Simplify the binding of the guest state to the CPU when returning to the
host by storing a copy of the structure used to pass the state in the
vCPU data and initialising it during guest setup rather than creating a
new copy each time we exit the guest.
Signed-off-by: Mark Brown <[email protected]>
---
Changes in v2:
- Remove stale fp_state local.
- Changelog updates.
- Link to v1: https://lore.kernel.org/r/[email protected]
---
Mark Brown (2):
KVM: arm64: Rename variable for tracking ownership of FP state
KVM: arm64: Reuse struct cpu_fp_state to track the guest FP state
arch/arm64/include/asm/kvm_emulate.h | 4 ++--
arch/arm64/include/asm/kvm_host.h | 25 ++++++++++++-------------
arch/arm64/kvm/arm.c | 14 +++++++++++++-
arch/arm64/kvm/fpsimd.c | 31 ++++++-------------------------
arch/arm64/kvm/guest.c | 21 ++++++++++++++-------
arch/arm64/kvm/hyp/include/hyp/switch.h | 6 +++---
arch/arm64/kvm/hyp/nvhe/hyp-main.c | 9 +++++----
arch/arm64/kvm/hyp/nvhe/switch.c | 2 +-
arch/arm64/kvm/hyp/vhe/switch.c | 2 +-
arch/arm64/kvm/reset.c | 14 ++++++++------
10 files changed, 65 insertions(+), 63 deletions(-)
---
base-commit: 54be6c6c5ae8e0d93a6c4641cb7528eb0b6ba478
change-id: 20240226-kvm-arm64-group-fp-data-0ae363ce24fe
Best regards,
--
Mark Brown <[email protected]>
In preparation for refactoring how we store the actual FP state into a
single struct let's free up the name 'fp_state' which we currently use for
the variable where we track the ownership of the FP registers to something
a bit more specific to it's usage. While we're at it also move the enum
definition next to the rest of the FP state.
No functional changes.
Signed-off-by: Mark Brown <[email protected]>
---
arch/arm64/include/asm/kvm_emulate.h | 4 ++--
arch/arm64/include/asm/kvm_host.h | 14 +++++++-------
arch/arm64/kvm/arm.c | 2 +-
arch/arm64/kvm/fpsimd.c | 10 +++++-----
arch/arm64/kvm/hyp/include/hyp/switch.h | 6 +++---
arch/arm64/kvm/hyp/nvhe/hyp-main.c | 4 ++--
arch/arm64/kvm/hyp/nvhe/switch.c | 2 +-
arch/arm64/kvm/hyp/vhe/switch.c | 2 +-
8 files changed, 22 insertions(+), 22 deletions(-)
diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
index b804fe832184..1211d93aa712 100644
--- a/arch/arm64/include/asm/kvm_emulate.h
+++ b/arch/arm64/include/asm/kvm_emulate.h
@@ -593,7 +593,7 @@ static __always_inline u64 kvm_get_reset_cptr_el2(struct kvm_vcpu *vcpu)
val = (CPACR_EL1_FPEN_EL0EN | CPACR_EL1_FPEN_EL1EN);
if (!vcpu_has_sve(vcpu) ||
- (vcpu->arch.fp_state != FP_STATE_GUEST_OWNED))
+ (vcpu->arch.fp_owner != FP_STATE_GUEST_OWNED))
val |= CPACR_EL1_ZEN_EL1EN | CPACR_EL1_ZEN_EL0EN;
if (cpus_have_final_cap(ARM64_SME))
val |= CPACR_EL1_SMEN_EL1EN | CPACR_EL1_SMEN_EL0EN;
@@ -601,7 +601,7 @@ static __always_inline u64 kvm_get_reset_cptr_el2(struct kvm_vcpu *vcpu)
val = CPTR_NVHE_EL2_RES1;
if (vcpu_has_sve(vcpu) &&
- (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED))
+ (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED))
val |= CPTR_EL2_TZ;
if (cpus_have_final_cap(ARM64_SME))
val &= ~CPTR_EL2_TSM;
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 21c57b812569..e0fbba52f1d3 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -544,6 +544,13 @@ struct kvm_vcpu_arch {
unsigned int sve_max_vl;
u64 svcr;
+ /* Ownership of the FP regs */
+ enum {
+ FP_STATE_FREE,
+ FP_STATE_HOST_OWNED,
+ FP_STATE_GUEST_OWNED,
+ } fp_owner;
+
/* Stage 2 paging state used by the hardware on next switch */
struct kvm_s2_mmu *hw_mmu;
@@ -558,13 +565,6 @@ struct kvm_vcpu_arch {
/* Exception Information */
struct kvm_vcpu_fault_info fault;
- /* Ownership of the FP regs */
- enum {
- FP_STATE_FREE,
- FP_STATE_HOST_OWNED,
- FP_STATE_GUEST_OWNED,
- } fp_state;
-
/* Configuration flags, set once and for all before the vcpu can run */
u8 cflags;
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index a25265aca432..a2cba18effb2 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -377,7 +377,7 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
* Default value for the FP state, will be overloaded at load
* time if we support FP (pretty likely)
*/
- vcpu->arch.fp_state = FP_STATE_FREE;
+ vcpu->arch.fp_owner = FP_STATE_FREE;
/* Set up the timer */
kvm_timer_vcpu_init(vcpu);
diff --git a/arch/arm64/kvm/fpsimd.c b/arch/arm64/kvm/fpsimd.c
index 8c1d0d4853df..8dbd62d1e677 100644
--- a/arch/arm64/kvm/fpsimd.c
+++ b/arch/arm64/kvm/fpsimd.c
@@ -86,7 +86,7 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
* guest in kvm_arch_vcpu_ctxflush_fp() and override this to
* FP_STATE_FREE if the flag set.
*/
- vcpu->arch.fp_state = FP_STATE_HOST_OWNED;
+ vcpu->arch.fp_owner = FP_STATE_HOST_OWNED;
vcpu_clear_flag(vcpu, HOST_SVE_ENABLED);
if (read_sysreg(cpacr_el1) & CPACR_EL1_ZEN_EL0EN)
@@ -110,7 +110,7 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
* been saved, this is very unlikely to happen.
*/
if (read_sysreg_s(SYS_SVCR) & (SVCR_SM_MASK | SVCR_ZA_MASK)) {
- vcpu->arch.fp_state = FP_STATE_FREE;
+ vcpu->arch.fp_owner = FP_STATE_FREE;
fpsimd_save_and_flush_cpu_state();
}
}
@@ -126,7 +126,7 @@ void kvm_arch_vcpu_load_fp(struct kvm_vcpu *vcpu)
void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu)
{
if (test_thread_flag(TIF_FOREIGN_FPSTATE))
- vcpu->arch.fp_state = FP_STATE_FREE;
+ vcpu->arch.fp_owner = FP_STATE_FREE;
}
/*
@@ -142,7 +142,7 @@ void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu)
WARN_ON_ONCE(!irqs_disabled());
- if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED) {
+ if (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED) {
/*
* Currently we do not support SME guests so SVCR is
@@ -195,7 +195,7 @@ void kvm_arch_vcpu_put_fp(struct kvm_vcpu *vcpu)
isb();
}
- if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED) {
+ if (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED) {
if (vcpu_has_sve(vcpu)) {
__vcpu_sys_reg(vcpu, ZCR_EL1) = read_sysreg_el1(SYS_ZCR);
diff --git a/arch/arm64/kvm/hyp/include/hyp/switch.h b/arch/arm64/kvm/hyp/include/hyp/switch.h
index a038320cdb08..575c39847d40 100644
--- a/arch/arm64/kvm/hyp/include/hyp/switch.h
+++ b/arch/arm64/kvm/hyp/include/hyp/switch.h
@@ -42,7 +42,7 @@ extern struct kvm_exception_table_entry __stop___kvm_ex_table;
/* Check whether the FP regs are owned by the guest */
static inline bool guest_owns_fp_regs(struct kvm_vcpu *vcpu)
{
- return vcpu->arch.fp_state == FP_STATE_GUEST_OWNED;
+ return vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED;
}
/* Save the 32-bit only FPSIMD system register state */
@@ -370,7 +370,7 @@ static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
isb();
/* Write out the host state if it's in the registers */
- if (vcpu->arch.fp_state == FP_STATE_HOST_OWNED)
+ if (vcpu->arch.fp_owner == FP_STATE_HOST_OWNED)
__fpsimd_save_state(vcpu->arch.host_fpsimd_state);
/* Restore the guest state */
@@ -383,7 +383,7 @@ static bool kvm_hyp_handle_fpsimd(struct kvm_vcpu *vcpu, u64 *exit_code)
if (!(read_sysreg(hcr_el2) & HCR_RW))
write_sysreg(__vcpu_sys_reg(vcpu, FPEXC32_EL2), fpexc32_el2);
- vcpu->arch.fp_state = FP_STATE_GUEST_OWNED;
+ vcpu->arch.fp_owner = FP_STATE_GUEST_OWNED;
return true;
}
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 2385fd03ed87..85ea18227d33 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -39,7 +39,7 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
hyp_vcpu->vcpu.arch.cptr_el2 = host_vcpu->arch.cptr_el2;
hyp_vcpu->vcpu.arch.iflags = host_vcpu->arch.iflags;
- hyp_vcpu->vcpu.arch.fp_state = host_vcpu->arch.fp_state;
+ hyp_vcpu->vcpu.arch.fp_owner = host_vcpu->arch.fp_owner;
hyp_vcpu->vcpu.arch.debug_ptr = kern_hyp_va(host_vcpu->arch.debug_ptr);
hyp_vcpu->vcpu.arch.host_fpsimd_state = host_vcpu->arch.host_fpsimd_state;
@@ -64,7 +64,7 @@ static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
host_vcpu->arch.fault = hyp_vcpu->vcpu.arch.fault;
host_vcpu->arch.iflags = hyp_vcpu->vcpu.arch.iflags;
- host_vcpu->arch.fp_state = hyp_vcpu->vcpu.arch.fp_state;
+ host_vcpu->arch.fp_owner = hyp_vcpu->vcpu.arch.fp_owner;
host_cpu_if->vgic_hcr = hyp_cpu_if->vgic_hcr;
for (i = 0; i < hyp_cpu_if->used_lrs; ++i)
diff --git a/arch/arm64/kvm/hyp/nvhe/switch.c b/arch/arm64/kvm/hyp/nvhe/switch.c
index c50f8459e4fc..9f9404c9bbae 100644
--- a/arch/arm64/kvm/hyp/nvhe/switch.c
+++ b/arch/arm64/kvm/hyp/nvhe/switch.c
@@ -337,7 +337,7 @@ int __kvm_vcpu_run(struct kvm_vcpu *vcpu)
__sysreg_restore_state_nvhe(host_ctxt);
- if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED)
+ if (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED)
__fpsimd_save_fpexc32(vcpu);
__debug_switch_to_host(vcpu);
diff --git a/arch/arm64/kvm/hyp/vhe/switch.c b/arch/arm64/kvm/hyp/vhe/switch.c
index 1581df6aec87..17596586806c 100644
--- a/arch/arm64/kvm/hyp/vhe/switch.c
+++ b/arch/arm64/kvm/hyp/vhe/switch.c
@@ -258,7 +258,7 @@ static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
sysreg_restore_host_state_vhe(host_ctxt);
- if (vcpu->arch.fp_state == FP_STATE_GUEST_OWNED)
+ if (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED)
__fpsimd_save_fpexc32(vcpu);
__debug_switch_to_host(vcpu);
--
2.30.2
At present we store the various bits of floating point state individually
in struct kvm_vcpu_arch and construct a struct cpu_fp_state to share with
the host each time we exit the guest. Let's simplify this a little by
having a struct cpu_fp_state in the struct kvm_vcpu_arch and initialising
this while initialising the guest.
As part of this remove the separate variables used for the SVE register
storage and vector length information, just using the variables in the
struct cpu_fp_state directly.
Since cpu_fp_state stores pointers to variables to be updated as part of
saving we do still need some variables stored directly in the struct for
the FPSIMD registers, SVCR and the type of FP state saved. Due to the
embedding of the FPSIMD registers into ucontext which is stored directly in
the host's data and the future need to support KVM's system register view
of SVCR and FPMR unpicking these indirections would be more involved.
We initialise the structure when the vCPU is created and then update it
if SVE is enabled, this split initialisation mirrors the existing code
and helps avoid future modifications creating a situation where
partially initialised floating point state is exposed to userspace. The
need to offer configurability of the SVE vector length means we have to
have some reinitialisation.
No functional changes.
Signed-off-by: Mark Brown <[email protected]>
---
arch/arm64/include/asm/kvm_host.h | 11 +++++------
arch/arm64/kvm/arm.c | 12 ++++++++++++
arch/arm64/kvm/fpsimd.c | 21 +--------------------
arch/arm64/kvm/guest.c | 21 ++++++++++++++-------
arch/arm64/kvm/hyp/nvhe/hyp-main.c | 5 +++--
arch/arm64/kvm/reset.c | 14 ++++++++------
6 files changed, 43 insertions(+), 41 deletions(-)
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index e0fbba52f1d3..47bd769a26ff 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -539,9 +539,8 @@ struct kvm_vcpu_arch {
* floating point code saves the register state of a task it
* records which view it saved in fp_type.
*/
- void *sve_state;
+ struct cpu_fp_state fp_state;
enum fp_type fp_type;
- unsigned int sve_max_vl;
u64 svcr;
/* Ownership of the FP regs */
@@ -799,16 +798,16 @@ struct kvm_vcpu_arch {
/* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
-#define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.sve_state) + \
- sve_ffr_offset((vcpu)->arch.sve_max_vl))
+#define vcpu_sve_pffr(vcpu) (kern_hyp_va((vcpu)->arch.fp_state.sve_state) + \
+ sve_ffr_offset((vcpu)->arch.fp_state.sve_vl))
-#define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.sve_max_vl)
+#define vcpu_sve_max_vq(vcpu) sve_vq_from_vl((vcpu)->arch.fp_state.sve_vl)
#define vcpu_sve_state_size(vcpu) ({ \
size_t __size_ret; \
unsigned int __vcpu_vq; \
\
- if (WARN_ON(!sve_vl_valid((vcpu)->arch.sve_max_vl))) { \
+ if (WARN_ON(!sve_vl_valid((vcpu)->arch.fp_state.sve_vl))) { \
__size_ret = 0; \
} else { \
__vcpu_vq = vcpu_sve_max_vq(vcpu); \
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index a2cba18effb2..84cc0dbd9b14 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -379,6 +379,18 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
*/
vcpu->arch.fp_owner = FP_STATE_FREE;
+ /*
+ * Initial setup for FP state for sharing with host, if SVE is
+ * enabled additional configuration will be done.
+ *
+ * Currently we do not support SME guests so SVCR is always 0
+ * and we just need a variable to point to.
+ */
+ vcpu->arch.fp_state.st = &vcpu->arch.ctxt.fp_regs;
+ vcpu->arch.fp_state.fp_type = &vcpu->arch.fp_type;
+ vcpu->arch.fp_state.svcr = &vcpu->arch.svcr;
+ vcpu->arch.fp_state.to_save = FP_STATE_FPSIMD;
+
/* Set up the timer */
kvm_timer_vcpu_init(vcpu);
diff --git a/arch/arm64/kvm/fpsimd.c b/arch/arm64/kvm/fpsimd.c
index 8dbd62d1e677..fc270a2257d5 100644
--- a/arch/arm64/kvm/fpsimd.c
+++ b/arch/arm64/kvm/fpsimd.c
@@ -138,29 +138,10 @@ void kvm_arch_vcpu_ctxflush_fp(struct kvm_vcpu *vcpu)
*/
void kvm_arch_vcpu_ctxsync_fp(struct kvm_vcpu *vcpu)
{
- struct cpu_fp_state fp_state;
-
WARN_ON_ONCE(!irqs_disabled());
if (vcpu->arch.fp_owner == FP_STATE_GUEST_OWNED) {
-
- /*
- * Currently we do not support SME guests so SVCR is
- * always 0 and we just need a variable to point to.
- */
- fp_state.st = &vcpu->arch.ctxt.fp_regs;
- fp_state.sve_state = vcpu->arch.sve_state;
- fp_state.sve_vl = vcpu->arch.sve_max_vl;
- fp_state.sme_state = NULL;
- fp_state.svcr = &vcpu->arch.svcr;
- fp_state.fp_type = &vcpu->arch.fp_type;
-
- if (vcpu_has_sve(vcpu))
- fp_state.to_save = FP_STATE_SVE;
- else
- fp_state.to_save = FP_STATE_FPSIMD;
-
- fpsimd_bind_state_to_cpu(&fp_state);
+ fpsimd_bind_state_to_cpu(&vcpu->arch.fp_state);
clear_thread_flag(TIF_FOREIGN_FPSTATE);
}
diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
index aaf1d4939739..54e9d3b648f0 100644
--- a/arch/arm64/kvm/guest.c
+++ b/arch/arm64/kvm/guest.c
@@ -317,7 +317,7 @@ static int get_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
if (!vcpu_has_sve(vcpu))
return -ENOENT;
- if (WARN_ON(!sve_vl_valid(vcpu->arch.sve_max_vl)))
+ if (WARN_ON(!sve_vl_valid(vcpu->arch.fp_state.sve_vl)))
return -EINVAL;
memset(vqs, 0, sizeof(vqs));
@@ -344,7 +344,7 @@ static int set_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
if (kvm_arm_vcpu_sve_finalized(vcpu))
return -EPERM; /* too late! */
- if (WARN_ON(vcpu->arch.sve_state))
+ if (WARN_ON(vcpu->arch.fp_state.sve_state))
return -EINVAL;
if (copy_from_user(vqs, (const void __user *)reg->addr, sizeof(vqs)))
@@ -373,8 +373,11 @@ static int set_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
if (max_vq < SVE_VQ_MIN)
return -EINVAL;
- /* vcpu->arch.sve_state will be alloc'd by kvm_vcpu_finalize_sve() */
- vcpu->arch.sve_max_vl = sve_vl_from_vq(max_vq);
+ /*
+ * vcpu->arch.fp_state.sve_state will be alloc'd by
+ * kvm_vcpu_finalize_sve().
+ */
+ vcpu->arch.fp_state.sve_vl = sve_vl_from_vq(max_vq);
return 0;
}
@@ -403,7 +406,10 @@ static int set_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
*/
#define vcpu_sve_slices(vcpu) 1
-/* Bounds of a single SVE register slice within vcpu->arch.sve_state */
+/*
+ * Bounds of a single SVE register slice within
+ * vcpu->arch.fp_state.sve_state
+ */
struct sve_state_reg_region {
unsigned int koffset; /* offset into sve_state in kernel memory */
unsigned int klen; /* length in kernel memory */
@@ -499,7 +505,7 @@ static int get_sve_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
if (!kvm_arm_vcpu_sve_finalized(vcpu))
return -EPERM;
- if (copy_to_user(uptr, vcpu->arch.sve_state + region.koffset,
+ if (copy_to_user(uptr, vcpu->arch.fp_state.sve_state + region.koffset,
region.klen) ||
clear_user(uptr + region.klen, region.upad))
return -EFAULT;
@@ -525,7 +531,8 @@ static int set_sve_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
if (!kvm_arm_vcpu_sve_finalized(vcpu))
return -EPERM;
- if (copy_from_user(vcpu->arch.sve_state + region.koffset, uptr,
+ if (copy_from_user(vcpu->arch.fp_state.sve_state + region.koffset,
+ uptr,
region.klen))
return -EFAULT;
diff --git a/arch/arm64/kvm/hyp/nvhe/hyp-main.c b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
index 85ea18227d33..63971b801cf3 100644
--- a/arch/arm64/kvm/hyp/nvhe/hyp-main.c
+++ b/arch/arm64/kvm/hyp/nvhe/hyp-main.c
@@ -29,8 +29,9 @@ static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
hyp_vcpu->vcpu.arch.ctxt = host_vcpu->arch.ctxt;
- hyp_vcpu->vcpu.arch.sve_state = kern_hyp_va(host_vcpu->arch.sve_state);
- hyp_vcpu->vcpu.arch.sve_max_vl = host_vcpu->arch.sve_max_vl;
+ hyp_vcpu->vcpu.arch.fp_state.sve_state
+ = kern_hyp_va(host_vcpu->arch.fp_state.sve_state);
+ hyp_vcpu->vcpu.arch.fp_state.sve_vl = host_vcpu->arch.fp_state.sve_vl;
hyp_vcpu->vcpu.arch.hw_mmu = host_vcpu->arch.hw_mmu;
diff --git a/arch/arm64/kvm/reset.c b/arch/arm64/kvm/reset.c
index 68d1d05672bd..675b8925242f 100644
--- a/arch/arm64/kvm/reset.c
+++ b/arch/arm64/kvm/reset.c
@@ -75,7 +75,7 @@ int __init kvm_arm_init_sve(void)
static void kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
{
- vcpu->arch.sve_max_vl = kvm_sve_max_vl;
+ vcpu->arch.fp_state.sve_vl = kvm_sve_max_vl;
/*
* Userspace can still customize the vector lengths by writing
@@ -87,7 +87,7 @@ static void kvm_vcpu_enable_sve(struct kvm_vcpu *vcpu)
/*
* Finalize vcpu's maximum SVE vector length, allocating
- * vcpu->arch.sve_state as necessary.
+ * vcpu->arch.fp_state.sve_state as necessary.
*/
static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
{
@@ -96,7 +96,7 @@ static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
size_t reg_sz;
int ret;
- vl = vcpu->arch.sve_max_vl;
+ vl = vcpu->arch.fp_state.sve_vl;
/*
* Responsibility for these properties is shared between
@@ -118,7 +118,8 @@ static int kvm_vcpu_finalize_sve(struct kvm_vcpu *vcpu)
return ret;
}
- vcpu->arch.sve_state = buf;
+ vcpu->arch.fp_state.sve_state = buf;
+ vcpu->arch.fp_state.to_save = FP_STATE_SVE;
vcpu_set_flag(vcpu, VCPU_SVE_FINALIZED);
return 0;
}
@@ -149,7 +150,7 @@ bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu)
void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
{
- void *sve_state = vcpu->arch.sve_state;
+ void *sve_state = vcpu->arch.fp_state.sve_state;
kvm_vcpu_unshare_task_fp(vcpu);
kvm_unshare_hyp(vcpu, vcpu + 1);
@@ -162,7 +163,8 @@ void kvm_arm_vcpu_destroy(struct kvm_vcpu *vcpu)
static void kvm_vcpu_reset_sve(struct kvm_vcpu *vcpu)
{
if (vcpu_has_sve(vcpu))
- memset(vcpu->arch.sve_state, 0, vcpu_sve_state_size(vcpu));
+ memset(vcpu->arch.fp_state.sve_state, 0,
+ vcpu_sve_state_size(vcpu));
}
static void kvm_vcpu_enable_ptrauth(struct kvm_vcpu *vcpu)
--
2.30.2
On Thu, 29 Feb 2024 21:47:35 +0000,
Mark Brown <[email protected]> wrote:
>
> At present we store the various bits of floating point state individually
> in struct kvm_vcpu_arch and construct a struct cpu_fp_state to share with
> the host each time we exit the guest. Let's simplify this a little by
> having a struct cpu_fp_state in the struct kvm_vcpu_arch and initialising
> this while initialising the guest.
This structure is only useful to the physical CPU we run on, and does
not capture anything that is related to the guest state. Why should it
live in the vcpu structure, duplicating things we already have?
This is just making things even more opaque.
If you need to add such a structure so that you can know what to
save/restore on context switch, then attach it to the per-CPU data
structure we already have.
M.
--
Without deviation from the norm, progress is not possible.
On Sat, Mar 02, 2024 at 11:30:51AM +0000, Marc Zyngier wrote:
> Mark Brown <[email protected]> wrote:
> > At present we store the various bits of floating point state individually
> > in struct kvm_vcpu_arch and construct a struct cpu_fp_state to share with
> > the host each time we exit the guest. Let's simplify this a little by
> > having a struct cpu_fp_state in the struct kvm_vcpu_arch and initialising
> > this while initialising the guest.
> This structure is only useful to the physical CPU we run on, and does
> not capture anything that is related to the guest state. Why should it
> live in the vcpu structure, duplicating things we already have?
You were previously complaining that we were having to bind too much
floating point state each time we bind a guest to a CPU, the goal here
is to address this concern by filling in the structure at startup and
then reusing it. As noted in the commit log given that this state
includes system registers there are difficulties in trying to rearrange
things so everything is in a single structure, I'm not sure that having
to special case the FP registers would be a win there.
> This is just making things even more opaque.
> If you need to add such a structure so that you can know what to
> save/restore on context switch, then attach it to the per-CPU data
> structure we already have.
I'm having a hard time understanding what you're thinking of here, this
is mainly about addressing whatever it is that's concering you but I'm
not sure I have a good handle on what that is other than just a general
concern that there's a lot of FP state which needs binding. If we put
something in the per-CPU state without reorganising the data a lot it'll
look very similar to the code you were complaining about, the current
code is doing roughly what you suggest already.
Personally I'm not overly concerned about the current state of the
world and find it to be a reasonable tradeoff given what's going on,
this is trying to address your feedback.