x86/fpu/kvm: Sanitize the FPU guest/user handling
cover x86/fpu/kvm: Sanitize the FPU guest/user handling
Currently KVM allocates two FPU structs which are used for saving the user state of the vCPU thread and restoring the guest state when entering vcpu_run() and doing the reverse operation before leaving vcpu_run().
With the new fpstate mechanism this can be reduced to one extra buffer by swapping the fpstate pointer in current::thread::fpu. This makes also the upcoming support for AMX and XFD simpler because then fpstate information (features, sizes, xfd) are always consistent and it does not require any nasty workarounds.
The following series cleans that up and replaces the current scheme with a single guest state which is switched in when entering vcpu_run() and switched out before leaving it.
The rework is valuable even without AMX/XFD because it consumes less memory and when swapping the fpstates there is no memory copy required when TIF_NEED_LOAD_FPU is set on the going out fpstate.
The series is based on:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git x86/fpu-3
and is now part of the full AMX series:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git x86/fpu
On top of that I've integrated the KVM reallocation mechanism into:
git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git x86/fpu-kvm
The latter builds, boots and runs KVM guests, but that reallocation functionality is obviously completely untested. I want to share this with KVM folks so they can start to look how to integrate their XFD/XCR0 and reallocation scheme as discussed and outlined here:
https://lore.kernel.org/r/87mtn93u58.ffs@tglx
and the related thread. It's a tiny incremental update on top of x86/fpu (6 files changed, 183 insertions(+), 36 deletions(-)) which reuses the host side mechanisms.
Thanks,
tglx
--- include/asm/fpu/api.h | 19 ++++++-- include/asm/fpu/types.h | 44 ++++++++++++++++++- include/asm/kvm_host.h | 7 --- kernel/fpu/core.c | 110 +++++++++++++++++++++++++++++++++--------------- kvm/svm/svm.c | 7 +-- kvm/x86.c | 88 ++++++++++---------------------------- 6 files changed, 165 insertions(+), 110 deletions(-)
PATCH 1/4 x86/fpu: Prepare for sanitizing KVM FPU code
For the upcoming AMX support it's necessary to do a proper integration with KVM. To avoid more nasty hackery in KVM which violate encapsulation extend struct fpu and fpstate so the fpstate switching can be consolidated and simplified.
Currently KVM allocates two FPU structs which are used for saving the user state of the vCPU thread and restoring the guest state when entering vcpu_run() and doing the reverse operation before leaving vcpu_run().
With the new fpstate mechanism this can be reduced to one extra buffer by swapping the fpstate pointer in current::thread::fpu. This makes the upcoming support for AMX and XFD simpler because then fpstate information (features, sizes, xfd) are always consistent and it does not require any nasty workarounds.
Add fpu::__task_fpstate to save the regular fpstate pointer while the task is inside vcpu_run(). Add some state fields to fpstate to indicate the nature of the state.
--- arch/x86/include/asm/fpu/types.h | 44 ++++++++++++++++++++++++++++++++++++++++- 1 file changed, 43 insertions(+), 1 deletion(-)
---
arch/x86/include/asm/fpu/types.h +43 -1
@@ -322,8 +322,32 @@ struct fpstate {/* @user_xfeatures: xfeatures valid in UABI buffers */ u64 user_xfeatures; +/* @is_valloc: Indicator for dynamically allocated state */ + unsigned int is_valloc : 1; + +/* @is_guest: Indicator for guest state (KVM) */ + unsigned int is_guest : 1; + +/* +* @is_confidential: Indicator for KVM confidential mode. +* The FPU registers are restored by the +* vmentry firmware from encrypted guest +* memory. On vmexit the FPU registers are +* saved by firmware to encrypted guest memory +* and the registers are scrubbed before +* returning to the host. So there is no +* content which is worth saving and restoring +* The fpstate has to be there so that +* preemption and softirq FPU usage works. +* without special casing. +*/ + unsigned int is_confidential : 1; + +/* @in_use: State is in use */ + unsigned int in_use : 1; +/* @regs: The register state union for all supported formats */ -union fpregs_state regs; +union fpregs_state regs;/* @regs is dynamically sized! Don't add anything after @regs! */ } __attribute__ ((aligned (64)));
@@ -364,6 +388,14 @@ struct fpu {struct fpstate *fpstate;/* +* @__task_fpstate: +* +* Pointer to an inactive struct fpstate. Initialized to NULL. Is +* used only for KVM support to swap out the regular task fpstate. +*/ +struct fpstate *__task_fpstate; + +/* * @__fpstate: * * Initial in-memory storage for FPU registers which are saved in
@@ -379,6 +411,16 @@ struct fpu { };/* +* Guest pseudo FPU container +*/ +struct fpu_guest { +/* +* @fpstate: Pointer to the allocated guest fpstate +*/ +struct fpstate *fpstate; +}; + +/* * FPU state configuration data. Initialized at boot time. Read only after init. */ struct fpu_state_config {
PATCH 2/4 x86/fpu: Provide infrastructure for KVM FPU cleanup
In reply to Thomas GleixnerFor the upcoming AMX support it's necessary to do a proper integration with KVM. Currently KVM allocates two FPU structs which are used for saving the user state of the vCPU thread and restoring the guest state when entering vcpu_run() and doing the reverse operation before leaving vcpu_run().
With the new fpstate mechanism this can be reduced to one extra buffer by swapping the fpstate pointer in current::thread::fpu. This makes the upcoming support for AMX and XFD simpler because then fpstate information (features, sizes, xfd) are always consistent and it does not require any nasty workarounds.
Provide:
- An allocator which initializes the state properly
- A replacement for the existing FPU swap mechanim
Aside of the reduced memory foot print, this also makes state switching more efficient when TIF_FPU_NEED_LOAD is set. It does not require a memcpy as the state is already correct in the to be swapped out fpstate.
The existing interfaces will be removed once KVM is converted over.
--- arch/x86/include/asm/fpu/api.h | 13 ++++++- arch/x86/kernel/fpu/core.c | 88 ++++++++++++++++++++++++++++++++++++++++--- 2 files changed, 95 insertions(+), 6 deletions(-)
---
arch/x86/include/asm/fpu/api.h +13 -0
@@ -135,9 +135,22 @@ extern void fpu_init_fpstate_user(struct fpu *fpu);extern void fpstate_clear_xstate_component (struct fpstate *fps,unsigned int xfeature);/* KVM specific functions */ +extern bool fpu_alloc_guest_fpstate (struct fpu_guest *gfpu); +extern void fpu_free_guest_fpstate (struct fpu_guest *gfpu); +extern int fpu_swap_kvm_fpstate (struct fpu_guest *gfpu,bool enter_guest,u64 restore_mask);extern void fpu_swap_kvm_fpu (struct fpu *save,struct fpu *rstor,u64 restore_mask);extern int fpu_copy_kvm_uabi_to_fpstate (struct fpu *fpu,const void *buf,u64 xcr0,u32 *pkru);extern void fpu_copy_fpstate_to_kvm_uabi (struct fpu *fpu,void *buf,unsigned int size,u32 pkru); +static inline void fpstate_set_confidential (struct fpu_guest *gfpu) +{ + gfpu->fpstate->is_confidential = true; +} + +static inline bool fpstate_is_confidential (struct fpu_guest *gfpu) +{ +return gfpu->fpstate->is_confidential; +} +#endif /* _ASM_X86_FPU_API_H */
arch/x86/kernel/fpu/core.c +82 -6
104 lines not on this page · raw message
PATCH 3/4 x86/kvm: Convert FPU handling to a single swap buffer
In reply to Thomas GleixnerFor the upcoming AMX support it's necessary to do a proper integration with KVM. Currently KVM allocates two FPU structs which are used for saving the user state of the vCPU thread and restoring the guest state when entering vcpu_run() and doing the reverse operation before leaving vcpu_run().
With the new fpstate mechanism this can be reduced to one extra buffer by swapping the fpstate pointer in current::thread::fpu. This makes the upcoming support for AMX and XFD simpler because then fpstate information (features, sizes, xfd) are always consistent and it does not require any nasty workarounds.
Convert the KVM FPU code over to this new scheme.
--- arch/x86/include/asm/fpu/api.h | 4 +- arch/x86/include/asm/kvm_host.h | 7 +--- arch/x86/kernel/fpu/core.c | 16 ++++---- arch/x86/kvm/svm/svm.c | 7 +-- arch/x86/kvm/x86.c | 88 +++++++++++------------------------------- 5 files changed, 40 insertions(+), 82 deletions(-)
---
arch/x86/include/asm/fpu/api.h +2 -2
@@ -140,8 +140,8 @@ extern void fpu_free_guest_fpstate(struct fpu_guest *gfpu);extern int fpu_swap_kvm_fpstate (struct fpu_guest *gfpu,bool enter_guest,u64 restore_mask);extern void fpu_swap_kvm_fpu (struct fpu *save,struct fpu *rstor,u64 restore_mask); -extern int fpu_copy_kvm_uabi_to_fpstate (struct fpu *fpu,const void *buf,u64 xcr0,u32 *pkru); -extern void fpu_copy_fpstate_to_kvm_uabi (struct fpu *fpu,void *buf,unsigned int size,u32 pkru); +extern void fpu_copy_guest_fpstate_to_uabi (struct fpu_guest *gfpu,void *buf,unsigned int size,u32 pkru); +extern int fpu_copy_uabi_to_guest_fpstate (struct fpu_guest *gfpu,const void *buf,u64 xcr0,u32 *vpkru);static inline void fpstate_set_confidential (struct fpu_guest *gfpu) {
arch/x86/include/asm/kvm_host.h +2 -5
@@ -691,11 +691,10 @@ struct kvm_vcpu_arch {* * Note that while the PKRU state lives inside the fpu registers, * it is switched out separately at VMENTER and VMEXIT time. The -* "guest_fpu" state here contains the guest FPU context, with the +* "guest_fpstate" state here contains the guest FPU context, with the * host PRKU bits. */ -struct fpu *user_fpu; -struct fpu *guest_fpu; +struct fpu_guest guest_fpu;u64 xcr0;u64 guest_supported_xcr0;
@@ -1685,8 +1684,6 @@ void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);int kvm_task_switch (struct kvm_vcpu *vcpu,u16 tss_selector,int idt_index,int reason,bool has_error_code,u32 error_code); -void kvm_free_guest_fpu (struct kvm_vcpu *vcpu); -void kvm_post_set_cr0 (struct kvm_vcpu *vcpu,unsigned long old_cr0,unsigned long cr0);void kvm_post_set_cr4 (struct kvm_vcpu *vcpu,unsigned long old_cr4,unsigned long cr4);int kvm_set_cr0 (struct kvm_vcpu *vcpu,unsigned long cr0);
arch/x86/kernel/fpu/core.c +8 -8
@@ -271,10 +271,10 @@ void fpu_swap_kvm_fpu(struct fpu *save, struct fpu *rstor, u64 restore_mask) }EXPORT_SYMBOL_GPL (fpu_swap_kvm_fpu); -void fpu_copy_fpstate_to_kvm_uabi (struct fpu *fpu,void *buf, -unsigned int size,u32 pkru) +void fpu_copy_guest_fpstate_to_uabi (struct fpu_guest *gfpu,void *buf, +unsigned int size,u32 pkru) { -struct fpstate *kstate = fpu->fpstate; +struct fpstate *kstate = gfpu->fpstate;union fpregs_state *ustate = buf;struct membuf mb = { .p = buf, .left = size };
@@ -287,12 +287,12 @@ void fpu_copy_fpstate_to_kvm_uabi(struct fpu *fpu, void *buf, ustate->xsave.header.xfeatures =XFEATURE_MASK_FPSSE ; } } -EXPORT_SYMBOL_GPL (fpu_copy_fpstate_to_kvm_uabi); +EXPORT_SYMBOL_GPL (fpu_copy_guest_fpstate_to_uabi); -int fpu_copy_kvm_uabi_to_fpstate (struct fpu *fpu,const void *buf,u64 xcr0, -u32 *vpkru) +int fpu_copy_uabi_to_guest_fpstate (struct fpu_guest *gfpu,const void *buf, +u64 xcr0,u32 *vpkru) { -struct fpstate *kstate = fpu->fpstate; +struct fpstate *kstate = gfpu->fpstate;const union fpregs_state *ustate = buf;struct pkru_state *xpkru;int ret;
@@ -323,7 +323,7 @@ int fpu_copy_kvm_uabi_to_fpstate(struct fpu *fpu, const void *buf, u64 xcr0,xstate_init_xcomp_bv (&kstate->regs.xsave, fpu_user_cfg.max_features); return 0; } -EXPORT_SYMBOL_GPL (fpu_copy_kvm_uabi_to_fpstate); +EXPORT_SYMBOL_GPL (fpu_copy_uabi_to_guest_fpstate);#endif /* CONFIG_KVM */ void kernel_fpu_begin_mask (unsigned int kfpu_mask)
arch/x86/kvm/svm/svm.c +4 -3
@@ -36,6 +36,7 @@#include <asm/spec-ctrl.h> #include <asm/cpu_device_id.h> #include <asm/traps.h> +#include <asm/fpu/api.h> #include <asm/virtext.h> #include "trace.h"
@@ -1346,10 +1347,10 @@ static int svm_create_vcpu(struct kvm_vcpu *vcpu)/* * SEV-ES guests maintain an encrypted version of their FPU * state which is restored and saved on VMRUN and VMEXIT. -* Free the fpu structure to prevent KVM from attempting to -* access the FPU state. +* Mark vcpu->arch.guest_fpu->fpstate as scratch so it won't +* do xsave/xrstor on it. */ -kvm_free_guest_fpu (vcpu); +fpstate_set_confidential (&vcpu->arch.guest_fpu); } err =avic_init_vcpu (svm);
arch/x86/kvm/x86.c +24 -64
197 lines not on this page · raw message
Paolo Bonzini
maintainerRe: [patch 3/4] x86/kvm: Convert FPU handling to a single swap buffer
Quoted 6 lines · Thomas Gleixner
*/
arch/x86/kvm/x86.c +1
-2-fpu_swap_kvm_fpu (vcpu->arch.user_fpu, vcpu->arch.guest_fpu, - ~XFEATURE_MASK_PKRU ); +fpu_swap_kvm_fpstate (&vcpu->arch.guest_fpu, true, ~XFEATURE_MASK_PKRU );trace_kvm_fpu (1 ); }
arch/x86/kvm/x86.c static int complete_emulated_mmio(struct kvm_vcpu *vcpu) · line 9789
Quoted 10 lines
arch/x86/kvm/x86.c +1
-5/* When vcpu_run ends, restore user space FPU context. */ static void kvm_put_guest_fpu (struct kvm_vcpu *vcpu) { -/* -* Guests with protected state have guest_fpu == NULL which makes -* swap only restore the host state. -*/ -fpu_swap_kvm_fpu (vcpu->arch.guest_fpu, vcpu->arch.user_fpu, ~0ULL ); +fpu_swap_kvm_fpstate (&vcpu->arch.guest_fpu, false, ~0ULL );
The restore mask can be ~XFEATURE_MASK_PKRU in this case tool this way it's constant and you can drop the third argument to the function.
Also perhaps it could be useful to add an
if (WARN_ON_ONCE(cur_fps->is_guest == enter_guest))
return;
at the top of fpu_swap_kvm_fpstate, since the is_guest member (at least for now?) is only used for such kind of assertion.
Paolo
PATCH 4/4 x86/fpu: Remove old KVM FPU interface
In reply to Thomas GleixnerNo more users.
--- arch/x86/include/asm/fpu/api.h | 2 -- arch/x86/kernel/fpu/core.c | 32 -------------------------------- 2 files changed, 34 deletions(-)
---
arch/x86/include/asm/fpu/api.h +0 -2
@@ -131,14 +131,12 @@ static inline void fpstate_init_soft(struct swregs_state *soft) {} DECLARE_PER_CPU(struct fpu *, fpu_fpregs_owner_ctx);/* fpstate-related functions which are exported to KVM */ -extern void fpu_init_fpstate_user (struct fpu *fpu);extern void fpstate_clear_xstate_component (struct fpstate *fps,unsigned int xfeature);/* KVM specific functions */ extern bool fpu_alloc_guest_fpstate (struct fpu_guest *gfpu);extern void fpu_free_guest_fpstate (struct fpu_guest *gfpu);extern int fpu_swap_kvm_fpstate (struct fpu_guest *gfpu,bool enter_guest,u64 restore_mask); -extern void fpu_swap_kvm_fpu (struct fpu *save,struct fpu *rstor,u64 restore_mask);extern void fpu_copy_guest_fpstate_to_uabi (struct fpu_guest *gfpu,void *buf,unsigned int size,u32 pkru);extern int fpu_copy_uabi_to_guest_fpstate (struct fpu_guest *gfpu,const void *buf,u64 xcr0,u32 *vpkru);
arch/x86/kernel/fpu/core.c +0 -32
@@ -248,29 +248,6 @@ int fpu_swap_kvm_fpstate(struct fpu_guest *guest_fpu, bool enter_guest, }EXPORT_SYMBOL_GPL (fpu_swap_kvm_fpstate); -void fpu_swap_kvm_fpu (struct fpu *save,struct fpu *rstor,u64 restore_mask) -{ -fpregs_lock (); - -if (save) { -struct fpstate *fpcur = current->thread.fpu.fpstate; - -if (test_thread_flag (TIF_NEED_FPU_LOAD )) -memcpy (&save->fpstate->regs, &fpcur->regs, fpcur->size); -else -save_fpregs_to_fpstate (save); - } - -if (rstor) { - restore_mask &=XFEATURE_MASK_FPSTATE ; -restore_fpregs_from_fpstate (rstor->fpstate, restore_mask); - } - -fpregs_mark_activate (); -fpregs_unlock (); -} -EXPORT_SYMBOL_GPL (fpu_swap_kvm_fpu); -void fpu_copy_guest_fpstate_to_uabi (struct fpu_guest *gfpu,void *buf,unsigned int size,u32 pkru) {
@@ -440,15 +417,6 @@ void fpstate_reset(struct fpu *fpu)__fpstate_reset (fpu->fpstate); } -#if IS_ENABLED (CONFIG_KVM ) -void fpu_init_fpstate_user (struct fpu *fpu) -{ -fpstate_reset (fpu); -fpstate_init_user (fpu->fpstate); -} -EXPORT_SYMBOL_GPL (fpu_init_fpstate_user); -#endif -/* Clone current's FPU state on fork */ int fpu_clone (struct task_struct *dst) {
Quoted 3 lines · Thomas Gleixner
The latter builds, boots and runs KVM guests, but that reallocation functionality is obviously completely untested.
Compiled and booted on bare-metal and KVM (guest with the same kernel). No dmesg regression. No selftest regression.
Tested-by Chang S. Bae <[email protected]>
Thanks,
Chang
Paolo Bonzini
maintainerQuoted 9 lines · Chang Seok
Quoted 3 lines · Thomas Gleixner
The latter builds, boots and runs KVM guests, but that reallocation functionality is obviously completely untested.
Compiled and booted on bare-metal and KVM (guest with the same kernel). No dmesg regression. No selftest regression.
Tested-by Chang S. Bae <[email protected]>
Same here. Thanks, Chang Seok!
Paolo