Independent per-CPU data section for nVHE
Subsystems (3)
v3 cover Independent per-CPU data section for nVHE
Introduce '.hyp.data..percpu' as part of ongoing effort to make nVHE hyp code self-contained and independent of the rest of the kernel.
Main benefits:
* independent nVHE per-CPU data section that can be unmapped from host,
* more robust linking of nVHE hyp code,
* no need for hyp-specific macros to access per-CPU variables.
The series is structured as follows:
- patch 1: Improve existing hyp build rules. This could be sent and merged independently of per-CPU but this series builds on it.
- patches 2-3: Minor cleanups.
- patches 4-5: Replace hyp helpers for accessing per-CPU variables with common helpers modified to work correctly in hyp. Per-CPU variables can now be accessed with one API anywhere.
- patches 6-8: Where VHE and nVHE use per-CPU variables defined in kernel proper, move their definitions to hyp/ where they are duplicated and owned by VHE/nVHE, respectively. Non-VHE hyp code now refers only to per-CPU variables defined in its source files. Helpers are added so that kernel proper can continue to access nVHE hyp variables, same way as it does with other nVHE symbols.
- patches 9-11: Introduce '.hyp.data..percpu' ELF section and allocate memory for every CPU core during KVM init. All nVHE per-CPU state is now grouped together in ELF and in memory. Introducing a new per-CPU variable does not require adding new memory mappings any more. nVHE hyp code cannot accidentally refer to kernel-proper per-CPU data as it only has the pointer to its own per-CPU memory.
Patches are rebased on v5.9-rc5 and available in branch 'topic/percpu-v3' at:
https://android-kvm.googlesource.com/linux
For maintainers: In case of interest, there are patches that remove the need for redefining macros under DEBUG_PREEMPT available at the same repo, branch 'topic/percpu-v3-debug-preempt'. Since they are non-trivial, I am not going to post them here so late in the 5.10 window. I plan to post them for 5.11 when they will also be useful for other patches.
Changes v2 -> v3:
* Use PERCPU_INPUT in hyp.ld instead of modifying PERCPU_SECTION
* Only pass linker script once to LD (fix error message)
* Renamed '.hyp.o' to '.nvhe.o'
* Use __KVM_VHE_HYPERVISOR__ to select TPIDR_EL2 instead of alternatives
* Move all prefixing-related macros to hyp_image.h
Changes v1 -> v2:
* partially link hyp code, add linker script
David Brazdil (11):
kvm: arm64: Partially link nVHE hyp code, simplify HYPCOPY
kvm: arm64: Move nVHE hyp namespace macros to hyp_image.h
kvm: arm64: Only define __kvm_ex_table for CONFIG_KVM
kvm: arm64: Remove __hyp_this_cpu_read
kvm: arm64: Remove hyp_adr/ldr_this_cpu
kvm: arm64: Add helpers for accessing nVHE hyp per-cpu vars
kvm: arm64: Duplicate arm64_ssbd_callback_required for nVHE hyp
kvm: arm64: Create separate instances of kvm_host_data for VHE/nVHE
kvm: arm64: Mark hyp stack pages reserved
kvm: arm64: Set up hyp percpu data for nVHE
kvm: arm64: Remove unnecessary hyp mappings
arch/arm64/include/asm/assembler.h | 27 ++++-- arch/arm64/include/asm/hyp_image.h | 36 +++++++ arch/arm64/include/asm/kvm_asm.h | 82 ++++++++-------- arch/arm64/include/asm/kvm_host.h | 2 +- arch/arm64/include/asm/kvm_mmu.h | 19 ++-- arch/arm64/include/asm/percpu.h | 33 ++++++- arch/arm64/include/asm/sections.h | 1 + arch/arm64/kernel/image-vars.h | 4 - arch/arm64/kernel/vmlinux.lds.S | 13 +++ arch/arm64/kvm/arm.c | 109 +++++++++++++++++++--- arch/arm64/kvm/hyp/hyp-entry.S | 2 +- arch/arm64/kvm/hyp/include/hyp/debug-sr.h | 4 +- arch/arm64/kvm/hyp/include/hyp/switch.h | 8 +- arch/arm64/kvm/hyp/nvhe/Makefile | 60 ++++++------ arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 19 ++++ arch/arm64/kvm/hyp/nvhe/switch.c | 8 +- arch/arm64/kvm/hyp/vhe/switch.c | 5 +- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 4 +- arch/arm64/kvm/pmu.c | 13 ++- 19 files changed, 320 insertions(+), 129 deletions(-) create mode 100644 arch/arm64/include/asm/hyp_image.h create mode 100644 arch/arm64/kvm/hyp/nvhe/hyp.lds.S
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 1/11 kvm: arm64: Partially link nVHE hyp code, simplify HYPCOPY
Relying on objcopy to prefix the ELF section names of the nVHE hyp code is brittle and prevents us from using wildcards to match specific section names.
Improve the build rules by partially linking all '.nvhe.o' files and prefixing their ELF section names using a linker script. Continue using objcopy for prefixing ELF symbol names.
One immediate advantage of this approach is that all subsections matching a pattern can be merged into a single prefixed section, eg. .text and .text.* can be linked into a single '.hyp.text'. This removes the need for -fno-reorder-functions on GCC and will be useful in the future too: LTO builds use .text subsections, compilers routinely generate .rodata subsections, etc.
Partially linking all hyp code into a single object file also makes it easier to analyze.
--- arch/arm64/include/asm/hyp_image.h | 24 ++++++++++++ arch/arm64/kvm/hyp/nvhe/Makefile | 60 ++++++++++++++++-------------- arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 13 +++++++ 3 files changed, 70 insertions(+), 27 deletions(-) create mode 100644 arch/arm64/include/asm/hyp_image.h create mode 100644 arch/arm64/kvm/hyp/nvhe/hyp.lds.S
arch/arm64/include/asm/hyp_image.h new file +24 -0
@@ -0,0 +1,24 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* Copyright (C) 2020 Google LLC. +* Written by David Brazdil <[email protected]> +*/ + +#ifndef __ARM64_HYP_IMAGE_H__ +#define __ARM64_HYP_IMAGE_H__ + +#ifdef LINKER_SCRIPT + +/* +* KVM nVHE ELF section names are prefixed with .hyp, to separate them +* from the kernel proper. +*/ +#define HYP_SECTION_NAME (NAME ) .hyp##NAME + +/* Defines an ELF hyp section from input section @NAME and its subsections. */ +#define HYP_SECTION (NAME ) \ + HYP_SECTION_NAME(NAME) : { *(NAME NAME##.[0-9a-zA-Z_]*) } + +#endif /* LINKER_SCRIPT */ + +#endif /* __ARM64_HYP_IMAGE_H__ */
arch/arm64/kvm/hyp/nvhe/Makefile +33 -27
73 lines not on this page · raw message
arch/arm64/kvm/hyp/nvhe/hyp.lds.S new file +13 -0
@@ -0,0 +1,13 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * Copyright (C) 2020 Google LLC. + * Written by David Brazdil <[email protected]> + * + * Linker script used for partial linking of nVHE EL2 object files. + */ + +#include <asm/hyp_image.h> + +SECTIONS { + HYP_SECTION(.text) +}
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 2/11 kvm: arm64: Move nVHE hyp namespace macros to hyp_image.h
In reply to David BrazdilMinor cleanup to move all macros related to prefixing nVHE hyp section and symbol names into one place: hyp_image.h.
--- arch/arm64/include/asm/hyp_image.h | 12 ++++++++++++ arch/arm64/include/asm/kvm_asm.h | 8 +------- arch/arm64/kernel/image-vars.h | 2 -- 3 files changed, 13 insertions(+), 9 deletions(-)
arch/arm64/include/asm/hyp_image.h +12 -0
@@ -7,6 +7,12 @@#ifndef __ARM64_HYP_IMAGE_H__#define __ARM64_HYP_IMAGE_H__ +/* +* KVM nVHE code has its own symbol namespace prefixed with __kvm_nvhe_, +* to separate it from the kernel proper. +*/ +#define kvm_nvhe_sym (sym) __kvm_nvhe_##sym +#ifdef LINKER_SCRIPT /*
@@ -19,6 +25,12 @@#define HYP_SECTION (NAME ) \ HYP_SECTION_NAME(NAME) : { *(NAME NAME##.[0-9a-zA-Z_]*) } +/* +* Defines a linker script alias of a kernel-proper symbol referenced by +* KVM nVHE hyp code. +*/ +#define KVM_NVHE_ALIAS (sym) kvm_nvhe_sym(sym) = sym; +#endif /* LINKER_SCRIPT */ #endif /* __ARM64_HYP_IMAGE_H__ */
arch/arm64/include/asm/kvm_asm.h +1 -7
@@ -7,6 +7,7 @@#ifndef __ARM_KVM_ASM_H__#define __ARM_KVM_ASM_H__ +#include <asm/hyp_image.h> #include <asm/virt.h> #define VCPU_WORKAROUND_2_FLAG_SHIFT 0
@@ -42,13 +43,6 @@#include <linux/mm.h> -/* -* Translate name of a symbol defined in nVHE hyp to the name seen -* by kernel proper. All nVHE symbols are prefixed by the build system -* to avoid clashes with the VHE variants. -*/ -#define kvm_nvhe_sym (sym) __kvm_nvhe_##sym -#define DECLARE_KVM_VHE_SYM (sym) extern char sym[]#define DECLARE_KVM_NVHE_SYM (sym) extern char kvm_nvhe_sym(sym)[]
arch/arm64/kernel/image-vars.h +0 -2
@@ -61,8 +61,6 @@ __efistub__ctype = _ctype;* memory mappings. */ -#define KVM_NVHE_ALIAS (sym) __kvm_nvhe_##sym = sym; -/* Alternative callbacks for init-time patching of nVHE hyp code. */ KVM_NVHE_ALIAS (arm64_enable_wa2_handling);KVM_NVHE_ALIAS (kvm_patch_vector_branch);
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 3/11 kvm: arm64: Only define __kvm_ex_table for CONFIG_KVM
In reply to David BrazdilMinor cleanup that only creates __kvm_ex_table ELF section and related symbols if CONFIG_KVM is enabled.
--- arch/arm64/kernel/vmlinux.lds.S | 4 ++++ 1 file changed, 4 insertions(+)
arch/arm64/kernel/vmlinux.lds.S +4 -0
@@ -21,11 +21,15 @@ ENTRY(_text) jiffies = jiffies_64; +#ifdef CONFIG_KVM #define HYPERVISOR_EXTABLE \ . = ALIGN(SZ_8); \ __start___kvm_ex_table = .; \ *(__kvm_ex_table) \ __stop___kvm_ex_table = .; +#else /* CONFIG_KVM */ +#define HYPERVISOR_EXTABLE +#endif #define HYPERVISOR_TEXT \ /* \
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 4/11 kvm: arm64: Remove __hyp_this_cpu_read
In reply to David Brazdilthis_cpu_ptr is meant for use in kernel proper because it selects between TPIDR_EL1/2 based on nVHE/VHE. __hyp_this_cpu_ptr was used in hyp to always select TPIDR_EL2. Unify all users behind this_cpu_ptr and friends by selecting _EL2 register under __KVM_VHE/NVHE_HYPERVISOR__.
Under CONFIG_DEBUG_PREEMPT, the kernel helpers perform a preemption check which is omitted by the hyp helpers. Preserve the behavior for nVHE by overriding the corresponding macros under __KVM_NVHE_HYPERVISOR__. Extend the checks into VHE hyp code.
--- arch/arm64/include/asm/kvm_asm.h | 20 -------------- arch/arm64/include/asm/percpu.h | 33 +++++++++++++++++++++-- arch/arm64/kvm/hyp/include/hyp/debug-sr.h | 4 +-- arch/arm64/kvm/hyp/include/hyp/switch.h | 8 +++--- arch/arm64/kvm/hyp/nvhe/switch.c | 2 +- arch/arm64/kvm/hyp/vhe/switch.c | 2 +- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 4 +-- 7 files changed, 41 insertions(+), 32 deletions(-)
arch/arm64/include/asm/kvm_asm.h +0 -20
@@ -143,26 +143,6 @@ extern char __smccc_workaround_1_smc[__SMCCC_WORKAROUND_1_SMC_SZ]; addr; \ }) -/* -* Home-grown __this_cpu_{ptr,read} variants that always work at HYP, -* provided that sym is really a *symbol* and not a pointer obtained from -* a data structure. As for SHIFT_PERCPU_PTR(), the creative casting keeps -* sparse quiet. -*/ -#define __hyp_this_cpu_ptr (sym) \ - ({ \ - void *__ptr; \ - __verify_pcpu_ptr(&sym); \ - __ptr = hyp_symbol_addr(sym); \ - __ptr += read_sysreg(tpidr_el2); \ - (typeof(sym) __kernel __force *)__ptr; \ - }) - -#define __hyp_this_cpu_read (sym) \ - ({ \ - *__hyp_this_cpu_ptr(sym); \ - }) -#define __KVM_EXTABLE (from, to) \ " .pushsection __kvm_ex_table, \"a\"\n" \ " .align 3\n" \
arch/arm64/include/asm/percpu.h +31 -2
@@ -19,7 +19,21 @@ static inline void set_my_cpu_offset(unsigned long off) :: "r" (off) : "memory"); } -static inline unsigned long __my_cpu_offset (void ) +static inline unsigned long __hyp_my_cpu_offset (void ) +{ +unsigned long off; + +/* +* We want to allow caching the value, so avoid using volatile and +* instead use a fake stack read to hazard against barrier(). +*/ + asm("mrs %0, tpidr_el2" :"=r" (off) : +"Q" (*(const unsigned long *)current_stack_pointer)); + +return off; +} + +static inline unsigned long __kern_my_cpu_offset (void ) {unsigned long off;
@@ -35,7 +49,12 @@ static inline unsigned long __my_cpu_offset(void)return off; } -#define __my_cpu_offset __my_cpu_offset() + +#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) +#define __my_cpu_offset __hyp_my_cpu_offset() +#else +#define __my_cpu_offset __kern_my_cpu_offset() +#endif #define PERCPU_RW_OPS (sz) \ static inline unsigned long __percpu_read_##sz(void *ptr) \
@@ -227,4 +246,14 @@ PERCPU_RET_OP(add, add, ldadd)#include <asm-generic/percpu.h> +/* Redefine macros for nVHE hyp under DEBUG_PREEMPT to avoid its dependencies. */ +#if defined(__KVM_NVHE_HYPERVISOR__) && defined(CONFIG_DEBUG_PREEMPT ) +#undef this_cpu_ptr +#define this_cpu_ptr raw_cpu_ptr +#undef __this_cpu_read +#define __this_cpu_read raw_cpu_read +#undef __this_cpu_write +#define __this_cpu_write raw_cpu_write +#endif +#endif /* __ASM_PERCPU_H */
arch/arm64/kvm/hyp/include/hyp/debug-sr.h +2 -2
@@ -135,7 +135,7 @@ static inline void __debug_switch_to_guest_common(struct kvm_vcpu *vcpu)if (!(vcpu->arch.flags &KVM_ARM64_DEBUG_DIRTY ))return ; - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt; guest_ctxt = &vcpu->arch.ctxt; host_dbg = &vcpu->arch.host_debug_state.regs; guest_dbg =kern_hyp_va (vcpu->arch.debug_ptr);
@@ -154,7 +154,7 @@ static inline void __debug_switch_to_host_common(struct kvm_vcpu *vcpu)if (!(vcpu->arch.flags &KVM_ARM64_DEBUG_DIRTY ))return ; - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt; guest_ctxt = &vcpu->arch.ctxt; host_dbg = &vcpu->arch.host_debug_state.regs; guest_dbg =kern_hyp_va (vcpu->arch.debug_ptr);
arch/arm64/kvm/hyp/include/hyp/switch.h +4 -4
@@ -386,7 +386,7 @@ static inline bool __hyp_handle_ptrauth(struct kvm_vcpu *vcpu) !esr_is_ptrauth_trap(kvm_vcpu_get_esr(vcpu)))return false; - ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt;__ptrauth_save_key (ctxt,APIA );__ptrauth_save_key (ctxt,APIB );__ptrauth_save_key (ctxt,APDA );
@@ -495,7 +495,7 @@ static inline void __set_guest_arch_workaround_state(struct kvm_vcpu *vcpu)* guest wants it disabled, so be it... */ if (__needs_ssbd_off (vcpu) && -__hyp_this_cpu_read (arm64_ssbd_callback_required)) +__this_cpu_read (arm64_ssbd_callback_required))arm_smccc_1_1_smc (ARM_SMCCC_ARCH_WORKAROUND_2 ,0 ,NULL );#endif }
@@ -507,7 +507,7 @@ static inline void __set_host_arch_workaround_state(struct kvm_vcpu *vcpu)* If the guest has disabled the workaround, bring it back on. */ if (__needs_ssbd_off (vcpu) && -__hyp_this_cpu_read (arm64_ssbd_callback_required)) +__this_cpu_read (arm64_ssbd_callback_required))arm_smccc_1_1_smc (ARM_SMCCC_ARCH_WORKAROUND_2 ,1 ,NULL );#endif }
@@ -521,7 +521,7 @@ static inline void __kvm_unexpected_el2_exception(void) entry =hyp_symbol_addr (__start___kvm_ex_table); end =hyp_symbol_addr (__stop___kvm_ex_table); - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt;while (entry < end) { addr = (unsigned long )&entry->insn + entry->insn;
arch/arm64/kvm/hyp/nvhe/switch.c +1 -1
@@ -175,7 +175,7 @@ int __kvm_vcpu_run(struct kvm_vcpu *vcpu) vcpu =kern_hyp_va (vcpu); - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt; host_ctxt->__hyp_running_vcpu = vcpu; guest_ctxt = &vcpu->arch.ctxt;
arch/arm64/kvm/hyp/vhe/switch.c +1 -1
@@ -108,7 +108,7 @@ static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)struct kvm_cpu_context *guest_ctxt;u64 exit_code; - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt; host_ctxt->__hyp_running_vcpu = vcpu; guest_ctxt = &vcpu->arch.ctxt;
arch/arm64/kvm/hyp/vhe/sysreg-sr.c +2 -2
@@ -66,7 +66,7 @@ void kvm_vcpu_load_sysregs_vhe(struct kvm_vcpu *vcpu)struct kvm_cpu_context *guest_ctxt = &vcpu->arch.ctxt;struct kvm_cpu_context *host_ctxt; - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt;__sysreg_save_user_state (host_ctxt);/*
@@ -100,7 +100,7 @@ void kvm_vcpu_put_sysregs_vhe(struct kvm_vcpu *vcpu)struct kvm_cpu_context *guest_ctxt = &vcpu->arch.ctxt;struct kvm_cpu_context *host_ctxt; - host_ctxt = &__hyp_this_cpu_ptr (kvm_host_data)->host_ctxt; + host_ctxt = &this_cpu_ptr (&kvm_host_data)->host_ctxt;deactivate_traps_vhe_put ();__sysreg_save_el1_state (guest_ctxt);
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 5/11 kvm: arm64: Remove hyp_adr/ldr_this_cpu
In reply to David BrazdilThe hyp_adr/ldr_this_cpu helpers were introduced for use in hyp code because they always needed to use TPIDR_EL2 for base, while adr/ldr_this_cpu from kernel proper would select between TPIDR_EL2 and _EL1 based on VHE/nVHE.
Simplify this now that the hyp mode case can be handled using the __KVM_VHE/NVHE_HYPERVISOR__ macros.
--- arch/arm64/include/asm/assembler.h | 27 +++++++++++++++++---------- arch/arm64/include/asm/kvm_asm.h | 14 +------------- arch/arm64/kvm/hyp/hyp-entry.S | 2 +- 3 files changed, 19 insertions(+), 24 deletions(-)
arch/arm64/include/asm/assembler.h +17 -10
@@ -218,6 +218,21 @@ lr .req x30 // link register str \src, [\tmp, :lo12:\sym] .endm + /* + * @dst: destination register (32 or 64 bit wide) + */ + .macro this_cpu_offset, dst +#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) + mrs \dst, tpidr_el2 +#else +alternative_if_not ARM64_HAS_VIRT_HOST_EXTN + mrs \dst, tpidr_el1 +alternative_else + mrs \dst, tpidr_el2 +alternative_endif +#endif + .endm +/* * @dst: Result of per_cpu(sym, smp_processor_id()) (can be SP) * @sym: The name of the per-cpu variable
@@ -226,11 +241,7 @@ lr .req x30 // link register .macro adr_this_cpu, dst, sym, tmp adrp \tmp, \sym add \dst, \tmp, #:lo12:\sym -alternative_if_not ARM64_HAS_VIRT_HOST_EXTN - mrs \tmp, tpidr_el1 -alternative_else - mrs \tmp, tpidr_el2 -alternative_endif + this_cpu_offset \tmp add \dst, \dst, \tmp .endm
@@ -241,11 +252,7 @@ alternative_endif*/ .macro ldr_this_cpu dst, sym, tmp adr_l \dst, \sym -alternative_if_not ARM64_HAS_VIRT_HOST_EXTN - mrs \tmp, tpidr_el1 -alternative_else - mrs \tmp, tpidr_el2 -alternative_endif + this_cpu_offset \tmp ldr \dst, [\dst, \tmp] .endm
arch/arm64/include/asm/kvm_asm.h +1 -13
@@ -173,20 +173,8 @@ extern char __smccc_workaround_1_smc[__SMCCC_WORKAROUND_1_SMC_SZ];#else /* __ASSEMBLY__ */ -.macro hyp_adr_this_cpu reg, sym, tmp - adr_l \reg, \sym - mrs \tmp, tpidr_el2 - add \reg, \reg, \tmp -.endm - -.macro hyp_ldr_this_cpu reg, sym, tmp - adr_l \reg, \sym - mrs \tmp, tpidr_el2 - ldr \reg, [\reg, \tmp] -.endm - .macro get_host_ctxt reg, tmp - hyp_adr_this_cpu \reg, kvm_host_data, \tmp + adr_this_cpu \reg, kvm_host_data, \tmp add \reg, \reg, #HOST_DATA_CONTEXT .endm
arch/arm64/kvm/hyp/hyp-entry.S +1 -1
@@ -132,7 +132,7 @@ alternative_cb_end str x0, [x2, #VCPU_WORKAROUND_FLAGS] /* Check that we actually need to perform the call */ - hyp_ldr_this_cpu x0, arm64_ssbd_callback_required, x2 + ldr_this_cpu x0, arm64_ssbd_callback_required, x2 cbz x0, wa2_end mov w0, #ARM_SMCCC_ARCH_WORKAROUND_2
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 6/11 kvm: arm64: Add helpers for accessing nVHE hyp per-cpu vars
In reply to David BrazdilDefining a per-CPU variable in hyp/nvhe will result in its name being prefixed with __kvm_nvhe_. Add helpers for declaring these variables in kernel proper and accessing them with this_cpu_ptr and per_cpu_ptr.
--- arch/arm64/include/asm/kvm_asm.h | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)
arch/arm64/include/asm/kvm_asm.h +23 -2
@@ -54,9 +54,21 @@DECLARE_KVM_VHE_SYM (sym); \ DECLARE_KVM_NVHE_SYM(sym) +#define DECLARE_KVM_VHE_PER_CPU (type, sym) \ + DECLARE_PER_CPU(type, sym) +#define DECLARE_KVM_NVHE_PER_CPU (type, sym) \ + DECLARE_PER_CPU(type, kvm_nvhe_sym(sym)) + +#define DECLARE_KVM_HYP_PER_CPU (type, sym) \ + DECLARE_KVM_VHE_PER_CPU(type, sym); \ + DECLARE_KVM_NVHE_PER_CPU(type, sym) +#define CHOOSE_VHE_SYM (sym) sym#define CHOOSE_NVHE_SYM (sym) kvm_nvhe_sym(sym) +#define this_cpu_ptr_nvhe (sym) this_cpu_ptr(&kvm_nvhe_sym(sym)) +#define per_cpu_ptr_nvhe (sym, cpu) per_cpu_ptr(&kvm_nvhe_sym(sym), cpu) +#ifndef __KVM_NVHE_HYPERVISOR__/* * BIG FAT WARNINGS:
@@ -69,12 +81,21 @@* - Don't let the nVHE hypervisor have access to this, as it will * pick the *wrong* symbol (yes, it runs at EL2...). */ -#define CHOOSE_HYP_SYM (sym) (is_kernel_in_hyp_mode() ? CHOOSE_VHE_SYM(sym) \ +#define CHOOSE_HYP_SYM (sym) (is_kernel_in_hyp_mode() \ + ? CHOOSE_VHE_SYM(sym) \ : CHOOSE_NVHE_SYM(sym)) +#define this_cpu_ptr_hyp (sym) (is_kernel_in_hyp_mode() \ + ? this_cpu_ptr(&sym) \ + : this_cpu_ptr_nvhe(sym)) +#define per_cpu_ptr_hyp (sym, cpu) (is_kernel_in_hyp_mode() \ + ? per_cpu_ptr(&sym, cpu) \ + : per_cpu_ptr_nvhe(sym, cpu))#else /* The nVHE hypervisor shouldn't even try to access anything */ extern void *__nvhe_undefined_symbol; -#define CHOOSE_HYP_SYM (sym) __nvhe_undefined_symbol +#define CHOOSE_HYP_SYM (sym) __nvhe_undefined_symbol +#define this_cpu_ptr_hyp (sym) (&__nvhe_undefined_symbol) +#define per_cpu_ptr_hyp (sym, cpu) (&__nvhe_undefined_symbol)#endif /* Translate a kernel address @ptr into its equivalent linear mapping */
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 7/11 kvm: arm64: Duplicate arm64_ssbd_callback_required for nVHE hyp
In reply to David BrazdilHyp keeps track of which cores require SSBD callback by accessing a kernel-proper global variable. Create an nVHE symbol of the same name and copy the value from kernel proper to nVHE at KVM init time.
Done in preparation for separating percpu memory owned by kernel proper and nVHE.
--- arch/arm64/include/asm/kvm_mmu.h | 10 +++++++--- arch/arm64/kernel/image-vars.h | 1 - arch/arm64/kvm/arm.c | 2 +- arch/arm64/kvm/hyp/nvhe/switch.c | 3 +++ 4 files changed, 11 insertions(+), 5 deletions(-)
arch/arm64/include/asm/kvm_mmu.h +7 -3
@@ -529,23 +529,27 @@ static inline int kvm_map_vectors(void)#ifdef CONFIG_ARM64_SSBD DECLARE_PER_CPU_READ_MOSTLY (u64, arm64_ssbd_callback_required); +DECLARE_KVM_NVHE_PER_CPU (u64, arm64_ssbd_callback_required); -static inline int hyp_map_aux_data (void ) +static inline int hyp_init_aux_data (void ) {int cpu, err;for_each_possible_cpu (cpu) {u64 *ptr; - ptr =per_cpu_ptr (&arm64_ssbd_callback_required, cpu); + ptr =per_cpu_ptr_nvhe (arm64_ssbd_callback_required, cpu); err =create_hyp_mappings (ptr, ptr +1 ,PAGE_HYP );if (err)return err; + +/* Copy value from kernel to hyp. */ + *ptr =per_cpu (arm64_ssbd_callback_required, cpu); }return 0 ; }#else -static inline int hyp_map_aux_data (void ) +static inline int hyp_init_aux_data (void ) {return 0 ; }
arch/arm64/kernel/image-vars.h +0 -1
@@ -67,7 +67,6 @@ KVM_NVHE_ALIAS(kvm_patch_vector_branch);KVM_NVHE_ALIAS (kvm_update_va_mask);/* Global kernel state accessed by nVHE hyp code. */ -KVM_NVHE_ALIAS (arm64_ssbd_callback_required);KVM_NVHE_ALIAS (kvm_host_data);KVM_NVHE_ALIAS (kvm_vgic_global_state);
arch/arm64/kvm/arm.c +1 -1
@@ -1552,7 +1552,7 @@ static int init_hyp_mode(void) } } - err =hyp_map_aux_data (); + err =hyp_init_aux_data ();if (err)kvm_err ("Cannot map host auxiliary data: %d\n" , err);
arch/arm64/kvm/hyp/nvhe/switch.c +3 -0
@@ -27,6 +27,9 @@#include <asm/processor.h> #include <asm/thread_info.h> +/* Non-VHE copy of the kernel symbol. */ +DEFINE_PER_CPU_READ_MOSTLY (u64, arm64_ssbd_callback_required); +static void __activate_traps (struct kvm_vcpu *vcpu) {u64 val;
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 8/11 kvm: arm64: Create separate instances of kvm_host_data for VHE/nVHE
In reply to David BrazdilHost CPU context is stored in a global per-cpu variable kvm_host_data. In preparation for introducing independent per-CPU region for nVHE hyp, create two separate instances of kvm_host_data, one for VHE and one for nVHE.
--- arch/arm64/include/asm/kvm_host.h | 2 +- arch/arm64/kernel/image-vars.h | 1 - arch/arm64/kvm/arm.c | 5 ++--- arch/arm64/kvm/hyp/nvhe/switch.c | 3 +++ arch/arm64/kvm/hyp/vhe/switch.c | 3 +++ arch/arm64/kvm/pmu.c | 8 ++++---- 6 files changed, 13 insertions(+), 9 deletions(-)
arch/arm64/include/asm/kvm_host.h +1 -1
@@ -565,7 +565,7 @@ void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);struct kvm_vcpu *kvm_mpidr_to_vcpu (struct kvm *kvm,unsigned long mpidr); -DECLARE_PER_CPU (kvm_host_data_t, kvm_host_data); +DECLARE_KVM_HYP_PER_CPU (kvm_host_data_t, kvm_host_data);static inline void kvm_init_host_cpu_context (struct kvm_cpu_context *cpu_ctxt) {
arch/arm64/kernel/image-vars.h +0 -1
@@ -67,7 +67,6 @@ KVM_NVHE_ALIAS(kvm_patch_vector_branch);KVM_NVHE_ALIAS (kvm_update_va_mask);/* Global kernel state accessed by nVHE hyp code. */ -KVM_NVHE_ALIAS (kvm_host_data);KVM_NVHE_ALIAS (kvm_vgic_global_state);/* Kernel constant needed to compute idmap addresses. */
arch/arm64/kvm/arm.c +2 -3
@@ -46,7 +46,6 @@ __asm__(".arch_extension virt" );#endif -DEFINE_PER_CPU (kvm_host_data_t, kvm_host_data); static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);/* The VMID used in the VTTBR */
@@ -1308,7 +1307,7 @@ static void cpu_hyp_reset(void)static void cpu_hyp_reinit (void ) { -kvm_init_host_cpu_context (&this_cpu_ptr (&kvm_host_data)->host_ctxt); +kvm_init_host_cpu_context (&this_cpu_ptr_hyp (kvm_host_data)->host_ctxt);cpu_hyp_reset ();
@@ -1543,7 +1542,7 @@ static int init_hyp_mode(void)for_each_possible_cpu (cpu) {kvm_host_data_t *cpu_data; - cpu_data =per_cpu_ptr (&kvm_host_data, cpu); + cpu_data =per_cpu_ptr_hyp (kvm_host_data, cpu); err =create_hyp_mappings (cpu_data, cpu_data +1 ,PAGE_HYP );if (err) {
arch/arm64/kvm/hyp/nvhe/switch.c +3 -0
@@ -30,6 +30,9 @@/* Non-VHE copy of the kernel symbol. */ DEFINE_PER_CPU_READ_MOSTLY (u64, arm64_ssbd_callback_required); +/* Non-VHE instance of kvm_host_data. */ +DEFINE_PER_CPU (kvm_host_data_t, kvm_host_data); +static void __activate_traps (struct kvm_vcpu *vcpu) {u64 val;
arch/arm64/kvm/hyp/vhe/switch.c +3 -0
@@ -28,6 +28,9 @@const char __hyp_panic_string[] ="HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n" ; +/* VHE instance of kvm_host_data. */ +DEFINE_PER_CPU (kvm_host_data_t, kvm_host_data); +static void __activate_traps (struct kvm_vcpu *vcpu) {u64 val;
arch/arm64/kvm/pmu.c +4 -4
@@ -31,7 +31,7 @@ static bool kvm_pmu_switch_needed(struct perf_event_attr *attr)*/ void kvm_set_pmu_events (u32 set,struct perf_event_attr *attr) { -struct kvm_host_data *ctx =this_cpu_ptr (&kvm_host_data); +struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data);if (!kvm_pmu_switch_needed (attr))return ;
@@ -47,7 +47,7 @@ void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr)*/ void kvm_clr_pmu_events (u32 clr) { -struct kvm_host_data *ctx =this_cpu_ptr (&kvm_host_data); +struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); ctx->pmu_events.events_host &= ~clr; ctx->pmu_events.events_guest &= ~clr;
@@ -173,7 +173,7 @@ void kvm_vcpu_pmu_restore_guest(struct kvm_vcpu *vcpu)return ;preempt_disable (); - host =this_cpu_ptr (&kvm_host_data); + host =this_cpu_ptr_hyp (kvm_host_data); events_guest = host->pmu_events.events_guest; events_host = host->pmu_events.events_host;
@@ -193,7 +193,7 @@ void kvm_vcpu_pmu_restore_host(struct kvm_vcpu *vcpu)if (!has_vhe ())return ; - host =this_cpu_ptr (&kvm_host_data); + host =this_cpu_ptr_hyp (kvm_host_data); events_guest = host->pmu_events.events_guest; events_host = host->pmu_events.events_host;
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 10/11 kvm: arm64: Set up hyp percpu data for nVHE
In reply to David BrazdilAdd hyp percpu section to linker script and rename the corresponding ELF sections of hyp/nvhe object files. This moves all nVHE-specific percpu variables to the new hyp percpu section.
Allocate sufficient amount of memory for all percpu hyp regions at global KVM init time and create corresponding hyp mappings.
The base addresses of hyp percpu regions are kept in a dynamically allocated array in the kernel.
Add NULL checks in PMU event-reset code as it may run before KVM memory is initialized.
--- arch/arm64/include/asm/kvm_asm.h | 19 +++++++++-- arch/arm64/include/asm/sections.h | 1 + arch/arm64/kernel/vmlinux.lds.S | 9 +++++ arch/arm64/kvm/arm.c | 56 +++++++++++++++++++++++++++++-- arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 6 ++++ arch/arm64/kvm/pmu.c | 5 ++- 6 files changed, 90 insertions(+), 6 deletions(-)
arch/arm64/include/asm/kvm_asm.h +17 -2
@@ -66,8 +66,23 @@#define CHOOSE_VHE_SYM (sym) sym#define CHOOSE_NVHE_SYM (sym) kvm_nvhe_sym(sym) -#define this_cpu_ptr_nvhe (sym) this_cpu_ptr(&kvm_nvhe_sym(sym)) -#define per_cpu_ptr_nvhe (sym, cpu) per_cpu_ptr(&kvm_nvhe_sym(sym), cpu) +/* Array of percpu base addresses. Length of the array is nr_cpu_ids. */ +extern unsigned long *kvm_arm_hyp_percpu_base; + +/* +* Compute pointer to a symbol defined in nVHE percpu region. +* Returns NULL if percpu memory has not been allocated yet. +*/ +#define this_cpu_ptr_nvhe (sym) per_cpu_ptr_nvhe(sym, smp_processor_id()) +#define per_cpu_ptr_nvhe (sym, cpu) \ + ({ \ + unsigned long base, off; \ + base = kvm_arm_hyp_percpu_base \ + ? kvm_arm_hyp_percpu_base[cpu] : 0; \ + off = (unsigned long)&kvm_nvhe_sym(sym) - \ + (unsigned long)&kvm_nvhe_sym(__per_cpu_start); \ + base ? (typeof(kvm_nvhe_sym(sym))*)(base + off) : NULL; \ + })#ifndef __KVM_NVHE_HYPERVISOR__/*
arch/arm64/include/asm/sections.h +1 -0
@@ -18,5 +18,6 @@ extern char __exittext_begin[], __exittext_end[];extern char __irqentry_text_start[], __irqentry_text_end[];extern char __mmuoff_data_start[], __mmuoff_data_end[];extern char __entry_tramp_text_start[], __entry_tramp_text_end[]; +extern char __kvm_nvhe___per_cpu_start[], __kvm_nvhe___per_cpu_end[];#endif /* __ASM_SECTIONS_H */
arch/arm64/kernel/vmlinux.lds.S +9 -0
@@ -9,6 +9,7 @@ #include <asm-generic/vmlinux.lds.h> #include <asm/cache.h> +#include <asm/hyp_image.h> #include <asm/kernel-pgtable.h> #include <asm/memory.h> #include <asm/page.h>
@@ -27,8 +28,15 @@ jiffies = jiffies_64; __start___kvm_ex_table = .; \ *(__kvm_ex_table) \ __stop___kvm_ex_table = .; + +#define HYPERVISOR_PERCPU_SECTION \ + . = ALIGN(PAGE_SIZE); \ + HYP_SECTION_NAME(.data..percpu) : { \ + *(HYP_SECTION_NAME(.data..percpu)) \ + } #else /* CONFIG_KVM */ #define HYPERVISOR_EXTABLE +#define HYPERVISOR_PERCPU_SECTION #endif #define HYPERVISOR_TEXT \
@@ -194,6 +202,7 @@ SECTIONS } PERCPU_SECTION(L1_CACHE_BYTES) + HYPERVISOR_PERCPU_SECTION .rela.dyn : ALIGN(8) { *(.rela .rela*)
arch/arm64/kvm/arm.c +53 -3
93 lines not on this page · raw message
arch/arm64/kvm/hyp/nvhe/hyp.lds.S +6 -0
@@ -7,7 +7,13 @@ */ #include <asm/hyp_image.h> +#include <asm-generic/vmlinux.lds.h> +#include <asm/cache.h> +#include <asm/memory.h> SECTIONS { HYP_SECTION(.text) + HYP_SECTION_NAME(.data..percpu) : { + PERCPU_INPUT(L1_CACHE_BYTES) + } }
arch/arm64/kvm/pmu.c +4 -1
@@ -33,7 +33,7 @@ void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); -if (!kvm_pmu_switch_needed (attr)) +if (!ctx || !kvm_pmu_switch_needed (attr))return ;if (!attr->exclude_host)
@@ -49,6 +49,9 @@ void kvm_clr_pmu_events(u32 clr) {struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); +if (!ctx) +return ; + ctx->pmu_events.events_host &= ~clr; ctx->pmu_events.events_guest &= ~clr; }
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 9/11 kvm: arm64: Mark hyp stack pages reserved
In reply to David BrazdilIn preparation for unmapping hyp pages from host stage-2, allocate/free hyp stack using new helpers which automatically mark the pages reserved.
--- arch/arm64/kvm/arm.c | 49 ++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 47 insertions(+), 2 deletions(-)
arch/arm64/kvm/arm.c +47 -2
@@ -1456,13 +1456,58 @@ static int init_subsystems(void)return err; } +/* +* Alloc pages and mark them reserved so the kernel never tries to +* take them away from the hypervisor. +*/ +static unsigned long alloc_hyp_pages (gfp_t flags,unsigned int order) +{ +struct page *page; +unsigned long i; + + page =alloc_pages (flags, order); +if (!page) +return 0 ; + +for (i =0 ; i < (1ul << order); ++i) +mark_page_reserved (page + i); + +return (unsigned long )page_address (page); +} + +static unsigned long alloc_hyp_page (gfp_t flags) +{ +return alloc_hyp_pages (flags,0 ); +} + +/* +* Free pages which were previously marked reserved for the hypervisor. +*/ +static void free_hyp_pages (unsigned long addr,unsigned int order) +{ +unsigned long i; +struct page *page; + +if (!addr) +return ; + + page =virt_to_page (addr); +for (i =0 ; i < (1ul << order); ++i) +free_reserved_page (page + i); +} + +static void free_hyp_page (unsigned long addr) +{ +return free_hyp_pages (addr,0 ); +} +static void teardown_hyp_mode (void ) {int cpu;free_hyp_pgds ();for_each_possible_cpu (cpu) -free_page (per_cpu (kvm_arm_hyp_stack_page, cpu)); +free_hyp_page (per_cpu (kvm_arm_hyp_stack_page, cpu)); }/**
@@ -1486,7 +1531,7 @@ static int init_hyp_mode(void)for_each_possible_cpu (cpu) {unsigned long stack_page; - stack_page =__get_free_page (GFP_KERNEL ); + stack_page =alloc_hyp_page (GFP_KERNEL );if (!stack_page) { err = -ENOMEM ; goto out_err;
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
PATCH v3 11/11 kvm: arm64: Remove unnecessary hyp mappings
In reply to David BrazdilWith all nVHE per-CPU variables being part of the hyp per-CPU region, mapping them individual is not necessary any longer. They are mapped to hyp as part of the overall per-CPU region.
--- arch/arm64/include/asm/kvm_mmu.h | 19 +++++-------------- arch/arm64/kvm/arm.c | 17 +---------------- 2 files changed, 6 insertions(+), 30 deletions(-)
arch/arm64/include/asm/kvm_mmu.h +5 -14
@@ -531,28 +531,19 @@ static inline int kvm_map_vectors(void)DECLARE_PER_CPU_READ_MOSTLY (u64, arm64_ssbd_callback_required);DECLARE_KVM_NVHE_PER_CPU (u64, arm64_ssbd_callback_required); -static inline int hyp_init_aux_data (void ) +static inline void hyp_init_aux_data (void ) { -int cpu, err; +int cpu; +/* Copy arm64_ssbd_callback_required values from kernel to hyp. */ for_each_possible_cpu (cpu) { -u64 *ptr; +u64 *ptr =per_cpu_ptr_nvhe (arm64_ssbd_callback_required, cpu); - ptr =per_cpu_ptr_nvhe (arm64_ssbd_callback_required, cpu); - err =create_hyp_mappings (ptr, ptr +1 ,PAGE_HYP ); -if (err) -return err; - -/* Copy value from kernel to hyp. */ *ptr =per_cpu (arm64_ssbd_callback_required, cpu); } -return 0 ; }#else -static inline int hyp_init_aux_data (void ) -{ -return 0 ; -} +static inline void hyp_init_aux_data (void ) {}#endif #define kvm_phys_to_vttbr (addr) phys_to_ttbr(addr)
arch/arm64/kvm/arm.c +1 -16
@@ -1634,22 +1634,7 @@ static int init_hyp_mode(void) } } -for_each_possible_cpu (cpu) { -kvm_host_data_t *cpu_data; - - cpu_data =per_cpu_ptr_hyp (kvm_host_data, cpu); - err =create_hyp_mappings (cpu_data, cpu_data +1 ,PAGE_HYP ); - -if (err) { -kvm_err ("Cannot map host CPU state: %d\n" , err); - goto out_err; - } - } - - err =hyp_init_aux_data (); -if (err) -kvm_err ("Cannot map host auxiliary data: %d\n" , err); - +hyp_init_aux_data ();return 0 ; out_err:
Signature · 6 lines
2.28.0.618.gf4bc123cb7-goog _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
Re: [PATCH v3 01/11] kvm: arm64: Partially link nVHE hyp code, simplify HYPCOPY
In reply to David BrazdilQuoted 43 lines · David Brazdil
Relying on objcopy to prefix the ELF section names of the nVHE hyp code is brittle and prevents us from using wildcards to match specific section names.
Improve the build rules by partially linking all '.nvhe.o' files and prefixing their ELF section names using a linker script. Continue using objcopy for prefixing ELF symbol names.
One immediate advantage of this approach is that all subsections matching a pattern can be merged into a single prefixed section, eg. .text and .text.* can be linked into a single '.hyp.text'. This removes the need for -fno-reorder-functions on GCC and will be useful in the future too: LTO builds use .text subsections, compilers routinely generate .rodata subsections, etc.
Partially linking all hyp code into a single object file also makes it easier to analyze.
Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/hyp_image.h | 24 ++++++++++++ arch/arm64/kvm/hyp/nvhe/Makefile | 60 ++++++++++++++++-------------- arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 13 +++++++ 3 files changed, 70 insertions(+), 27 deletions(-) create mode 100644 arch/arm64/include/asm/hyp_image.h create mode 100644 arch/arm64/kvm/hyp/nvhe/hyp.lds.Sarch/arm64/include/asm/hyp_image.h new file +10
-0@@ -0,0 +1,24 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* Copyright (C) 2020 Google LLC. +* Written by David Brazdil <[email protected]> +*/ + +#ifndef __ARM64_HYP_IMAGE_H__ +#define __ARM64_HYP_IMAGE_H__ + +#ifdef LINKER_SCRIPT
arch/arm64/include/asm/hyp_image.h line 20
Quoted 10 lines
arch/arm64/include/asm/hyp_image.h new file +10
-0+ +/* +* KVM nVHE ELF section names are prefixed with .hyp, to separate them +* from the kernel proper. +*/ +#define HYP_SECTION_NAME (NAME ) .hyp##NAME + +/* Defines an ELF hyp section from input section @NAME and its subsections. */ +#define HYP_SECTION (NAME ) \ + HYP_SECTION_NAME(NAME) : { *(NAME NAME##.[0-9a-zA-Z_]*) }
I still don't get why we can't just use NAME ## .* for the regex here. That matches what we do elsewhere for linker script wildcarding, e.g. .rodata.*, .init.text.* ... in asm-generic/vmlinux.lds.h. Why is it different for these sections?
Quoted 32 lines
arch/arm64/include/asm/hyp_image.h new file +4
-0+ +#endif /* LINKER_SCRIPT */ + +#endif /* __ARM64_HYP_IMAGE_H__ */ arch/arm64/kvm/hyp/nvhe/Makefile +13
-4@@ -10,40 +10,46 @@ obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o obj-y += ../vgic-v3-sr.o ../aarch32.o ../vgic-v2-cpuif-proxy.o ../entry.o \ ../fpsimd.o ../hyp-entry.o -obj-y := $(patsubst %.o,%.hyp.o,$(obj-y)) -extra-y := $(patsubst %.hyp.o,%.hyp.tmp.o,$(obj-y)) +## +## Build rules for compiling nVHE hyp code +## Output of this folder is `kvm_nvhe.o`, a partially linked object +## file containing all nVHE hyp code and data. +## -$(obj)/%.hyp.tmp.o: $(src)/%.c FORCE +hyp-obj := $(patsubst %.o,%.nvhe.o,$(obj-y)) +obj-y := kvm_nvhe.o +extra-y := $(hyp-obj) kvm_nvhe.tmp.o hyp.lds + +# 1) Compile all source files to `.nvhe.o` object files. The file extension +# avoids file name clashes for files shared with VHE. +$(obj)/%.nvhe.o: $(src)/%.c FORCE $(call if_changed_rule,cc_o_c) -$(obj)/%.hyp.tmp.o: $(src)/%.S FORCE +$(obj)/%.nvhe.o: $(src)/%.S FORCE $(call if_changed_rule,as_o_S)
arch/arm64/kvm/hyp/nvhe/Makefile obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o · line 32
Quoted 10 lines
arch/arm64/kvm/hyp/nvhe/Makefile +3
-6-$(obj)/%.hyp.o: $(obj)/%.hyp.tmp.o FORCE - $(call if_changed,hypcopy) -# Disable reordering functions by GCC (enabled at -O2). -# This pass puts functions into '.text.*' sections to aid the linker -# in optimizing ELF layout. See HYPCOPY comment below for more info. -ccflags-y += $(call cc-option,-fno-reorder-functions) +# 2) Compile linker script. +$(obj)/hyp.lds: $(src)/hyp.lds.S FORCE + $(call if_changed_dep,cpp_lds_S)
You need a .gitignore file listing hyp.lds, otherwise some idiot will end up committing it. I definitely didn't do that when playing around with this series. Nope. Not at all.
With that, and the regex resolved:
Will
Re: [PATCH v3 02/11] kvm: arm64: Move nVHE hyp namespace macros to hyp_image.h
In reply to David BrazdilQuoted 9 lines · David Brazdil
Minor cleanup to move all macros related to prefixing nVHE hyp section and symbol names into one place:
hyp_image.h.Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/hyp_image.h | 12 ++++++++++++ arch/arm64/include/asm/kvm_asm.h | 8 +------- arch/arm64/kernel/image-vars.h | 2 -- 3 files changed, 13 insertions(+), 9 deletions(-)
Will
Will Deacon
maintainerRe: [PATCH v3 04/11] kvm: arm64: Remove __hyp_this_cpu_read
In reply to David BrazdilQuoted 60 lines · David Brazdil
this_cpu_ptris meant for use in kernel proper because it selects between TPIDR_EL1/2 based on nVHE/VHE.__hyp_this_cpu_ptrwas used in hyp to always selectTPIDR_EL2. Unify all users behindthis_cpu_ptrand friends by selecting_EL2register under __KVM_VHE/NVHE_HYPERVISOR__.Under
CONFIG_DEBUG_PREEMPT, the kernel helpers perform a preemption check which is omitted by the hyp helpers. Preserve the behavior for nVHE by overriding the corresponding macros under__KVM_NVHE_HYPERVISOR__. Extend the checks into VHE hyp code.Acked-by Andrew Scull <[email protected]>Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/kvm_asm.h | 20 -------------- arch/arm64/include/asm/percpu.h | 33 +++++++++++++++++++++-- arch/arm64/kvm/hyp/include/hyp/debug-sr.h | 4 +-- arch/arm64/kvm/hyp/include/hyp/switch.h | 8 +++--- arch/arm64/kvm/hyp/nvhe/switch.c | 2 +- arch/arm64/kvm/hyp/vhe/switch.c | 2 +- arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 4 +-- 7 files changed, 41 insertions(+), 32 deletions(-)arch/arm64/include/asm/kvm_asm.h +0
-20@@ -143,26 +143,6 @@ extern char __smccc_workaround_1_smc[__SMCCC_WORKAROUND_1_SMC_SZ]; addr; \ }) -/* -* Home-grown __this_cpu_{ptr,read} variants that always work at HYP, -* provided that sym is really a *symbol* and not a pointer obtained from -* a data structure. As for SHIFT_PERCPU_PTR(), the creative casting keeps -* sparse quiet. -*/ -#define __hyp_this_cpu_ptr (sym) \ - ({ \ - void *__ptr; \ - __verify_pcpu_ptr(&sym); \ - __ptr = hyp_symbol_addr(sym); \ - __ptr += read_sysreg(tpidr_el2); \ - (typeof(sym) __kernel __force *)__ptr; \ - }) - -#define __hyp_this_cpu_read (sym) \ - ({ \ - *__hyp_this_cpu_ptr(sym); \ - }) -#define __KVM_EXTABLE (from, to) \ " .pushsection __kvm_ex_table, \"a\"\n" \ " .align 3\n" \arch/arm64/include/asm/percpu.h +0
-0@@ -19,7 +19,21 @@ static inline void set_my_cpu_offset(unsigned long off) :: "r" (off) : "memory"); }
arch/arm64/include/asm/percpu.h static inline void set_my_cpu_offset(unsigned long off) · line 29
Quoted 10 lines
arch/arm64/include/asm/percpu.h +8
-1-static inline unsigned long __my_cpu_offset (void ) +static inline unsigned long __hyp_my_cpu_offset (void ) +{ +unsigned long off; + +/* +* We want to allow caching the value, so avoid using volatile and +* instead use a fake stack read to hazard against barrier(). +*/
I don't think we need to copy/paste the comment...
arch/arm64/include/asm/percpu.h static inline void set_my_cpu_offset(unsigned long off) · line 31
Quoted 2 lines
arch/arm64/include/asm/percpu.h +2
-0+ asm("mrs %0, tpidr_el2" :"=r" (off) : +"Q" (*(const unsigned long *)current_stack_pointer));
... especially given that we're not preemptible at EL2 with nVHE, maybe we don't need to play this trick at all because we're always going to be on the same CPU. So we could actually just do:
return read_sysreg(tpidr_el2);
which is much better, and the comment should say something to that effect.
Quoted 9 lines
arch/arm64/include/asm/percpu.h +5
-0+ +return off; +} + +static inline unsigned long __kern_my_cpu_offset (void ) {unsigned long off;
arch/arm64/include/asm/percpu.h static inline unsigned long __my_cpu_offset(void) · line 54
Quoted 7 lines
arch/arm64/include/asm/percpu.h +3
-1@@ -35,7 +49,12 @@ static inline unsigned long __my_cpu_offset(void)return off; } -#define __my_cpu_offset __my_cpu_offset() + +#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) +#define __my_cpu_offset __hyp_my_cpu_offset()
Why would VHE code need to use this? Especially in light of my preemption comments above, shouldn't it now be using __kern_my_cpu_offset()?
Will
Re: [PATCH v3 03/11] kvm: arm64: Only define __kvm_ex_table for CONFIG_KVM
In reply to David BrazdilQuoted 7 lines · David Brazdil
Minor cleanup that only creates
__kvm_ex_tableELF section and related symbols ifCONFIG_KVMis enabled.Signed-off-by David Brazdil <[email protected]>--- arch/arm64/kernel/vmlinux.lds.S | 4 ++++ 1 file changed, 4 insertions(+)
It's also useful because we're about to add another entry here, so:
Will
Re: [PATCH v3 05/11] kvm: arm64: Remove hyp_adr/ldr_this_cpu
In reply to David BrazdilQuoted 27 lines · David Brazdil
The hyp_adr/ldr_this_cpu helpers were introduced for use in hyp code because they always needed to use
TPIDR_EL2for base, while adr/ldr_this_cpu from kernel proper would select betweenTPIDR_EL2and_EL1based on VHE/nVHE.Simplify this now that the hyp mode case can be handled using the __KVM_VHE/NVHE_HYPERVISOR__ macros.
Acked-by Andrew Scull <[email protected]>Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/assembler.h | 27 +++++++++++++++++---------- arch/arm64/include/asm/kvm_asm.h | 14 +------------- arch/arm64/kvm/hyp/hyp-entry.S | 2 +- 3 files changed, 19 insertions(+), 24 deletions(-)arch/arm64/include/asm/assembler.h +3
-0@@ -218,6 +218,21 @@ lr .req x30 // link register str \src, [\tmp, :lo12:\sym] .endm +/* +* @dst: destination register (32 or 64 bit wide) +*/
arch/arm64/include/asm/assembler.h lr .req x30 // link register · line 233
Quoted 10 lines
arch/arm64/include/asm/assembler.h +10
-0+ .macro this_cpu_offset, dst +#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) + mrs \dst, tpidr_el2 +#else +alternative_if_not ARM64_HAS_VIRT_HOST_EXTN + mrs \dst, tpidr_el1 +alternative_else + mrs \dst, tpidr_el2 +alternative_endif +#endif
Cosmetic, but I think it would be cleaner just to define two variants of the macro here:
#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__)
.macro this_cpu_offset, dst
mrs \dst, tpidr_el2
.endm
#else
.macro this_cpu_offset, dstalternative_if_not ARM64_HAS_VIRT_HOST_EXTN
mrs \dst, tpidr_el1alternative_else
mrs \dst, tpidr_el2alternative_endif
.endm
#endif
(and should we have a shorthand __HYPERVISOR define to avoid the NVHE || VHE logic?)
With that:
Will
Re: [PATCH v3 06/11] kvm: arm64: Add helpers for accessing nVHE hyp per-cpu vars
In reply to David BrazdilQuoted 21 lines · David Brazdil
Defining a per-CPU variable in hyp/nvhe will result in its name being prefixed with
__kvm_nvhe_. Add helpers for declaring these variables in kernel proper and accessing them withthis_cpu_ptrandper_cpu_ptr.Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/kvm_asm.h | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-)arch/arm64/include/asm/kvm_asm.h +4
-0@@ -54,9 +54,21 @@DECLARE_KVM_VHE_SYM (sym); \ DECLARE_KVM_NVHE_SYM(sym) +#define DECLARE_KVM_VHE_PER_CPU (type, sym) \ + DECLARE_PER_CPU(type, sym) +#define DECLARE_KVM_NVHE_PER_CPU (type, sym) \ + DECLARE_PER_CPU(type, kvm_nvhe_sym(sym))
arch/arm64/include/asm/kvm_asm.h line 70
Quoted 10 lines
arch/arm64/include/asm/kvm_asm.h +7
-0+ +#define DECLARE_KVM_HYP_PER_CPU (type, sym) \ + DECLARE_KVM_VHE_PER_CPU(type, sym); \ + DECLARE_KVM_NVHE_PER_CPU(type, sym) +#define CHOOSE_VHE_SYM (sym) sym#define CHOOSE_NVHE_SYM (sym) kvm_nvhe_sym(sym) +#define this_cpu_ptr_nvhe (sym) this_cpu_ptr(&kvm_nvhe_sym(sym)) +#define per_cpu_ptr_nvhe (sym, cpu) per_cpu_ptr(&kvm_nvhe_sym(sym), cpu)
nit: I'd probably stick a _sym suffix on these macros, to make it clear that they're just munging the symbol name rather than doing some completely different pcpu implementation.
THat said, do you expect these to be used outside of the pcpu implementation? If not, I suggest some underscores as a prefix as well.
Quoted 7 lines
arch/arm64/include/asm/kvm_asm.h +0
-0#ifndef __KVM_NVHE_HYPERVISOR__/* * BIG FAT WARNINGS: @@ -69,12 +81,21 @@* - Don't let the nVHE hypervisor have access to this, as it will * pick the *wrong* symbol (yes, it runs at EL2...). */
arch/arm64/include/asm/kvm_asm.h line 92
Quoted 10 lines
arch/arm64/include/asm/kvm_asm.h +8
-1-#define CHOOSE_HYP_SYM (sym) (is_kernel_in_hyp_mode() ? CHOOSE_VHE_SYM(sym) \ +#define CHOOSE_HYP_SYM (sym) (is_kernel_in_hyp_mode() \ + ? CHOOSE_VHE_SYM(sym) \ : CHOOSE_NVHE_SYM(sym)) +#define this_cpu_ptr_hyp (sym) (is_kernel_in_hyp_mode() \ + ? this_cpu_ptr(&sym) \ + : this_cpu_ptr_nvhe(sym)) +#define per_cpu_ptr_hyp (sym, cpu) (is_kernel_in_hyp_mode() \ + ? per_cpu_ptr(&sym, cpu) \ + : per_cpu_ptr_nvhe(sym, cpu))
is_kernel_in_hyp_mode() reads a system register to determine the current exception level, so this doesn't seem like something we should be doing everytime here. Perhaps is_kernel_in_hyp_mode() should avoid read_sysreg() and instead use a non-volatile asm to allow the result to be cached by the compiler. Hmm.
But I think that can be tackled as a future patch, so with the naming nits resolved:
Will
Re: [PATCH v3 08/11] kvm: arm64: Create separate instances of kvm_host_data for VHE/nVHE
In reply to David BrazdilQuoted 62 lines · David Brazdil
Host CPU context is stored in a global per-cpu variable
kvm_host_data. In preparation for introducing independent per-CPU region for nVHE hyp, create two separate instances ofkvm_host_data, one for VHE and one for nVHE.Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/kvm_host.h | 2 +- arch/arm64/kernel/image-vars.h | 1 - arch/arm64/kvm/arm.c | 5 ++--- arch/arm64/kvm/hyp/nvhe/switch.c | 3 +++ arch/arm64/kvm/hyp/vhe/switch.c | 3 +++ arch/arm64/kvm/pmu.c | 8 ++++---- 6 files changed, 13 insertions(+), 9 deletions(-)arch/arm64/include/asm/kvm_host.h +1
-1@@ -565,7 +565,7 @@ void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome);struct kvm_vcpu *kvm_mpidr_to_vcpu (struct kvm *kvm,unsigned long mpidr); -DECLARE_PER_CPU (kvm_host_data_t, kvm_host_data); +DECLARE_KVM_HYP_PER_CPU (kvm_host_data_t, kvm_host_data);static inline void kvm_init_host_cpu_context (struct kvm_cpu_context *cpu_ctxt) {arch/arm64/kernel/image-vars.h +0
-1@@ -67,7 +67,6 @@ KVM_NVHE_ALIAS(kvm_patch_vector_branch);KVM_NVHE_ALIAS (kvm_update_va_mask);/* Global kernel state accessed by nVHE hyp code. */ -KVM_NVHE_ALIAS (kvm_host_data);KVM_NVHE_ALIAS (kvm_vgic_global_state);/* Kernel constant needed to compute idmap addresses. */ arch/arm64/kvm/arm.c +1
-2@@ -46,7 +46,6 @@ __asm__(".arch_extension virt" );#endif -DEFINE_PER_CPU (kvm_host_data_t, kvm_host_data); static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);/* The VMID used in the VTTBR */ @@ -1308,7 +1307,7 @@ static void cpu_hyp_reset(void)static void cpu_hyp_reinit (void ) { -kvm_init_host_cpu_context (&this_cpu_ptr (&kvm_host_data)->host_ctxt); +kvm_init_host_cpu_context (&this_cpu_ptr_hyp (kvm_host_data)->host_ctxt);cpu_hyp_reset ();
arch/arm64/kvm/arm.c static int init_hyp_mode(void) · line 1545
Quoted 5 lines
arch/arm64/kvm/arm.c +1
-1@@ -1543,7 +1542,7 @@ static int init_hyp_mode(void)for_each_possible_cpu (cpu) {kvm_host_data_t *cpu_data; - cpu_data =per_cpu_ptr (&kvm_host_data, cpu); + cpu_data =per_cpu_ptr_hyp (kvm_host_data, cpu);
I stand by my earlier comment to add _sym here, given that the ampersand gets dropped from the argument.
So assuming you do that in the earlier patch:
Will
Will Deacon
maintainerRe: [PATCH v3 07/11] kvm: arm64: Duplicate arm64_ssbd_callback_required for nVHE hyp
In reply to David BrazdilQuoted 31 lines · David Brazdil
Hyp keeps track of which cores require SSBD callback by accessing a kernel-proper global variable. Create an nVHE symbol of the same name and copy the value from kernel proper to nVHE at KVM init time.
Done in preparation for separating percpu memory owned by kernel proper and nVHE.
Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/kvm_mmu.h | 10 +++++++--- arch/arm64/kernel/image-vars.h | 1 - arch/arm64/kvm/arm.c | 2 +- arch/arm64/kvm/hyp/nvhe/switch.c | 3 +++ 4 files changed, 11 insertions(+), 5 deletions(-)arch/arm64/include/asm/kvm_mmu.h +2
-1@@ -529,23 +529,27 @@ static inline int kvm_map_vectors(void)#ifdef CONFIG_ARM64_SSBD DECLARE_PER_CPU_READ_MOSTLY (u64, arm64_ssbd_callback_required); +DECLARE_KVM_NVHE_PER_CPU (u64, arm64_ssbd_callback_required); -static inline int hyp_map_aux_data (void ) +static inline int hyp_init_aux_data (void ) {int cpu, err;for_each_possible_cpu (cpu) {
arch/arm64/include/asm/kvm_mmu.h static inline int kvm_map_vectors(void) · line 547
Quoted 10 lines
arch/arm64/include/asm/kvm_mmu.h +4
-1u64 *ptr; - ptr =per_cpu_ptr (&arm64_ssbd_callback_required, cpu); + ptr =per_cpu_ptr_nvhe (arm64_ssbd_callback_required, cpu); err =create_hyp_mappings (ptr, ptr +1 ,PAGE_HYP );if (err)return err; + +/* Copy value from kernel to hyp. */ + *ptr =per_cpu (arm64_ssbd_callback_required, cpu);
Hmm. Is this correct for late arriving CPUs, where we don't know whether a callback is required at the point we do the copy?
That sounds fiddly to resolve, but this _might_ all be moot because I'm about to post a series that allows us to remove the hyp mapping of this variable entirely. So leave this for now, but maybe stick a comment in that it doesn't work for late cpus.
Will
Will Deacon
maintainerRe: [PATCH v3 09/11] kvm: arm64: Mark hyp stack pages reserved
In reply to David BrazdilQuoted 2 lines · David Brazdil
In preparation for unmapping hyp pages from host stage-2, allocate/free hyp stack using new helpers which automatically mark the pages reserved.
Given that this series doesn't get us that the point of having a stage-2 for the host, I don't see why we need this part yet.
Will
Will Deacon
maintainerRe: [PATCH v3 10/11] kvm: arm64: Set up hyp percpu data for nVHE
In reply to David BrazdilQuoted 31 lines · David Brazdil
Add hyp percpu section to linker script and rename the corresponding ELF sections of hyp/nvhe object files. This moves all nVHE-specific percpu variables to the new hyp percpu section.
Allocate sufficient amount of memory for all percpu hyp regions at global KVM init time and create corresponding hyp mappings.
The base addresses of hyp percpu regions are kept in a dynamically allocated array in the kernel.
Add NULL checks in PMU event-reset code as it may run before KVM memory is initialized.
Signed-off-by David Brazdil <[email protected]>--- arch/arm64/include/asm/kvm_asm.h | 19 +++++++++-- arch/arm64/include/asm/sections.h | 1 + arch/arm64/kernel/vmlinux.lds.S | 9 +++++ arch/arm64/kvm/arm.c | 56 +++++++++++++++++++++++++++++-- arch/arm64/kvm/hyp/nvhe/hyp.lds.S | 6 ++++ arch/arm64/kvm/pmu.c | 5 ++- 6 files changed, 90 insertions(+), 6 deletions(-)arch/arm64/include/asm/kvm_asm.h +0
-0@@ -66,8 +66,23 @@#define CHOOSE_VHE_SYM (sym) sym#define CHOOSE_NVHE_SYM (sym) kvm_nvhe_sym(sym)
arch/arm64/include/asm/kvm_asm.h line 76
Quoted 10 lines
arch/arm64/include/asm/kvm_asm.h +8
-2-#define this_cpu_ptr_nvhe (sym) this_cpu_ptr(&kvm_nvhe_sym(sym)) -#define per_cpu_ptr_nvhe (sym, cpu) per_cpu_ptr(&kvm_nvhe_sym(sym), cpu) +/* Array of percpu base addresses. Length of the array is nr_cpu_ids. */ +extern unsigned long *kvm_arm_hyp_percpu_base; + +/* +* Compute pointer to a symbol defined in nVHE percpu region. +* Returns NULL if percpu memory has not been allocated yet. +*/ +#define this_cpu_ptr_nvhe (sym) per_cpu_ptr_nvhe(sym, smp_processor_id())
Don't you run into similar problems here with the pcpu accessors when CONFIG_DEBUG_PREEMPT=y? I'm worried we can end up with an unresolved reference to debug_smp_processor_id().
Quoted 72 lines
arch/arm64/include/asm/kvm_asm.h +9
-0+#define per_cpu_ptr_nvhe (sym, cpu) \ + ({ \ + unsigned long base, off; \ + base = kvm_arm_hyp_percpu_base \ + ? kvm_arm_hyp_percpu_base[cpu] : 0; \ + off = (unsigned long)&kvm_nvhe_sym(sym) - \ + (unsigned long)&kvm_nvhe_sym(__per_cpu_start); \ + base ? (typeof(kvm_nvhe_sym(sym))*)(base + off) : NULL; \ + })#ifndef __KVM_NVHE_HYPERVISOR__/* arch/arm64/include/asm/sections.h +1
-0@@ -18,5 +18,6 @@ extern char __exittext_begin[], __exittext_end[];extern char __irqentry_text_start[], __irqentry_text_end[];extern char __mmuoff_data_start[], __mmuoff_data_end[];extern char __entry_tramp_text_start[], __entry_tramp_text_end[]; +extern char __kvm_nvhe___per_cpu_start[], __kvm_nvhe___per_cpu_end[];#endif /* __ASM_SECTIONS_H */ arch/arm64/kernel/vmlinux.lds.S +9
-0@@ -9,6 +9,7 @@ #include <asm-generic/vmlinux.lds.h> #include <asm/cache.h> +#include <asm/hyp_image.h> #include <asm/kernel-pgtable.h> #include <asm/memory.h> #include <asm/page.h>@@ -27,8 +28,15 @@ jiffies = jiffies_64; __start___kvm_ex_table = .; \ *(__kvm_ex_table) \ __stop___kvm_ex_table = .; + +#define HYPERVISOR_PERCPU_SECTION \ + . = ALIGN(PAGE_SIZE); \ + HYP_SECTION_NAME(.data..percpu) : { \ + *(HYP_SECTION_NAME(.data..percpu)) \ + } #else /* CONFIG_KVM */ #define HYPERVISOR_EXTABLE +#define HYPERVISOR_PERCPU_SECTION #endif #define HYPERVISOR_TEXT \@@ -194,6 +202,7 @@ SECTIONS } PERCPU_SECTION(L1_CACHE_BYTES) + HYPERVISOR_PERCPU_SECTION .rela.dyn : ALIGN(8) { *(.rela .rela*)arch/arm64/kvm/arm.c +1
-0@@ -47,6 +47,7 @@ __asm__(".arch_extension virt");#endif static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); +unsigned long *kvm_arm_hyp_percpu_base;/* The VMID used in the VTTBR */ static atomic64_t kvm_vmid_gen =ATOMIC64_INIT (1 );
arch/arm64/kvm/arm.c long kvm_arch_vm_ioctl(struct file *filp, · line 1262
Quoted 4 lines
arch/arm64/kvm/arm.c +1
-0@@ -1258,6 +1259,15 @@ long kvm_arch_vm_ioctl(struct file *filp, } } +#define kvm_hyp_percpu_base (cpu) ((unsigned long)per_cpu_ptr_nvhe(__per_cpu_start, cpu))
Having both kvm_arm_hyp_percpu_base and kvm_hyp_percpu_base be so completely different is a recipe for disaster! Please can you rename one/both of them to make it clearer what they represent?
Quoted 14 lines
arch/arm64/kvm/pmu.c +1
-1@@ -33,7 +33,7 @@ void kvm_set_pmu_events(u32 set, struct perf_event_attr *attr) {struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); -if (!kvm_pmu_switch_needed (attr)) +if (!ctx || !kvm_pmu_switch_needed (attr))return ;if (!attr->exclude_host)
arch/arm64/kvm/pmu.c void kvm_clr_pmu_events(u32 clr) · line 53
Quoted 5 lines
arch/arm64/kvm/pmu.c +2
-0@@ -49,6 +49,9 @@ void kvm_clr_pmu_events(u32 clr) {struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); +if (!ctx) +return ;
I guess this should only happen if KVM failed to initialise or something, right? (e.g. if we were booted at EL1). If so, I'm wondering whether it would be better to do something like:
if (!is_hyp_mode_available()) return; WARN_ON_ONCE(!ctx);
so that any unexpected NULL pointer there screams loudly, rather than causes the register switch to be silently ignored. What do you think?
Will
Re: [PATCH v3 04/11] kvm: arm64: Remove __hyp_this_cpu_read
In reply to Will DeaconHi Will,
Quoted 18 lines
arch/arm64/include/asm/percpu.h static inline void set_my_cpu_offset(unsigned long off) · line 29
Quoted 8 lines
arch/arm64/include/asm/percpu.h +8
-0+static inline unsigned long __hyp_my_cpu_offset (void ) +{ +unsigned long off; + +/* +* We want to allow caching the value, so avoid using volatile and +* instead use a fake stack read to hazard against barrier(). +*/ I don't think we need to copy/paste the comment...
arch/arm64/include/asm/percpu.h static inline void set_my_cpu_offset(unsigned long off) · line 31
Quoted 2 lines
arch/arm64/include/asm/percpu.h +2
-0+ asm("mrs %0, tpidr_el2" :"=r" (off) : +"Q" (*(const unsigned long *)current_stack_pointer));... especially given that we're not preemptible at EL2 with nVHE, maybe we don't need to play this trick at all because we're always going to be on the same CPU. So we could actually just do:
Quoted 3 lines
return read_sysreg(tpidr_el2);which is much better, and the comment should say something to that effect.
I must be misinterpreting the comment. I understood that it enables the compiler optimizing multiple reads of TPIDR by avoiding 'asm volatile' (signaling that the value does not change between reads). So what exactly does it do?
read_sysreg expands to 'asm volatile' but I have no problem with priotizing readability over a micro-optimization.
Quoted 5 lines
arch/arm64/include/asm/percpu.h static inline unsigned long __my_cpu_offset(void) · line 54
Quoted 2 lines
arch/arm64/include/asm/percpu.h +2
-0+#if defined(__KVM_NVHE_HYPERVISOR__) || defined(__KVM_VHE_HYPERVISOR__) +#define __my_cpu_offset __hyp_my_cpu_offset()Why would VHE code need to use this? Especially in light of my preemption comments above, shouldn't it now be using
__kern_my_cpu_offset()?
During v2 review Andrew Scull pointed out we can avoid alternatives on VHE code by using __hyp_my_cpu_offset for it as well. Obviously if __hyp_my_cpu_offset becomes nVHE-specific, we can always move VHE back to __kern. This was just about saving a few cycles during boot.
David
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
Re: [PATCH v3 05/11] kvm: arm64: Remove hyp_adr/ldr_this_cpu
In reply to Will DeaconHi Will,
Quoted 16 lines
Cosmetic, but I think it would be cleaner just to define two variants of the macro here:
#if
defined(__KVM_NVHE_HYPERVISOR__)||defined(__KVM_VHE_HYPERVISOR__)
.macrothis_cpu_offset, dst
mrs \dst,tpidr_el2
.endm
#else
.macrothis_cpu_offset, dstalternative_if_notARM64_HAS_VIRT_HOST_EXTN
mrs \dst,tpidr_el1alternative_else
mrs \dst,tpidr_el2alternative_endif
.endm
#endif
Sure.
Quoted 2 lines
(and should we have a shorthand
__HYPERVISORdefine to avoid the NVHE || VHE logic?)
Happy to add this but let's agree on the details.
* name: just __HYPERVISOR or __KVM_HYPERVISOR__?
* defined where? I'm wary of defining it in a header file because then sombody will forget to add it and 'ifdef HYP' will be skipped. So I'd put this as another '-D__HYPERVISOR' in the build rules. Do you agree?
David
Will Deacon
maintainerRe: [PATCH v3 05/11] kvm: arm64: Remove hyp_adr/ldr_this_cpu
Quoted 17 lines · David Brazdil
Quoted 16 lines
Cosmetic, but I think it would be cleaner just to define two variants of the macro here:
#if
defined(__KVM_NVHE_HYPERVISOR__)||defined(__KVM_VHE_HYPERVISOR__)
.macrothis_cpu_offset, dst
mrs \dst,tpidr_el2
.endm
#else
.macrothis_cpu_offset, dstalternative_if_notARM64_HAS_VIRT_HOST_EXTN
mrs \dst,tpidr_el1alternative_else
mrs \dst,tpidr_el2alternative_endif
.endm
#endif
Quoted 1 line
Sure.
Ta.
Quoted 5 lines
Quoted 2 lines
(and should we have a shorthand
__HYPERVISORdefine to avoid the NVHE || VHE logic?)Happy to add this but let's agree on the details.
* name: just__HYPERVISORor__KVM_HYPERVISOR__?
Hey, I leave the hard decisions to others!
Quoted 3 lines
* defined where? I'm wary of defining it in a header file because then sombody will forget to add it and 'ifdef HYP' will be skipped. So I'd put this as another '-D__HYPERVISOR' in the build rules. Do you agree?
Hmm, that's a good point. Leave that part as-is for now then and we can tackle it separately if it comes up again.
Will
Re: [PATCH v3 01/11] kvm: arm64: Partially link nVHE hyp code, simplify HYPCOPY
In reply to Will DeaconQuoted 10 lines
arch/arm64/include/asm/hyp_image.h line 20
Quoted 5 lines
arch/arm64/include/asm/hyp_image.h new file +5
-0+#define HYP_SECTION_NAME (NAME ) .hyp##NAME + +/* Defines an ELF hyp section from input section @NAME and its subsections. */ +#define HYP_SECTION (NAME ) \ + HYP_SECTION_NAME(NAME) : { *(NAME NAME##.[0-9a-zA-Z_]*) }I still don't get why we can't just use NAME ## .* for the regex here. That matches what we do elsewhere for linker script wildcarding, e.g. .rodata.*, .init.text.* ... in
asm-generic/vmlinux.lds.h. Why is it different for these sections?
Shouldn't matter. I copied it from a script in the Clang LTO branch. Will align it with what's in the main arm64 script.
Quoted 7 lines
arch/arm64/kvm/hyp/nvhe/Makefile obj-y := timer-sr.o sysreg-sr.o debug-sr.o switch.o tlb.o hyp-init.o · line 32
Quoted 3 lines
arch/arm64/kvm/hyp/nvhe/Makefile +3
-0+# 2) Compile linker script. +$(obj)/hyp.lds: $(src)/hyp.lds.S FORCE + $(call if_changed_dep,cpp_lds_S)You need a .gitignore file listing
hyp.lds, otherwise some idiot will end up committing it. I definitely didn't do that when playing around with this series. Nope. Not at all.
Thanks for noticing, I never compile in-tree. Added
Re: [PATCH v3 07/11] kvm: arm64: Duplicate arm64_ssbd_callback_required for nVHE hyp
In reply to Will DeaconQuoted 14 lines
arch/arm64/include/asm/kvm_mmu.h static inline int kvm_map_vectors(void) · line 547
Quoted 10 lines
arch/arm64/include/asm/kvm_mmu.h +4
-1u64 *ptr; - ptr =per_cpu_ptr (&arm64_ssbd_callback_required, cpu); + ptr =per_cpu_ptr_nvhe (arm64_ssbd_callback_required, cpu); err =create_hyp_mappings (ptr, ptr +1 ,PAGE_HYP );if (err)return err; + +/* Copy value from kernel to hyp. */ + *ptr =per_cpu (arm64_ssbd_callback_required, cpu);Hmm. Is this correct for late arriving CPUs, where we don't know whether a callback is required at the point we do the copy?
Quoted 4 lines
That sounds fiddly to resolve, but this _might_ all be moot because I'm about to post a series that allows us to remove the hyp mapping of this variable entirely. So leave this for now, but maybe stick a comment in that it doesn't work for late cpus.
Ah, good point. I'll move the value copy at the end of cpu_init_hyp_mode(). It must be known at that point. And if your series gets rid of this completely, even better.
Re: [PATCH v3 09/11] kvm: arm64: Mark hyp stack pages reserved
In reply to Will DeaconQuoted 8 lines · Will Deacon
Quoted 2 lines · David Brazdil
In preparation for unmapping hyp pages from host stage-2, allocate/free hyp stack using new helpers which automatically mark the pages reserved.
Given that this series doesn't get us that the point of having a stage-2 for the host, I don't see why we need this part yet.
Will
Dropped
Re: [PATCH v3 10/11] kvm: arm64: Set up hyp percpu data for nVHE
In reply to Will DeaconQuoted 6 lines
Quoted 6 lines
arch/arm64/include/asm/kvm_asm.h +4
-2-#define this_cpu_ptr_nvhe (sym) this_cpu_ptr(&kvm_nvhe_sym(sym)) -#define per_cpu_ptr_nvhe (sym, cpu) per_cpu_ptr(&kvm_nvhe_sym(sym), cpu) +/* Array of percpu base addresses. Length of the array is nr_cpu_ids. */ +extern unsigned long *kvm_arm_hyp_percpu_base; + +/*
Quoted 8 lines
Quoted 4 lines
+ * Compute pointer to a symbol defined in nVHE percpu region.
+ * Returns NULL if percpu memory has not been allocated yet.
+ */
+#definethis_cpu_ptr_nvhe(sym)per_cpu_ptr_nvhe(sym, smp_processor_id())Don't you run into similar problems here with the pcpu accessors when
CONFIG_DEBUG_PREEMPT=y? I'm worried we can end up with an unresolved reference todebug_smp_processor_id().
Fortunately not. This now doesn't use the generic macros at all.
Quoted 11 lines
arch/arm64/kvm/arm.c long kvm_arch_vm_ioctl(struct file *filp, · line 1262
Quoted 7 lines
/* The VMID used in the VTTBR */
staticatomic64_tkvm_vmid_gen=ATOMIC64_INIT(1);arch/arm64/kvm/arm.c +1
-0@@ -1258,6 +1259,15 @@ long kvm_arch_vm_ioctl(struct file *filp, } } +#define kvm_hyp_percpu_base (cpu) ((unsigned long)per_cpu_ptr_nvhe(__per_cpu_start, cpu))Having both
kvm_arm_hyp_percpu_baseandkvm_hyp_percpu_basebe so completely different is a recipe for disaster! Please can you rename one/both of them to make it clearer what they represent?
I am heavily simplifying this code in v4. Got rid of this macro completely, so hopefully there will be no confusion.
Quoted 16 lines
arch/arm64/kvm/pmu.c void kvm_clr_pmu_events(u32 clr) · line 53
Quoted 11 lines
arch/arm64/kvm/pmu.c +3
-1-if (!kvm_pmu_switch_needed (attr)) +if (!ctx || !kvm_pmu_switch_needed (attr))return ;if (!attr->exclude_host)@@ -49,6 +49,9 @@ void kvm_clr_pmu_events(u32 clr) {struct kvm_host_data *ctx =this_cpu_ptr_hyp (kvm_host_data); +if (!ctx) +return ;I guess this should only happen if KVM failed to initialise or something, right? (e.g. if we were booted at EL1). If so, I'm wondering whether it would be better to do something like:
Quoted 7 lines
if (!is_hyp_mode_available()) return; WARN_ON_ONCE(!ctx);so that any unexpected NULL pointer there screams loudly, rather than causes the register switch to be silently ignored. What do you think?
Unfortunately, this happens on every boot. I don't fully understand how the boot order is determined, so please correct me if this makes no sense, but kvm_clr_pmu_events is called as part of CPUHP_AP_PERF_ARM_STARTING. The first time that happens is before KVM initialized (tested from inserting BUG_ON(!ctx)). That's not a problem, the per-CPU memory is there and it's all zeroes. It becomes a problem with this patch because the per-CPU memory is not there *yet*.