ correct if PCIDs are in play (which I don't think they are on AMD).
A flush_tlb_global() is probably more appropriate. Better yet, is there a reason not to use flush_tlb_kernel_range()? I don't think it's necessary to whack the entire TLB for one PTE set.
Re: [PATCH 21/70] x86/boot/compressed/64: Add function to map a page unencrypted
Quoted 8 lines · Dave Hansen
It *never* flushes global pages. For a generic function like this, that seems pretty dangerous because the PTEs it goes after could quite easily be Global. It's also not _obviously_ correct if PCIDs are in play (which I don't think they are on AMD).
A
flush_tlb_global()is probably more appropriate. Better yet, is there a reason not to useflush_tlb_kernel_range()? I don't think it's necessary to whack the entire TLB for one PTE set.
This code runs before the actual kernel image is decompressed, so there is no PCID and no global pages (I think CR4.PGE is still 0). So a cr3-write is enough to flush the TLB. Also the TLB-flush helpers of the running kernel are not available here.
Regards,
Joerg
Re: [PATCH 21/70] x86/boot/compressed/64: Add function to map a page unencrypted
Quoted 6 lines · Joerg Roedel
Quoted 5 lines · Dave Hansen
It *never* flushes global pages. For a generic function like this, that seems pretty dangerous because the PTEs it goes after could quite easily be Global. It's also not _obviously_ correct if PCIDs are in play (which I don't think they are on AMD).
Quoted 8 lines
Quoted 3 lines
A
flush_tlb_global()is probably more appropriate. Better yet, is there a reason not to useflush_tlb_kernel_range()? I don't think it's necessary to whack the entire TLB for one PTE set.This code runs before the actual kernel image is decompressed, so there is no PCID and no global pages (I think CR4.PGE is still 0). So a cr3-write is enough to flush the TLB. Also the TLB-flush helpers of the running kernel are not available here.
Geez, I always forget about the compressed code. :) Good point about PCIDs.
In any case, I thought this all came through initialize_identity_maps(), which does, for instance:
mapping_info.page_flag = __PAGE_KERNEL_LARGE_EXEC | sme_me_mask;
Where:
#define __PAGE_KERNEL_LARGE_EXEC (__PP|__RW| 0|___A| 0|___D|_PSE|___G)
That looks like it has the Global bit set. Does that not apply here somehow?
Re: [PATCH 21/70] x86/boot/compressed/64: Add function to map a page unencrypted
Quoted 11 lines · Dave Hansen
In any case, I thought this all came through
initialize_identity_maps(), which does, for instance:mapping_info.page_flag = __PAGE_KERNEL_LARGE_EXEC | sme_me_mask;Where:
#define
__PAGE_KERNEL_LARGE_EXEC(__PP|__RW| 0|___A| 0|___D|_PSE|___G)That looks like it has the Global bit set. Does that not apply here somehow?
No, as the value of %cr4 at boot is 0x00000020, so PGE is not set and global pages are not enabled. It wouldn't make sense anyhow, as global pages only make sense when there are more than one address space, which is not the case that early in boot.
Regards,
Joerg
PATCH v2 22/70 x86/boot/compressed/64: Setup GHCB Based VC Exception handler
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Install an exception handler for #VC exception that uses a GHCB. Also add the infrastructure for handling different exit-codes by decoding the instruction that caused the exception and error handling.
--- arch/x86/Kconfig | 1 + arch/x86/boot/compressed/idt_64.c | 4 + arch/x86/boot/compressed/idt_handlers_64.S | 1 + arch/x86/boot/compressed/misc.h | 1 + arch/x86/boot/compressed/sev-es.c | 94 ++++++++++++++ arch/x86/include/asm/sev-es.h | 33 +++++ arch/x86/include/uapi/asm/svm.h | 1 + arch/x86/kernel/sev-es-shared.c | 142 +++++++++++++++++++++ 8 files changed, 277 insertions(+)
arch/x86/Kconfig +1 -0
@@ -1526,6 +1526,7 @@ config AMD_MEM_ENCRYPT select DYNAMIC_PHYSICAL_MASK select ARCH_USE_MEMREMAP_PROT select ARCH_HAS_FORCE_DMA_UNENCRYPTED + select INSTRUCTION_DECODER ---help--- Say yes to enable support for the encryption of system memory. This requires an AMD processor that supports Secure Memory
arch/x86/boot/compressed/idt_64.c +4 -0
@@ -45,5 +45,9 @@ void load_stage2_idt(void)set_idt_entry (X86_TRAP_PF , boot_pf_handler); +#ifdef CONFIG_AMD_MEM_ENCRYPT +set_idt_entry (X86_TRAP_VC , boot_stage2_vc_handler); +#endif +load_boot_idt (&boot_idt_desc); }
arch/x86/boot/compressed/idt_handlers_64.S +1 -0
@@ -78,4 +78,5 @@ EXCEPTION_HANDLER boot_pf_handler do_boot_page_fault error_code=1 #ifdef CONFIG_AMD_MEM_ENCRYPT EXCEPTION_HANDLER boot_stage1_vc_handler vc_no_ghcb_handler error_code=1 +EXCEPTION_HANDLER boot_stage2_vc_handler boot_vc_handler error_code=1 #endif
arch/x86/boot/compressed/misc.h +1 -0
@@ -143,5 +143,6 @@ extern struct desc_ptr boot_idt_desc;/* IDT Entry Points */ void boot_pf_handler (void );void boot_stage1_vc_handler (void ); +void boot_stage2_vc_handler (void );#endif /* BOOT_COMPRESSED_MISC_H */
arch/x86/boot/compressed/sev-es.c +94 -0
114 lines not on this page · raw message
arch/x86/include/asm/sev-es.h +33 -0
@@ -9,7 +9,14 @@#define __ASM_ENCRYPTED_STATE_H#include <linux/types.h> +#include <asm/insn.h> +#define GHCB_SEV_INFO 0x001UL +#define GHCB_SEV_INFO_REQ 0x002UL +#define GHCB_INFO (v) ((v) & 0xfffUL) +#define GHCB_PROTO_MAX (v) (((v) >> 48) & 0xffffUL) +#define GHCB_PROTO_MIN (v) (((v) >> 32) & 0xffffUL) +#define GHCB_PROTO_OUR 0x0001UL#define GHCB_SEV_CPUID_REQ 0x004UL#define GHCB_CPUID_REQ_EAX 0#define GHCB_CPUID_REQ_EBX 1
@@ -21,10 +28,36 @@#define GHCB_SEV_CPUID_RESP 0x005UL#define GHCB_SEV_TERMINATE 0x100UL +#define GHCB_SEV_ES_REASON_GENERAL_REQUEST 0 +#define GHCB_SEV_ES_REASON_PROTOCOL_UNSUPPORTED 1#define GHCB_SEV_GHCB_RESP_CODE (v) ((v) & 0xfff)#define VMGEXIT () { asm volatile("rep; vmmcall\n\r"); } +enum es_result { +ES_OK ,/* All good */ +ES_UNSUPPORTED ,/* Requested operation not supported */ +ES_VMM_ERROR ,/* Unexpected state from the VMM */ +ES_DECODE_FAILED ,/* Instruction decoding failed */ +ES_EXCEPTION ,/* Instruction caused exception */ +ES_RETRY ,/* Retry instruction emulation */ +}; + +struct es_fault_info { +unsigned long vector; +unsigned long error_code; +unsigned long cr2; +}; + +struct pt_regs ; + +/* ES instruction emulation context */ +struct es_em_ctxt { +struct pt_regs *regs; +struct insn insn; +struct es_fault_info fi; +}; +static inline u64 lower_bits (u64 val,unsigned int bits) {u64 mask = (1ULL << bits) -1 ;
arch/x86/include/uapi/asm/svm.h +1 -0
@@ -29,6 +29,7 @@#define SVM_EXIT_WRITE_DR6 0x036#define SVM_EXIT_WRITE_DR7 0x037#define SVM_EXIT_EXCP_BASE 0x040 +#define SVM_EXIT_LAST_EXCP 0x05f#define SVM_EXIT_INTR 0x060#define SVM_EXIT_NMI 0x061#define SVM_EXIT_SMI 0x062
arch/x86/kernel/sev-es-shared.c +142 -0
151 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 23/70 x86/sev-es: Add support for handling IOIO exceptions
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Add support for decoding and handling #VC exceptions for IOIO events.
--- arch/x86/boot/compressed/sev-es.c | 32 +++++ arch/x86/kernel/sev-es-shared.c | 202 ++++++++++++++++++++++++++++++ 2 files changed, 234 insertions(+)
arch/x86/boot/compressed/sev-es.c +32 -0
@@ -18,6 +18,35 @@struct ghcb boot_ghcb_page__aligned(PAGE_SIZE) ;struct ghcb *boot_ghcb; +/* +* Copy a version of this function here - insn-eval.c can't be used in +* pre-decompression code. +*/ +static bool insn_rep_prefix (struct insn *insn) +{ +int i; + +insn_get_prefixes (insn); + +for (i =0 ; i < insn->prefixes.nbytes; i++) { +insn_byte_t p = insn->prefixes.bytes[i]; + +if (p ==0xf2 || p ==0xf3 ) +return true; + } + +return false; +} + +/* +* Only a dummy for insn_get_seg_base() - Early boot-code is 64bit only and +* doesn't use segments. +*/ +static unsigned long insn_get_seg_base (struct pt_regs *regs,int seg_reg_idx) +{ +return 0UL ; +} +static inline u64 sev_es_rd_ghcb_msr (void ) {unsigned long low, high;
@@ -117,6 +146,9 @@ void boot_vc_handler(struct pt_regs *regs, unsigned long exit_code) goto finish;switch (exit_code) { +case SVM_EXIT_IOIO : + result =vc_handle_ioio (boot_ghcb, &ctxt); +break ;default : result =ES_UNSUPPORTED ;break ;
arch/x86/kernel/sev-es-shared.c +202 -0
205 lines not on this page · raw message
Signature · 1 line
2.17.1
Re: [PATCH 23/70] x86/sev-es: Add support for handling IOIO exceptions
Quoted 205 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
Add support for decoding and handling #VC exceptions for IOIO events.
Signed-off-by Tom Lendacky <[email protected]>[ [email protected]: Adapted code to #VC handling framework ]Co-developed-by Joerg Roedel <[email protected]>Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/boot/compressed/sev-es.c | 32 +++++ arch/x86/kernel/sev-es-shared.c | 202 ++++++++++++++++++++++++++++++ 2 files changed, 234 insertions(+)arch/x86/boot/compressed/sev-es.c +32
-0@@ -18,6 +18,35 @@struct ghcb boot_ghcb_page__aligned(PAGE_SIZE) ;struct ghcb *boot_ghcb; +/* +* Copy a version of this function here - insn-eval.c can't be used in +* pre-decompression code. +*/ +static bool insn_rep_prefix (struct insn *insn) +{ +int i; + +insn_get_prefixes (insn); + +for (i =0 ; i < insn->prefixes.nbytes; i++) { +insn_byte_t p = insn->prefixes.bytes[i]; + +if (p ==0xf2 || p ==0xf3 ) +return true; + } + +return false; +} + +/* +* Only a dummy for insn_get_seg_base() - Early boot-code is 64bit only and +* doesn't use segments. +*/ +static unsigned long insn_get_seg_base (struct pt_regs *regs,int seg_reg_idx) +{ +return 0UL ; +} +static inline u64 sev_es_rd_ghcb_msr (void ) {unsigned long low, high;@@ -117,6 +146,9 @@ void boot_vc_handler(struct pt_regs *regs, unsigned long exit_code) goto finish;switch (exit_code) { +case SVM_EXIT_IOIO : + result =vc_handle_ioio (boot_ghcb, &ctxt); +break ;default : result =ES_UNSUPPORTED ;break ;arch/x86/kernel/sev-es-shared.c +134
-0@@ -205,3 +205,205 @@ static enum es_result vc_insn_string_write(struct es_em_ctxt *ctxt,return ret; } + +#define IOIO_TYPE_STR BIT(2) +#define IOIO_TYPE_IN 1 +#define IOIO_TYPE_INS (IOIO_TYPE_IN | IOIO_TYPE_STR) +#define IOIO_TYPE_OUT 0 +#define IOIO_TYPE_OUTS (IOIO_TYPE_OUT | IOIO_TYPE_STR) + +#define IOIO_REP BIT(3) + +#define IOIO_ADDR_64 BIT(9) +#define IOIO_ADDR_32 BIT(8) +#define IOIO_ADDR_16 BIT(7) + +#define IOIO_DATA_32 BIT(6) +#define IOIO_DATA_16 BIT(5) +#define IOIO_DATA_8 BIT(4) + +#define IOIO_SEG_ES (0 << 10) +#define IOIO_SEG_DS (3 << 10) + +static enum es_result vc_ioio_exitinfo (struct es_em_ctxt *ctxt,u64 *exitinfo) +{ +struct insn *insn = &ctxt->insn; + *exitinfo =0 ; + +switch (insn->opcode.bytes[0 ]) { +/* INS opcodes */ +case 0x6c : +case 0x6d : + *exitinfo |=IOIO_TYPE_INS ; + *exitinfo |=IOIO_SEG_ES ; + *exitinfo |= (ctxt->regs->dx &0xffff ) <<16 ; +break ; + +/* OUTS opcodes */ +case 0x6e : +case 0x6f : + *exitinfo |=IOIO_TYPE_OUTS ; + *exitinfo |=IOIO_SEG_DS ; + *exitinfo |= (ctxt->regs->dx &0xffff ) <<16 ; +break ; + +/* IN immediate opcodes */ +case 0xe4 : +case 0xe5 : + *exitinfo |=IOIO_TYPE_IN ; + *exitinfo |= insn->immediate.value <<16 ; +break ; + +/* OUT immediate opcodes */ +case 0xe6 : +case 0xe7 : + *exitinfo |=IOIO_TYPE_OUT ; + *exitinfo |= insn->immediate.value <<16 ; +break ; + +/* IN register opcodes */ +case 0xec : +case 0xed : + *exitinfo |=IOIO_TYPE_IN ; + *exitinfo |= (ctxt->regs->dx &0xffff ) <<16 ; +break ; + +/* OUT register opcodes */ +case 0xee : +case 0xef : + *exitinfo |=IOIO_TYPE_OUT ; + *exitinfo |= (ctxt->regs->dx &0xffff ) <<16 ; +break ; + +default : +return ES_DECODE_FAILED ; + } + +switch (insn->opcode.bytes[0 ]) { +case 0x6c : +case 0x6e : +case 0xe4 : +case 0xe6 : +case 0xec : +case 0xee : +/* Single byte opcodes */ + *exitinfo |=IOIO_DATA_8 ; +break ; +default : +/* Length determined by instruction parsing */ + *exitinfo |= (insn->opnd_bytes ==2 ) ?IOIO_DATA_16 + :IOIO_DATA_32 ; + } +switch (insn->addr_bytes) { +case 2 : + *exitinfo |=IOIO_ADDR_16 ; +break ; +case 4 : + *exitinfo |=IOIO_ADDR_32 ; +break ; +case 8 : + *exitinfo |=IOIO_ADDR_64 ; +break ; + } + +if (insn_rep_prefix (insn)) + *exitinfo |=IOIO_REP ; + +return ES_OK ; +} + +static enum es_result vc_handle_ioio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +{ +struct pt_regs *regs = ctxt->regs; +u64 exit_info_1, exit_info_2; +enum es_result ret; + + ret =vc_ioio_exitinfo (ctxt, &exit_info_1); +if (ret !=ES_OK ) +return ret; + +if (exit_info_1 &IOIO_TYPE_STR ) { +int df = (regs->flags &X86_EFLAGS_DF ) ?-1 :1 ; +unsigned int io_bytes, exit_bytes; +unsigned int ghcb_count, op_count; +unsigned long es_base; +u64 sw_scratch; + +/* +* For the string variants with rep prefix the amount of in/out +* operations per #VC exception is limited so that the kernel +* has a chance to take interrupts an re-schedule while the +* instruction is emulated. +*/ + io_bytes = (exit_info_1 >>4 ) &0x7 ; + ghcb_count =sizeof (ghcb->shared_buffer) / io_bytes; + + op_count = (exit_info_1 &IOIO_REP ) ? regs->cx :1 ;
arch/x86/kernel/sev-es-shared.c static enum es_result vc_insn_string_write(struct es_em_ctxt *ctxt, · line 351
Quoted 10 lines
arch/x86/kernel/sev-es-shared.c +10
-0+ exit_info_2 =min (op_count, ghcb_count); + exit_bytes = exit_info_2 * io_bytes; + + es_base =insn_get_seg_base (ctxt->regs,INAT_SEG_REG_ES ); + +if (!(exit_info_1 &IOIO_TYPE_IN )) { + ret =vc_insn_string_read (ctxt, + (void *)(es_base + regs->si), + ghcb->shared_buffer, io_bytes, + exit_info_2, df);
The last argument to vc_insn_string_read() is "bool backwards" which in this case it appears will always be true?
Quoted 62 lines
arch/x86/kernel/sev-es-shared.c +58
-0+if (ret) +return ret; + } + + sw_scratch =__pa (ghcb) + offsetof(struct ghcb , shared_buffer); +ghcb_set_sw_scratch (ghcb, sw_scratch); + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_IOIO , + exit_info_1, exit_info_2); +if (ret !=ES_OK ) +return ret; + +/* Everything went well, write back results */ +if (exit_info_1 &IOIO_TYPE_IN ) { + ret =vc_insn_string_write (ctxt, + (void *)(es_base + regs->di), + ghcb->shared_buffer, io_bytes, + exit_info_2, df); +if (ret) +return ret; + +if (df) + regs->di -= exit_bytes; +else + regs->di += exit_bytes; + }else { +if (df) + regs->si -= exit_bytes; +else + regs->si += exit_bytes; + } + +if (exit_info_1 &IOIO_REP ) + regs->cx -= exit_info_2; + + ret = regs->cx ?ES_RETRY :ES_OK ; + + }else { +int bits = (exit_info_1 &0x70 ) >>1 ; +u64 rax =0 ; + +if (!(exit_info_1 &IOIO_TYPE_IN )) + rax =lower_bits (regs->ax, bits); + +ghcb_set_rax (ghcb, rax); + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_IOIO , exit_info_1,0 ); +if (ret !=ES_OK ) +return ret; + +if (exit_info_1 &IOIO_TYPE_IN ) { +if (!ghcb_is_valid_rax (ghcb)) +return ES_VMM_ERROR ; + regs->ax =lower_bits (ghcb->save.rax, bits); + } + } + +return ret; +}Signature · 1 line
2.17.1
Re: [PATCH 23/70] x86/sev-es: Add support for handling IOIO exceptions
Quoted 12 lines · David Rientjes
arch/x86/kernel/sev-es-shared.c static enum es_result vc_insn_string_write(struct es_em_ctxt *ctxt, · line 351
Quoted 8 lines · Joerg Roedel
arch/x86/kernel/sev-es-shared.c +7
-0+if (exit_info_1 &IOIO_TYPE_STR ) { +int df = (regs->flags &X86_EFLAGS_DF ) ?-1 :1 ; +if (!(exit_info_1 &IOIO_TYPE_IN )) { + ret =vc_insn_string_read (ctxt, + (void *)(es_base + regs->si), + ghcb->shared_buffer, io_bytes, + exit_info_2, df);The last argument to
vc_insn_string_read()is "bool backwards" which in this case it appears will always be true?
Right, thanks, good catch, I'll fix this. Seems to be a leftover from a previous version.
Regards,
Joerg
PATCH v2 24/70 x86/fpu: Move xgetbv()/xsetbv() into separate header
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The xgetbv() function is needed in pre-decompression boot code, but asm/fpu/internal.h can't be included there directly. Doing so opens the door to include-hell due to various include-magic in boot/compressed/misc.h.
Avoid that by moving xgetbv()/xsetbv() to a separate header file and include this instead.
--- arch/x86/include/asm/fpu/internal.h | 29 +------------------------- arch/x86/include/asm/fpu/xcr.h | 32 +++++++++++++++++++++++++++++ 2 files changed, 33 insertions(+), 28 deletions(-) create mode 100644 arch/x86/include/asm/fpu/xcr.h
arch/x86/include/asm/fpu/internal.h +1 -28
@@ -19,6 +19,7 @@#include <asm/user.h> #include <asm/fpu/api.h> #include <asm/fpu/xstate.h> +#include <asm/fpu/xcr.h> #include <asm/cpufeature.h> #include <asm/trace/fpu.h>
@@ -614,32 +615,4 @@ static inline void switch_fpu_finish(struct fpu *new_fpu) }__write_pkru (pkru_val); } - -/* -* MXCSR and XCR definitions: -*/ - -extern unsigned int mxcsr_feature_mask; - -#define XCR_XFEATURE_ENABLED_MASK 0x00000000 - -static inline u64 xgetbv (u32 index) -{ -u32 eax, edx; - - asmvolatile (".byte 0x0f,0x01,0xd0" /* xgetbv */ - :"=a" (eax),"=d" (edx) - :"c" (index)); -return eax + ((u64 )edx <<32 ); -} - -static inline void xsetbv (u32 index,u64 value) -{ -u32 eax = value; -u32 edx = value >>32 ; - - asmvolatile (".byte 0x0f,0x01,0xd1" /* xsetbv */ - : :"a" (eax),"d" (edx),"c" (index)); -} -#endif /* _ASM_X86_FPU_INTERNAL_H */
arch/x86/include/asm/fpu/xcr.h new file +32 -0
@@ -0,0 +1,32 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _ASM_X86_FPU_XCR_H +#define _ASM_X86_FPU_XCR_H + +/* +* MXCSR and XCR definitions: +*/ + +extern unsigned int mxcsr_feature_mask; + +#define XCR_XFEATURE_ENABLED_MASK 0x00000000 + +static inline u64 xgetbv (u32 index) +{ +u32 eax, edx; + + asmvolatile (".byte 0x0f,0x01,0xd0" /* xgetbv */ + :"=a" (eax),"=d" (edx) + :"c" (index)); +return eax + ((u64 )edx <<32 ); +} + +static inline void xsetbv (u32 index,u64 value) +{ +u32 eax = value; +u32 edx = value >>32 ; + + asmvolatile (".byte 0x0f,0x01,0xd1" /* xsetbv */ + : :"a" (eax),"d" (edx),"c" (index)); +} + +#endif /* _ASM_X86_FPU_XCR_H */
Signature · 1 line
2.17.1
PATCH v2 25/70 x86/sev-es: Add CPUID handling to #VC handler
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Handle #VC exceptions caused by CPUID instructions. These happen in early boot code when the KASLR code checks for RDTSC.
--- arch/x86/boot/compressed/sev-es.c | 4 ++++ arch/x86/kernel/sev-es-shared.c | 35 +++++++++++++++++++++++++++++++ 2 files changed, 39 insertions(+)
arch/x86/boot/compressed/sev-es.c +4 -0
@@ -10,6 +10,7 @@#include <asm/sev-es.h> #include <asm/trap_defs.h> #include <asm/msr-index.h> +#include <asm/fpu/xcr.h> #include <asm/ptrace.h> #include <asm/svm.h>
@@ -149,6 +150,9 @@ void boot_vc_handler(struct pt_regs *regs, unsigned long exit_code)case SVM_EXIT_IOIO : result =vc_handle_ioio (boot_ghcb, &ctxt);break ; +case SVM_EXIT_CPUID : + result =vc_handle_cpuid (boot_ghcb, &ctxt); +break ;default : result =ES_UNSUPPORTED ;break ;
arch/x86/kernel/sev-es-shared.c +35 -0
@@ -407,3 +407,38 @@ static enum es_result vc_handle_ioio(struct ghcb *ghcb, struct es_em_ctxt *ctxt)return ret; } + +static enum es_result vc_handle_cpuid (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +struct pt_regs *regs = ctxt->regs; +u32 cr4 =native_read_cr4 (); +enum es_result ret; + +ghcb_set_rax (ghcb, regs->ax); +ghcb_set_rcx (ghcb, regs->cx); + +if (cr4 &X86_CR4_OSXSAVE ) +/* Safe to read xcr0 */ +ghcb_set_xcr0 (ghcb,xgetbv (XCR_XFEATURE_ENABLED_MASK )); +else +/* xgetbv will cause #GP - use reset value for xcr0 */ +ghcb_set_xcr0 (ghcb,1 ); + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_CPUID ,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +if (!(ghcb_is_valid_rax (ghcb) && +ghcb_is_valid_rbx (ghcb) && +ghcb_is_valid_rcx (ghcb) && +ghcb_is_valid_rdx (ghcb))) +return ES_VMM_ERROR ; + + regs->ax = ghcb->save.rax; + regs->bx = ghcb->save.rbx; + regs->cx = ghcb->save.rcx; + regs->dx = ghcb->save.rdx; + +return ES_OK ; +}
Signature · 1 line
2.17.1
PATCH v2 26/70 x86/idt: Move IDT to data segment
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
With SEV-ES, exception handling is needed very early, even before the kernel has cleared the bss segment. In order to prevent clearing the currently used IDT, move the IDT to the data segment.
--- arch/x86/kernel/idt.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/kernel/idt.c +6 -2
@@ -165,8 +165,12 @@ static const __initconst struct idt_data dbg_idts[] = { };#endif -/* Must be page-aligned because the real IDT is used in a fixmap. */ -gate_desc idt_table[IDT_ENTRIES] __page_aligned_bss; +/* +* Must be page-aligned because the real IDT is used in a fixmap. +* Also needs to be in the .data segment, because the idt_table is +* needed before the kernel clears the .bss segment. +*/ +gate_desc idt_table[IDT_ENTRIES] __page_aligned_data;struct desc_ptr idt_descr__ro_after_init = { .size = (IDT_ENTRIES *2 *sizeof (unsigned long )) -1 ,
Signature · 1 line
2.17.1
PATCH v2 27/70 x86/idt: Split idt_data setup out of set_intr_gate()
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The code to setup idt_data is needed for early exception handling, but set_intr_gate() can't be used that early because it has pv-ops in its code path, which don't work that early.
Split out the idt_data initialization part from set_intr_gate() so that it can be used separatly.
--- arch/x86/kernel/idt.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)
arch/x86/kernel/idt.c +14 -8
@@ -231,18 +231,24 @@ idt_setup_from_table(gate_desc *idt, const struct idt_data *t, int size, bool sy } } +static void init_idt_data (struct idt_data *data,unsigned int n, +const void *addr) +{ +BUG_ON (n >0xFF ); + +memset (data,0 ,sizeof (*data)); + data->vector = n; + data->addr = addr; + data->segment = __KERNEL_CS; + data->bits.type =GATE_INTERRUPT ; + data->bits.p =1 ; +} +static void set_intr_gate (unsigned int n,const void *addr) {struct idt_data data; -BUG_ON (n >0xFF ); - -memset (&data,0 ,sizeof (data)); - data.vector = n; - data.addr = addr; - data.segment = __KERNEL_CS; - data.bits.type =GATE_INTERRUPT ; - data.bits.p =1 ; +init_idt_data (&data, n, addr);idt_setup_from_table (idt_table, &data,1 , false); }
Signature · 1 line
2.17.1
PATCH v2 28/70 x86/idt: Move two function from k/idt.c to i/a/desc.h
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Move these two functions from kernel/idt.c to include/asm/desc.h:
* init_idt_data()
* idt_init_desc()
These functions are needed to setup IDT entries very early and need to be called from head64.c. To be usable this early these functions need to be compiled without instrumentation and the stack-protector feature. These features need to be kept enabled for kernel/idt.c, so head64.c must use its own versions.
--- arch/x86/include/asm/desc.h | 27 +++++++++++++++++++++++++ arch/x86/include/asm/desc_defs.h | 7 +++++++ arch/x86/kernel/idt.c | 34 -------------------------------- 3 files changed, 34 insertions(+), 34 deletions(-)
arch/x86/include/asm/desc.h +27 -0
@@ -389,6 +389,33 @@ static inline void set_desc_limit(struct desc_struct *desc, unsigned long limit)void update_intr_gate (unsigned int n,const void *addr);void alloc_intr_gate (unsigned int n,const void *addr); +static inline void init_idt_data (struct idt_data *data,unsigned int n, +const void *addr) +{ +BUG_ON (n >0xFF ); + +memset (data,0 ,sizeof (*data)); + data->vector = n; + data->addr = addr; + data->segment = __KERNEL_CS; + data->bits.type =GATE_INTERRUPT ; + data->bits.p =1 ; +} + +static inline void idt_init_desc (gate_desc *gate,const struct idt_data *d) +{ +unsigned long addr = (unsigned long ) d->addr; + + gate->offset_low = (u16 ) addr; + gate->segment = (u16 ) d->segment; + gate->bits = d->bits; + gate->offset_middle = (u16) (addr >>16 ); +#ifdef CONFIG_X86_64 + gate->offset_high = (u32) (addr >>32 ); + gate->reserved =0 ; +#endif +} +extern unsigned long system_vectors[];#ifdef CONFIG_X86_64
arch/x86/include/asm/desc_defs.h +7 -0
@@ -74,6 +74,13 @@ struct idt_bits { p :1 ; } __attribute__((packed)); +struct idt_data { +unsigned int vector; +unsigned int segment; +struct idt_bits bits; +const void *addr; +}; +struct gate_struct {u16 offset_low;u16 segment;
arch/x86/kernel/idt.c +0 -34
@@ -9,13 +9,6 @@#include <asm/desc.h> #include <asm/hw_irq.h> -struct idt_data { -unsigned int vector; -unsigned int segment; -struct idt_bits bits; -const void *addr; -}; -#define DPL0 0x0#define DPL3 0x3
@@ -204,20 +197,6 @@ const struct desc_ptr debug_idt_descr = { };#endif -static inline void idt_init_desc (gate_desc *gate,const struct idt_data *d) -{ -unsigned long addr = (unsigned long ) d->addr; - - gate->offset_low = (u16 ) addr; - gate->segment = (u16 ) d->segment; - gate->bits = d->bits; - gate->offset_middle = (u16) (addr >>16 ); -#ifdef CONFIG_X86_64 - gate->offset_high = (u32) (addr >>32 ); - gate->reserved =0 ; -#endif -} -static void idt_setup_from_table (gate_desc *idt,const struct idt_data *t,int size,bool sys) {
@@ -231,19 +210,6 @@ idt_setup_from_table(gate_desc *idt, const struct idt_data *t, int size, bool sy } } -static void init_idt_data (struct idt_data *data,unsigned int n, -const void *addr) -{ -BUG_ON (n >0xFF ); - -memset (data,0 ,sizeof (*data)); - data->vector = n; - data->addr = addr; - data->segment = __KERNEL_CS; - data->bits.type =GATE_INTERRUPT ; - data->bits.p =1 ; -} -static void set_intr_gate (unsigned int n,const void *addr) {struct idt_data data;
Signature · 1 line
2.17.1
PATCH v2 29/70 x86/head/64: Install boot GDT
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Handling exceptions during boot requires a working GDT. The kernel GDT is not yet ready for use, so install a temporary boot GDT.
--- arch/x86/kernel/head_64.S | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)
arch/x86/kernel/head_64.S +32 -0
@@ -72,6 +72,26 @@ SYM_CODE_START_NOALIGN(startup_64) /* Set up the stack for verify_cpu(), similar to initial_stack below */ leaq (__end_init_task - SIZEOF_PTREGS)(%rip), %rsp + /* Setup boot GDT descriptor and load boot GDT */ + leaq boot_gdt(%rip), %rax + movq %rax, boot_gdt_base(%rip) + lgdt boot_gdt_descr(%rip) + + /* New GDT is live - reload data segment registers */ + movl $__KERNEL_DS, %eax + movl %eax, %ds + movl %eax, %ss + movl %eax, %es + + /* Now switch to __KERNEL_CS so IRET works reliably */ + pushq $__KERNEL_CS + leaq .Lon_kernel_cs(%rip), %rax + pushq %rax + lretq + +.Lon_kernel_cs: + UNWIND_HINT_EMPTY + /* Sanitize CPU configuration */ call verify_cpu
@@ -480,6 +500,18 @@ SYM_DATA_LOCAL(early_gdt_descr_base, .quad INIT_PER_CPU_VAR(gdt_page)) SYM_DATA(phys_base, .quad 0x0) EXPORT_SYMBOL(phys_base) +/* Boot GDT used when kernel addresses are not mapped yet */ +SYM_DATA_LOCAL(boot_gdt_descr, .word boot_gdt_end - boot_gdt) +SYM_DATA_LOCAL(boot_gdt_base, .quad 0) +SYM_DATA_START(boot_gdt) + .quad 0 + .quad 0x00cf9a000000ffff /* __KERNEL32_CS */ + .quad 0x00af9a000000ffff /* __KERNEL_CS */ + .quad 0x00cf92000000ffff /* __KERNEL_DS */ + .quad 0x0080890000000000 /* TS descriptor */ + .quad 0x0000000000000000 /* TS continued */ +SYM_DATA_END_LABEL(boot_gdt, SYM_L_LOCAL, boot_gdt_end) + #include "../../x86/xen/xen-head.S" __PAGE_ALIGNED_BSS
Signature · 1 line
2.17.1
PATCH v2 30/70 x86/head/64: Reload GDT after switch to virtual addresses
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Reload the GDT after switching to virtual addresses to make sure it will not go away when the lower mappings are removed.
--- arch/x86/kernel/head_64.S | 5 +++++ 1 file changed, 5 insertions(+)
arch/x86/kernel/head_64.S +5 -0
@@ -163,6 +163,11 @@ SYM_CODE_START(secondary_startup_64) 1: UNWIND_HINT_EMPTY + /* Setup boot GDT descriptor and load boot GDT */ + leaq boot_gdt(%rip), %rax + movq %rax, boot_gdt_base(%rip) + lgdt boot_gdt_descr(%rip) + /* Check if nx is implemented */ movl $0x80000001, %eax cpuid
Signature · 1 line
2.17.1
PATCH v2 31/70 x86/head/64: Load segment registers earlier
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Make sure segments are properly set up before setting up an IDT and doing anything that might cause a #VC exception. This is later needed for early exception handling.
--- arch/x86/kernel/head_64.S | 52 +++++++++++++++++++-------------------- 1 file changed, 26 insertions(+), 26 deletions(-)
arch/x86/kernel/head_64.S +26 -26
@@ -168,6 +168,32 @@ SYM_CODE_START(secondary_startup_64) movq %rax, boot_gdt_base(%rip) lgdt boot_gdt_descr(%rip) + /* set up data segments */ + xorl %eax,%eax + movl %eax,%ds + movl %eax,%ss + movl %eax,%es + + /* + * We don't really need to load %fs or %gs, but load them anyway + * to kill any stale realmode selectors. This allows execution + * under VT hardware. + */ + movl %eax,%fs + movl %eax,%gs + + /* Set up %gs. + * + * The base of %gs always points to fixed_percpu_data. If the + * stack protector canary is enabled, it is located at %gs:40. + * Note that, on SMP, the boot cpu uses init data section until + * the per cpu areas are set up. + */ + movl $MSR_GS_BASE,%ecx + movl initial_gs(%rip),%eax + movl initial_gs+4(%rip),%edx + wrmsr + /* Check if nx is implemented */ movl $0x80000001, %eax cpuid
@@ -203,32 +229,6 @@ SYM_CODE_START(secondary_startup_64) */ lgdt early_gdt_descr(%rip) - /* set up data segments */ - xorl %eax,%eax - movl %eax,%ds - movl %eax,%ss - movl %eax,%es - - /* - * We don't really need to load %fs or %gs, but load them anyway - * to kill any stale realmode selectors. This allows execution - * under VT hardware. - */ - movl %eax,%fs - movl %eax,%gs - - /* Set up %gs. - * - * The base of %gs always points to fixed_percpu_data. If the - * stack protector canary is enabled, it is located at %gs:40. - * Note that, on SMP, the boot cpu uses init data section until - * the per cpu areas are set up. - */ - movl $MSR_GS_BASE,%ecx - movl initial_gs(%rip),%eax - movl initial_gs+4(%rip),%edx - wrmsr - /* rsi is pointer to real mode structure with interesting info. pass it to C */ movq %rsi, %rdi
Signature · 1 line
2.17.1
PATCH v2 32/70 x86/head/64: Switch to initial stack earlier
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Make sure there is a stack once the kernel runs from virual addresses. At this stage any secondary CPU which boots will have lost its stack because the kernel switched to a new page-table which does not map the real-mode stack anymore.
This is also needed for handling early #VC exceptions caused by instructions like CPUID.
--- arch/x86/kernel/head_64.S | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)
arch/x86/kernel/head_64.S +6 -3
@@ -194,6 +194,12 @@ SYM_CODE_START(secondary_startup_64) movl initial_gs+4(%rip),%edx wrmsr + /* + * Setup a boot time stack - Any secondary CPU will have lost its stack + * by now because the cr3-switch above unmaps the real-mode stack + */ + movq initial_stack(%rip), %rsp + /* Check if nx is implemented */ movl $0x80000001, %eax cpuid
@@ -214,9 +220,6 @@ SYM_CODE_START(secondary_startup_64) /* Make changes effective */ movq %rax, %cr0 - /* Setup a boot time stack */ - movq initial_stack(%rip), %rsp - /* zero EFLAGS after setting rsp */ pushq $0 popfq
Signature · 1 line
2.17.1
PATCH v2 33/70 x86/head/64: Build k/head64.c with -fno-stack-protector
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The code inserted by the stack protector does not work in the early boot environment because it uses the GS segment, at least with memory encryption enabled. Make sure the early code is compiled without this feature enabled.
--- arch/x86/kernel/Makefile | 4 ++++ 1 file changed, 4 insertions(+)
arch/x86/kernel/Makefile +4 -0
@@ -36,6 +36,10 @@ ifdef CONFIG_FRAME_POINTER OBJECT_FILES_NON_STANDARD_ftrace_$(BITS).o := y endif +# make sure head64.c is built without stack protector +nostackp := $(call cc-option, -fno-stack-protector) +CFLAGS_head64.o := $(nostackp) + # If instrumentation of this dir is enabled, boot hangs during first second. # Probably could be more selective here, but note that files related to irqs, # boot, dumpstack/stacktrace, etc are either non-interesting or can lead to
Signature · 1 line
2.17.1
PATCH v2 34/70 x86/head/64: Load IDT earlier
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Load the IDT right after switching to virtual addresses in head_64.S so that the kernel can handle #VC exceptions.
--- arch/x86/kernel/head64.c | 15 +++++++++++++++ arch/x86/kernel/head_64.S | 17 +++++++++++++++++ 2 files changed, 32 insertions(+)
arch/x86/kernel/head64.c +15 -0
@@ -489,3 +489,18 @@ void __init x86_64_start_reservations(char *real_mode_data)start_kernel (); } + +void __head early_idt_setup_early_handler(unsigned long physaddr) +{ +gate_desc *idt =fixup_pointer (idt_table, physaddr); +int i; + +for (i =0 ; i <NUM_EXCEPTION_VECTORS ; i++) { +struct idt_data data; +gate_desc desc; + +init_idt_data (&data, i, early_idt_handler_array[i]); +idt_init_desc (&desc, &data); +native_write_idt_entry (idt, i, &desc); + } +}
arch/x86/kernel/head_64.S +17 -0
@@ -104,6 +104,20 @@ SYM_CODE_START_NOALIGN(startup_64) leaq _text(%rip), %rdi pushq %rsi call __startup_64 + /* Save return value */ + pushq %rax + + /* + * Load IDT with early handlers - needed for SEV-ES + * Do this here because this must only happen on the boot CPU + * and the code below is shared with secondary CPU bringup. + */ + leaq _text(%rip), %rdi + call early_idt_setup_early_handler + + /* Restore __startup_64 return value*/ + popq %rax + /* Restore pointer to real_mode_data */ popq %rsi /* Form the CR3 value being sure to include the CR3 modifier */
@@ -200,6 +214,9 @@ SYM_CODE_START(secondary_startup_64) */ movq initial_stack(%rip), %rsp + /* Load IDT */ + lidt idt_descr(%rip) + /* Check if nx is implemented */ movl $0x80000001, %eax cpuid
Signature · 1 line
2.17.1
PATCH v2 35/70 x86/head/64: Move early exception dispatch to C code
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Move the assembly coded dispatch between page-faults and all other exceptions to C code to make it easier to maintain and extend.
Also change the return-type of early_make_pgtable() to bool and make it static.
--- arch/x86/include/asm/pgtable.h | 2 +- arch/x86/include/asm/setup.h | 1 - arch/x86/kernel/head64.c | 19 +++++++++++++++---- arch/x86/kernel/head_64.S | 11 +---------- 4 files changed, 17 insertions(+), 16 deletions(-)
arch/x86/include/asm/pgtable.h +1 -1
@@ -27,7 +27,7 @@#include <asm/fpu/api.h> extern pgd_t early_top_pgt[PTRS_PER_PGD ]; -int __init __early_make_pgtable (unsigned long address,pmdval_t pmd); +bool __init __early_make_pgtable (unsigned long address,pmdval_t pmd);void ptdump_walk_pgd_level (struct seq_file *m,struct mm_struct *mm); void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
arch/x86/include/asm/setup.h +0 -1
@@ -48,7 +48,6 @@ extern void reserve_standard_io_resources(void);extern void i386_reserve_resources (void );extern unsigned long __startup_64 (unsigned long physaddr,struct boot_params *bp);extern unsigned long __startup_secondary_64 (void ); -extern int early_make_pgtable (unsigned long address);#ifdef CONFIG_X86_INTEL_MID extern void x86_intel_mid_early_setup (void );
arch/x86/kernel/head64.c +15 -4
@@ -36,6 +36,8 @@#include <asm/microcode.h> #include <asm/kasan.h> #include <asm/fixmap.h> +#include <asm/extable.h> +#include <asm/trap_defs.h> /* * Manage page tables very early on.
@@ -297,7 +299,7 @@ static void __init reset_early_page_tables(void) }/* Create a new PMD entry */ -int __init __early_make_pgtable (unsigned long address,pmdval_t pmd) +bool __init __early_make_pgtable (unsigned long address,pmdval_t pmd) {unsigned long physaddr = address - __PAGE_OFFSET;pgdval_t pgd, *pgd_p;
@@ -307,7 +309,7 @@ int __init __early_make_pgtable(unsigned long address, pmdval_t pmd)/* Invalid address or early pgt is done ? */ if (physaddr >=MAXMEM ||read_cr3_pa () !=__pa_nodebug (early_top_pgt)) -return -1 ; +return false; again: pgd_p = &early_top_pgt[pgd_index (address)].pgd;
@@ -364,10 +366,10 @@ int __init __early_make_pgtable(unsigned long address, pmdval_t pmd) } pmd_p[pmd_index (address)] = pmd; -return 0 ; +return true; } -int __init early_make_pgtable (unsigned long address) +static bool __init early_make_pgtable (unsigned long address) {unsigned long physaddr = address - __PAGE_OFFSET;pmdval_t pmd;
@@ -377,6 +379,15 @@ int __init early_make_pgtable(unsigned long address)return __early_make_pgtable (address, pmd); } +void __init early_exception (struct pt_regs *regs,int trapnr) +{ +if (trapnr ==X86_TRAP_PF && +early_make_pgtable (native_read_cr2 ())) +return ; + +early_fixup_exception (regs, trapnr); +} +/* Don't add a printk in there. printk relies on the PDA which is not initialized yet. */ static void __init clear_bss (void )
arch/x86/kernel/head_64.S +1 -10
@@ -363,18 +363,9 @@ SYM_CODE_START_LOCAL(early_idt_handler_common) pushq %r15 /* pt_regs->r15 */ UNWIND_HINT_REGS - cmpq $14,%rsi /* Page fault? */ - jnz 10f - GET_CR2_INTO(%rdi) /* can clobber %rax if pv */ - call early_make_pgtable - andl %eax,%eax - jz 20f /* All good */ - -10: movq %rsp,%rdi /* RDI = pt_regs; RSI is already trapnr */ - call early_fixup_exception + call early_exception -20: decl early_recursion_flag(%rip) jmp restore_regs_and_return_to_kernel SYM_CODE_END(early_idt_handler_common)
Signature · 1 line
2.17.1
PATCH v2 36/70 x86/sev-es: Add SEV-ES Feature Detection
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add the sev_es_active function for checking whether SEV-ES is enabled. Also cache the value of MSR_AMD64_SEV at boot to speed up the feature checking in the running code.
--- arch/x86/include/asm/mem_encrypt.h | 3 +++ arch/x86/include/asm/msr-index.h | 2 ++ arch/x86/mm/mem_encrypt.c | 11 ++++++++++- arch/x86/mm/mem_encrypt_identity.c | 3 +++ 4 files changed, 18 insertions(+), 1 deletion(-)
arch/x86/include/asm/mem_encrypt.h +3 -0
@@ -19,6 +19,7 @@#ifdef CONFIG_AMD_MEM_ENCRYPT extern u64 sme_me_mask; +extern u64 sev_status;extern bool sev_enabled; void sme_encrypt_execute(unsigned long encrypted_kernel_vaddr,
@@ -49,6 +50,7 @@ void __init mem_encrypt_free_decrypted_mem(void);bool sme_active (void );bool sev_active (void ); +bool sev_es_active (void );#define __bss_decrypted __attribute__((__section__(".bss..decrypted")))
@@ -71,6 +73,7 @@ static inline void __init sme_enable(struct boot_params *bp) { }static inline bool sme_active (void ) {return false; }static inline bool sev_active (void ) {return false; } +static inline bool sev_es_active (void ) {return false; }static inline int __init early_set_memory_decrypted (unsigned long vaddr,unsigned long size) {return 0 ; }
arch/x86/include/asm/msr-index.h +2 -0
@@ -435,7 +435,9 @@#define MSR_AMD64_SEV_ES_GHCB 0xc0010130#define MSR_AMD64_SEV 0xc0010131#define MSR_AMD64_SEV_ENABLED_BIT 0 +#define MSR_AMD64_SEV_ES_ENABLED_BIT 1#define MSR_AMD64_SEV_ENABLED BIT_ULL(MSR_AMD64_SEV_ENABLED_BIT) +#define MSR_AMD64_SEV_ES_ENABLED BIT_ULL(MSR_AMD64_SEV_ES_ENABLED_BIT)#define MSR_AMD64_VIRT_SPEC_CTRL 0xc001011f
arch/x86/mm/mem_encrypt.c +10 -1
@@ -38,7 +38,9 @@* section is later cleared. */ u64 sme_me_mask __section(.data) = 0; +u64 sev_status __section(.data) = 0;EXPORT_SYMBOL (sme_me_mask); +EXPORT_SYMBOL (sev_status);DEFINE_STATIC_KEY_FALSE (sev_enable_key);EXPORT_SYMBOL_GPL (sev_enable_key);
@@ -347,9 +349,16 @@ bool sme_active(void)bool sev_active (void ) { -return sme_me_mask && sev_enabled; +return !!(sev_status &MSR_AMD64_SEV_ENABLED ); } +bool sev_es_active (void ) +{ +return !!(sev_status &MSR_AMD64_SEV_ES_ENABLED ); +} +EXPORT_SYMBOL_GPL (sev_es_active); + +/* Override for DMA direct allocation check - ARCH_HAS_FORCE_DMA_UNENCRYPTED */ bool force_dma_unencrypted (struct device *dev) {
arch/x86/mm/mem_encrypt_identity.c +3 -0
@@ -540,6 +540,9 @@ void __init sme_enable(struct boot_params *bp)if (!(msr &MSR_AMD64_SEV_ENABLED ))return ; +/* Save SEV_STATUS to avoid reading MSR again */ + sev_status = msr; +/* SEV state cannot be controlled by a command line option */ sme_me_mask = me_mask; sev_enabled = true;
Signature · 1 line
2.17.1
PATCH v2 37/70 x86/sev-es: Compile early handler code into kernel image
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Setup sev-es.c and include the code from the pre-decompression stage to also build it into the image of the running kernel. Temporarily add __maybe_unused annotations to avoid build warnings until the functions get used.
--- arch/x86/kernel/Makefile | 1 + arch/x86/kernel/sev-es-shared.c | 21 +++-- arch/x86/kernel/sev-es.c | 162 ++++++++++++++++++++++++++++++++ 3 files changed, 174 insertions(+), 10 deletions(-) create mode 100644 arch/x86/kernel/sev-es.c
arch/x86/kernel/Makefile +1 -0
@@ -147,6 +147,7 @@ obj-$(CONFIG_UNWINDER_ORC) += unwind_orc.o obj-$(CONFIG_UNWINDER_FRAME_POINTER) += unwind_frame.o obj-$(CONFIG_UNWINDER_GUESS) += unwind_guess.o +obj-$(CONFIG_AMD_MEM_ENCRYPT) += sev-es.o ### # 64 bit specific files ifeq ($(CONFIG_X86_64),y)
arch/x86/kernel/sev-es-shared.c +11 -10
@@ -9,7 +9,7 @@* and is included directly into both code-bases. */ -static void sev_es_terminate (unsigned int reason) +static void __maybe_unused sev_es_terminate (unsigned int reason) {/* Request Guest Termination from Hypvervisor */ sev_es_wr_ghcb_msr (GHCB_SEV_TERMINATE );
@@ -19,7 +19,7 @@ static void sev_es_terminate(unsigned int reason) asmvolatile ("hlt\n" : : :"memory" ); } -static bool sev_es_negotiate_protocol (void ) +static bool __maybe_unused sev_es_negotiate_protocol (void ) {u64 val;
@@ -38,7 +38,7 @@ static bool sev_es_negotiate_protocol(void)return true; } -static void vc_ghcb_invalidate (struct ghcb *ghcb) +static void __maybe_unused vc_ghcb_invalidate (struct ghcb *ghcb) {memset (ghcb->save.valid_bitmap,0 ,sizeof (ghcb->save.valid_bitmap)); }
@@ -50,9 +50,9 @@ static bool vc_decoding_needed(unsigned long exit_code) exit_code <= SVM_EXIT_LAST_EXCP); } -static enum es_result vc_init_em_ctxt (struct es_em_ctxt *ctxt, -struct pt_regs *regs, -unsigned long exit_code) +static enum es_result __maybe_unused vc_init_em_ctxt (struct es_em_ctxt *ctxt, +struct pt_regs *regs, +unsigned long exit_code) {enum es_result ret =ES_OK ;
@@ -65,7 +65,7 @@ static enum es_result vc_init_em_ctxt(struct es_em_ctxt *ctxt,return ret; } -static void vc_finish_insn (struct es_em_ctxt *ctxt) +static void __maybe_unused vc_finish_insn (struct es_em_ctxt *ctxt) { ctxt->regs->ip += ctxt->insn.length; }
@@ -312,7 +312,8 @@ static enum es_result vc_ioio_exitinfo(struct es_em_ctxt *ctxt, u64 *exitinfo)return ES_OK ; } -static enum es_result vc_handle_ioio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +static enum es_result __maybe_unused +vc_handle_ioio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) {struct pt_regs *regs = ctxt->regs;u64 exit_info_1, exit_info_2;
@@ -408,8 +409,8 @@ static enum es_result vc_handle_ioio(struct ghcb *ghcb, struct es_em_ctxt *ctxt)return ret; } -static enum es_result vc_handle_cpuid (struct ghcb *ghcb, -struct es_em_ctxt *ctxt) +static enum es_result __maybe_unused vc_handle_cpuid (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) {struct pt_regs *regs = ctxt->regs;u32 cr4 =native_read_cr4 ();
arch/x86/kernel/sev-es.c new file +162 -0
162 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 38/70 x86/sev-es: Setup early #VC handler
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Setup an early handler for #VC exceptions. There is no GHCB mapped yet, so just re-use the vc_no_ghcb_handler. It can only handle CPUID exit-codes, but that should be enough to get the kernel through verify_cpu() and __startup_64() until it runs on virtual addresses.
--- arch/x86/include/asm/desc.h | 1 + arch/x86/include/asm/processor.h | 1 + arch/x86/include/asm/sev-es.h | 2 ++ arch/x86/kernel/head64.c | 17 +++++++++++++++ arch/x86/kernel/head_64.S | 36 ++++++++++++++++++++++++++++++++ arch/x86/kernel/idt.c | 10 +++++++++ 6 files changed, 67 insertions(+)
arch/x86/include/asm/desc.h +1 -0
@@ -388,6 +388,7 @@ static inline void set_desc_limit(struct desc_struct *desc, unsigned long limit)void update_intr_gate (unsigned int n,const void *addr);void alloc_intr_gate (unsigned int n,const void *addr); +void set_early_idt_handler (gate_desc *idt,int n,void *handler);static inline void init_idt_data (struct idt_data *data,unsigned int n,const void *addr)
arch/x86/include/asm/processor.h +1 -0
@@ -768,6 +768,7 @@ extern int sysenter_setup(void);/* Defined in head.S */ extern struct desc_ptr early_gdt_descr; +extern struct desc_ptr early_idt_descr;extern void switch_to_new_gdt (int );extern void load_direct_gdt (int );
arch/x86/include/asm/sev-es.h +2 -0
@@ -75,4 +75,6 @@ static inline u64 copy_lower_bits(u64 out, u64 in, unsigned int bits) return out; } +extern void early_vc_handler (void ); +#endif
arch/x86/kernel/head64.c +17 -0
@@ -38,6 +38,7 @@#include <asm/fixmap.h> #include <asm/extable.h> #include <asm/trap_defs.h> +#include <asm/sev-es.h> /* * Manage page tables very early on.
@@ -515,3 +516,19 @@ void __head early_idt_setup_early_handler(unsigned long physaddr)native_write_idt_entry (idt, i, &desc); } } + +void __head early_idt_setup(unsigned long physbase) +{ +gate_desc *idt =fixup_pointer (idt_table, physbase); +void __maybe_unused *handler; + +#ifdef CONFIG_AMD_MEM_ENCRYPT +/* VMM Communication Exception */ + handler =fixup_pointer (early_vc_handler, physbase); +set_early_idt_handler (idt,X86_TRAP_VC , handler); +#endif + +/* Initialize IDT descriptor and load IDT */ + early_idt_descr.address = (unsigned long )idt; +native_load_idt (&early_idt_descr); +}
arch/x86/kernel/head_64.S +36 -0
@@ -92,6 +92,12 @@ SYM_CODE_START_NOALIGN(startup_64) .Lon_kernel_cs: UNWIND_HINT_EMPTY + /* Setup IDT - Needed for SEV-ES */ + leaq _text(%rip), %rdi + pushq %rsi + call early_idt_setup + popq %rsi + /* Sanitize CPU configuration */ call verify_cpu
@@ -370,6 +376,33 @@ SYM_CODE_START_LOCAL(early_idt_handler_common) jmp restore_regs_and_return_to_kernel SYM_CODE_END(early_idt_handler_common) +#ifdef CONFIG_AMD_MEM_ENCRYPT +/* + * VC Exception handler used during very early boot. The + * early_idt_handler_array can't be used because it returns via the + * paravirtualized INTERRUPT_RETURN and pv-ops don't work that early. + */ +SYM_CODE_START_NOALIGN(early_vc_handler) + UNWIND_HINT_IRET_REGS offset=8 + + /* Build pt_regs */ + PUSH_AND_CLEAR_REGS + + /* Call C handler */ + movq %rsp, %rdi + movq ORIG_RAX(%rsp), %rsi + call vc_no_ghcb_handler + + /* Unwind pt_regs */ + POP_REGS + + /* Remove Error Code */ + addq $8, %rsp + + /* Pure iret required here - don't use INTERRUPT_RETURN */ + iretq +SYM_CODE_END(early_vc_handler) +#endif #define SYM_DATA_START_PAGE_ALIGNED(name) \ SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
@@ -511,6 +544,9 @@ SYM_DATA_END(level1_fixmap_pgt) SYM_DATA(early_gdt_descr, .word GDT_ENTRIES*8-1) SYM_DATA_LOCAL(early_gdt_descr_base, .quad INIT_PER_CPU_VAR(gdt_page)) +SYM_DATA(early_idt_descr, .word NUM_EXCEPTION_VECTORS * 16) +SYM_DATA_LOCAL(early_idt_descr_base, .quad 0) + .align 16 /* This must match the first entry in level2_kernel_pgt */ SYM_DATA(phys_base, .quad 0x0)
arch/x86/kernel/idt.c +10 -0
@@ -341,3 +341,13 @@ void alloc_intr_gate(unsigned int n, const void *addr)if (!test_and_set_bit (n, system_vectors))set_intr_gate (n, addr); } + +void set_early_idt_handler (gate_desc *idt,int n,void *handler) +{ +struct idt_data data; +gate_desc desc; + +init_idt_data (&data, n, handler); +idt_init_desc (&desc, &data); +native_write_idt_entry (idt, n, &desc); +}
Signature · 1 line
2.17.1
PATCH v2 39/70 x86/sev-es: Setup GHCB based boot #VC handler
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add the infrastructure to handle #VC exceptions when the kernel runs on virtual addresses and has a GHCB mapped. This handler will be used until the runtime #VC handler takes over.
--- arch/x86/include/asm/segment.h | 2 +- arch/x86/include/asm/sev-es.h | 1 + arch/x86/kernel/head64.c | 6 ++ arch/x86/kernel/sev-es-shared.c | 14 ++-- arch/x86/kernel/sev-es.c | 116 ++++++++++++++++++++++++++++++++ arch/x86/mm/extable.c | 1 + 6 files changed, 132 insertions(+), 8 deletions(-)
arch/x86/include/asm/segment.h +1 -1
@@ -230,7 +230,7 @@#define NUM_EXCEPTION_VECTORS 32/* Bitmask of exception vectors which push an error code on the stack: */ -#define EXCEPTION_ERRCODE_MASK 0x00027d00 +#define EXCEPTION_ERRCODE_MASK 0x20027d00#define GDT_SIZE (GDT_ENTRIES*8)#define GDT_ENTRY_TLS_ENTRIES 3
arch/x86/include/asm/sev-es.h +1 -0
@@ -76,5 +76,6 @@ static inline u64 copy_lower_bits(u64 out, u64 in, unsigned int bits) }extern void early_vc_handler (void ); +extern bool boot_vc_exception (struct pt_regs *regs);#endif
arch/x86/kernel/head64.c +6 -0
@@ -386,6 +386,12 @@ void __init early_exception(struct pt_regs *regs, int trapnr) early_make_pgtable(native_read_cr2()))return ; +#ifdef CONFIG_AMD_MEM_ENCRYPT +if (trapnr ==X86_TRAP_VC && +boot_vc_exception (regs)) +return ; +#endif +early_fixup_exception (regs, trapnr); }
arch/x86/kernel/sev-es-shared.c +7 -7
@@ -9,7 +9,7 @@* and is included directly into both code-bases. */ -static void __maybe_unused sev_es_terminate (unsigned int reason) +static void sev_es_terminate (unsigned int reason) {/* Request Guest Termination from Hypvervisor */ sev_es_wr_ghcb_msr (GHCB_SEV_TERMINATE );
@@ -19,7 +19,7 @@ static void __maybe_unused sev_es_terminate(unsigned int reason) asmvolatile ("hlt\n" : : :"memory" ); } -static bool __maybe_unused sev_es_negotiate_protocol (void ) +static bool sev_es_negotiate_protocol (void ) {u64 val;
@@ -38,7 +38,7 @@ static bool __maybe_unused sev_es_negotiate_protocol(void)return true; } -static void __maybe_unused vc_ghcb_invalidate (struct ghcb *ghcb) +static void vc_ghcb_invalidate (struct ghcb *ghcb) {memset (ghcb->save.valid_bitmap,0 ,sizeof (ghcb->save.valid_bitmap)); }
@@ -50,9 +50,9 @@ static bool vc_decoding_needed(unsigned long exit_code) exit_code <= SVM_EXIT_LAST_EXCP); } -static enum es_result __maybe_unused vc_init_em_ctxt (struct es_em_ctxt *ctxt, -struct pt_regs *regs, -unsigned long exit_code) +static enum es_result vc_init_em_ctxt (struct es_em_ctxt *ctxt, +struct pt_regs *regs, +unsigned long exit_code) {enum es_result ret =ES_OK ;
@@ -65,7 +65,7 @@ static enum es_result __maybe_unused vc_init_em_ctxt(struct es_em_ctxt *ctxt,return ret; } -static void __maybe_unused vc_finish_insn (struct es_em_ctxt *ctxt) +static void vc_finish_insn (struct es_em_ctxt *ctxt) { ctxt->regs->ip += ctxt->insn.length; }
arch/x86/kernel/sev-es.c +116 -0
134 lines not on this page · raw message
arch/x86/mm/extable.c +1 -0
@@ -5,6 +5,7 @@#include <xen/xen.h> #include <asm/fpu/internal.h> +#include <asm/sev-es.h> #include <asm/traps.h> #include <asm/kdebug.h>
Signature · 1 line
2.17.1
PATCH v2 40/70 x86/sev-es: Setup per-cpu GHCBs for the runtime handler
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
The runtime handler needs a GHCB per CPU. Set them up and map them unencrypted.
--- arch/x86/include/asm/mem_encrypt.h | 2 ++ arch/x86/kernel/sev-es.c | 28 +++++++++++++++++++++++++++- arch/x86/kernel/traps.c | 3 +++ 3 files changed, 32 insertions(+), 1 deletion(-)
arch/x86/include/asm/mem_encrypt.h +2 -0
@@ -48,6 +48,7 @@ int __init early_set_memory_encrypted(unsigned long vaddr, unsigned long size);void __init mem_encrypt_init (void );void __init mem_encrypt_free_decrypted_mem (void ); +void __init sev_es_init_ghcbs (void );bool sme_active (void );bool sev_active (void );bool sev_es_active (void );
@@ -71,6 +72,7 @@ static inline void __init sme_early_init(void) { }static inline void __init sme_encrypt_kernel (struct boot_params *bp) { }static inline void __init sme_enable (struct boot_params *bp) { } +static inline void sev_es_init_ghcbs (void ) { }static inline bool sme_active (void ) {return false; }static inline bool sev_active (void ) {return false; }static inline bool sev_es_active (void ) {return false; }
arch/x86/kernel/sev-es.c +27 -1
@@ -8,8 +8,11 @@*/ #include <linux/sched/debug.h> /* For show_regs() */ -#include <linux/kernel.h> +#include <linux/percpu-defs.h> +#include <linux/mem_encrypt.h> #include <linux/printk.h> +#include <linux/set_memory.h> +#include <linux/kernel.h> #include <linux/mm.h> #include <asm/trap_defs.h>
@@ -29,6 +32,9 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);*/ struct ghcb __initdata *boot_ghcb; +/* Runtime GHCB pointers */ +static struct ghcb __percpu *ghcb_page; +/* Needed in vc_early_vc_forward_exception */ extern void early_exception (struct pt_regs *regs,int trapnr);
@@ -197,6 +203,26 @@ static bool __init sev_es_setup_ghcb(void)return true; } +void sev_es_init_ghcbs (void ) +{ +int cpu; + +if (!sev_es_active ()) +return ; + +/* Allocate GHCB pages */ + ghcb_page =__alloc_percpu (sizeof (struct ghcb ),PAGE_SIZE ); + +/* Initialize per-cpu GHCB pages */ +for_each_possible_cpu (cpu) { +struct ghcb *ghcb = (struct ghcb *)per_cpu_ptr (ghcb_page, cpu); + +set_memory_decrypted ((unsigned long )ghcb, +sizeof (*ghcb) >>PAGE_SHIFT ); +memset (ghcb,0 ,sizeof (*ghcb)); + } +} +static void __init vc_early_vc_forward_exception (struct es_em_ctxt *ctxt) {int trapnr = ctxt->fi.vector;
arch/x86/kernel/traps.c +3 -0
@@ -918,6 +918,9 @@ void __init trap_init(void)/* Init cpu_entry_area before IST entries are set up */ setup_cpu_entry_areas (); +/* Init GHCB memory pages when running as an SEV-ES guest */ +sev_es_init_ghcbs (); +idt_setup_traps ();/*
Signature · 1 line
2.17.1
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 32 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
The runtime handler needs a GHCB per CPU. Set them up and map them unencrypted.
Signed-off-by Tom Lendacky <[email protected]>Signed-off-by Joerg Roedel <[email protected]>---
arch/x86/include/asm/mem_encrypt.h | 2 ++ arch/x86/kernel/sev-es.c | 28 +++++++++++++++++++++++++++- arch/x86/kernel/traps.c | 3 +++ 3 files changed, 32 insertions(+), 1 deletion(-)arch/x86/kernel/sev-es.c +10
-0@@ -197,6 +203,26 @@ static bool __init sev_es_setup_ghcb(void)return true; } +void sev_es_init_ghcbs (void ) +{ +int cpu; + +if (!sev_es_active ()) +return ; + +/* Allocate GHCB pages */ + ghcb_page =__alloc_percpu (sizeof (struct ghcb ),PAGE_SIZE ); +
arch/x86/kernel/sev-es.c static bool __init sev_es_setup_ghcb(void) · line 225
Quoted 10 lines
arch/x86/kernel/sev-es.c +10
-0+/* Initialize per-cpu GHCB pages */ +for_each_possible_cpu (cpu) { +struct ghcb *ghcb = (struct ghcb *)per_cpu_ptr (ghcb_page, cpu); + +set_memory_decrypted ((unsigned long )ghcb, +sizeof (*ghcb) >>PAGE_SHIFT ); +memset (ghcb,0 ,sizeof (*ghcb)); + } +} +
set_memory_decrypted needs to check the return value. I see it consistently return ENOMEM. I've traced that back to split_large_page in arch/x86/mm/pat/set_memory.c.
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 39 lines · Mike Stunes
Quoted 38 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
The runtime handler needs a GHCB per CPU. Set them up and map them unencrypted.
Signed-off-by Tom Lendacky <[email protected]>Signed-off-by Joerg Roedel <[email protected]>---
arch/x86/include/asm/mem_encrypt.h | 2 ++ arch/x86/kernel/sev-es.c | 28 +++++++++++++++++++++++++++- arch/x86/kernel/traps.c | 3 +++ 3 files changed, 32 insertions(+), 1 deletion(-)arch/x86/kernel/sev-es.c +16
-0@@ -197,6 +203,26 @@ static bool __init sev_es_setup_ghcb(void)return true; } +void sev_es_init_ghcbs (void ) +{ +int cpu; + +if (!sev_es_active ()) +return ; + +/* Allocate GHCB pages */ + ghcb_page =__alloc_percpu (sizeof (struct ghcb ),PAGE_SIZE ); + +/* Initialize per-cpu GHCB pages */ +for_each_possible_cpu (cpu) { +struct ghcb *ghcb = (struct ghcb *)per_cpu_ptr (ghcb_page, cpu); + +set_memory_decrypted ((unsigned long )ghcb, +sizeof (*ghcb) >>PAGE_SHIFT );
Quoted 8 lines
arch/x86/kernel/sev-es.c static bool __init sev_es_setup_ghcb(void) · line 225
Quoted 4 lines
arch/x86/kernel/sev-es.c +4
-0+memset (ghcb,0 ,sizeof (*ghcb)); + } +} +
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.
At that point the guest won't be able to communicate with the hypervisor, too. Maybe we should BUG() here to terminate further processing?
Thanks,
Tom
Quoted 1 line
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 7 lines · Tom Lendacky
Quoted 3 lines
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.At that point the guest won't be able to communicate with the hypervisor, too. Maybe we should
BUG()here to terminate further processing?
Escalating an -ENOMEM into a crashed kernel seems a bit extreme. Granted, the guest may be in an unrecoverable state, but the host doesn't need to be too.
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 5 lines · Dave Hansen
Quoted 4 lines · Tom Lendacky
Quoted 3 lines
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.
Quoted 8 lines
Quoted 3 lines
At that point the guest won't be able to communicate with the hypervisor, too. Maybe we should
BUG()here to terminate further processing?Escalating an
-ENOMEMinto a crashed kernel seems a bit extreme. Granted, the guest may be in an unrecoverable state, but the host doesn't need to be too.
The host wouldn't be. This only happens in a guest, so it would be just causing the guest kernel to panic early in the boot.
Thanks,
Tom
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 12 lines · Tom Lendacky
Quoted 9 lines · Dave Hansen
Quoted 7 lines · Tom Lendacky
Quoted 3 lines
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.At that point the guest won't be able to communicate with the hypervisor, too. Maybe we should
BUG()here to terminate further processing?
Quoted 6 lines
Quoted 4 lines
Escalating an
-ENOMEMinto a crashed kernel seems a bit extreme. Granted, the guest may be in an unrecoverable state, but the host doesn't need to be too.The host wouldn't be. This only happens in a guest, so it would be just causing the guest kernel to panic early in the boot.
And I should add that it would only impact an SEV-ES guest.
Thanks,
Tom
Quoted 4 lines
Thanks,
Tom
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
In reply to Tom LendackyQuoted 2 lines · Tom Lendacky
At that point the guest won't be able to communicate with the hypervisor, too. Maybe we should
BUG()here to terminate further processing?
We could talk to the hypervisor, there is still the boot-GHCB in the bss-decrypted section. But there is nothing that could be done here anyway besides terminating the guest.
Regards,
Joerg
Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
In reply to Mike StunesHi Mike,
Quoted 3 lines · Mike Stunes
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.
I agree that the return code needs to be checked. But I wonder why this happens. The split_large_page() function returns -ENOMEM when alloc_pages() fails. Do you boot the guest with minal RAM assigned?
Regards,
Joerg
Re: Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 7 lines · Joerg Roedel
Hi Mike,
Quoted 3 lines · Mike Stunes
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.
Quoted 8 lines
I agree that the return code needs to be checked. But I wonder why this happens. The
split_large_page()function returns-ENOMEMwhenalloc_pages()fails. Do you boot the guest with minal RAM assigned?Regards,
Joerg
I just want to add some context around this. The call path that lead to the failure is like the following:
__alloc_pages_slowpath __alloc_pages_nodemask alloc_pages_current alloc_pages split_large_page __change_page_attr __change_page_attr_set_clr __set_memory_enc_dec set_memory_decrypted sev_es_init_ghcbs trap_init -> before mm_init (in init/main.c) start_kernel x86_64_start_reservations x86_64_start_kernel secondary_startup_64
At this time, mem_init hasn't been called yet (which would be called by mm_init). Thus, the free pages are still owned by memblock. It's in mem_init (x86/mm/init_64.c) that memblock_free_all gets called and free pages are released.
During testing, I've also noticed that debug_pagealloc=1 will make the issue disappear. That's because with debug_pagealloc=1, probe_page_size_mask in x86/mm/init.c will not allow large pages (2M/1G). Therefore, no split_large_page would happen. Similarly, if CPU doesn't have X86_FEATURE_PSE, there won't be large pages either.
Any thoughts? Maybe split_large_page should get pages from memblock at early boot?
Bo
Re: Re: [PATCH 40/70] x86/sev-es: Setup per-cpu GHCBs for the runtime handler
Quoted 41 lines · Bo Gan
Quoted 7 lines · Joerg Roedel
Hi Mike,
Quoted 3 lines · Mike Stunes
set_memory_decryptedneeds to check the return value. I see it consistently return ENOMEM. I've traced that back tosplit_large_pageinarch/x86/mm/pat/set_memory.c.Quoted 8 lines
I agree that the return code needs to be checked. But I wonder why this happens. The
split_large_page()function returns-ENOMEMwhenalloc_pages()fails. Do you boot the guest with minal RAM assigned?Regards,
Joerg
I just want to add some context around this. The call path that lead to the failure is like the following:
__alloc_pages_slowpath __alloc_pages_nodemask alloc_pages_current alloc_pages split_large_page __change_page_attr __change_page_attr_set_clr __set_memory_enc_dec set_memory_decrypted sev_es_init_ghcbs trap_init -> before mm_init (in init/main.c) start_kernel x86_64_start_reservations x86_64_start_kernel secondary_startup_64At this time,
mem_inithasn't been called yet (which would be called bymm_init). Thus, the free pages are still owned by memblock. It's inmem_init(x86/mm/init_64.c) thatmemblock_free_allgets called and free pages are released.
Quoted 8 lines
During testing, I've also noticed that debug_pagealloc=1 will make the issue disappear. That's because with debug_pagealloc=1,
probe_page_size_maskinx86/mm/init.cwill not allow large pages (2M/1G). Therefore, nosplit_large_pagewould happen. Similarly, if CPU doesn't haveX86_FEATURE_PSE, there won't be large pages either.Any thoughts? Maybe
split_large_pageshould get pages from memblock at early boot?
Thanks for you analysis. I fixed it (verified by Mike) by using early_set_memory_decrypted() instead of set_memory_decrypted(). I still wonder why I didn't see that issue on my kernel. It has DEBUG_PAGEALLOC=y set, but it is not enabled by default and I also didn't pass the command-line parameter.
Regards,
Joerg
PATCH v2 41/70 x86/sev-es: Add Runtime #VC Exception Handler
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Add the handler for #VC exceptions invoked at runtime.
--- arch/x86/entry/entry_64.S | 4 ++ arch/x86/include/asm/traps.h | 7 ++++ arch/x86/kernel/idt.c | 4 +- arch/x86/kernel/sev-es.c | 77 +++++++++++++++++++++++++++++++++++- 4 files changed, 90 insertions(+), 2 deletions(-)
arch/x86/entry/entry_64.S +4 -0
@@ -1210,6 +1210,10 @@ idtentry async_page_fault do_async_page_fault has_error_code=1 read_cr2=1 idtentry machine_check do_mce has_error_code=0 paranoid=1 #endif +#ifdef CONFIG_AMD_MEM_ENCRYPT +idtentry vmm_communication do_vmm_communication has_error_code=1 +#endif + /* * Save all registers in pt_regs, and switch gs if needed. * Use slow, but surefire "are we in kernel?" check.
arch/x86/include/asm/traps.h +7 -0
@@ -35,6 +35,9 @@ asmlinkage void alignment_check(void);#ifdef CONFIG_X86_MCE asmlinkage void machine_check(void);#endif /* CONFIG_X86_MCE */ +#ifdef CONFIG_AMD_MEM_ENCRYPT +asmlinkage void vmm_communication(void); +#endif asmlinkage void simd_coprocessor_error(void);#if defined(CONFIG_X86_64 ) && defined(CONFIG_XEN_PV )
@@ -93,6 +96,10 @@ dotraplinkage void do_alignment_check(struct pt_regs *regs, long error_code); dotraplinkage void do_machine_check(struct pt_regs *regs, long error_code);#endif dotraplinkage void do_simd_coprocessor_error(struct pt_regs *regs, long error_code); +#ifdef CONFIG_AMD_MEM_ENCRYPT +dotraplinkage void do_vmm_communication_error(struct pt_regs *regs, +long error_code); +#endif #ifdef CONFIG_X86_32 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code);#endif
arch/x86/kernel/idt.c +3 -1
@@ -88,8 +88,10 @@ static const __initconst struct idt_data def_idts[] = {#ifdef CONFIG_X86_MCE INTG(X86_TRAP_MC, &machine_check), #endif - SYSG(X86_TRAP_OF, overflow), +#ifdef CONFIG_AMD_MEM_ENCRYPT + INTG(X86_TRAP_VC, vmm_communication), +#endif #if defined(CONFIG_IA32_EMULATION ) SYSG(IA32_SYSCALL_VECTOR, entry_INT80_compat),#elif defined(CONFIG_X86_32 )
arch/x86/kernel/sev-es.c +76 -1
89 lines not on this page · raw message
Signature · 1 line
2.17.1
Andy Lutomirski
maintainerRe: [PATCH 41/70] x86/sev-es: Add Runtime #VC Exception Handler
Quoted 4 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
Add the handler for #VC exceptions invoked at runtime.
If I read this correctly, this does not use IST. If that's true, I don't see how this can possibly work. There at least two nasty cases that come to mind:
1. SYSCALL followed by NMI. The NMI IRET hack gets to #VC and we explode. This is fixable by getting rid of the NMI EFLAGS.TF hack.
2. tools/testing/selftests/x86/mov_ss_trap_64. User code does MOV (addr), SS; SYSCALL, where addr has a data breakpoint. We get #DB promoted to #VC with no stack.
Re: [PATCH 41/70] x86/sev-es: Add Runtime #VC Exception Handler
Quoted 4 lines · Andy Lutomirski
Quoted 3 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
Quoted 8 lines
Quoted 1 line
Add the handler for #VC exceptions invoked at runtime.
If I read this correctly, this does not use IST. If that's true, I don't see how this can possibly work. There at least two nasty cases that come to mind:
1. SYSCALL followed by NMI. The NMI IRET hack gets to #VC and we explode. This is fixable by getting rid of the NMI EFLAGS.TF hack.
Not an issue in this patch-set, the confusion comes from the fact that I left some parts of the single-step-over-iret code in the patch. But it is not used. The NMI handling in this patch-set sends the NMI-complete message before the IRET, when the kernel is still in a safe environment (kernel stack, kernel cr3).
Quoted 3 lines
2.
tools/testing/selftests/x86/mov_ss_trap_64. User code does MOV (addr), SS; SYSCALL, where addr has a data breakpoint. We get #DB promoted to #VC with no stack.
Also not an issue, as debugging is not supported at the moment in SEV-ES guests (hardware has no way yet to save/restore the debug registers across #VMEXITs). But this will change with future hardware. If you look at the implementation for dr7 read/write events, you see that the dr7 value is cached and returned, but does not make it to the hardware dr7.
I though about using IST for the #VC handler, but the implications for nesting #VC handlers made me decide against it. But for future hardware that supports debugging inside SEV-ES guests it will be an issue. I'll think about how to fix the problem, it probably has to be IST :(
Regards,
Joerg
Andy Lutomirski
maintainerRe: [PATCH 41/70] x86/sev-es: Add Runtime #VC Exception Handler
Quoted 12 lines · Joerg Roedel
Quoted 10 lines · Andy Lutomirski
Quoted 4 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
Add the handler for #VC exceptions invoked at runtime.
If I read this correctly, this does not use IST. If that's true, I don't see how this can possibly work. There at least two nasty cases that come to mind:
Quoted 8 lines
Quoted 2 lines
1. SYSCALL followed by NMI. The NMI IRET hack gets to #VC and we explode. This is fixable by getting rid of the NMI EFLAGS.TF hack.
Not an issue in this patch-set, the confusion comes from the fact that I left some parts of the single-step-over-iret code in the patch. But it is not used. The NMI handling in this patch-set sends the NMI-complete message before the IRET, when the kernel is still in a safe environment (kernel stack, kernel cr3).
Got it!
Quoted 10 lines
Quoted 4 lines
2.
tools/testing/selftests/x86/mov_ss_trap_64. User code does MOV (addr), SS; SYSCALL, where addr has a data breakpoint. We get #DB promoted to #VC with no stack.Also not an issue, as debugging is not supported at the moment in SEV-ES guests (hardware has no way yet to save/restore the debug registers across #VMEXITs). But this will change with future hardware. If you look at the implementation for dr7 read/write events, you see that the dr7 value is cached and returned, but does not make it to the hardware dr7.
Eek. This would probably benefit from some ptrace / perf logic to prevent the kernel or userspace from thinking that debugging works.
I guess this means that #DB only happens due to TF or INT01. I suppose this is probably okay.
Quoted 5 lines
I though about using IST for the #VC handler, but the implications for nesting #VC handlers made me decide against it. But for future hardware that supports debugging inside SEV-ES guests it will be an issue. I'll think about how to fix the problem, it probably has to be IST :(
Or future generations could have enough hardware support for debugging that #DB doesn't need to be intercepted or can be re-injected correctly with the #DB vector.
Quoted 4 lines
Regards,
Joerg
Re: [PATCH 41/70] x86/sev-es: Add Runtime #VC Exception Handler
Quoted 3 lines · Andy Lutomirski
Or future generations could have enough hardware support for debugging that #DB doesn't need to be intercepted or can be re-injected correctly with the #DB vector.
Yeah, the problem is, the GHCB spec suggests the single-step-over-iret way to re-enable the NMI window and requires intercepting #DB for it. So the hypervisor probably still has to intercept it, even when debug support is added some day. I need to think more about this.
Regards,
Joerg
PATCH v2 42/70 x86/sev-es: Support nested #VC exceptions
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Handle #VC exceptions that happen while the GHCB is in use. This can happen when an NMI happens in the #VC exception handler and the NMI handler causes a #VC exception itself. Save the contents of the GHCB when nesting is detected and restore it when the GHCB is no longer used.
--- arch/x86/kernel/sev-es.c | 63 +++++++++++++++++++++++++++++++++++++--- 1 file changed, 59 insertions(+), 4 deletions(-)
arch/x86/kernel/sev-es.c +59 -4
@@ -32,9 +32,57 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);*/ struct ghcb __initdata *boot_ghcb; +struct ghcb_state { +struct ghcb *ghcb; +}; +/* Runtime GHCB pointers */ static struct ghcb __percpu *ghcb_page; +/* +* Mark the per-cpu GHCB as in-use to detect nested #VC exceptions. +* There is no need for it to be atomic, because nothing is written to the GHCB +* between the read and the write of ghcb_active. So it is safe to use it when a +* nested #VC exception happens before the write. +*/ +static DEFINE_PER_CPU(bool, ghcb_active); + +static struct ghcb *sev_es_get_ghcb (struct ghcb_state *state) +{ +struct ghcb *ghcb = (struct ghcb *)this_cpu_ptr (ghcb_page); +bool *active =this_cpu_ptr (&ghcb_active); + +if (unlikely (*active)) { +/* GHCB is already in use - save its contents */ + + state->ghcb =kzalloc (sizeof (struct ghcb ),GFP_ATOMIC ); +if (!state->ghcb) +return NULL ; + + *state->ghcb = *ghcb; + }else { + state->ghcb =NULL ; + *active = true; + } + +return ghcb; +} + +static void sev_es_put_ghcb (struct ghcb_state *state) +{ +bool *active =this_cpu_ptr (&ghcb_active); +struct ghcb *ghcb = (struct ghcb *)this_cpu_ptr (ghcb_page); + +if (state->ghcb) { +/* Restore saved state and free backup memory */ + *ghcb = *state->ghcb; +kfree (state->ghcb); + state->ghcb =NULL ; + }else { + *active = false; + } +} +/* Needed in vc_early_vc_forward_exception */ extern void early_exception (struct pt_regs *regs,int trapnr);
@@ -272,6 +320,7 @@ static void vc_forward_exception(struct es_em_ctxt *ctxt) dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exit_code) { +struct ghcb_state state;struct es_em_ctxt ctxt;enum es_result result;struct ghcb *ghcb;
@@ -282,14 +331,20 @@ dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exit* keep the IRQs disabled to protect us against concurrent TLB flushes. */ - ghcb = (struct ghcb *)this_cpu_ptr (ghcb_page); - -vc_ghcb_invalidate (ghcb); - result =vc_init_em_ctxt (&ctxt, regs, exit_code); + ghcb =sev_es_get_ghcb (&state); +if (!ghcb) { +/* This can only fail on an allocation error, so just retry */ + result =ES_RETRY ; + }else { +vc_ghcb_invalidate (ghcb); + result =vc_init_em_ctxt (&ctxt, regs, exit_code); + }if (result ==ES_OK ) result =vc_handle_exitcode (&ctxt, ghcb, exit_code); +sev_es_put_ghcb (&state); +/* Done - now check the result */ switch (result) {case ES_OK :
Signature · 1 line
2.17.1
Andy Lutomirski
maintainerRe: [PATCH 42/70] x86/sev-es: Support nested #VC exceptions
Quoted 38 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Handle #VC exceptions that happen while the GHCB is in use. This can happen when an NMI happens in the #VC exception handler and the NMI handler causes a #VC exception itself. Save the contents of the GHCB when nesting is detected and restore it when the GHCB is no longer used.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/kernel/sev-es.c | 63 +++++++++++++++++++++++++++++++++++++--- 1 file changed, 59 insertions(+), 4 deletions(-)arch/x86/kernel/sev-es.c +13
-0@@ -32,9 +32,57 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);*/ struct ghcb __initdata *boot_ghcb; +struct ghcb_state { +struct ghcb *ghcb; +}; +/* Runtime GHCB pointers */ static struct ghcb __percpu *ghcb_page; +/* +* Mark the per-cpu GHCB as in-use to detect nested #VC exceptions. +* There is no need for it to be atomic, because nothing is written to the GHCB +* between the read and the write of ghcb_active. So it is safe to use it when a +* nested #VC exception happens before the write. +*/ +static DEFINE_PER_CPU(bool, ghcb_active); + +static struct ghcb *sev_es_get_ghcb (struct ghcb_state *state)
arch/x86/kernel/sev-es.c struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE); · line 60
Quoted 10 lines
arch/x86/kernel/sev-es.c +10
-0+{ +struct ghcb *ghcb = (struct ghcb *)this_cpu_ptr (ghcb_page); +bool *active =this_cpu_ptr (&ghcb_active); + +if (unlikely (*active)) { +/* GHCB is already in use - save its contents */ + + state->ghcb =kzalloc (sizeof (struct ghcb ),GFP_ATOMIC ); +if (!state->ghcb) +return NULL ;
This can't possibly end well. Maybe have a little percpu list of GHCBs and make sure there are enough for any possible nesting?
Also, I admit confusion. Isn't the GHCB required to be unencrypted? How does that work with kzalloc()?
Re: [PATCH 42/70] x86/sev-es: Support nested #VC exceptions
Quoted 2 lines · Andy Lutomirski
This can't possibly end well. Maybe have a little percpu list of GHCBs and make sure there are enough for any possible nesting?
Yeah, it is not entirely robust yet. Without NMI nesting the number of possible #VC nesting levels should be limited. At least one backup GHCB pre-allocated is probably a good idea.
Quoted 2 lines
Also, I admit confusion. Isn't the GHCB required to be unencrypted? How does that work with
kzalloc()?
Yes, but the kzalloc'ed ghcb is just the backup space for the real GHCB, which is mapped unencrypted. The contents of the unencrypted GHCB is copied to the backup and restored on return, so that the interrupted #VC handler finds the GHCB unmodified.
Regards,
Joerg
PATCH v2 43/70 x86/sev-es: Wire up existing #VC exit-code handlers
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Re-use the handlers for CPUID and IOIO caused #VC exceptions in the early boot handler.
--- arch/x86/kernel/sev-es-shared.c | 7 +++---- arch/x86/kernel/sev-es.c | 6 ++++++ 2 files changed, 9 insertions(+), 4 deletions(-)
arch/x86/kernel/sev-es-shared.c +3 -4
@@ -312,8 +312,7 @@ static enum es_result vc_ioio_exitinfo(struct es_em_ctxt *ctxt, u64 *exitinfo)return ES_OK ; } -static enum es_result __maybe_unused -vc_handle_ioio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +static enum es_result vc_handle_ioio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) {struct pt_regs *regs = ctxt->regs;u64 exit_info_1, exit_info_2;
@@ -409,8 +408,8 @@ vc_handle_ioio(struct ghcb *ghcb, struct es_em_ctxt *ctxt)return ret; } -static enum es_result __maybe_unused vc_handle_cpuid (struct ghcb *ghcb, -struct es_em_ctxt *ctxt) +static enum es_result vc_handle_cpuid (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) {struct pt_regs *regs = ctxt->regs;u32 cr4 =native_read_cr4 ();
arch/x86/kernel/sev-es.c +6 -0
@@ -289,6 +289,12 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,enum es_result result;switch (exit_code) { +case SVM_EXIT_CPUID : + result =vc_handle_cpuid (ghcb, ctxt); +break ; +case SVM_EXIT_IOIO : + result =vc_handle_ioio (ghcb, ctxt); +break ;default :/* * Unexpected #VC exception
Signature · 1 line
2.17.1
PATCH v2 44/70 x86/sev-es: Handle instruction fetches from user-space
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
When a #VC exception is triggered by user-space the instruction decoder needs to read the instruction bytes from user addresses. Enhance vc_decode_insn() to safely fetch kernel and user instructions.
--- arch/x86/kernel/sev-es.c | 31 ++++++++++++++++++++++--------- 1 file changed, 22 insertions(+), 9 deletions(-)
arch/x86/kernel/sev-es.c +22 -9
@@ -114,17 +114,30 @@ static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)enum es_result ret;int res; - res =vc_fetch_insn_kernel (ctxt, buffer); -if (unlikely (res == -EFAULT )) { - ctxt->fi.vector =X86_TRAP_PF ; - ctxt->fi.error_code =0 ; - ctxt->fi.cr2 = ctxt->regs->ip; -return ES_EXCEPTION ; +if (!user_mode (ctxt->regs)) { + res =vc_fetch_insn_kernel (ctxt, buffer); +if (unlikely (res == -EFAULT )) { + ctxt->fi.vector =X86_TRAP_PF ; + ctxt->fi.error_code =0 ; + ctxt->fi.cr2 = ctxt->regs->ip; +return ES_EXCEPTION ; + } + +insn_init (&ctxt->insn, buffer,MAX_INSN_SIZE - res,1 ); +insn_get_length (&ctxt->insn); + }else { + res =insn_fetch_from_user (ctxt->regs, buffer); +if (res ==0 ) { + ctxt->fi.vector =X86_TRAP_PF ; + ctxt->fi.cr2 = ctxt->regs->ip; + ctxt->fi.error_code =X86_PF_INSTR |X86_PF_USER ; +return ES_EXCEPTION ; + } + +if (!insn_decode (ctxt->regs, &ctxt->insn, buffer, res)) +return ES_DECODE_FAILED ; } -insn_init (&ctxt->insn, buffer,MAX_INSN_SIZE - res,1 ); -insn_get_length (&ctxt->insn); - ret = ctxt->insn.immediate.got ?ES_OK :ES_DECODE_FAILED ;return ret;
Signature · 1 line
2.17.1
PATCH v2 45/70 x86/sev-es: Harden runtime #VC handler for exceptions from user-space
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Send SIGBUS to the user-space process that caused the #VC exception instead of killing the machine. Also ratelimit the error messages so that user-space can't flood the kernel log.
--- arch/x86/kernel/sev-es.c | 32 +++++++++++++++++++++++--------- 1 file changed, 23 insertions(+), 9 deletions(-)
arch/x86/kernel/sev-es.c +23 -9
@@ -370,16 +370,16 @@ dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exitvc_finish_insn (&ctxt);break ;case ES_UNSUPPORTED : -pr_emerg ("Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n" , - exit_code, regs->ip); +pr_err_ratelimited ("Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n" , + exit_code, regs->ip); goto fail;case ES_VMM_ERROR : -pr_emerg ("PANIC: Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n" , - exit_code, regs->ip); +pr_err_ratelimited ("Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n" , + exit_code, regs->ip); goto fail;case ES_DECODE_FAILED : -pr_emerg ("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n" , - exit_code, regs->ip); +pr_err_ratelimited ("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n" , + exit_code, regs->ip); goto fail;case ES_EXCEPTION :vc_forward_exception (&ctxt);
@@ -394,10 +394,24 @@ dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exitreturn ; fail: -show_regs (regs); +if (user_mode (regs)) { +/* +* Do not kill the machine if user-space triggered the +* exception. Send SIGBUS instead and let user-space deal with +* it. +*/ +force_sig_fault (SIGBUS ,BUS_OBJERR , (void __user *)0 ); + }else { +/* Show some debug info */ +show_regs (regs); -while (true) -halt (); +/* Ask hypervisor to sev_es_terminate */ +sev_es_terminate (GHCB_SEV_ES_REASON_GENERAL_REQUEST ); + +/* If that fails and we get here - just halt the machine */ +while (true) +halt (); + } }bool __init boot_vc_exception (struct pt_regs *regs)
Signature · 1 line
2.17.1
PATCH v2 46/70 x86/sev-es: Filter exceptions not supported from user-space
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Currently only CPUID caused #VC exceptions are supported from user-space. Filter the others out early.
--- arch/x86/kernel/sev-es.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)
arch/x86/kernel/sev-es.c +24 -0
@@ -318,6 +318,26 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,return result; } +static enum es_result vc_context_filter (struct pt_regs *regs,long exit_code) +{ +enum es_result r =ES_OK ; + +if (user_mode (regs)) { +switch (exit_code) { +/* List of #VC exit-codes we support in user-space */ + case SVM_EXIT_EXCP_BASE ... SVM_EXIT_LAST_EXCP: +case SVM_EXIT_CPUID : + r =ES_OK ; +break ; +default : + r =ES_UNSUPPORTED ; +break ; + } + } + +return r; +} +static void vc_forward_exception (struct es_em_ctxt *ctxt) {long error_code = ctxt->fi.error_code;
@@ -359,6 +379,10 @@ dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exit result =vc_init_em_ctxt (&ctxt, regs, exit_code); } +/* Check if the exception is supported in the context we came from. */ +if (result ==ES_OK ) + result =vc_context_filter (regs, exit_code); +if (result ==ES_OK ) result =vc_handle_exitcode (&ctxt, ghcb, exit_code);
Signature · 1 line
2.17.1
PATCH v2 47/70 x86/sev-es: Handle MMIO events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Add handler for VC exceptions caused by MMIO intercepts. These intercepts come along as nested page faults on pages with reserved bits set.
TODO:
- Add return values of helper functions
- Check permissions on page-table walks
- Check data segments
--- arch/x86/boot/compressed/sev-es.c | 5 + arch/x86/include/uapi/asm/svm.h | 5 + arch/x86/kernel/sev-es.c | 188 ++++++++++++++++++++++++++++++ 3 files changed, 198 insertions(+)
arch/x86/boot/compressed/sev-es.c +5 -0
@@ -100,6 +100,11 @@ static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,return ES_OK ; } +static phys_addr_t vc_slow_virt_to_phys (struct ghcb *ghcb,long vaddr) +{ +return (phys_addr_t )vaddr; +} +#undef __init#undef __pa#define __init
arch/x86/include/uapi/asm/svm.h +5 -0
@@ -81,6 +81,11 @@#define SVM_EXIT_AVIC_INCOMPLETE_IPI 0x401#define SVM_EXIT_AVIC_UNACCELERATED_ACCESS 0x402 +/* SEV-ES software-defined VMGEXIT events */ +#define SVM_VMGEXIT_MMIO_READ 0x80000001 +#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_UNSUPPORTED_EVENT 0x8000ffff +#define SVM_EXIT_ERR -1#define SVM_EXIT_REASONS \
arch/x86/kernel/sev-es.c +188 -0
206 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 48/70 x86/sev-es: Handle MMIO String Instructions
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add handling for emulation the MOVS instruction on MMIO regions, as done by the memcpy_toio() and memcpy_fromio() functions.
--- arch/x86/kernel/sev-es.c | 78 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 78 insertions(+)
arch/x86/kernel/sev-es.c +78 -0
@@ -415,6 +415,74 @@ static enum es_result vc_handle_mmio_twobyte_ops(struct ghcb *ghcb,return ret; } +/* +* The MOVS instruction has two memory operands, which raises the +* problem that it is not known whether the access to the source or the +* destination caused the #VC exception (and hence whether an MMIO read +* or write operation needs to be emulated). +* +* Instead of playing games with walking page-tables and trying to guess +* whether the source or destination is an MMIO range, this code splits +* the move into two operations, a read and a write with only one +* memory operand. This will cause a nested #VC exception on the MMIO +* address which can then be handled. +* +* This implementation has the benefit that it also supports MOVS where +* source _and_ destination are MMIO regions. +* +* It will slow MOVS on MMIO down a lot, but in SEV-ES guests it is a +* rare operation. If it turns out to be a performance problem the split +* operations can be moved to memcpy_fromio() and memcpy_toio(). +*/ +static enum es_result vc_handle_mmio_movs (struct es_em_ctxt *ctxt, +unsigned int bytes) +{ +unsigned long ds_base, es_base; +unsigned char *src, *dst; +unsigned char buffer[8 ]; +enum es_result ret; +bool rep; +int off; + + ds_base =insn_get_seg_base (ctxt->regs,INAT_SEG_REG_DS ); + es_base =insn_get_seg_base (ctxt->regs,INAT_SEG_REG_ES ); + +if (ds_base ==-1L || es_base ==-1L ) { + ctxt->fi.vector =X86_TRAP_GP ; + ctxt->fi.error_code =0 ; +return ES_EXCEPTION ; + } + + src = ds_base + (unsigned char *)ctxt->regs->si; + dst = es_base + (unsigned char *)ctxt->regs->di; + + ret =vc_read_mem (ctxt, src, buffer, bytes); +if (ret !=ES_OK ) +return ret; + + ret =vc_write_mem (ctxt, dst, buffer, bytes); +if (ret !=ES_OK ) +return ret; + +if (ctxt->regs->flags &X86_EFLAGS_DF ) + off = -bytes; +else + off = bytes; + + ctxt->regs->si += off; + ctxt->regs->di += off; + + rep =insn_rep_prefix (&ctxt->insn); + +if (rep) + ctxt->regs->cx -=1 ; + +if (!rep || ctxt->regs->cx ==0 ) +return ES_OK ; +else +return ES_RETRY ; +} +static enum es_result vc_handle_mmio (struct ghcb *ghcb,struct es_em_ctxt *ctxt) {
@@ -469,6 +537,16 @@ static enum es_result vc_handle_mmio(struct ghcb *ghcb,memcpy (reg_data, ghcb->shared_buffer, bytes);break ; +/* MOVS instruction */ +case 0xa4 : + bytes =1 ; +/* Fallthrough */ +case 0xa5 : +if (!bytes) + bytes = insn->opnd_bytes; + + ret =vc_handle_mmio_movs (ctxt, bytes); +break ;/* Two-Byte Opcodes */ case 0x0f : ret =vc_handle_mmio_twobyte_ops (ghcb, ctxt);
Signature · 1 line
2.17.1
PATCH v2 49/70 x86/sev-es: Handle MSR events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by RDMSR/WRMSR instructions.
--- arch/x86/kernel/sev-es.c | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+)
arch/x86/kernel/sev-es.c +32 -0
@@ -262,6 +262,35 @@ static phys_addr_t vc_slow_virt_to_phys(struct ghcb *ghcb, long vaddr)/* Include code shared with pre-decompression boot stage */ #include "sev-es-shared.c" +static enum es_result vc_handle_msr (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +{ +struct pt_regs *regs = ctxt->regs; +enum es_result ret; +bool write; +u64 exit_info_1; + + write = (ctxt->insn.opcode.bytes[1 ] ==0x30 ); + +ghcb_set_rcx (ghcb, regs->cx); +if (write) { +ghcb_set_rax (ghcb, regs->ax); +ghcb_set_rdx (ghcb, regs->dx); + exit_info_1 =1 ; + }else { + exit_info_1 =0 ; + } + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_MSR , exit_info_1,0 ); +if (ret !=ES_OK ) +return ret; +else if (!write) { + regs->ax = ghcb->save.rax; + regs->dx = ghcb->save.rdx; + } + +return ret; +} +/* * This function runs on the first #VC exception after the kernel * switched to virtual addresses.
@@ -571,6 +600,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_IOIO : result =vc_handle_ioio (ghcb, ctxt);break ; +case SVM_EXIT_MSR : + result =vc_handle_msr (ghcb, ctxt); +break ;case SVM_EXIT_NPF : result =vc_handle_mmio (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 50/70 x86/sev-es: Handle DR7 read/write events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Add code to handle #VC exceptions on DR7 register reads and writes. This is needed early because show_regs() reads DR7 to print it out.
[ [email protected]: - Adapt to #VC handling framework
- Support early usage ]
--- arch/x86/kernel/sev-es.c | 87 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 83 insertions(+), 4 deletions(-)
arch/x86/kernel/sev-es.c +83 -4
@@ -23,6 +23,8 @@#include <asm/traps.h> #include <asm/svm.h> +#define DR7_RESET_VALUE 0x400 +/* For early boot hypervisor communication in SEV-ES enabled guests */ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);
@@ -31,6 +33,7 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);* cleared */ struct ghcb __initdata *boot_ghcb; +static DEFINE_PER_CPU(unsigned long, cached_dr7) = DR7_RESET_VALUE;struct ghcb_state {struct ghcb *ghcb;
@@ -359,6 +362,21 @@ static long *vc_insn_get_reg(struct es_em_ctxt *ctxt)return reg_array + offset; } +static long *vc_insn_get_rm (struct es_em_ctxt *ctxt) +{ +long *reg_array; +int offset; + + reg_array = (long *)ctxt->regs; + offset =insn_get_modrm_rm_off (&ctxt->insn, ctxt->regs); + +if (offset <0 ) +return NULL ; + + offset /=sizeof (long ); + +return reg_array + offset; +}static enum es_result vc_do_mmio (struct ghcb *ghcb,struct es_em_ctxt *ctxt,unsigned int bytes,bool read) {
@@ -587,13 +605,74 @@ static enum es_result vc_handle_mmio(struct ghcb *ghcb,return ret; } +static enum es_result vc_handle_dr7_write (struct ghcb *ghcb, +struct es_em_ctxt *ctxt, +bool early) +{ +long val, *reg =vc_insn_get_rm (ctxt); +enum es_result ret; + +if (!reg) +return ES_DECODE_FAILED ; + + val = *reg; + +/* Upper 32 bits must be written as zeroes */ +if (val >>32 ) { + ctxt->fi.vector =X86_TRAP_GP ; + ctxt->fi.error_code =0 ; +return ES_EXCEPTION ; + } + +/* Clear out other reservered bits and set bit 10 */ + val = (val &0xffff23ffL ) |BIT (10 ); + +/* Early non-zero writes to DR7 are not supported */ +if (early && (val & ~DR7_RESET_VALUE )) +return ES_UNSUPPORTED ; + +/* Using a value of 0 for ExitInfo1 means RAX holds the value */ +ghcb_set_rax (ghcb, val); + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_WRITE_DR7 ,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +this_cpu_write (cached_dr7, *reg); + +return ES_OK ; +} + +static enum es_result vc_handle_dr7_read (struct ghcb *ghcb, +struct es_em_ctxt *ctxt, +bool early) +{ +long *reg =vc_insn_get_rm (ctxt); + +if (!reg) +return ES_DECODE_FAILED ; + +if (early) + *reg =DR7_RESET_VALUE ; +else + *reg =this_cpu_read (cached_dr7); + +return ES_OK ; +} +static enum es_result vc_handle_exitcode (struct es_em_ctxt *ctxt, -struct ghcb *ghcb, -unsigned long exit_code) +struct ghcb *ghcb, +unsigned long exit_code, +bool early) {enum es_result result;switch (exit_code) { +case SVM_EXIT_READ_DR7 : + result =vc_handle_dr7_read (ghcb, ctxt, early); +break ; +case SVM_EXIT_WRITE_DR7 : + result =vc_handle_dr7_write (ghcb, ctxt, early); +break ;case SVM_EXIT_CPUID : result =vc_handle_cpuid (ghcb, ctxt);break ;
@@ -682,7 +761,7 @@ dotraplinkage void do_vmm_communication(struct pt_regs *regs, unsigned long exit result =vc_context_filter (regs, exit_code);if (result ==ES_OK ) - result =vc_handle_exitcode (&ctxt, ghcb, exit_code); + result =vc_handle_exitcode (&ctxt, ghcb, exit_code, false);sev_es_put_ghcb (&state);
@@ -750,7 +829,7 @@ bool __init boot_vc_exception(struct pt_regs *regs) result =vc_init_em_ctxt (&ctxt, regs, exit_code);if (result ==ES_OK ) - result =vc_handle_exitcode (&ctxt, boot_ghcb, exit_code); + result =vc_handle_exitcode (&ctxt, boot_ghcb, exit_code, true);/* Done - now check the result */ switch (result) {
Signature · 1 line
2.17.1
PATCH v2 51/70 x86/sev-es: Handle WBINVD Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by WBINVD instructions.
--- arch/x86/kernel/sev-es.c | 9 +++++++++ 1 file changed, 9 insertions(+)
arch/x86/kernel/sev-es.c +9 -0
@@ -659,6 +659,12 @@ static enum es_result vc_handle_dr7_read(struct ghcb *ghcb,return ES_OK ; } +static enum es_result vc_handle_wbinvd (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_WBINVD ,0 ,0 ); +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -682,6 +688,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_MSR : result =vc_handle_msr (ghcb, ctxt);break ; +case SVM_EXIT_WBINVD : + result =vc_handle_wbinvd (ghcb, ctxt); +break ;case SVM_EXIT_NPF : result =vc_handle_mmio (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 52/70 x86/sev-es: Handle RDTSC Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by RDTSC instructions.
--- arch/x86/kernel/sev-es.c | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)
arch/x86/kernel/sev-es.c +20 -0
@@ -665,6 +665,23 @@ static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_WBINVD ,0 ,0 ); } +static enum es_result vc_handle_rdtsc (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +{ +enum es_result ret; + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_RDTSC ,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb))) +return ES_VMM_ERROR ; + + ctxt->regs->ax = ghcb->save.rax; + ctxt->regs->dx = ghcb->save.rdx; + +return ES_OK ; +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -679,6 +696,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_WRITE_DR7 : result =vc_handle_dr7_write (ghcb, ctxt, early);break ; +case SVM_EXIT_RDTSC : + result =vc_handle_rdtsc (ghcb, ctxt); +break ;case SVM_EXIT_CPUID : result =vc_handle_cpuid (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 53/70 x86/sev-es: Handle RDPMC Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by RDPMC instructions.
--- arch/x86/kernel/sev-es.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)
arch/x86/kernel/sev-es.c +22 -0
@@ -682,6 +682,25 @@ static enum es_result vc_handle_rdtsc(struct ghcb *ghcb, struct es_em_ctxt *ctxtreturn ES_OK ; } +static enum es_result vc_handle_rdpmc (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +{ +enum es_result ret; + +ghcb_set_rcx (ghcb, ctxt->regs->cx); + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_RDPMC ,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb))) +return ES_VMM_ERROR ; + + ctxt->regs->ax = ghcb->save.rax; + ctxt->regs->dx = ghcb->save.rdx; + +return ES_OK ; +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -699,6 +718,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_RDTSC : result =vc_handle_rdtsc (ghcb, ctxt);break ; +case SVM_EXIT_RDPMC : + result =vc_handle_rdpmc (ghcb, ctxt); +break ;case SVM_EXIT_CPUID : result =vc_handle_cpuid (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 54/70 x86/sev-es: Handle INVD Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by INVD instructions. Since Linux should never use INVD, just mark it as unsupported.
--- arch/x86/kernel/sev-es.c | 4 ++++ 1 file changed, 4 insertions(+)
arch/x86/kernel/sev-es.c +4 -0
@@ -721,6 +721,10 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_RDPMC : result =vc_handle_rdpmc (ghcb, ctxt);break ; +case SVM_EXIT_INVD : +pr_err_ratelimited ("#VC exception for INVD??? Seriously???\n" ); + result =ES_UNSUPPORTED ; +break ;case SVM_EXIT_CPUID : result =vc_handle_cpuid (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 55/70 x86/sev-es: Handle RDTSCP Events
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Extend the RDTSC handler to also handle RDTSCP events.
--- arch/x86/kernel/sev-es.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)
arch/x86/kernel/sev-es.c +11 -4
@@ -665,19 +665,25 @@ static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_WBINVD ,0 ,0 ); } -static enum es_result vc_handle_rdtsc (struct ghcb *ghcb,struct es_em_ctxt *ctxt) +static enum es_result vc_handle_rdtsc (struct ghcb *ghcb, +struct es_em_ctxt *ctxt, +unsigned long exit_code) { +bool rdtscp = (exit_code ==SVM_EXIT_RDTSCP );enum es_result ret; - ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_RDTSC ,0 ,0 ); + ret =sev_es_ghcb_hv_call (ghcb, ctxt, exit_code,0 ,0 );if (ret !=ES_OK )return ret; -if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb))) +if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb) && + (!rdtscp ||ghcb_is_valid_rcx (ghcb))))return ES_VMM_ERROR ; ctxt->regs->ax = ghcb->save.rax; ctxt->regs->dx = ghcb->save.rdx; +if (rdtscp) + ctxt->regs->cx = ghcb->save.rcx;return ES_OK ; }
@@ -716,7 +722,8 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, result =vc_handle_dr7_write (ghcb, ctxt, early);break ;case SVM_EXIT_RDTSC : - result =vc_handle_rdtsc (ghcb, ctxt); +case SVM_EXIT_RDTSCP : + result =vc_handle_rdtsc (ghcb, ctxt, exit_code);break ;case SVM_EXIT_RDPMC : result =vc_handle_rdpmc (ghcb, ctxt);
Signature · 1 line
2.17.1
PATCH 1/1 Allow RDTSC and RDTSCP from userspace
Hi Joerg,
I needed to allow RDTSC(P) from userspace and in early boot in order to get userspace started properly. Patch below.
---
SEV-ES guests will need to execute rdtsc and rdtscp from userspace and during early boot. Move the rdtsc(p) #VC handler into common code and extend the #VC handlers.
--- arch/x86/boot/compressed/sev-es.c | 4 ++++ arch/x86/kernel/sev-es-shared.c | 23 +++++++++++++++++++++++ arch/x86/kernel/sev-es.c | 25 ++----------------------- 3 files changed, 29 insertions(+), 23 deletions(-)
arch/x86/boot/compressed/sev-es.c +4 -0
@@ -158,6 +158,10 @@ void boot_vc_handler(struct pt_regs *regs, unsigned long exit_code)case SVM_EXIT_CPUID : result =vc_handle_cpuid (boot_ghcb, &ctxt);break ; +case SVM_EXIT_RDTSC : +case SVM_EXIT_RDTSCP : + result =vc_handle_rdtsc (boot_ghcb, &ctxt, exit_code); +break ;default : result =ES_UNSUPPORTED ;break ;
arch/x86/kernel/sev-es-shared.c +23 -0
@@ -442,3 +442,26 @@ static enum es_result vc_handle_cpuid(struct ghcb *ghcb,return ES_OK ; } + +static enum es_result vc_handle_rdtsc (struct ghcb *ghcb, +struct es_em_ctxt *ctxt, +unsigned long exit_code) +{ +bool rdtscp = (exit_code ==SVM_EXIT_RDTSCP ); +enum es_result ret; + + ret =sev_es_ghcb_hv_call (ghcb, ctxt, exit_code,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb) && + (!rdtscp ||ghcb_is_valid_rcx (ghcb)))) +return ES_VMM_ERROR ; + + ctxt->regs->ax = ghcb->save.rax; + ctxt->regs->dx = ghcb->save.rdx; +if (rdtscp) + ctxt->regs->cx = ghcb->save.rcx; + +return ES_OK ; +}
arch/x86/kernel/sev-es.c +2 -23
@@ -815,29 +815,6 @@ static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_WBINVD ,0 ,0 ); } -static enum es_result vc_handle_rdtsc (struct ghcb *ghcb, -struct es_em_ctxt *ctxt, -unsigned long exit_code) -{ -bool rdtscp = (exit_code ==SVM_EXIT_RDTSCP ); -enum es_result ret; - - ret =sev_es_ghcb_hv_call (ghcb, ctxt, exit_code,0 ,0 ); -if (ret !=ES_OK ) -return ret; - -if (!(ghcb_is_valid_rax (ghcb) &&ghcb_is_valid_rdx (ghcb) && - (!rdtscp ||ghcb_is_valid_rcx (ghcb)))) -return ES_VMM_ERROR ; - - ctxt->regs->ax = ghcb->save.rax; - ctxt->regs->dx = ghcb->save.rdx; -if (rdtscp) - ctxt->regs->cx = ghcb->save.rcx; - -return ES_OK ; -} -static enum es_result vc_handle_rdpmc (struct ghcb *ghcb,struct es_em_ctxt *ctxt) {enum es_result ret;
@@ -1001,6 +978,8 @@ static enum es_result vc_context_filter(struct pt_regs *regs, long exit_code)/* List of #VC exit-codes we support in user-space */ case SVM_EXIT_EXCP_BASE ... SVM_EXIT_LAST_EXCP:case SVM_EXIT_CPUID : +case SVM_EXIT_RDTSC : +case SVM_EXIT_RDTSCP : r =ES_OK ;break ;default :
Signature · 1 line
2.26.1
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 7 lines · Mike Stunes
I needed to allow RDTSC(P) from userspace and in early boot in order to get userspace started properly. Patch below.
---
SEV-ES guests will need to execute rdtsc and rdtscp from userspace and during early boot. Move the rdtsc(p) #VC handler into common code and extend the #VC handlers.
Do SEV-ES guests _always_ #VC on rdtsc(p)?
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 10 lines · Dave Hansen
Quoted 7 lines · Mike Stunes
I needed to allow RDTSC(P) from userspace and in early boot in order to get userspace started properly. Patch below.
---
SEV-ES guests will need to execute rdtsc and rdtscp from userspace and during early boot. Move the rdtsc(p) #VC handler into common code and extend the #VC handlers.Do SEV-ES guests _always_ #VC on rdtsc(p)?
Only if the hypervisor is intercepting those instructions.
Thanks,
Tom
Quoted 1 line
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 5 lines · Tom Lendacky
Quoted 4 lines · Dave Hansen
Quoted 3 lines · Mike Stunes
I needed to allow RDTSC(P) from userspace and in early boot in order to get userspace started properly. Patch below.
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
---
SEV-ES guests will need to execute rdtsc and rdtscp from userspace and during early boot. Move the rdtsc(p) #VC handler into common code and extend the #VC handlers.Do SEV-ES guests _always_ #VC on rdtsc(p)?
Only if the hypervisor is intercepting those instructions.
Ahh, so any instruction that can have an instruction intercept set potentially needs to be able to tolerate a #VC? Those instruction intercepts are under the control of the (untrusted relative to the guest) hypervisor, right?
From the main sev-es series:
+#ifdef CONFIG_AMD_MEM_ENCRYPT
+idtentry vmm_communication do_vmm_communication has_error_code=1
+#endif
Since this is set as non-paranoid, that both limits the instructions that can be used in entry paths *and* limits the future architecture from being able add instructions that a current SEV-ES guest doesn't know about. Does SEV-ES have versioning so guests can tell if they might be subject to new interrupt intercepts for which they are not prepared? I didn't see anything obvious in section 15.35 of the manual.
There's also a nugget in the manual that says:
Quoted 2 lines
Similarly, the hypervisor should avoid setting intercept bits for events that would occur in the #VC handler (such as IRET).
That's a fun point because it means that the (untrusted) hypervisor can cause endless faults. I *guess* we have mitigation for this with our stack guard pages, but it's still a bit nasty that the hypervisor can arbitrarily land a guest in the double-fault handler.
It just all seems a bit weak for the hypervisor to be considered untrusted. But, it's _certainly_ a steep in the right direction from SEV.
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Hi Dave,
Review
Quoted 10 lines · Dave Hansen
Ahh, so any instruction that can have an instruction intercept set potentially needs to be able to tolerate a #VC? Those instruction intercepts are under the control of the (untrusted relative to the guest) hypervisor, right?
Quoted 1 line
From the main sev-es series:
(quoted diff) +3
-0+#ifdef CONFIG_AMD_MEM_ENCRYPT +idtentry vmm_communication do_vmm_communication has_error_code=1 +#endif
The next version of the patch-set (which I will hopefully have ready next week) will have this changed. The #VC exception handler uses an IST stack and is set to paranoid=1 and shift_ist. The IST stacks for the #VC handler are only allocated when SEV-ES is active.
Quoted 7 lines
That's a fun point because it means that the (untrusted) hypervisor can cause endless faults. I *guess* we have mitigation for this with our stack guard pages, but it's still a bit nasty that the hypervisor can arbitrarily land a guest in the double-fault handler.
It just all seems a bit weak for the hypervisor to be considered untrusted. But, it's _certainly_ a steep in the right direction from SEV.
Yeah, a malicious hypervisor can do bad things to an SEV-ES VM, but it can't easily steal its secrets from memory or registers. The #VC handler does its best to just crash the VM if unexpected hypervisor behavior is detected.
Regards,
Joerg
Andy Lutomirski
maintainerRe: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 11 lines · Joerg Roedel
Hi Dave,
Quoted 7 lines · Dave Hansen
Ahh, so any instruction that can have an instruction intercept set potentially needs to be able to tolerate a #VC? Those instruction intercepts are under the control of the (untrusted relative to the guest) hypervisor, right?
Quoted 1 line
From the main sev-es series:
Quoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+#ifdef CONFIG_AMD_MEM_ENCRYPT +idtentry vmm_communication do_vmm_communication has_error_code=1 +#endifThe next version of the patch-set (which I will hopefully have ready next week) will have this changed. The #VC exception handler uses an IST stack and is set to paranoid=1 and
shift_ist. The IST stacks for the #VC handler are only allocated when SEV-ES is active.
shift_ist is gross. What's it for? If it's not needed, I'd rather not use it, and I eventually want to get rid of it for #DB as well.
--Andy
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 2 lines · Andy Lutomirski
shift_istis gross. What's it for? If it's not needed, I'd rather not use it, and I eventually want to get rid of it for #DB as well.
The #VC handler needs to be able to nest, there is no way around that for various reasons, the two most important ones are:
1. The #VC -> NMI -> #VC case. #VCs can happen in the NMI handler, for example (but not exclusivly) for RDPMC.
2. In case of an error the #VC handler needs to print out error information by calling one of the printk wrappers. Those will end up doing IO to some console/serial port/whatever which will also cause #VC exceptions to emulate the access to the output devices.
Using shift_ist is perfect for that, the only problem is the race condition with the NMI handler, as shift_ist does not work well with exceptions that can also trigger within the NMI handler. But I have taken care of that for #VC.
Regards,
Joerg
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 20 lines · Joerg Roedel
Quoted 2 lines · Andy Lutomirski
shift_istis gross. What's it for? If it's not needed, I'd rather not use it, and I eventually want to get rid of it for #DB as well.The #VC handler needs to be able to nest, there is no way around that for various reasons, the two most important ones are:
1. The #VC -> NMI -> #VC case. #VCs can happen in the NMI handler, for example (but not exclusivly) for RDPMC.
2. In case of an error the #VC handler needs to print out error information by calling one of the printk wrappers. Those will end up doing IO to some console/serial port/whatever which will also cause #VC exceptions to emulate the access to the output devices.
Quoted 5 lines
Using
shift_istis perfect for that, the only problem is the race condition with the NMI handler, asshift_istdoes not work well with exceptions that can also trigger within the NMI handler. But I have taken care of that for #VC.
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VC
Kaboom.
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Quoted 5 lines
Regards,
Joerg
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 11 lines · Andy Lutomirski
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VCKaboom.
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Right, I dealt with that by unconditionally shifting/unshifting the #VC IST entry in do_nmi() (thanks to Davin Kaplan for the idea). It might cause one of the IST stacks to be unused during nesting, but that is fine. The stack memory for #VC is only allocated when SEV-ES is active (in an SEV-ES VM).
Regards,
Joerg
Andy Lutomirski
maintainerRe: [PATCH] Allow RDTSC and RDTSCP from userspace
Quoted 14 lines · Joerg Roedel
Quoted 11 lines · Andy Lutomirski
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VCKaboom.
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Quoted 5 lines
Right, I dealt with that by unconditionally shifting/unshifting the #VC IST entry in
do_nmi()(thanks to Davin Kaplan for the idea). It might cause one of the IST stacks to be unused during nesting, but that is fine. The stack memory for #VC is only allocated when SEV-ES is active (in an SEV-ES VM).
Blech. It probably works, but still, yuck. It's a bit sad that we seem to be growing more and more poorly designed happens-anywhere exception types at an alarming rate. We seem to have #NM, #MC, #VC, #HV, and #DB. This doesn't really scale.
--Andy
Andy Lutomirski
maintainerShould SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 22 lines · Andy Lutomirski
Quoted 14 lines · Joerg Roedel
Quoted 11 lines · Andy Lutomirski
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VCKaboom.
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Quoted 5 lines
Right, I dealt with that by unconditionally shifting/unshifting the #VC IST entry in
do_nmi()(thanks to Davin Kaplan for the idea). It might cause one of the IST stacks to be unused during nesting, but that is fine. The stack memory for #VC is only allocated when SEV-ES is active (in an SEV-ES VM).
Quoted 4 lines
Blech. It probably works, but still, yuck. It's a bit sad that we seem to be growing more and more poorly designed happens-anywhere exception types at an alarming rate. We seem to have #NM, #MC, #VC, #HV, and #DB. This doesn't really scale.
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
So I think we have two choices:
1. Use IST for #VC and deal with all the mess that entails.
2. Say that we SEV-ES client support on Rome and Naples is for development only and do a quick boot-time check for whether #DB is intercepted. (Just set TF and see what vector we get.) If #DB is intercepted, print a very loud warning and refuse to boot unless some special sev_es.insecure_development_mode or similar option is set.
#2 results in simpler and more robust entry code. #1 is more secure.
So my question is: will anyone actually use SEV-ES in production on Rome or Naples? As I understand it, it's not really ready for prime time on those chips. And do we care if the combination of a malicious hypervisor and malicious guest userspace on Milan can compromise the guest kernel? I don't think SEV-ES is really mean to resist a concerted effort by the hypervisor to compromise the guest.
--Andy
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 40 lines · Andy Lutomirski
Quoted 18 lines · Andy Lutomirski
Quoted 9 lines · Joerg Roedel
Quoted 8 lines · Andy Lutomirski
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VCKaboom.
Quoted 8 lines
Quoted 3 lines
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Right, I dealt with that by unconditionally shifting/unshifting the #VC IST entry in
do_nmi()(thanks to Davin Kaplan for the idea). It might cause one of the IST stacks to be unused during nesting, but that is fine. The stack memory for #VC is only allocated when SEV-ES is active (in an SEV-ES VM).Quoted 4 lines
Blech. It probably works, but still, yuck. It's a bit sad that we seem to be growing more and more poorly designed happens-anywhere exception types at an alarming rate. We seem to have #NM, #MC, #VC, #HV, and #DB. This doesn't really scale.
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
So I think we have two choices:
1. Use IST for #VC and deal with all the mess that entails.
2. Say that we SEV-ES client support on Rome and Naples is for development only and do a quick boot-time check for whether #DB is intercepted. (Just set TF and see what vector we get.) If #DB is intercepted, print a very loud warning and refuse to boot unless some special sev_es.insecure_development_mode or similar option is set.
Quoted 8 lines
#2 results in simpler and more robust entry code. #1 is more secure.
So my question is: will anyone actually use SEV-ES in production on Rome or Naples? As I understand it, it's not really ready for prime time on those chips. And do we care if the combination of a malicious hypervisor and malicious guest userspace on Milan can compromise the guest kernel? I don't think SEV-ES is really mean to resist a concerted effort by the hypervisor to compromise the guest.
More specifically, it is mandatory for hypervisors to intercept #DB to defend against CVE-2015-8104, unless they're willing to trust the guest not to tickle that corner case.
This is believed fixed with SEV-SNP to allow the encrypted guest to use debugging functionality without posing a DoS risk to the host. In this case, the hypervisor is expected not to intercept #DB.
If #DB is intercepted, and #VC doesn't use IST, malicious userspace can cause problems with a movss-delayed breakpoint over SYSCALL.
The question basically whether it is worth going to the effort of making #VC IST and all the problems that entails, to cover one corner case in earlier hardware.
Ultimately, this depends on whether anyone plans to put SEV-ES into production on pre SEV-SNP hardware, and if developers using pre-SEV-SNP hardware are happy with "don't run malicious userspace" or "don't run malicious kernels and skip the #DB intercept" as a fair tradeoff to avoid the #VC IST fun.
~Andrew
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Andy LutomirskiQuoted 45 lines · Andy Lutomirski
Quoted 22 lines · Andy Lutomirski
Quoted 14 lines · Joerg Roedel
Quoted 11 lines · Andy Lutomirski
I assume the race you mean is:
#VC
Immediate NMI before IST gets shifted
#VCKaboom.
How are you dealing with this? Ultimately, I think that NMI will need to turn off IST before engaging in any funny business. Let me ponder this a bit.
Quoted 5 lines
Right, I dealt with that by unconditionally shifting/unshifting the #VC IST entry in
do_nmi()(thanks to Davin Kaplan for the idea). It might cause one of the IST stacks to be unused during nesting, but that is fine. The stack memory for #VC is only allocated when SEV-ES is active (in an SEV-ES VM).Quoted 4 lines
Blech. It probably works, but still, yuck. It's a bit sad that we seem to be growing more and more poorly designed happens-anywhere exception types at an alarming rate. We seem to have #NM, #MC, #VC, #HV, and #DB. This doesn't really scale.
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
So I think we have two choices:
1. Use IST for #VC and deal with all the mess that entails.
2. Say that we SEV-ES client support on Rome and Naples is for development only and do a quick boot-time check for whether #DB is intercepted. (Just set TF and see what vector we get.) If #DB is intercepted, print a very loud warning and refuse to boot unless some special sev_es.insecure_development_mode or similar option is set.
Quoted 5 lines
#2 results in simpler and more robust entry code. #1 is more secure.
So my question is: will anyone actually use SEV-ES in production on Rome or Naples? As I understand it, it's not really ready for prime time on those chips. And do we care if the combination of a malicious
Naples was limited in the number of encryption keys available for guests (15), but Rome increased that significantly (509). SEV-ES is ready on those chips - Rome more so with the increased key count given the requirement that SEV and SEV-ES guests have non-overlapping ASID ranges (which corresponds to key usage).
Thanks,
Tom
Quoted 6 lines
hypervisor and malicious guest userspace on Milan can compromise the guest kernel? I don't think SEV-ES is really mean to resist a concerted effort by the hypervisor to compromise the guest.
--Andy
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Andy LutomirskiQuoted 6 lines · Andy Lutomirski
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
The reason for #VC being IST is not only #DB, but also SEV-SNP. SNP adds page ownership tracking between guest and host, so that the hypervisor can't remap guest pages without the guest noticing.
If there is a violation of ownership, which can happen at any memory access, there will be a #VC exception to notify the guest. And as this can happen anywhere, for example on a carefully crafted stack page set by userspace before doing SYSCALL, the only robust choice for #VC is to use IST.
Regards,
Joerg
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 11 lines · Joerg Roedel
Quoted 6 lines · Andy Lutomirski
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
The reason for #VC being IST is not only #DB, but also SEV-SNP. SNP adds page ownership tracking between guest and host, so that the hypervisor can't remap guest pages without the guest noticing.
Quoted 5 lines
If there is a violation of ownership, which can happen at any memory access, there will be a #VC exception to notify the guest. And as this can happen anywhere, for example on a carefully crafted stack page set by userspace before doing SYSCALL, the only robust choice for #VC is to use IST.
The kernel won't ever touch the guest stack before restoring %rsp in the syscall path, but the (minimum 2) memory accesses required to save the user %rsp and load the kernel stack may be subject to #VC exceptions, as are instruction fetches at the head of the SYSCALL path.
So yes - #VC needs IST.
Sorry for the noise. (That said, it is unfortunate that the hypervisor messing with the memory backing the guest #VC handler results in an infinite loop, rather than an ability to cleanly terminate.)
~Andrew
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Joerg RoedelQuoted 12 lines · Joerg Roedel
Quoted 6 lines · Andy Lutomirski
I have a somewhat serious question: should we use IST for #VC at all? As I understand it, Rome and Naples make it mandatory for hypervisors to intercept #DB, which means that, due to the MOV SS mess, it's sort of mandatory to use IST for #VC. But Milan fixes the #DB issue, so, if we're running under a sufficiently sensible hypervisor, we don't need IST for #VC.
The reason for #VC being IST is not only #DB, but also SEV-SNP. SNP adds page ownership tracking between guest and host, so that the hypervisor can't remap guest pages without the guest noticing.
Quoted 5 lines
If there is a violation of ownership, which can happen at any memory access, there will be a #VC exception to notify the guest. And as this can happen anywhere, for example on a carefully crafted stack page set by userspace before doing SYSCALL, the only robust choice for #VC is to use IST.
So what happens if this #VC triggers on the first access to the #VC stack, because the malicious host has craftily mucked with only the #VC IST stack page?
Or on the NMI IST stack, then we get #VC in NMI before the NMI can fix you up.
AFAICT all of that is non-recoverable.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 10 lines · Peter Zijlstra
On Tue, Apr 28, 2020 at 09:55:12AM +0200, Joerg Roedel wrote:
So what happens if this #VC triggers on the first access to the #VC stack, because the malicious host has craftily mucked with only the #VC IST stack page?
Or on the NMI IST stack, then we get #VC in NMI before the NMI can fix you up.
AFAICT all of that is non-recoverable.
I am not 100% sure, but I think if the #VC stack page is not validated, the #VC should be promoted to a #DF.
Note that this is an issue only with secure nested paging (SNP), which is not enabled yet with this patch-set. When it gets enabled a stack recursion check in the #VC handler is needed which panics the VM. That also fixes the #VC-in-early-NMI problem.
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 4 lines · Joerg Roedel
Note that this is an issue only with secure nested paging (SNP), which is not enabled yet with this patch-set. When it gets enabled a stack recursion check in the #VC handler is needed which panics the VM. That also fixes the #VC-in-early-NMI problem.
But you cannot do a recursion check in #VC, because the NMI can happen on the first instruction of #VC, before we can increment our counter, and then the #VC can happen on NMI because the IST stack is a goner, and we're fscked again (or on a per-cpu variable we touch in our elaborate NMI setup, etc..).
There is no way I can see SNP-#VC 'work'. The best I can come up with is 'mostly', but do you like your bridges/dikes/etc.. to be mostly ok? Or do you want a guarantee they'll actually work?
I'll keep repeating this, x86_64 exceptions are a trainwreck, and IST in specific is utter crap.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 7 lines · Peter Zijlstra
On Tue, Jun 23, 2020 at 01:30:07PM +0200, Joerg Roedel wrote:
But you cannot do a recursion check in #VC, because the NMI can happen on the first instruction of #VC, before we can increment our counter, and then the #VC can happen on NMI because the IST stack is a goner, and we're fscked again (or on a per-cpu variable we touch in our elaborate NMI setup, etc..).
No, the recursion check is fine, because overwriting an already used IST stack doesn't matter (as long as it can be detected) if we are going to panic anyway. It doesn't matter because the kernel will not leave the currently running handler anymore.
I agree there is no way to keep the system running if that happens, but that is also not what is wanted. If stack recursion happens, something malicious from the HV side is going on, and all the kernel needs to be able to is to safely and reliably detect the situation and panic the VM to prevent any data corruption or loss or even leakage.
Quoted 2 lines
I'll keep repeating this,
x86_64exceptions are a trainwreck, and IST in specific is utter crap.
I agree, but don't forget the most prominent underlying reason for IST:
The SYSCALL gap. If SYSCALL would switch stacks most of those issues would not exist. IST would still be needed because there are no task gates in x86-64, but still...
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 9 lines · Joerg Roedel
Quoted 7 lines · Peter Zijlstra
On Tue, Jun 23, 2020 at 01:30:07PM +0200, Joerg Roedel wrote:
But you cannot do a recursion check in #VC, because the NMI can happen on the first instruction of #VC, before we can increment our counter, and then the #VC can happen on NMI because the IST stack is a goner, and we're fscked again (or on a per-cpu variable we touch in our elaborate NMI setup, etc..).
Quoted 4 lines
No, the recursion check is fine, because overwriting an already used IST stack doesn't matter (as long as it can be detected) if we are going to panic anyway. It doesn't matter because the kernel will not leave the currently running handler anymore.
You only have that guarantee when any SNP #VC from kernel is an automatic panic. But in that case, what's the point of having the recursion count?
Quoted 7 lines
Quoted 2 lines
I'll keep repeating this,
x86_64exceptions are a trainwreck, and IST in specific is utter crap.I agree, but don't forget the most prominent underlying reason for IST:
The SYSCALL gap. If SYSCALL would switch stacks most of those issues would not exist. IST would still be needed because there are no task gates in x86-64, but still...
We could all go back to int80 ;-) /me runs like heck
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 9 lines · Peter Zijlstra
Quoted 4 lines · Joerg Roedel
No, the recursion check is fine, because overwriting an already used IST stack doesn't matter (as long as it can be detected) if we are going to panic anyway. It doesn't matter because the kernel will not leave the currently running handler anymore.
You only have that guarantee when any SNP #VC from kernel is an automatic panic. But in that case, what's the point of having the recursion count?
It is not a recursion count, it is a stack-recursion check. Basically walk down the stack and look if your current stack is already in use. Yes, this can be optimized, but that is what is needed.
IIRC the current prototype code for SNP just pre-validates all memory in the VM and doesn't support moving pages around on the host. So any #VC SNP exception would be fatal, yes.
In a scenario with on-demand validation of guest pages and support for guest-assisted page-moving on the HV side it would be more complicated. Basically all memory that is accessed during #VC exception handling must stay validated at all times, including the IST stack.
So saying this, I don't understand why _all_ SNP #VC exceptions from kernel space must be fatal?
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 14 lines · Joerg Roedel
Quoted 3 lines · Peter Zijlstra
You only have that guarantee when any SNP #VC from kernel is an automatic panic. But in that case, what's the point of having the recursion count?
It is not a recursion count, it is a stack-recursion check. Basically walk down the stack and look if your current stack is already in use. Yes, this can be optimized, but that is what is needed.
IIRC the current prototype code for SNP just pre-validates all memory in the VM and doesn't support moving pages around on the host. So any #VC SNP exception would be fatal, yes.
Quoted 7 lines
In a scenario with on-demand validation of guest pages and support for guest-assisted page-moving on the HV side it would be more complicated. Basically all memory that is accessed during #VC exception handling must stay validated at all times, including the IST stack.
So saying this, I don't understand why _all_ SNP #VC exceptions from kernel space must be fatal?
Ah, because I hadn't thought of the stack-recursion check.
So basically when your exception frame points to your own IST, you die. That sounds like something we should have in generic IST code.
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 2 lines · Peter Zijlstra
So basically when your exception frame points to your own IST, you die. That sounds like something we should have in generic IST code.
Something like this... #DF already dies and NMI is 'magic'
--- arch/x86/entry/common.c | 7 +++++++ arch/x86/include/asm/idtentry.h | 12 +++++++++++- 2 files changed, 18 insertions(+), 1 deletion(-)
arch/x86/entry/common.c +7 -0
@@ -742,6 +742,13 @@ noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore)__nmi_exit (); } +noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +} +#ifdef CONFIG_XEN_PV #ifndef CONFIG_PREEMPTION /*
arch/x86/include/asm/idtentry.h +11 -1
@@ -19,6 +19,8 @@ void idtentry_exit_cond_rcu(struct pt_regs *regs, bool rcu_exit);bool idtentry_enter_nmi (struct pt_regs *regs);void idtentry_exit_nmi (struct pt_regs *regs,bool irq_state); +void idtentry_validate_ist (struct pt_regs *regs); +/** * DECLARE_IDTENTRY - Declare functions for simple IDT entry points * No error code pushed by hardware
@@ -322,7 +324,15 @@ static __always_inline void __##func(struct pt_regs *regs)* Maps to DEFINE_IDTENTRY_RAW */ #define DEFINE_IDTENTRY_IST (func) \ - DEFINE_IDTENTRY_RAW(func) +static __always_inline void __##func(struct pt_regs *regs); \ + \ +__visible noinstr void func(struct pt_regs *regs) \ +{ \ + idtentry_validate_ist(regs); \ + __##func(regs); \ +} \ + \ +static __always_inline void __##func(struct pt_regs *regs)/** * DEFINE_IDTENTRY_NOIST - Emit code for NOIST entry points which
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
arch/x86/entry/common.c noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore) · line 750
Quoted 6 lines · Peter Zijlstra
arch/x86/entry/common.c +6
-0+noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +}
Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down regs->sp too. The dumpstack code already has some logic for this.
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 10 lines · Joerg Roedel
arch/x86/entry/common.c noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore) · line 750
Quoted 6 lines · Peter Zijlstra
arch/x86/entry/common.c +6
-0+noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +}Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down
regs->sptoo.
That shouldn't be possible with the current code, I think.
Quoted 1 line
The dumpstack code already has some logic for this.
Reliability of that depends on the unwinder, I wouldn't want the guess uwinder to OOPS me by accident.
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 12 lines · Peter Zijlstra
Quoted 10 lines · Joerg Roedel
arch/x86/entry/common.c noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore) · line 750
Quoted 6 lines · Peter Zijlstra
arch/x86/entry/common.c +6
-0+noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +}Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down
regs->sptoo.
Quoted 1 line
That shouldn't be possible with the current code, I think.
To clarify, we have: NMI, MCE, DB and DF.
DF (with the exception of ESPFIX) is fatal.
MCE from kernel is fatal (which is what makes the MCE in NMI 'work') NMI and DB clear DR7, which avoids DB in NMI.
So that leaves: NMI in DB, and that works.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Peter ZijlstraQuoted 12 lines · Peter Zijlstra
Quoted 10 lines · Joerg Roedel
arch/x86/entry/common.c noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore) · line 750
Quoted 6 lines · Peter Zijlstra
arch/x86/entry/common.c +6
-0+noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +}Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down
regs->sptoo.
Quoted 1 line
That shouldn't be possible with the current code, I think.
Not with the current code, but possibly with SNP #VC exceptions:
-> First #VC -> NMI before VC handler switched off its IST stack (now on NMI IST stack) -> Second SNP #VC exception before the NMI handler did the #VC stack check (because HV messed around with some pages touched there).
In the second #VC you use the same IST stack as in the first #VC, but the the NMI-stack in-between.
Quoted 2 lines
Reliability of that depends on the unwinder, I wouldn't want the guess uwinder to OOPS me by accident.
It doesn't use the full unwinder, it just assumes that there is a pt_regs struct at the top of every kernel stack and walks through them until SP points to a user-space stack.
As long as the assumption that there is a pt_regs struct on top of every stack holds, this should be safe. The assumption might be wrong when an exception happens during SYSCALL/SYSENTER entry, when the return frame is not written by hardware.
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 5 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
Reliability of that depends on the unwinder, I wouldn't want the guess uwinder to OOPS me by accident.
Quoted 8 lines
It doesn't use the full unwinder, it just assumes that there is a
pt_regsstruct at the top of every kernel stack and walks through them until SP points to a user-space stack.As long as the assumption that there is a
pt_regsstruct on top of every stack holds, this should be safe. The assumption might be wrong when an exception happens during SYSCALL/SYSENTER entry, when the return frame is not written by hardware.
The IRQ and SoftIRQ stacks don't have that I think. Only the task and exception stacks.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Peter ZijlstraQuoted 11 lines · Peter Zijlstra
Quoted 9 lines · Joerg Roedel
arch/x86/entry/common.c noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore) · line 750
Quoted 6 lines · Peter Zijlstra
arch/x86/entry/common.c +6
-0+noinstr void idtentry_validate_ist(struct pt_regs *regs) +{ +if ((regs->sp & ~(EXCEPTION_STKSZ -1 )) == + (_RET_IP_ & ~(EXCEPTION_STKSZ -1 ))) +die ("IST stack recursion" , regs,0 ); +}Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down
regs->sptoo.That shouldn't be possible with the current code, I think.
NMI; #MC; Anything which IRET but isn't fatal - #DB, or #BP from patching, #GP from *_safe(), etc; NMI
Sure its a corner case, but did you hear that IST is evil?
~Andrew
P.S. did you also hear that with Rowhammer, userspace has a nonzero quantity of control over generating #MC, depending on how ECC is configured on the platform.
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 11 lines · Andrew Cooper
Quoted 5 lines · Peter Zijlstra
Quoted 2 lines · Joerg Roedel
Yes, this is a start, it doesn't cover the case where the NMI stack is in-between, so I think you need to walk down
regs->sptoo.That shouldn't be possible with the current code, I think.
NMI; #MC; Anything which IRET but isn't fatal - #DB, or #BP from patching, #GP from
*_safe(), etc; NMISure its a corner case, but did you hear that IST is evil?
Isn't current #MC unconditionally fatal from kernel? But yes, I was sorta aware people want that changed.
And yes, NMI can recurse, mostly on #BP and #PF. Like I wrote, its broken vs #MC.
But Joerg was talking about IST recursion with NMI in the middle, something like: #DB, NMI, #DB, and not already being fatal. This one in particular is ruled out by #DB itself clearing DR7 (but NMI would also do that).
Quoted 3 lines
P.S. did you also hear that with Rowhammer, userspace has a nonzero quantity of control over generating #MC, depending on how ECC is configured on the platform.
Yes, excellent stuff.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 2 lines · Peter Zijlstra
Isn't current #MC unconditionally fatal from kernel? But yes, I was sorta aware people want that changed.
Not unconditionally. copy_to_iter_mcsafe() is a good example of one thing we _can_ handle.
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 6 lines · Dave Hansen
Quoted 2 lines · Peter Zijlstra
Isn't current #MC unconditionally fatal from kernel? But yes, I was sorta aware people want that changed.
Not unconditionally.
copy_to_iter_mcsafe()is a good example of one thing we _can_ handle.
Urgh, I thought that stuff was still pending.
Anyway, the important thing is that it is fatal if we hit early NMI. Which I think still holds.
Borislav Petkov
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Andrew CooperQuoted 3 lines · Andrew Cooper
P.S. did you also hear that with Rowhammer, userspace has a nonzero quantity of control over generating #MC, depending on how ECC is configured on the platform.
Where does that #MC point to? Can it control for which address to flip the bits for, i.e., make the #MC appear it has been generated for an address in kernel space?
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Joerg RoedelQuoted 9 lines · Joerg Roedel
Quoted 8 lines · Peter Zijlstra
On Tue, Apr 28, 2020 at 09:55:12AM +0200, Joerg Roedel wrote:
So what happens if this #VC triggers on the first access to the #VC stack, because the malicious host has craftily mucked with only the #VC IST stack page?Or on the NMI IST stack, then we get #VC in NMI before the NMI can fix you up.
Quoted 8 lines
Quoted 1 line
AFAICT all of that is non-recoverable.
I am not 100% sure, but I think if the #VC stack page is not validated, the #VC should be promoted to a #DF.
Note that this is an issue only with secure nested paging (SNP), which is not enabled yet with this patch-set. When it gets enabled a stack recursion check in the #VC handler is needed which panics the VM. That also fixes the #VC-in-early-NMI problem.
There are cases which are definitely non-recoverable.
For both ES and SNP, a malicious hypervisor can mess with the guest physmap to make the the NMI, #VC and #DF stacks all alias.
For ES, this had better result in the #DF handler deciding that crashing is the way out, whereas for SNP, this had better escalate to Shutdown.
What matters here is the security model in SNP. The hypervisor is relied upon for availability (because it could simply refuse to schedule the VM), but market/business forces will cause it to do its best to keep the VM running. Therefore, the securely model is simply(?) that the hypervisor can't do anything to undermine the confidentiality or integrity of the VM.
Crashing out hard if the hypervisor is misbehaving is acceptable. In a cloud, I as a customer would (threaten to?) take my credit card elsewhere, while for enterprise, I'd shout at my virtualisation vendor until a fix happened (also perhaps threaten to take my credit card elsewhere).
Therefore, it is reasonable to build on the expectation that the hypervisor won't be messing around with remapping stacks/etc. Most #VC's are synchronous with guest actions (they equate to actions which would have caused a VMExit), so you can safely reason about when the first #VC might occur, presuming no funny business with the frames backing any memory operands touched.
~Andrew
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 9 lines · Andrew Cooper
There are cases which are definitely non-recoverable.
For both ES and SNP, a malicious hypervisor can mess with the guest physmap to make the the NMI, #VC and #DF stacks all alias.
For ES, this had better result in the #DF handler deciding that crashing is the way out, whereas for SNP, this had better escalate to Shutdown.
Crashing out hard if the hypervisor is misbehaving is acceptable.
Then I'm thinking the only sensible option is to crash hard for any SNP #VC from kernel mode.
Sadly that doesn't help with #VC needing to be IST :-( IST is such a frigging nightmare.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 7 lines · Peter Zijlstra
Quoted 5 lines · Andrew Cooper
There are cases which are definitely non-recoverable.
For both ES and SNP, a malicious hypervisor can mess with the guest physmap to make the the NMI, #VC and #DF stacks all alias.
Quoted 8 lines
Quoted 3 lines
For ES, this had better result in the #DF handler deciding that crashing is the way out, whereas for SNP, this had better escalate to Shutdown. Crashing out hard if the hypervisor is misbehaving is acceptable.
Then I'm thinking the only sensible option is to crash hard for any SNP #VC from kernel mode.
Sadly that doesn't help with #VC needing to be IST :-( IST is such a frigging nightmare.
I presume you mean any #VC caused by RMP faults (i.e. something went wrong with the memory owner/etc metadata) ?
If so, then yes. Any failure here is a bug in the kernel or hypervisor (and needs fixing) or a malicious hypervisor and the guest should terminate for its own safety.
~Andrew
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Borislav Petkov
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Andrew CooperQuoted 5 lines · Andrew Cooper
Crashing out hard if the hypervisor is misbehaving is acceptable. In a cloud, I as a customer would (threaten to?) take my credit card elsewhere, while for enterprise, I'd shout at my virtualisation vendor until a fix happened (also perhaps threaten to take my credit card elsewhere).
This is called customer, credit-card-enforced bug fixing.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Andy LutomirskiHi Andy,
Quoted 1 line · Andy Lutomirski
1. Use IST for #VC and deal with all the mess that entails.
With the removal of IST shifting I wonder what you would suggest on how to best implement an NMI-safe IST handler with nesting support.
My current plan is to implement an IST handler which switches itself off the IST stack as soon as possible, freeing it for re-use.
The flow would be roughly like this upon entering the handler;
build_pt_regs();
RSP = pt_regs->sp;
if (RSP in VC_IST_stack)
error("unallowed nesting")
if (RSP in current_kernel_stack)
RSP = round_down_to_8(RSP)
else
RSP = current_top_of_stack() // non-ist kernel stack
copy_pt_regs(pt_regs, RSP);
switch_stack_to(RSP);To make this NMI safe, the NMI handler needs some logic too. Upon entering NMI, it needs to check the return RSP, and if it is in the #VC IST stack, it must do the above flow by itself and update the return RSP and RIP. It needs to take into account the case when PT_REGS is not fully populated on the return side.
Alternativly the NMI handler could safe/restore the contents of the #VC IST stack or just switch to a special #VC-in-NMI IST stack.
All in all it could get complicated, and imho shift_ist would have been simpler, but who am I anyway...
Or maybe you have a better idea how to implement this, so I'd like to hear your opinion first before I spend too many days implementing something.
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 37 lines · Joerg Roedel
Hi Andy,
Quoted 1 line · Andy Lutomirski
1. Use IST for #VC and deal with all the mess that entails.
With the removal of IST shifting I wonder what you would suggest on how to best implement an NMI-safe IST handler with nesting support.
My current plan is to implement an IST handler which switches itself off the IST stack as soon as possible, freeing it for re-use.
The flow would be roughly like this upon entering the handler;
build_pt_regs(); RSP = pt_regs->sp; if (RSP in VC_IST_stack) error("unallowed nesting") if (RSP in current_kernel_stack) RSP = round_down_to_8(RSP) else RSP = current_top_of_stack() // non-ist kernel stack copy_pt_regs(pt_regs, RSP); switch_stack_to(RSP);To make this NMI safe, the NMI handler needs some logic too. Upon entering NMI, it needs to check the return RSP, and if it is in the #VC IST stack, it must do the above flow by itself and update the return RSP and RIP. It needs to take into account the case when
PT_REGSis not fully populated on the return side.Alternativly the NMI handler could safe/restore the contents of the #VC IST stack or just switch to a special #VC-in-NMI IST stack.
Quoted 6 lines
All in all it could get complicated, and imho
shift_istwould have been simpler, but who am I anyway...Or maybe you have a better idea how to implement this, so I'd like to hear your opinion first before I spend too many days implementing something.
OK, excuse my ignorance, but I'm not seeing how that IST shifting nonsense would've helped in the first place.
If I understand correctly the problem is:
<#VC> shift IST <NMI> ... does stuff <#VC> # again, safe because the shift
But what happens if you get the NMI before your IST adjustment?
<#VC>
<NMI>
... does stuff
<#VC> # again, happily wrecks your earlier #VC
shift IST # whoopsy, too late
Either way around we get to fix this up in NMI (and any other IST exception that can happen while in #VC, hello #MC). And more complexity there is the very last thing we need :-(
There's no way you can fix up the IDT without getting an NMI first.
This entire exception model is fundamentally buggered :-/
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Hi Peter,
Quoted 10 lines · Peter Zijlstra
Quoted 3 lines · Joerg Roedel
Or maybe you have a better idea how to implement this, so I'd like to hear your opinion first before I spend too many days implementing something.
OK, excuse my ignorance, but I'm not seeing how that IST shifting nonsense would've helped in the first place.
If I understand correctly the problem is:
Quoted 7 lines
<#VC> shift IST <NMI> ... does stuff <#VC> # again, safe because the shiftBut what happens if you get the NMI before your IST adjustment?
The v3 patchset implements an unconditional shift of the #VC IST entry in the NMI handler, before it can trigger a #VC exception.
Quoted 3 lines
Either way around we get to fix this up in NMI (and any other IST exception that can happen while in #VC, hello #MC). And more complexity there is the very last thing we need :-(
Yes, in whatever way this gets implemented, it needs some fixup in the NMI handler. But that can happen in C code, so it does not make the assembly more complex, at least.
Quoted 1 line
There's no way you can fix up the IDT without getting an NMI first.
Not sure what you mean by this.
Quoted 1 line
This entire exception model is fundamentally buggered :-/
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 19 lines · Joerg Roedel
Hi Peter,
Quoted 16 lines · Peter Zijlstra
Quoted 3 lines · Joerg Roedel
Or maybe you have a better idea how to implement this, so I'd like to hear your opinion first before I spend too many days implementing something.
OK, excuse my ignorance, but I'm not seeing how that IST shifting nonsense would've helped in the first place.
If I understand correctly the problem is:
<#VC> shift IST <NMI> ... does stuff <#VC> # again, safe because the shift
Quoted 4 lines
Quoted 1 line
But what happens if you get the NMI before your IST adjustment?
The v3 patchset implements an unconditional shift of the #VC IST entry in the NMI handler, before it can trigger a #VC exception.
Going by that other thread -- where you said that any memory access can trigger a #VC, there just isn't such a guarantee.
Quoted 11 lines
Quoted 3 lines
Either way around we get to fix this up in NMI (and any other IST exception that can happen while in #VC, hello #MC). And more complexity there is the very last thing we need :-(
Yes, in whatever way this gets implemented, it needs some fixup in the NMI handler. But that can happen in C code, so it does not make the assembly more complex, at least.
Quoted 1 line
There's no way you can fix up the IDT without getting an NMI first.
Not sure what you mean by this.
I was talking about the case where #VC would try and fix up its own IST.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 7 lines · Peter Zijlstra
Quoted 2 lines · Joerg Roedel
The v3 patchset implements an unconditional shift of the #VC IST entry in the NMI handler, before it can trigger a #VC exception.
Going by that other thread -- where you said that any memory access can trigger a #VC, there just isn't such a guarantee.
As I wrote in the other mail, this can only happen when SNP gets enabled (which is follow-on work to this) and is handled by a stack recursion check in the #VC handler.
The reason I mentioned the #VC-anywhere case is to make it more clear why #VC needs an IST handler.
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 9 lines · Joerg Roedel
Quoted 7 lines · Peter Zijlstra
Quoted 2 lines · Joerg Roedel
The v3 patchset implements an unconditional shift of the #VC IST entry in the NMI handler, before it can trigger a #VC exception.
Going by that other thread -- where you said that any memory access can trigger a #VC, there just isn't such a guarantee.
Quoted 6 lines
As I wrote in the other mail, this can only happen when SNP gets enabled (which is follow-on work to this) and is handled by a stack recursion check in the #VC handler.
The reason I mentioned the #VC-anywhere case is to make it more clear why #VC needs an IST handler.
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Besides that, I am not a fan of delegating problems I already see coming to future-Joerg and future-Peter, but if at all possible deal with them now and be safe later.
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 7 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Didn't I read somewhere that that is only so for Rome/Naples but not for the later chips (Milan) which have #DB pass-through?
Quoted 3 lines
Besides that, I am not a fan of delegating problems I already see coming to future-Joerg and future-Peter, but if at all possible deal with them now and be safe later.
Well, we could just say no :-) At some point in the very near future this house of cards is going to implode.
We're talking about the 3rd case where the only reason things 'work' is because we'll have to panic():
- #MC
- #DB with BUS LOCK DEBUG EXCEPTION
- #VC SNP
(and it ain't a happy accident they're all IST)
Did someone forget to pass the 'ISTs are *EVIL*' memo to the hardware folks? How come we're getting more and more of them? (/me puts fingers in ears and goes la-la-la-la in anticipation of Andrew mentioning CET)
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 11 lines · Peter Zijlstra
Quoted 7 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Didn't I read somewhere that that is only so for Rome/Naples but not for the later chips (Milan) which have #DB pass-through?
Probably, not sure which chips will get debug register virtualization under SEV-ES. But even when it is supported, the HV can (and sometimes will) intercept #DB, which then causes it to be promoted to #VC.
Quoted 4 lines
We're talking about the 3rd case where the only reason things 'work' is because we'll have to
panic():- #MC
Okay, #MC is special and can only be handled on a best-effort basis, as #MC could happen anytime, also while already executing the #MC handler.
Quoted 1 line
- #DB with BUS LOCK DEBUG EXCEPTION
If I understand the problem correctly, this can be solved by moving off the IST stack to the current task stack in the #DB handler, like I plan to do for #VC, no?
Quoted 1 line
- #VC SNP
This has to panic for other reasons that can't be worked around. It boils down to detecting that the HV is doing something fishy and bail out to avoid further harm (like in the #MC handler).
Regards,
Joerg
Peter Zijlstra
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 7 lines · Joerg Roedel
Quoted 4 lines
We're talking about the 3rd case where the only reason things 'work' is because we'll have to
panic():- #MC
Okay, #MC is special and can only be handled on a best-effort basis, as #MC could happen anytime, also while already executing the #MC handler.
I think the hardware has a MCE-mask bit somewhere. Flaky though because clearing it isn't 'atomic' with IRET, so there's a 'funny' window.
It also interacts really bad with the NMI handler. If we get an #MC early in the NMI, where we hard-rely on the NMI-mask being set to set-up the recursion stack, then the #MC IRET will clear the NMI-mask, and we're toast.
Andy has wild and crazy ideas, but I don't think we need more crazy here.
#VC SNP has a similar problem vs NMI, that needs to die() irrespective of the #VC IST recursion.
Quoted 5 lines
Quoted 1 line
- #DB with BUS LOCK DEBUG EXCEPTION
If I understand the problem correctly, this can be solved by moving off the IST stack to the current task stack in the #DB handler, like I plan to do for #VC, no?
Hmm, probably. Would take a bit of care, but should be doable.
Quoted 5 lines
Quoted 1 line
- #VC SNP
This has to panic for other reasons that can't be worked around. It boils down to detecting that the HV is doing something fishy and bail out to avoid further harm (like in the #MC handler).
Right, but it doesn't take away that IST-any-time vectors are fundamentally screwy.
Both the MCE and NMI have masks that are, as per the above, differently funny, but the other ISTs do not. Also, even if they had masks, the interaction between them is still screwy.
#VC would've been so much better if it would've had a mask bit somewhere, then at least we could've had the exception entry covered. Another #VC with the mask set should probably result in #DF or Shutdown, but that's all water under the bridge I suspect.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 9 lines · Peter Zijlstra
Quoted 6 lines · Joerg Roedel
Quoted 4 lines
We're talking about the 3rd case where the only reason things 'work' is because we'll have to
panic():- #MC
Okay, #MC is special and can only be handled on a best-effort basis, as #MC could happen anytime, also while already executing the #MC handler.
I think the hardware has a MCE-mask bit somewhere. Flaky though because clearing it isn't 'atomic' with IRET, so there's a 'funny' window.
MSR_MCG_STATUS.MCIP, and yes - any #MC before that point will immediately Shutdown. Any #MC between that point and IRET will clobber its IST stack and end up sad.
Quoted 7 lines
It also interacts really bad with the NMI handler. If we get an #MC early in the NMI, where we hard-rely on the NMI-mask being set to set-up the recursion stack, then the #MC IRET will clear the NMI-mask, and we're toast.
Andy has wild and crazy ideas, but I don't think we need more crazy here.
Want, certainly not. Need, maybe :-/
Quoted 5 lines
Quoted 4 lines
Quoted 1 line
- #DB with BUS LOCK DEBUG EXCEPTION
If I understand the problem correctly, this can be solved by moving off the IST stack to the current task stack in the #DB handler, like I plan to do for #VC, no?
Hmm, probably. Would take a bit of care, but should be doable.
Andy and I have spent some time considering this.
Having exactly one vector move off IST and onto an in-use task-stack is doable, I think, so long as it can sort out self-recursion protections.
Having more than one vector do this is very complicated. You've got to take care to walk up the list of IST-nesting to see if any interrupted context is in the middle of trying to copy themselves onto the stack, so you don't clobber the frame they're in the middle of copying.
~Andrew
Borislav Petkov
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 3 lines · Andrew Cooper
MSR_MCG_STATUS.MCIP, and yes - any #MC before that point will immediately Shutdown. Any #MC between that point and IRET will clobber its IST stack and end up sad.
Well, at some point we should simply accept that we're living a little on the edge. That is, until we get IRET with a mask of to-reenable vectors which has #MC, NMI and whatever else vectors. It would be even better if that mask were configurable.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
In reply to Peter ZijlstraQuoted 9 lines · Peter Zijlstra
Quoted 6 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Didn't I read somewhere that that is only so for Rome/Naples but not for the later chips (Milan) which have #DB pass-through?
I don't know about hardware timelines, but some future part can now opt in to having debug registers as part of the encrypted state, and swapped by VMExit, which would make debug facilities generally usable, and supposedly safe to the #DB infinite loop issues, at which point the hypervisor need not intercept #DB for safety reasons.
Its worth nothing that on current parts, the hypervisor can set up debug facilities on behalf of the guest (or behind its back) as the DR state is unencrypted, but that attempting to intercept #DB will redirect to #VC inside the guest and cause fun. (Also spare a thought for 32bit kernels which have to cope with userspace singlestepping the SYSENTER path with every #DB turning into #VC.)
Quoted 5 lines
Quoted 3 lines
Besides that, I am not a fan of delegating problems I already see coming to future-Joerg and future-Peter, but if at all possible deal with them now and be safe later.
Well, we could just say no :-) At some point in the very near future this house of cards is going to implode.
What currently exists is a picture of a house of cards in front of something which has fallen down.
Quoted 2 lines
Did someone forget to pass the 'ISTs are *EVIL*' memo to the hardware folks? How come we're getting more and more of them?
I have tried to get this point across. Then again - its far easier for the software folk in the same company as the hardware folk to make this point.
Quoted 2 lines
(/me puts fingers
in ears and goes la-la-la-la in anticipation of Andrew mentioning CET)
I wasn't going to bring it up, but seeing as you have - while there are prohibitively-complicating issues preventing it from working on native, I don't see any point even considering it for the mess which is #VC, or the even bigger mess which is #HV.
~Andrew
Andy Lutomirski
maintainerRe: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 18 lines · Andrew Cooper
Quoted 9 lines · Peter Zijlstra
Quoted 6 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Didn't I read somewhere that that is only so for Rome/Naples but not for the later chips (Milan) which have #DB pass-through?
I don't know about hardware timelines, but some future part can now opt in to having debug registers as part of the encrypted state, and swapped by VMExit, which would make debug facilities generally usable, and supposedly safe to the #DB infinite loop issues, at which point the hypervisor need not intercept #DB for safety reasons.
Quoted 6 lines
Its worth nothing that on current parts, the hypervisor can set up debug facilities on behalf of the guest (or behind its back) as the DR state is unencrypted, but that attempting to intercept #DB will redirect to #VC inside the guest and cause fun. (Also spare a thought for 32bit kernels which have to cope with userspace singlestepping the SYSENTER path with every #DB turning into #VC.)
What do you mean 32-bit? 64-bit kernels have exactly the same problem. At least the stack is okay, though.
Anyway, since I'm way behind on this thread, here are some thoughts:
First, I plan to implement actual precise recursion detection for the IST stacks. We'll be able to reliably panic when unallowed recursion happens.
Second, I don't object *that* strongly to switching to a second #VC stack if an NMI or MCE happens, but we really need to make sure we cover *all* the bases. And #VC is distressingly close to "happens at all kinds of unfortunate times and the guest doesn't actually have much ability to predice it" right now. So we have #VC + #DB + #VC, #VC + NMI + #VC, #VC + MCE + #VC, and even worse options. So doing the shift in a reliable way is not necessarily possible in a clean way.
Let me contemplate. And maybe produce some code soon.
Re: Should SEV-ES #VC use IST? (Re: [PATCH] Allow RDTSC and RDTSCP from userspace)
Quoted 17 lines · Andy Lutomirski
Quoted 16 lines · Andrew Cooper
Quoted 9 lines · Peter Zijlstra
Quoted 6 lines · Joerg Roedel
Quoted 2 lines · Peter Zijlstra
If SNP is the sole reason #VC needs to be IST, then I'd strongly urge you to only make it IST if/when you try and make SNP happen, not before.
It is not the only reason, when ES guests gain debug register support then #VC also needs to be IST, because #DB can be promoted into #VC then, and as #DB is IST for a reason, #VC needs to be too.
Didn't I read somewhere that that is only so for Rome/Naples but not for the later chips (Milan) which have #DB pass-through?
I don't know about hardware timelines, but some future part can now opt in to having debug registers as part of the encrypted state, and swapped by VMExit, which would make debug facilities generally usable, and supposedly safe to the #DB infinite loop issues, at which point the hypervisor need not intercept #DB for safety reasons.
Quoted 8 lines
Quoted 6 lines
Its worth nothing that on current parts, the hypervisor can set up debug facilities on behalf of the guest (or behind its back) as the DR state is unencrypted, but that attempting to intercept #DB will redirect to #VC inside the guest and cause fun. (Also spare a thought for 32bit kernels which have to cope with userspace singlestepping the SYSENTER path with every #DB turning into #VC.)
What do you mean 32-bit? 64-bit kernels have exactly the same problem. At least the stack is okay, though.
:)
AMD-like CPUs disallow SYSENTER/SYSEXIT in Long Mode, and raise #UD, even from a compatibility mode segment.
64bit kernels only have this problem on Intel-like CPUs.
(It is a massive shame that between everyone's attempts, there are 0 "fast system call" instructions with sane semantics, but it is several decades late to fix this problem...)
~Andrew
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
In reply to Joerg RoedelQuoted 11 lines · Joerg Roedel
Quoted 7 lines
That's a fun point because it means that the (untrusted) hypervisor can cause endless faults. I *guess* we have mitigation for this with our stack guard pages, but it's still a bit nasty that the hypervisor can arbitrarily land a guest in the double-fault handler.
It just all seems a bit weak for the hypervisor to be considered untrusted. But, it's _certainly_ a steep in the right direction from SEV.
Yeah, a malicious hypervisor can do bad things to an SEV-ES VM, but it can't easily steal its secrets from memory or registers. The #VC handler does its best to just crash the VM if unexpected hypervisor behavior is detected.
This is the kind of design information that would be very useful to reviewers. Will some of this information make it into the cover letter eventually? Or, Documentation/?
Also, for the security purists, an SEV-ES host is still trusted (in the same TCB as the guest). Truly guest-untrusted VMMs won't be available until SEV-SNP, right?
Re: [PATCH] Allow RDTSC and RDTSCP from userspace
In reply to Mike StunesHi Mike,
Quoted 2 lines · Mike Stunes
I needed to allow RDTSC(P) from userspace and in early boot in order to get userspace started properly. Patch below.
Thanks, but this is not needed anymore. I removed the vc_context_filter from the code. The emulation code is now capable of safely handling any exception from user-space.
Regards,
Joerg
PATCH v2 56/70 x86/sev-es: Handle MONITOR/MONITORX Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by MONITOR and MONITORX instructions.
--- arch/x86/kernel/sev-es.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)
arch/x86/kernel/sev-es.c +19 -0
@@ -707,6 +707,22 @@ static enum es_result vc_handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxtreturn ES_OK ; } +static enum es_result vc_handle_monitor (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +phys_addr_t monitor_pa; +pgd_t *pgd; + + pgd =__va (read_cr3_pa ()); + monitor_pa =vc_slow_virt_to_phys (ghcb, ctxt->regs->ax); + +ghcb_set_rax (ghcb, monitor_pa); +ghcb_set_rcx (ghcb, ctxt->regs->cx); +ghcb_set_rdx (ghcb, ctxt->regs->dx); + +return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_MONITOR ,0 ,0 ); +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -744,6 +760,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_WBINVD : result =vc_handle_wbinvd (ghcb, ctxt);break ; +case SVM_EXIT_MONITOR : + result =vc_handle_monitor (ghcb, ctxt); +break ;case SVM_EXIT_NPF : result =vc_handle_mmio (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 57/70 x86/sev-es: Handle MWAIT/MWAITX Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by MWAIT and MWAITX instructions.
--- arch/x86/kernel/sev-es.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
arch/x86/kernel/sev-es.c +12 -0
@@ -723,6 +723,15 @@ static enum es_result vc_handle_monitor(struct ghcb *ghcb,return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_MONITOR ,0 ,0 ); } +static enum es_result vc_handle_mwait (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +ghcb_set_rax (ghcb, ctxt->regs->ax); +ghcb_set_rcx (ghcb, ctxt->regs->cx); + +return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_MWAIT ,0 ,0 ); +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -763,6 +772,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_MONITOR : result =vc_handle_monitor (ghcb, ctxt);break ; +case SVM_EXIT_MWAIT : + result =vc_handle_mwait (ghcb, ctxt); +break ;case SVM_EXIT_NPF : result =vc_handle_mmio (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 58/70 x86/sev-es: Handle VMMCALL Events
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement a handler for #VC exceptions caused by VMMCALL instructions. This patch is only a starting point, VMMCALL emulation under SEV-ES needs further hypervisor-specific changes to provide additional state.
--- arch/x86/kernel/sev-es.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)
arch/x86/kernel/sev-es.c +23 -0
@@ -732,6 +732,26 @@ static enum es_result vc_handle_mwait(struct ghcb *ghcb,return sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_MWAIT ,0 ,0 ); } +static enum es_result vc_handle_vmmcall (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +enum es_result ret; + +ghcb_set_rax (ghcb, ctxt->regs->ax); +ghcb_set_cpl (ghcb,user_mode (ctxt->regs) ?3 :0 ); + + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_VMMCALL ,0 ,0 ); +if (ret !=ES_OK ) +return ret; + +if (!ghcb_is_valid_rax (ghcb)) +return ES_VMM_ERROR ; + + ctxt->regs->ax = ghcb->save.rax; + +return ES_OK ; +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -766,6 +786,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_MSR : result =vc_handle_msr (ghcb, ctxt);break ; +case SVM_EXIT_VMMCALL : + result =vc_handle_vmmcall (ghcb, ctxt); +break ;case SVM_EXIT_WBINVD : result =vc_handle_wbinvd (ghcb, ctxt);break ;
Signature · 1 line
2.17.1
PATCH v2 59/70 x86/sev-es: Handle #AC Events
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Implement a handler for #VC exceptions caused by #AC exceptions. The #AC exception is just forwarded to do_alignment_check() and not pushed down to the hypervisor, as requested by the SEV-ES GHCB Standardization Specification.
--- arch/x86/kernel/sev-es.c | 4 ++++ 1 file changed, 4 insertions(+)
arch/x86/kernel/sev-es.c +4 -0
@@ -766,6 +766,10 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_WRITE_DR7 : result =vc_handle_dr7_write (ghcb, ctxt, early);break ; +case SVM_EXIT_EXCP_BASE +X86_TRAP_AC : +do_alignment_check (ctxt->regs,0 ); + result =ES_RETRY ; +break ;case SVM_EXIT_RDTSC :case SVM_EXIT_RDTSCP : result =vc_handle_rdtsc (ghcb, ctxt, exit_code);
Signature · 1 line
2.17.1
PATCH v2 60/70 x86/sev-es: Handle #DB Events
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Handle #VC exceptions caused by #DB exceptions in the guest. Do not forward them to the hypervisor and handle them with do_debug() instead.
--- arch/x86/kernel/sev-es.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
arch/x86/kernel/sev-es.c +12 -0
@@ -752,6 +752,15 @@ static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,return ES_OK ; } +static enum es_result vc_handle_db_exception (struct ghcb *ghcb, +struct es_em_ctxt *ctxt) +{ +do_debug (ctxt->regs,0 ); + +/* Exception event, do not advance RIP */ +return ES_RETRY ; +} + static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt, struct ghcb *ghcb, unsigned long exit_code,
@@ -766,6 +775,9 @@ static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,case SVM_EXIT_WRITE_DR7 : result =vc_handle_dr7_write (ghcb, ctxt, early);break ; +case SVM_EXIT_EXCP_BASE +X86_TRAP_DB : + result =vc_handle_db_exception (ghcb, ctxt); +break ;case SVM_EXIT_EXCP_BASE +X86_TRAP_AC :do_alignment_check (ctxt->regs,0 ); result =ES_RETRY ;
Signature · 1 line
2.17.1
PATCH v2 61/70 x86/paravirt: Allow hypervisor specific VMMCALL handling under SEV-ES
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add two new paravirt callbacks to provide hypervisor specific processor state in the GHCB and to copy state from the hypervisor back to the processor.
--- arch/x86/include/asm/x86_init.h | 16 +++++++++++++++- arch/x86/kernel/sev-es.c | 12 ++++++++++++ 2 files changed, 27 insertions(+), 1 deletion(-)
arch/x86/include/asm/x86_init.h +15 -1
@@ -4,8 +4,10 @@#include <asm/bootparam.h> +struct ghcb ;struct mpc_bus ;struct mpc_cpu ; +struct pt_regs ;struct mpc_table ;struct cpuinfo_x86 ;
@@ -238,10 +240,22 @@ struct x86_legacy_features {/** * struct x86_hyper_runtime - x86 hypervisor specific runtime callbacks * -* @pin_vcpu: pin current vcpu to specified physical cpu (run rarely) +* @pin_vcpu: pin current vcpu to specified physical +* cpu (run rarely) +* @sev_es_hcall_prepare: Load additional hypervisor-specific +* state into the GHCB when doing a VMMCALL under +* SEV-ES. Called from the #VC exception handler. +* @sev_es_hcall_finish: Copies state from the GHCB back into the +* processor (or pt_regs). Also runs checks on the +* state returned from the hypervisor after a +* VMMCALL under SEV-ES. Needs to return 'false' +* if the checks fail. Called from the #VC +* exception handler. */ struct x86_hyper_runtime {void (*pin_vcpu)(int cpu); +void (*sev_es_hcall_prepare)(struct ghcb *ghcb,struct pt_regs *regs); +bool (*sev_es_hcall_finish)(struct ghcb *ghcb,struct pt_regs *regs); };/**
arch/x86/kernel/sev-es.c +12 -0
@@ -740,6 +740,9 @@ static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,ghcb_set_rax (ghcb, ctxt->regs->ax);ghcb_set_cpl (ghcb,user_mode (ctxt->regs) ?3 :0 ); +if (x86_platform.hyper.sev_es_hcall_prepare) + x86_platform.hyper.sev_es_hcall_prepare (ghcb, ctxt->regs); + ret =sev_es_ghcb_hv_call (ghcb, ctxt,SVM_EXIT_VMMCALL ,0 ,0 );if (ret !=ES_OK )return ret;
@@ -749,6 +752,15 @@ static enum es_result vc_handle_vmmcall(struct ghcb *ghcb, ctxt->regs->ax = ghcb->save.rax; +/* +* Call sev_es_hcall_finish() after regs->ax is already set. +* This allows the hypervisor handler to overwrite it again if +* necessary. +*/ +if (x86_platform.hyper.sev_es_hcall_finish && + !x86_platform.hyper.sev_es_hcall_finish (ghcb, ctxt->regs)) +return ES_VMM_ERROR ; +return ES_OK ; }
Signature · 1 line
2.17.1
PATCH v2 62/70 x86/kvm: Add KVM specific VMMCALL handling under SEV-ES
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Implement the callbacks to copy the processor state required by KVM to the GHCB.
[ [email protected]: - Split out of a larger patch
- Adapt to different callback functions ]
--- arch/x86/kernel/kvm.c | 35 +++++++++++++++++++++++++++++------ 1 file changed, 29 insertions(+), 6 deletions(-)
arch/x86/kernel/kvm.c +29 -6
@@ -34,6 +34,8 @@#include <asm/hypervisor.h> #include <asm/tlb.h> #include <asm/cpuidle_haltpoll.h> +#include <asm/ptrace.h> +#include <asm/svm.h> static int kvmapf =1 ;
@@ -729,13 +731,34 @@ static void __init kvm_init_platform(void) x86_platform.apic_post_init = kvm_apic_init; } +#if defined(CONFIG_AMD_MEM_ENCRYPT ) +static void kvm_sev_es_hcall_prepare (struct ghcb *ghcb,struct pt_regs *regs) +{ +/* RAX and CPL are already in the GHCB */ +ghcb_set_rbx (ghcb, regs->bx); +ghcb_set_rcx (ghcb, regs->cx); +ghcb_set_rdx (ghcb, regs->dx); +ghcb_set_rsi (ghcb, regs->si); +} + +static bool kvm_sev_es_hcall_finish (struct ghcb *ghcb,struct pt_regs *regs) +{ +/* No checking of the return state needed */ +return true; +} +#endif +const __initconst struct hypervisor_x86 x86_hyper_kvm = { - .name ="KVM" , - .detect = kvm_detect, - .type =X86_HYPER_KVM , - .init.guest_late_init = kvm_guest_init, - .init.x2apic_available = kvm_para_available, - .init.init_platform = kvm_init_platform, + .name ="KVM" , + .detect = kvm_detect, + .type =X86_HYPER_KVM , + .init.guest_late_init = kvm_guest_init, + .init.x2apic_available = kvm_para_available, + .init.init_platform = kvm_init_platform, +#if defined(CONFIG_AMD_MEM_ENCRYPT) + .runtime.sev_es_hcall_prepare = kvm_sev_es_hcall_prepare, + .runtime.sev_es_hcall_finish = kvm_sev_es_hcall_finish, +#endif };static __init int activate_jump_labels (void )
Signature · 1 line
2.17.1
Re: [PATCH 62/70] x86/kvm: Add KVM specific VMMCALL handling under SEV-ES
Quoted 29 lines · Joerg Roedel
From: Tom Lendacky <[email protected]>
Implement the callbacks to copy the processor state required by KVM to the GHCB.
Signed-off-by Tom Lendacky <[email protected]>[ [email protected]: - Split out of a larger patch
- Adapt to different callback functions ]Co-developed-by Joerg Roedel <[email protected]>Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/kernel/kvm.c | 35 +++++++++++++++++++++++++++++------ 1 file changed, 29 insertions(+), 6 deletions(-)arch/x86/kernel/kvm.c +2
-0@@ -34,6 +34,8 @@#include <asm/hypervisor.h> #include <asm/tlb.h> #include <asm/cpuidle_haltpoll.h> +#include <asm/ptrace.h> +#include <asm/svm.h> static int kvmapf =1 ;@@ -729,13 +731,34 @@ static void __init kvm_init_platform(void) x86_platform.apic_post_init = kvm_apic_init;
arch/x86/kernel/kvm.c static void __init kvm_init_platform(void) · line 741
Quoted 10 lines
arch/x86/kernel/kvm.c +8
-0} +#if defined(CONFIG_AMD_MEM_ENCRYPT ) +static void kvm_sev_es_hcall_prepare (struct ghcb *ghcb,struct pt_regs *regs) +{ +/* RAX and CPL are already in the GHCB */ +ghcb_set_rbx (ghcb, regs->bx); +ghcb_set_rcx (ghcb, regs->cx); +ghcb_set_rdx (ghcb, regs->dx); +ghcb_set_rsi (ghcb, regs->si);
Is it possible to check the hypercall from RAX and only copy the needed regs or is there a requirement that they must all be copied unconditionally?
Quoted 33 lines
arch/x86/kernel/kvm.c +19
-6+} + +static bool kvm_sev_es_hcall_finish (struct ghcb *ghcb,struct pt_regs *regs) +{ +/* No checking of the return state needed */ +return true; +} +#endif +const __initconst struct hypervisor_x86 x86_hyper_kvm = { - .name ="KVM" , - .detect = kvm_detect, - .type =X86_HYPER_KVM , - .init.guest_late_init = kvm_guest_init, - .init.x2apic_available = kvm_para_available, - .init.init_platform = kvm_init_platform, + .name ="KVM" , + .detect = kvm_detect, + .type =X86_HYPER_KVM , + .init.guest_late_init = kvm_guest_init, + .init.x2apic_available = kvm_para_available, + .init.init_platform = kvm_init_platform, +#if defined(CONFIG_AMD_MEM_ENCRYPT) + .runtime.sev_es_hcall_prepare = kvm_sev_es_hcall_prepare, + .runtime.sev_es_hcall_finish = kvm_sev_es_hcall_finish, +#endif };static __init int activate_jump_labels (void )Signature · 1 line
2.17.1
Re: [PATCH 62/70] x86/kvm: Add KVM specific VMMCALL handling under SEV-ES
Quoted 13 lines · David Rientjes
arch/x86/kernel/kvm.c static void __init kvm_init_platform(void) · line 741
Quoted 8 lines · Joerg Roedel
arch/x86/kernel/kvm.c +8
-0+#if defined(CONFIG_AMD_MEM_ENCRYPT ) +static void kvm_sev_es_hcall_prepare (struct ghcb *ghcb,struct pt_regs *regs) +{ +/* RAX and CPL are already in the GHCB */ +ghcb_set_rbx (ghcb, regs->bx); +ghcb_set_rcx (ghcb, regs->cx); +ghcb_set_rdx (ghcb, regs->dx); +ghcb_set_rsi (ghcb, regs->si);Is it possible to check the hypercall from RAX and only copy the needed regs or is there a requirement that they must all be copied unconditionally?
No, there is no such requirement. This could be optimized with hypercall specific knowledge as it is in the KVM code anyway.
Regards,
Joerg
PATCH v2 63/70 x86/vmware: Add VMware specific handling for VMMCALL under SEV-ES
In reply to Joerg RoedelFrom: Doug Covelli <[email protected]>
This change adds VMware specific handling for #VC faults caused by VMMCALL instructions.
--- arch/x86/kernel/cpu/vmware.c | 50 ++++++++++++++++++++++++++++++++---- 1 file changed, 45 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/vmware.c +45 -5
@@ -31,6 +31,7 @@#include <asm/timer.h> #include <asm/apic.h> #include <asm/vmware.h> +#include <asm/svm.h> #undef pr_fmt#define pr_fmt (fmt) "vmware: " fmt
@@ -263,10 +264,49 @@ static bool __init vmware_legacy_x2apic_available(void) (eax & (1 <<VMWARE_CMD_LEGACY_X2APIC )) !=0 ; } +#ifdef CONFIG_AMD_MEM_ENCRYPT +static void vmware_sev_es_hcall_prepare (struct ghcb *ghcb, +struct pt_regs *regs) +{ +/* Copy VMWARE specific Hypercall parameters to the GHCB */ +ghcb_set_rip (ghcb, regs->ip); +ghcb_set_rbx (ghcb, regs->bx); +ghcb_set_rcx (ghcb, regs->cx); +ghcb_set_rdx (ghcb, regs->dx); +ghcb_set_rsi (ghcb, regs->si); +ghcb_set_rdi (ghcb, regs->di); +ghcb_set_rbp (ghcb, regs->bp); +} + +static bool vmware_sev_es_hcall_finish (struct ghcb *ghcb,struct pt_regs *regs) +{ +if (!(ghcb_is_valid_rbx (ghcb) && +ghcb_is_valid_rcx (ghcb) && +ghcb_is_valid_rdx (ghcb) && +ghcb_is_valid_rsi (ghcb) && +ghcb_is_valid_rdi (ghcb) && +ghcb_is_valid_rbp (ghcb))) +return false; + + regs->bx = ghcb->save.rbx; + regs->cx = ghcb->save.rcx; + regs->dx = ghcb->save.rdx; + regs->si = ghcb->save.rsi; + regs->di = ghcb->save.rdi; + regs->bp = ghcb->save.rbp; + +return true; +} +#endif +const __initconst struct hypervisor_x86 x86_hyper_vmware = { - .name ="VMware" , - .detect = vmware_platform, - .type =X86_HYPER_VMWARE , - .init.init_platform = vmware_platform_setup, - .init.x2apic_available = vmware_legacy_x2apic_available, + .name ="VMware" , + .detect = vmware_platform, + .type =X86_HYPER_VMWARE , + .init.init_platform = vmware_platform_setup, + .init.x2apic_available = vmware_legacy_x2apic_available, +#ifdef CONFIG_AMD_MEM_ENCRYPT + .runtime.sev_es_hcall_prepare = vmware_sev_es_hcall_prepare, + .runtime.sev_es_hcall_finish = vmware_sev_es_hcall_finish, +#endif };
Signature · 1 line
2.17.1
Re: [PATCH 63/70] x86/vmware: Add VMware specific handling for VMMCALL under SEV-ES
Quoted 10 lines · Joerg Roedel
From: Doug Covelli <[email protected]>
This change adds VMware specific handling for #VC faults caused by VMMCALL instructions.
Signed-off-by Doug Covelli <[email protected]>Signed-off-by Tom Lendacky <[email protected]>[ [email protected]: - Adapt to different paravirt interface ]Co-developed-by Joerg Roedel <[email protected]>Signed-off-by Joerg Roedel <[email protected]>
Quoted 5 lines
--- arch/x86/kernel/cpu/vmware.c | 50 ++++++++++++++++++++++++++++++++ ---- 1 file changed, 45 insertions(+), 5 deletions(-)
PATCH v2 64/70 x86/realmode: Add SEV-ES specific trampoline entry point
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The code at the trampoline entry point is executed in real-mode. In real-mode #VC exceptions can't be handled, so anything that might cause such an exception must be avoided.
In the standard trampoline entry code this is the WBINVD instruction and the call to verify_cpu(), which are both not needed anyway when running as an SEV-ES guest.
--- arch/x86/include/asm/realmode.h | 3 +++ arch/x86/realmode/rm/header.S | 3 +++ arch/x86/realmode/rm/trampoline_64.S | 20 ++++++++++++++++++++ 3 files changed, 26 insertions(+)
arch/x86/include/asm/realmode.h +3 -0
@@ -21,6 +21,9 @@ struct real_mode_header {/* SMP trampoline */ u32 trampoline_start;u32 trampoline_header; +#ifdef CONFIG_AMD_MEM_ENCRYPT +u32 sev_es_trampoline_start; +#endif #ifdef CONFIG_X86_64 u32 trampoline_pgd;#endif
arch/x86/realmode/rm/header.S +3 -0
@@ -20,6 +20,9 @@ SYM_DATA_START(real_mode_header) /* SMP trampoline */ .long pa_trampoline_start .long pa_trampoline_header +#ifdef CONFIG_AMD_MEM_ENCRYPT + .long pa_sev_es_trampoline_start +#endif #ifdef CONFIG_X86_64 .long pa_trampoline_pgd; #endif
arch/x86/realmode/rm/trampoline_64.S +20 -0
@@ -56,6 +56,7 @@ SYM_CODE_START(trampoline_start) testl %eax, %eax # Check for return code jnz no_longmode +.Lswitch_to_protected: /* * GDT tables in non default location kernel can be beyond 16MB and * lgdt will not be able to load the address as in real mode default
@@ -80,6 +81,25 @@ no_longmode: jmp no_longmode SYM_CODE_END(trampoline_start) +#ifdef CONFIG_AMD_MEM_ENCRYPT +/* SEV-ES supports non-zero IP for entry points - no alignment needed */ +SYM_CODE_START(sev_es_trampoline_start) + cli # We should be safe anyway + + LJMPW_RM(1f) +1: + mov %cs, %ax # Code and data in the same place + mov %ax, %ds + mov %ax, %es + mov %ax, %ss + + # Setup stack + movl $rm_stack_end, %esp + + jmp .Lswitch_to_protected +SYM_CODE_END(sev_es_trampoline_start) +#endif /* CONFIG_AMD_MEM_ENCRYPT */ + #include "../kernel/verify_cpu.S" .section ".text32","ax"
Signature · 1 line
2.17.1
PATCH v2 65/70 x86/realmode: Setup AP jump table
In reply to Joerg RoedelFrom: Tom Lendacky <[email protected]>
Setup the AP jump table to point to the SEV-ES trampoline code so that the APs can boot.
[ [email protected]: - Adapted to different code base
- Moved AP table setup from SIPI sending path to real-mode setup code ]
--- arch/x86/include/asm/sev-es.h | 11 ++++++ arch/x86/include/uapi/asm/svm.h | 3 ++ arch/x86/kernel/sev-es.c | 66 +++++++++++++++++++++++++++++++++ arch/x86/realmode/init.c | 6 +++ 4 files changed, 86 insertions(+)
arch/x86/include/asm/sev-es.h +11 -0
@@ -78,4 +78,15 @@ static inline u64 copy_lower_bits(u64 out, u64 in, unsigned int bits)extern void early_vc_handler (void );extern bool boot_vc_exception (struct pt_regs *regs); +struct real_mode_header ; + +#ifdef CONFIG_AMD_MEM_ENCRYPT +int sev_es_setup_ap_jump_table (struct real_mode_header *rmh); +#else /* CONFIG_AMD_MEM_ENCRYPT */ +static inline int sev_es_setup_ap_jump_table (struct real_mode_header *rmh) +{ +return 0 ; +} +#endif /* CONFIG_AMD_MEM_ENCRYPT*/ +#endif
arch/x86/include/uapi/asm/svm.h +3 -0
@@ -84,6 +84,9 @@/* SEV-ES software-defined VMGEXIT events */ #define SVM_VMGEXIT_MMIO_READ 0x80000001#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_AP_JUMP_TABLE 0x80000005 +#define SVM_VMGEXIT_SET_AP_JUMP_TABLE 0 +#define SVM_VMGEXIT_GET_AP_JUMP_TABLE 1#define SVM_VMGEXIT_UNSUPPORTED_EVENT 0x8000ffff#define SVM_EXIT_ERR -1
arch/x86/kernel/sev-es.c +66 -0
84 lines not on this page · raw message
arch/x86/realmode/init.c +6 -0
@@ -9,6 +9,7 @@#include <asm/realmode.h> #include <asm/tlbflush.h> #include <asm/crash.h> +#include <asm/sev-es.h> struct real_mode_header *real_mode_header;u32 *trampoline_cr4_features;
@@ -107,6 +108,11 @@ static void __init setup_real_mode(void)if (sme_active ()) trampoline_header->flags |=TH_FLAGS_SME_ACTIVE ; +if (sev_es_active ()) { +if (sev_es_setup_ap_jump_table (real_mode_header)) +panic ("Failed to update SEV-ES AP Jump Table" ); + } + trampoline_pgd = (u64 *)__va (real_mode_header->trampoline_pgd); trampoline_pgd[0 ] = trampoline_pgd_entry.pgd; trampoline_pgd[511 ] = init_top_pgt[511 ].pgd;
Signature · 1 line
2.17.1
PATCH v2 66/70 x86/head/64: Don't call verify_cpu() on starting APs
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The APs are not ready to handle exceptions when verify_cpu() is called in secondary_startup_64.
--- arch/x86/include/asm/realmode.h | 1 + arch/x86/kernel/head_64.S | 1 + arch/x86/realmode/init.c | 6 ++++++ 3 files changed, 8 insertions(+)
arch/x86/include/asm/realmode.h +1 -0
@@ -69,6 +69,7 @@ extern unsigned char startup_32_smp[];extern unsigned char boot_gdt[];#else extern unsigned char secondary_startup_64[]; +extern unsigned char secondary_startup_64_no_verify[];#endif static inline size_t real_mode_size_needed (void )
arch/x86/kernel/head_64.S +1 -0
@@ -150,6 +150,7 @@ SYM_CODE_START(secondary_startup_64) /* Sanitize CPU configuration */ call verify_cpu +SYM_INNER_LABEL(secondary_startup_64_no_verify, SYM_L_GLOBAL) /* * Retrieve the modifier (SME encryption mask if SME is active) to be * added to the initial pgdir entry that will be programmed into CR3.
arch/x86/realmode/init.c +6 -0
@@ -109,6 +109,12 @@ static void __init setup_real_mode(void) trampoline_header->flags |=TH_FLAGS_SME_ACTIVE ;if (sev_es_active ()) { +/* +* Skip the call to verify_cpu() in secondary_startup_64 as it +* will cause #VC exceptions when the AP can't handle them yet. +*/ + trampoline_header->start = (u64 ) secondary_startup_64_no_verify; +if (sev_es_setup_ap_jump_table (real_mode_header))panic ("Failed to update SEV-ES AP Jump Table" ); }
Signature · 1 line
2.17.1
PATCH v2 67/70 x86/head/64: Rename start_cpu0
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
For SEV-ES this entry point will be used for restarting APs after they have been offlined. Remove the '0' from the name to reflect that.
--- arch/x86/include/asm/cpu.h | 2 +- arch/x86/kernel/head_32.S | 4 ++-- arch/x86/kernel/head_64.S | 6 +++--- arch/x86/kernel/smpboot.c | 4 ++-- 4 files changed, 8 insertions(+), 8 deletions(-)
arch/x86/include/asm/cpu.h +1 -1
@@ -29,7 +29,7 @@ struct x86_cpu {#ifdef CONFIG_HOTPLUG_CPU extern int arch_register_cpu (int num);extern void arch_unregister_cpu (int ); -extern void start_cpu0 (void ); +extern void start_cpu (void );#ifdef CONFIG_DEBUG_HOTPLUG_CPU0 extern int _debug_hotplug_cpu (int cpu,int action);#endif
arch/x86/kernel/head_32.S +2 -2
@@ -180,12 +180,12 @@ SYM_CODE_END(startup_32) * up already except stack. We just set up stack here. Then call * start_secondary(). */ -SYM_FUNC_START(start_cpu0) +SYM_FUNC_START(start_cpu) movl initial_stack, %ecx movl %ecx, %esp call *(initial_code) 1: jmp 1b -SYM_FUNC_END(start_cpu0) +SYM_FUNC_END(start_cpu) #endif /*
arch/x86/kernel/head_64.S +3 -3
@@ -299,15 +299,15 @@ SYM_CODE_END(secondary_startup_64) #ifdef CONFIG_HOTPLUG_CPU /* - * Boot CPU0 entry point. It's called from play_dead(). Everything has been set + * CPU entry point. It's called from play_dead(). Everything has been set * up already except stack. We just set up stack here. Then call * start_secondary() via .Ljump_to_C_code. */ -SYM_CODE_START(start_cpu0) +SYM_CODE_START(start_cpu) UNWIND_HINT_EMPTY movq initial_stack(%rip), %rsp jmp .Ljump_to_C_code -SYM_CODE_END(start_cpu0) +SYM_CODE_END(start_cpu) #endif /* Both SMP bootup and ACPI suspend change these variables */
arch/x86/kernel/smpboot.c +2 -2
@@ -1717,7 +1717,7 @@ static inline void mwait_play_dead(void)* If NMI wants to wake up CPU0, start CPU0. */ if (wakeup_cpu0 ()) -start_cpu0 (); +start_cpu (); } }
@@ -1732,7 +1732,7 @@ void hlt_play_dead(void)* If NMI wants to wake up CPU0, start CPU0. */ if (wakeup_cpu0 ()) -start_cpu0 (); +start_cpu (); } }
Signature · 1 line
2.17.1
PATCH v2 68/70 x86/sev-es: Support CPU offline/online
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Add a play_dead handler when running under SEV-ES. This is needed because the hypervisor can't deliver an SIPI request to restart the AP. Instead the kernel has to issue a VMGEXIT to halt the VCPU. When the hypervisor would deliver and SIPI is wakes up the VCPU instead.
--- arch/x86/include/uapi/asm/svm.h | 1 + arch/x86/kernel/sev-es.c | 58 +++++++++++++++++++++++++++++++++ 2 files changed, 59 insertions(+)
arch/x86/include/uapi/asm/svm.h +1 -0
@@ -84,6 +84,7 @@/* SEV-ES software-defined VMGEXIT events */ #define SVM_VMGEXIT_MMIO_READ 0x80000001#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_AP_HLT_LOOP 0x80000004#define SVM_VMGEXIT_AP_JUMP_TABLE 0x80000005#define SVM_VMGEXIT_SET_AP_JUMP_TABLE 0#define SVM_VMGEXIT_GET_AP_JUMP_TABLE 1
arch/x86/kernel/sev-es.c +58 -0
76 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v2 69/70 x86/cpufeature: Add SEV_ES_GUEST CPU Feature
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The feature bit will indicate whether the kernel runs as an SEV-ES guest. This can be used to apply alternatives at boot for SEV-ES guests and provides a way for user-space to detect whether it runs as an SEV-ES guest.
--- arch/x86/include/asm/cpufeatures.h | 1 + arch/x86/kernel/cpu/amd.c | 6 +++++- 2 files changed, 6 insertions(+), 1 deletion(-)
arch/x86/include/asm/cpufeatures.h +1 -0
@@ -235,6 +235,7 @@#define X86_FEATURE_VMCALL ( 8*32+18)/* "" Hypervisor supports the VMCALL instruction */ #define X86_FEATURE_VMW_VMMCALL ( 8*32+19)/* "" VMware prefers VMMCALL hypercall instruction */ #define X86_FEATURE_SEV_ES ( 8*32+20)/* AMD Secure Encrypted Virtualization - Encrypted State */ +#define X86_FEATURE_SEV_ES_GUEST ( 8*32+21)/* SEV-ES Guest */ /* Intel-defined CPU features, CPUID level 0x00000007:0 (EBX), word 9 */ #define X86_FEATURE_FSGSBASE ( 9*32+ 0)/* RDFSBASE, WRFSBASE, RDGSBASE, WRGSBASE instructions*/
arch/x86/kernel/cpu/amd.c +5 -1
@@ -485,7 +485,6 @@ static void early_init_amd_mc(struct cpuinfo_x86 *c)static void bsp_init_amd (struct cpuinfo_x86 *c) { -#ifdef CONFIG_X86_64 if (c->x86 >=0xf ) {unsigned long long tseg;
@@ -614,6 +613,11 @@ static void early_detect_mem_encrypt(struct cpuinfo_x86 *c)setup_clear_cpu_cap (X86_FEATURE_SEV );setup_clear_cpu_cap (X86_FEATURE_SEV_ES ); } + +if (!rdmsrl_safe (MSR_AMD64_SEV , &msr)) { +if (msr &MSR_AMD64_SEV_ES_ENABLED ) +set_cpu_cap (c,X86_FEATURE_SEV_ES_GUEST ); + } }static void early_init_amd (struct cpuinfo_x86 *c)
Signature · 1 line
2.17.1
PATCH v2 70/70 x86/sev-es: Add NMI state tracking
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
Keep NMI state in SEV-ES code so the kernel can re-enable NMIs for the vCPU when it reaches IRET.
--- arch/x86/entry/entry_64.S | 48 +++++++++++++++++++++++++++++++++ arch/x86/include/asm/sev-es.h | 27 +++++++++++++++++++ arch/x86/include/uapi/asm/svm.h | 1 + arch/x86/kernel/nmi.c | 8 ++++++ arch/x86/kernel/sev-es.c | 31 ++++++++++++++++++++- 5 files changed, 114 insertions(+), 1 deletion(-)
arch/x86/entry/entry_64.S +48 -0
@@ -38,6 +38,7 @@ #include <asm/export.h> #include <asm/frame.h> #include <asm/nospec-branch.h> +#include <asm/sev-es.h> #include <linux/err.h> #include "calling.h"
@@ -629,6 +630,13 @@ SYM_INNER_LABEL(swapgs_restore_regs_and_return_to_usermode, SYM_L_GLOBAL) ud2 1: #endif + + /* + * This code path is used by the NMI handler, so check if NMIs + * need to be re-enabled when running as an SEV-ES guest. + */ + SEV_ES_IRET_CHECK + POP_REGS pop_rdi=0 /*
@@ -1474,6 +1482,8 @@ SYM_CODE_START(nmi) movq $-1, %rsi call do_nmi + SEV_ES_NMI_COMPLETE + /* * Return back to user mode. We must *not* do the normal exit * work, because we don't want to enable interrupts.
@@ -1599,6 +1609,7 @@ nested_nmi_out: popq %rdx /* We are returning to kernel mode, so this cannot result in a fault. */ + SEV_ES_NMI_COMPLETE iretq first_nmi:
@@ -1687,6 +1698,12 @@ end_repeat_nmi: movq $-1, %rsi call do_nmi + /* + * When running as an SEV-ES guest, jump to the SEV-ES NMI IRET + * path. + */ + SEV_ES_NMI_COMPLETE + /* Always restore stashed CR3 value (see paranoid_entry) */ RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
@@ -1715,6 +1732,9 @@ nmi_restore: std movq $0, 5*8(%rsp) /* clear "NMI executing" */ +nmi_return: + UNWIND_HINT_IRET_REGS + /* * iretq reads the "iret" frame and exits the NMI stack in a * single instruction. We are returning to kernel mode, so this
@@ -1724,6 +1744,34 @@ nmi_restore: iretq SYM_CODE_END(nmi) +#ifdef CONFIG_AMD_MEM_ENCRYPT +SYM_CODE_START(sev_es_iret_user) + UNWIND_HINT_IRET_REGS offset=8 + /* + * The kernel jumps here directly from + * swapgs_restore_regs_and_return_to_usermode. %rsp points already to + * trampoline stack, but %cr3 is still from kernel. User-regs are live + * except %rdi. Switch to user CR3, restore user %rdi and user gs_base + * and single-step over IRET + */ + SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi + popq %rdi + SWAPGS + /* + * Enable single-stepping and execute IRET. When IRET is + * finished the resulting #DB exception will cause a #VC + * exception to be raised. The #VC exception handler will send a + * NMI-complete message to the hypervisor to re-open the NMI + * window. + */ +sev_es_iret_kernel: + pushf + btsq $X86_EFLAGS_TF_BIT, (%rsp) + popf + iretq +SYM_CODE_END(sev_es_iret_user) +#endif + #ifndef CONFIG_IA32_EMULATION /* * This handles SYSCALL from 32-bit code. There is no way to program
arch/x86/include/asm/sev-es.h +27 -0
@@ -8,6 +8,8 @@#ifndef __ASM_ENCRYPTED_STATE_H#define __ASM_ENCRYPTED_STATE_H +#ifndef __ASSEMBLY__ +#include <linux/types.h> #include <asm/insn.h>
@@ -82,11 +84,36 @@ struct real_mode_header; #ifdef CONFIG_AMD_MEM_ENCRYPT int sev_es_setup_ap_jump_table(struct real_mode_header *rmh); +void sev_es_nmi_enter(void); #else /* CONFIG_AMD_MEM_ENCRYPT */ static inline int sev_es_setup_ap_jump_table(struct real_mode_header *rmh) { return 0; } +static inline void sev_es_nmi_enter(void) { } +#endif /* CONFIG_AMD_MEM_ENCRYPT*/ + +#else /* !__ASSEMBLY__ */ + +#ifdef CONFIG_AMD_MEM_ENCRYPT +#define SEV_ES_NMI_COMPLETE \ + ALTERNATIVE "", "callq sev_es_nmi_complete", X86_FEATURE_SEV_ES_GUEST + +.macro SEV_ES_IRET_CHECK + ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_SEV_ES_GUEST + movq PER_CPU_VAR(sev_es_in_nmi), %rdi + testq %rdi, %rdi + jz .Lend_\@ + callq sev_es_nmi_complete +.Lend_\@: +.endm + +#else /* CONFIG_AMD_MEM_ENCRYPT */ +#define SEV_ES_NMI_COMPLETE +.macro SEV_ES_IRET_CHECK +.endm #endif /* CONFIG_AMD_MEM_ENCRYPT*/ +#endif /* __ASSEMBLY__ */ + #endif
arch/x86/include/uapi/asm/svm.h +1 -0
@@ -84,6 +84,7 @@/* SEV-ES software-defined VMGEXIT events */ #define SVM_VMGEXIT_MMIO_READ 0x80000001#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_NMI_COMPLETE 0x80000003#define SVM_VMGEXIT_AP_HLT_LOOP 0x80000004#define SVM_VMGEXIT_AP_JUMP_TABLE 0x80000005#define SVM_VMGEXIT_SET_AP_JUMP_TABLE 0
arch/x86/kernel/nmi.c +8 -0
@@ -37,6 +37,7 @@#include <asm/reboot.h> #include <asm/cache.h> #include <asm/nospec-branch.h> +#include <asm/sev-es.h> #define CREATE_TRACE_POINTS #include <trace/events/nmi.h>
@@ -510,6 +511,13 @@ NOKPROBE_SYMBOL(is_debug_stack); dotraplinkage notrace voiddo_nmi (struct pt_regs *regs,long error_code) { +/* +* For SEV-ES the kernel needs to track whether NMIs are blocked until +* IRET is reached, even when the CPU is offline. +*/ +if (sev_es_active ()) +sev_es_nmi_enter (); +if (IS_ENABLED (CONFIG_SMP ) &&cpu_is_offline (smp_processor_id ()))return ;
arch/x86/kernel/sev-es.c +30 -1
@@ -37,6 +37,7 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);*/ struct ghcb __initdata *boot_ghcb; static DEFINE_PER_CPU(unsigned long, cached_dr7) = DR7_RESET_VALUE; +DEFINE_PER_CPU (bool, sev_es_in_nmi) = false;struct ghcb_state {struct ghcb *ghcb;
@@ -270,6 +271,31 @@ static phys_addr_t vc_slow_virt_to_phys(struct ghcb *ghcb, long vaddr)/* Include code shared with pre-decompression boot stage */ #include "sev-es-shared.c" +void sev_es_nmi_enter (void ) +{ +this_cpu_write (sev_es_in_nmi, true); +} + +void sev_es_nmi_complete (void ) +{ +struct ghcb_state state; +struct ghcb *ghcb; + + ghcb =sev_es_get_ghcb (&state); + +vc_ghcb_invalidate (ghcb); +ghcb_set_sw_exit_code (ghcb,SVM_VMGEXIT_NMI_COMPLETE ); +ghcb_set_sw_exit_info_1 (ghcb,0 ); +ghcb_set_sw_exit_info_2 (ghcb,0 ); + +sev_es_wr_ghcb_msr (__pa (ghcb)); +VMGEXIT (); + +sev_es_put_ghcb (&state); + +this_cpu_write (sev_es_in_nmi, false); +} +static u64 sev_es_get_jump_table_addr (void ) {struct ghcb_state state;
@@ -891,7 +917,10 @@ static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,static enum es_result vc_handle_db_exception (struct ghcb *ghcb,struct es_em_ctxt *ctxt) { -do_debug (ctxt->regs,0 ); +if (this_cpu_read (sev_es_in_nmi)) +sev_es_nmi_complete (); +else +do_debug (ctxt->regs,0 );/* Exception event, do not advance RIP */ return ES_RETRY ;
Signature · 1 line
2.17.1
Andy Lutomirski
maintainerRe: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 5 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Keep NMI state in SEV-ES code so the kernel can re-enable NMIs for the vCPU when it reaches IRET.
IIRC I suggested just re-enabling NMI in C from do_nmi(). What was wrong with that approach?
Quoted 9 lines
arch/x86/entry/entry_64.S +9
-0+#ifdef CONFIG_AMD_MEM_ENCRYPT +SYM_CODE_START(sev_es_iret_user) + UNWIND_HINT_IRET_REGS offset=8 + /* + * The kernel jumps here directly from + * swapgs_restore_regs_and_return_to_usermode. %rsp points already to + * trampoline stack, but %cr3 is still from kernel. User-regs are live + * except %rdi. Switch to user CR3, restore user %rdi and user gs_base + * and single-step over IRET
arch/x86/entry/entry_64.S nmi_restore: · line 1765
Quoted 10 lines
arch/x86/entry/entry_64.S +10
-0+ */ + SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi + popq %rdi + SWAPGS + /* + * Enable single-stepping and execute IRET. When IRET is + * finished the resulting #DB exception will cause a #VC + * exception to be raised. The #VC exception handler will send a + * NMI-complete message to the hypervisor to re-open the NMI + * window.
This is distressing to say the least. The sequence if events is, roughly:
1. We're here with NMI masking in an unknown state because do_nmi() and any nested faults could have done IRET, at least architecturally. NMI could occur or it could not. I suppose that, on SEV-ES, as least on current CPUs, NMI is definitely masked. What about on newer CPUs? What if we migrate?
arch/x86/entry/entry_64.S nmi_restore: · line 1770
Quoted 5 lines
arch/x86/entry/entry_64.S +5
-0+ */ +sev_es_iret_kernel: + pushf + btsq $X86_EFLAGS_TF_BIT, (%rsp) + popf
Now we have TF on, NMIs (architecturally) in unknown state.
Quoted 1 line
+ iretq
This causes us to pop the NMI frame off the stack. Assuming the NMI restart logic is invoked (which is maybe impossible?), we get #DB, which presumably is actually delivered. And we end up on the #DB stack, which might already have been in use, so we have a potential increase in nesting. Also, #DB may be called from an unexpected context.
Now somehow #DB is supposed to invoke #VC, which is supposed to do the magic hypercall, and all of this is supposed to be safe? Or is #DB unconditionally redirected to #VC? What happens if we had no stack (e.g. we interrupted SYSCALL) or we were already in #VC to begin with?
I think there are two credible ways to approach this:
1. Just put the NMI unmask in do_nmi(). The kernel *already* knows how to handle running do_nmi() with NMIs unmasked. This is much, much simpler than your code.
2. Have an entirely separate NMI path for the SEV-ES-on-misdesigned-CPU case. And have very clear documentation for what prevents this code from being executed on future CPUs (Zen3?) that have this issue fixed for real?
This hybrid code is no good.
--Andy
Re: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Hi Andy,
Quoted 9 lines · Andy Lutomirski
Quoted 5 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Keep NMI state in SEV-ES code so the kernel can re-enable NMIs for the vCPU when it reaches IRET.
IIRC I suggested just re-enabling NMI in C from
do_nmi(). What was wrong with that approach?
If I understand the code correctly a nested NMI will just reset the interrupted NMI handler to start executing again at 'restart_nmi'. The interrupted NMI handler could be in the #VC handler, and it is not safe to just jump back to the start of the NMI handler from somewhere within the #VC handler.
So I decided to not allow NMI nesting for SEV-ES and only re-enable the NMI window when the first NMI returns. This is not implemented in this patch, but I will do that once Thomas' entry-code rewrite is upstream.
Quoted 6 lines
This causes us to pop the NMI frame off the stack. Assuming the NMI restart logic is invoked (which is maybe impossible?), we get #DB, which presumably is actually delivered. And we end up on the #DB stack, which might already have been in use, so we have a potential increase in nesting. Also, #DB may be called from an unexpected context.
An SEV-ES hypervisor is required to intercept #DB, which means that the #DB exception actually ends up being a #VC exception. So it will not end up on the #DB stack.
Quoted 4 lines
Now somehow #DB is supposed to invoke #VC, which is supposed to do the magic hypercall, and all of this is supposed to be safe? Or is #DB unconditionally redirected to #VC? What happens if we had no stack (e.g. we interrupted SYSCALL) or we were already in #VC to begin with?
Yeah, as I said above, the #DB is redirected to #VC, as the hypervisor has to intercept #DB.
The stack-problem is the one that prevents the Single-step-over-iret approach right now, because the NMI can hit while in kernel mode and on entry stack, which the generic entry code (besided NMI) does not handle. Getting a #VC exception there (like after an IRET to that state) breaks things.
Last, in this version of the patch-set the #VC handler became nesting-safe. It detects whether the per-cpu GHCB is in use and safes/restores its contents in this case.
Quoted 5 lines
I think there are two credible ways to approach this:
1. Just put the NMI unmask in
do_nmi(). The kernel *already* knows how to handle runningdo_nmi()with NMIs unmasked. This is much, much simpler than your code.
Right, and I thought about that, but the implication is that the complexity is moved somewhere else, namely into the #VC handler, which then has to be restartable.
Quoted 4 lines
2. Have an entirely separate NMI path for the SEV-ES-on-misdesigned-CPU case. And have very clear documentation for what prevents this code from being executed on future CPUs (Zen3?) that have this issue fixed for real?
That sounds like a good alternative, I will investigate this approach. The NMI handler should be much simpler as it doesn't need to allow NMI nesting. The question is, does the C code down the NMI path depend on the NMI handlers stack frame layout (e.g. the in-nmi flag)?
Regards,
Joerg
Andy Lutomirski
maintainerRe: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 11 lines · Joerg Roedel
Hi Andy,
Quoted 7 lines · Andy Lutomirski
Quoted 5 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Keep NMI state in SEV-ES code so the kernel can re-enable NMIs for the vCPU when it reaches IRET.
Quoted 8 lines
Quoted 2 lines
IIRC I suggested just re-enabling NMI in C from
do_nmi(). What was wrong with that approach?If I understand the code correctly a nested NMI will just reset the interrupted NMI handler to start executing again at '
restart_nmi'. The interrupted NMI handler could be in the #VC handler, and it is not safe to just jump back to the start of the NMI handler from somewhere within the #VC handler.
Nope. A nested NMI will reset the interrupted NMI's return frame to cause it to run again when it's done. I don't think this will have any real interaction with #VC. There's no longjmp() here.
Quoted 5 lines
So I decided to not allow NMI nesting for SEV-ES and only re-enable the NMI window when the first NMI returns. This is not implemented in this patch, but I will do that once Thomas' entry-code rewrite is upstream.
I certainly *like* preventing nesting, but I don't think we really want a whole alternate NMI path just for a couple of messed-up AMD generations. And the TF trick is not so pretty either.
Quoted 10 lines
Quoted 6 lines
This causes us to pop the NMI frame off the stack. Assuming the NMI restart logic is invoked (which is maybe impossible?), we get #DB, which presumably is actually delivered. And we end up on the #DB stack, which might already have been in use, so we have a potential increase in nesting. Also, #DB may be called from an unexpected context.
An SEV-ES hypervisor is required to intercept #DB, which means that the #DB exception actually ends up being a #VC exception. So it will not end up on the #DB stack.
With your patch set, #DB doesn't seem to end up on the #DB stack either.
Quoted 10 lines
Quoted 6 lines
I think there are two credible ways to approach this:
1. Just put the NMI unmask in
do_nmi(). The kernel *already* knows how to handle runningdo_nmi()with NMIs unmasked. This is much, much simpler than your code.Right, and I thought about that, but the implication is that the complexity is moved somewhere else, namely into the #VC handler, which then has to be restartable.
As above, I don't think there's an actual problem here.
Quoted 10 lines
Quoted 5 lines
2. Have an entirely separate NMI path for the SEV-ES-on-misdesigned-CPU case. And have very clear documentation for what prevents this code from being executed on future CPUs (Zen3?) that have this issue fixed for real?
That sounds like a good alternative, I will investigate this approach. The NMI handler should be much simpler as it doesn't need to allow NMI nesting. The question is, does the C code down the NMI path depend on the NMI handlers stack frame layout (e.g. the in-nmi flag)?
Nope. In particular, the 32-bit path doesn't have all this.
Re: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 3 lines · Andy Lutomirski
Nope. A nested NMI will reset the interrupted NMI's return frame to cause it to run again when it's done. I don't think this will have any real interaction with #VC. There's no
longjmp()here.
Ahh, so I misunderstood that part, in this case your proposal of sending the NMI-complete message right at the beginning of do_nmi() should work just fine. I will test this and see how it works out.
Quoted 3 lines
I certainly *like* preventing nesting, but I don't think we really want a whole alternate NMI path just for a couple of messed-up AMD generations. And the TF trick is not so pretty either.
Indeed, if it could be avoided, it should.
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
This causes us to pop the NMI frame off the stack. Assuming the NMI restart logic is invoked (which is maybe impossible?), we get #DB, which presumably is actually delivered. And we end up on the #DB stack, which might already have been in use, so we have a potential increase in nesting. Also, #DB may be called from an unexpected context.
Quoted 5 lines
Quoted 3 lines
An SEV-ES hypervisor is required to intercept #DB, which means that the #DB exception actually ends up being a #VC exception. So it will not end up on the #DB stack.
With your patch set, #DB doesn't seem to end up on the #DB stack either.
Right, it does not use the #DB stack or shift-ist stuff. Maybe it should, is this needed for anything else than making entry code debugable by kgdb?
Regards,
Joerg
Andy Lutomirski
maintainerRe: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 29 lines · Joerg Roedel
Quoted 3 lines · Andy Lutomirski
Nope. A nested NMI will reset the interrupted NMI's return frame to cause it to run again when it's done. I don't think this will have any real interaction with #VC. There's no
longjmp()here.Ahh, so I misunderstood that part, in this case your proposal of sending the NMI-complete message right at the beginning of
do_nmi()should work just fine. I will test this and see how it works out.Quoted 3 lines
I certainly *like* preventing nesting, but I don't think we really want a whole alternate NMI path just for a couple of messed-up AMD generations. And the TF trick is not so pretty either.
Indeed, if it could be avoided, it should.
Quoted 12 lines
Quoted 11 lines
Quoted 7 lines
This causes us to pop the NMI frame off the stack. Assuming the NMI restart logic is invoked (which is maybe impossible?), we get #DB, which presumably is actually delivered. And we end up on the #DB stack, which might already have been in use, so we have a potential increase in nesting. Also, #DB may be called from an unexpected context.
An SEV-ES hypervisor is required to intercept #DB, which means that the #DB exception actually ends up being a #VC exception. So it will not end up on the #DB stack.
Quoted 5 lines
Quoted 1 line
With your patch set, #DB doesn't seem to end up on the #DB stack either.
Right, it does not use the #DB stack or shift-ist stuff. Maybe it should, is this needed for anything else than making entry code debugable by kgdb?
AIUI the shift-ist stuff is because we aren't very good about the way that we handle tracing right now, and that can cause a limited degree of recursion. #DB uses IST for historical reasons that don't necessarily make sense. Right now, we need it for only one reason:
the MOV SS issue. IIRC this isn't actually triggerable without debugging enabled -- MOV SS with no breakpoint but TF on doesn't seem to malfunction quite as badly.
--Andy
Quoted 4 lines
Regards,
Joerg
Re: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 7 lines · Andy Lutomirski
AIUI the shift-ist stuff is because we aren't very good about the way that we handle tracing right now, and that can cause a limited degree of recursion. #DB uses IST for historical reasons that don't necessarily make sense. Right now, we need it for only one reason:
the MOV SS issue. IIRC this isn't actually triggerable without debugging enabled -- MOV SS with no breakpoint but TF on doesn't seem to malfunction quite as badly.
I had a look at the shift_ist stuff today and it looks like a good solution to the #VC nesting problem when it is turned into a #VC handler. The devil is in the details, of course, as 3 or 4 stacks for the #VC handler (per cpu) should only be allocated when actually running in an SEV-ES guest. Let's see how this works out in practice.
Regards,
Joerg
RFC PATCH 1/1 x86/sev-es: Handle NMI State
In reply to Andy LutomirskiQuoted 3 lines · Andy Lutomirski
1. Just put the NMI unmask in
do_nmi(). The kernel *already* knows how to handle runningdo_nmi()with NMIs unmasked. This is much, much simpler than your code.
Okay, attached is the updated patch which implements this approach. I tested it in an SEV-ES guest with 'perf top' running for a little more than 30 minutes and all looked good. I also removed the dead code from the patch.
From ec3b021c5d9130fd66e00d823c4fabc675c4b49e Mon Sep 17 00:00:00 2001
From: Joerg Roedel <[email protected]>
Date: Tue, 28 Jan 2020 17:31:05 +0100
Subject: [PATCH] x86/sev-es: Handle NMI State
When running under SEV-ES the kernel has to tell the hypervisor when to open the NMI window again after an NMI was injected. This is done with an NMI-complete message to the hypervisor.
Add code to the kernels NMI handler to send this message right at the beginning of do_nmi(). This always allows nesting NMIs.
--- arch/x86/include/asm/sev-es.h | 2 ++ arch/x86/include/uapi/asm/svm.h | 1 + arch/x86/kernel/nmi.c | 8 ++++++++ arch/x86/kernel/sev-es.c | 18 ++++++++++++++++++ 4 files changed, 29 insertions(+)
arch/x86/include/asm/sev-es.h +2 -0
@@ -82,11 +82,13 @@ struct real_mode_header;#ifdef CONFIG_AMD_MEM_ENCRYPT int sev_es_setup_ap_jump_table (struct real_mode_header *rmh); +void sev_es_nmi_complete (void );#else /* CONFIG_AMD_MEM_ENCRYPT */ static inline int sev_es_setup_ap_jump_table (struct real_mode_header *rmh) {return 0 ; } +static inline void sev_es_nmi_complete (void ) { }#endif /* CONFIG_AMD_MEM_ENCRYPT*/ #endif
arch/x86/include/uapi/asm/svm.h +1 -0
@@ -84,6 +84,7 @@/* SEV-ES software-defined VMGEXIT events */ #define SVM_VMGEXIT_MMIO_READ 0x80000001#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_NMI_COMPLETE 0x80000003#define SVM_VMGEXIT_AP_HLT_LOOP 0x80000004#define SVM_VMGEXIT_AP_JUMP_TABLE 0x80000005#define SVM_VMGEXIT_SET_AP_JUMP_TABLE 0
arch/x86/kernel/nmi.c +8 -0
@@ -37,6 +37,7 @@#include <asm/reboot.h> #include <asm/cache.h> #include <asm/nospec-branch.h> +#include <asm/sev-es.h> #define CREATE_TRACE_POINTS #include <trace/events/nmi.h>
@@ -510,6 +511,13 @@ NOKPROBE_SYMBOL(is_debug_stack); dotraplinkage notrace voiddo_nmi (struct pt_regs *regs,long error_code) { +/* +* Re-enable NMIs right here when running as an SEV-ES guest. This might +* cause nested NMIs, but those can be handled safely. +*/ +if (sev_es_active ()) +sev_es_nmi_complete (); +if (IS_ENABLED (CONFIG_SMP ) &&cpu_is_offline (smp_processor_id ()))return ;
arch/x86/kernel/sev-es.c +18 -0
@@ -270,6 +270,24 @@ static phys_addr_t vc_slow_virt_to_phys(struct ghcb *ghcb, long vaddr)/* Include code shared with pre-decompression boot stage */ #include "sev-es-shared.c" +void sev_es_nmi_complete (void ) +{ +struct ghcb_state state; +struct ghcb *ghcb; + + ghcb =sev_es_get_ghcb (&state); + +vc_ghcb_invalidate (ghcb); +ghcb_set_sw_exit_code (ghcb,SVM_VMGEXIT_NMI_COMPLETE ); +ghcb_set_sw_exit_info_1 (ghcb,0 ); +ghcb_set_sw_exit_info_2 (ghcb,0 ); + +sev_es_wr_ghcb_msr (__pa (ghcb)); +VMGEXIT (); + +sev_es_put_ghcb (&state); +} +static u64 sev_es_get_jump_table_addr (void ) {struct ghcb_state state;
Signature · 1 line
2.16.4
Re: [RFC PATCH v2.1] x86/sev-es: Handle NMI State
Quoted 8 lines · Joerg Roedel
Quoted 3 lines · Andy Lutomirski
1. Just put the NMI unmask in
do_nmi(). The kernel *already* knows how to handle runningdo_nmi()with NMIs unmasked. This is much, much simpler than your code.Okay, attached is the updated patch which implements this approach. I tested it in an SEV-ES guest with 'perf top' running for a little more than 30 minutes and all looked good. I also removed the dead code from the patch.
FWIW, perf plus the x86 selftests run in a big loop was my best way of stressing the NMI path when we mucked with it for PTI. The selftests make sure to hit some of the more rare entry/exit paths.
Re: [RFC PATCH v2.1] x86/sev-es: Handle NMI State
Quoted 3 lines · Dave Hansen
FWIW, perf plus the x86 selftests run in a big loop was my best way of stressing the NMI path when we mucked with it for PTI. The selftests make sure to hit some of the more rare entry/exit paths.
Yeah, I ran the x86 selftests in an SEV-ES guest on-top of these patches, that works. But doing this together with 'perf top' is also on the list of tests to do.
Regards,
Joerg
Re: [PATCH 70/70] x86/sev-es: Add NMI state tracking
In reply to Joerg RoedelHi!
Quoted 81 lines · Joerg Roedel
From: Joerg Roedel <[email protected]>
Keep NMI state in SEV-ES code so the kernel can re-enable NMIs for the vCPU when it reaches IRET.
Signed-off-by Joerg Roedel <[email protected]>--- arch/x86/entry/entry_64.S | 48 +++++++++++++++++++++++++++++++++ arch/x86/include/asm/sev-es.h | 27 +++++++++++++++++++ arch/x86/include/uapi/asm/svm.h | 1 + arch/x86/kernel/nmi.c | 8 ++++++ arch/x86/kernel/sev-es.c | 31 ++++++++++++++++++++- 5 files changed, 114 insertions(+), 1 deletion(-)arch/x86/entry/entry_64.S +20
-0@@ -38,6 +38,7 @@ #include <asm/export.h> #include <asm/frame.h> #include <asm/nospec-branch.h> +#include <asm/sev-es.h> #include <linux/err.h> #include "calling.h"@@ -629,6 +630,13 @@ SYM_INNER_LABEL(swapgs_restore_regs_and_return_to_usermode, SYM_L_GLOBAL) ud2 1: #endif + + /* + * This code path is used by the NMI handler, so check if NMIs + * need to be re-enabled when running as an SEV-ES guest. + */ + SEV_ES_IRET_CHECK + POP_REGS pop_rdi=0 /*@@ -1474,6 +1482,8 @@ SYM_CODE_START(nmi) movq $-1, %rsi call do_nmi + SEV_ES_NMI_COMPLETE + /* * Return back to user mode. We must *not* do the normal exit * work, because we don't want to enable interrupts.@@ -1599,6 +1609,7 @@ nested_nmi_out: popq %rdx /* We are returning to kernel mode, so this cannot result in a fault. */ + SEV_ES_NMI_COMPLETE iretq first_nmi:@@ -1687,6 +1698,12 @@ end_repeat_nmi: movq $-1, %rsi call do_nmi + /* + * When running as an SEV-ES guest, jump to the SEV-ES NMI IRET + * path. + */ + SEV_ES_NMI_COMPLETE + /* Always restore stashed CR3 value (see paranoid_entry) */ RESTORE_CR3 scratch_reg=%r15 save_reg=%r14@@ -1715,6 +1732,9 @@ nmi_restore: std movq $0, 5*8(%rsp) /* clear "NMI executing" */ +nmi_return: + UNWIND_HINT_IRET_REGS + /* * iretq reads the "iret" frame and exits the NMI stack in a * single instruction. We are returning to kernel mode, so this
arch/x86/entry/entry_64.S nmi_restore: · line 1749
Quoted 6 lines
arch/x86/entry/entry_64.S +2
-0@@ -1724,6 +1744,34 @@ nmi_restore: iretq SYM_CODE_END(nmi) +#ifdef CONFIG_AMD_MEM_ENCRYPT +SYM_CODE_START(sev_es_iret_user)
What makes kernel jump here? Can't see this referenced from anywhere?
Quoted 173 lines
arch/x86/entry/entry_64.S +26
-0+ UNWIND_HINT_IRET_REGS offset=8 + /* + * The kernel jumps here directly from + * swapgs_restore_regs_and_return_to_usermode. %rsp points already to + * trampoline stack, but %cr3 is still from kernel. User-regs are live + * except %rdi. Switch to user CR3, restore user %rdi and user gs_base + * and single-step over IRET + */ + SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi + popq %rdi + SWAPGS + /* + * Enable single-stepping and execute IRET. When IRET is + * finished the resulting #DB exception will cause a #VC + * exception to be raised. The #VC exception handler will send a + * NMI-complete message to the hypervisor to re-open the NMI + * window. + */ +sev_es_iret_kernel: + pushf + btsq $X86_EFLAGS_TF_BIT, (%rsp) + popf + iretq +SYM_CODE_END(sev_es_iret_user) +#endif + #ifndef CONFIG_IA32_EMULATION /** This handles SYSCALL from 32-bit code. There is no way to program
arch/x86/include/asm/sev-es.h +27
-0@@ -8,6 +8,8 @@#ifndef __ASM_ENCRYPTED_STATE_H#define __ASM_ENCRYPTED_STATE_H +#ifndef __ASSEMBLY__ +#include <linux/types.h> #include <asm/insn.h> @@ -82,11 +84,36 @@ struct real_mode_header; #ifdef CONFIG_AMD_MEM_ENCRYPT int sev_es_setup_ap_jump_table(struct real_mode_header *rmh); +void sev_es_nmi_enter(void); #else /* CONFIG_AMD_MEM_ENCRYPT */ static inline int sev_es_setup_ap_jump_table(struct real_mode_header *rmh) { return 0; } +static inline void sev_es_nmi_enter(void) { } +#endif /* CONFIG_AMD_MEM_ENCRYPT*/ + +#else /* !__ASSEMBLY__ */ + +#ifdef CONFIG_AMD_MEM_ENCRYPT +#define SEV_ES_NMI_COMPLETE \ + ALTERNATIVE "", "callq sev_es_nmi_complete", X86_FEATURE_SEV_ES_GUEST + +.macro SEV_ES_IRET_CHECK + ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_SEV_ES_GUEST + movq PER_CPU_VAR(sev_es_in_nmi), %rdi + testq %rdi, %rdi + jz .Lend_\@ + callq sev_es_nmi_complete +.Lend_\@: +.endm + +#else /* CONFIG_AMD_MEM_ENCRYPT */ +#define SEV_ES_NMI_COMPLETE +.macro SEV_ES_IRET_CHECK +.endm #endif /* CONFIG_AMD_MEM_ENCRYPT*/ +#endif /* __ASSEMBLY__ */ + #endifarch/x86/include/uapi/asm/svm.h +1
-0@@ -84,6 +84,7 @@/* SEV-ES software-defined VMGEXIT events */ #define SVM_VMGEXIT_MMIO_READ 0x80000001#define SVM_VMGEXIT_MMIO_WRITE 0x80000002 +#define SVM_VMGEXIT_NMI_COMPLETE 0x80000003#define SVM_VMGEXIT_AP_HLT_LOOP 0x80000004#define SVM_VMGEXIT_AP_JUMP_TABLE 0x80000005#define SVM_VMGEXIT_SET_AP_JUMP_TABLE 0arch/x86/kernel/nmi.c +8
-0@@ -37,6 +37,7 @@#include <asm/reboot.h> #include <asm/cache.h> #include <asm/nospec-branch.h> +#include <asm/sev-es.h> #define CREATE_TRACE_POINTS #include <trace/events/nmi.h> @@ -510,6 +511,13 @@ NOKPROBE_SYMBOL(is_debug_stack); dotraplinkage notrace voiddo_nmi (struct pt_regs *regs,long error_code) { +/* +* For SEV-ES the kernel needs to track whether NMIs are blocked until +* IRET is reached, even when the CPU is offline. +*/ +if (sev_es_active ()) +sev_es_nmi_enter (); +if (IS_ENABLED (CONFIG_SMP ) &&cpu_is_offline (smp_processor_id ()))return ;arch/x86/kernel/sev-es.c +30
-1@@ -37,6 +37,7 @@ struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);*/ struct ghcb __initdata *boot_ghcb; static DEFINE_PER_CPU(unsigned long, cached_dr7) = DR7_RESET_VALUE; +DEFINE_PER_CPU (bool, sev_es_in_nmi) = false;struct ghcb_state {struct ghcb *ghcb;@@ -270,6 +271,31 @@ static phys_addr_t vc_slow_virt_to_phys(struct ghcb *ghcb, long vaddr)/* Include code shared with pre-decompression boot stage */ #include "sev-es-shared.c" +void sev_es_nmi_enter (void ) +{ +this_cpu_write (sev_es_in_nmi, true); +} + +void sev_es_nmi_complete (void ) +{ +struct ghcb_state state; +struct ghcb *ghcb; + + ghcb =sev_es_get_ghcb (&state); + +vc_ghcb_invalidate (ghcb); +ghcb_set_sw_exit_code (ghcb,SVM_VMGEXIT_NMI_COMPLETE ); +ghcb_set_sw_exit_info_1 (ghcb,0 ); +ghcb_set_sw_exit_info_2 (ghcb,0 ); + +sev_es_wr_ghcb_msr (__pa (ghcb)); +VMGEXIT (); + +sev_es_put_ghcb (&state); + +this_cpu_write (sev_es_in_nmi, false); +} +static u64 sev_es_get_jump_table_addr (void ) {struct ghcb_state state;@@ -891,7 +917,10 @@ static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,static enum es_result vc_handle_db_exception (struct ghcb *ghcb,struct es_em_ctxt *ctxt) { -do_debug (ctxt->regs,0 ); +if (this_cpu_read (sev_es_in_nmi)) +sev_es_nmi_complete (); +else +do_debug (ctxt->regs,0 );/* Exception event, do not advance RIP */ return ES_RETRY ;
Re: [PATCH 70/70] x86/sev-es: Add NMI state tracking
Quoted 4 lines · Mika Penttilä
Quoted 1 line
+SYM_CODE_START(sev_es_iret_user)
What makes kernel jump here? Can't see this referenced from anywhere?
Sorry, it is just a left-over from a previous version of this patch (which implemented the single-step-over-iret). This label is not used anymore. The jump to it was in swapgs_restore_regs_and_return_to_usermode, after checking the sev_es_in_nmi flag.
Regards,
Joerg