Re: [PATCH 2/2] arm64: Support systems without FP/ASIMD

From: Ard Biesheuvel
Date: Tue Oct 25 2016 - 10:00:30 EST

Hi Suzuki,

On 25 October 2016 at 14:50, Suzuki K Poulose <suzuki.poulose@xxxxxxx> wrote:
> The arm64 kernel assumes that FP/ASIMD units are always present
> and accesses the FP/ASIMD specific registers unconditionally. This
> could cause problems when they are absent. This patch adds the
> support for kernel handling systems without FP/ASIMD by skipping the
> register access within the kernel. For kvm, we trap the accesses
> to FP/ASIMD and inject an undefined instruction exception to the VM.
>
> The callers of the exported kernel_neon_begin_parital() should
> make sure that the FP/ASIMD is supported.
>
> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> Cc: Will Deacon <will.deacon@xxxxxxx>
> Cc: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
> Cc: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> Signed-off-by: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
> ---
> ---
> arch/arm64/crypto/aes-ce-ccm-glue.c | 2 +-
> arch/arm64/crypto/aes-ce-cipher.c | 2 ++
> arch/arm64/crypto/ghash-ce-glue.c | 2 ++
> arch/arm64/crypto/sha1-ce-glue.c | 2 ++
> arch/arm64/include/asm/cpufeature.h | 8 +++++++-
> arch/arm64/include/asm/neon.h | 3 ++-
> arch/arm64/kernel/cpufeature.c | 15 +++++++++++++++
> arch/arm64/kernel/fpsimd.c | 14 ++++++++++++++
> arch/arm64/kvm/handle_exit.c | 11 +++++++++++
> arch/arm64/kvm/hyp/hyp-entry.S | 9 ++++++++-
> arch/arm64/kvm/hyp/switch.c | 5 ++++-
> 11 files changed, 68 insertions(+), 5 deletions(-)
>
> diff --git a/arch/arm64/crypto/aes-ce-ccm-glue.c b/arch/arm64/crypto/aes-ce-ccm-glue.c
> index f4bf2f2..d001b4e 100644
> --- a/arch/arm64/crypto/aes-ce-ccm-glue.c
> +++ b/arch/arm64/crypto/aes-ce-ccm-glue.c
> @@ -296,7 +296,7 @@ static struct aead_alg ccm_aes_alg = {
>
> static int __init aes_mod_init(void)
> {
> - if (!(elf_hwcap & HWCAP_AES))
> + if (!(elf_hwcap & HWCAP_AES) || !system_supports_fpsimd())

This looks weird to me. All crypto extensionsinstructions except CRC
operate strictly on FP/ASIMD registers, and so support for FP/ASIMD is
implied by having HWCAP_AES. In other words, I think it makes more
sense to sanity check that the info registers are consistent with each
other in core code than modifying each user (which for HWCAP_xxx
includes userland) to double check that the world is sane.

> return -ENODEV;
> return crypto_register_aead(&ccm_aes_alg);
> }
> diff --git a/arch/arm64/crypto/aes-ce-cipher.c b/arch/arm64/crypto/aes-ce-cipher.c
> index f7bd9bf..1a43be2 100644
> --- a/arch/arm64/crypto/aes-ce-cipher.c
> +++ b/arch/arm64/crypto/aes-ce-cipher.c
> @@ -253,6 +253,8 @@ static struct crypto_alg aes_alg = {
>
> static int __init aes_mod_init(void)
> {
> + if (!system_supports_fpsimd())
> + return -ENODEV;
> return crypto_register_alg(&aes_alg);
> }
>
> diff --git a/arch/arm64/crypto/ghash-ce-glue.c b/arch/arm64/crypto/ghash-ce-glue.c
> index 833ec1e..2bc518d 100644
> --- a/arch/arm64/crypto/ghash-ce-glue.c
> +++ b/arch/arm64/crypto/ghash-ce-glue.c
> @@ -144,6 +144,8 @@ static struct shash_alg ghash_alg = {
>
> static int __init ghash_ce_mod_init(void)
> {
> + if (!system_supports_fpsimd())
> + return -ENODEV;
> return crypto_register_shash(&ghash_alg);
> }
>
> diff --git a/arch/arm64/crypto/sha1-ce-glue.c b/arch/arm64/crypto/sha1-ce-glue.c
> index aefda98..9f3427a 100644
> --- a/arch/arm64/crypto/sha1-ce-glue.c
> +++ b/arch/arm64/crypto/sha1-ce-glue.c
> @@ -102,6 +102,8 @@ static struct shash_alg alg = {
>
> static int __init sha1_ce_mod_init(void)
> {
> + if (!system_supports_fpsimd())
> + return -ENODEV;
> return crypto_register_shash(&alg);
> }
>
> diff --git a/arch/arm64/include/asm/cpufeature.h b/arch/arm64/include/asm/cpufeature.h
> index ae5e994..63d739c 100644
> --- a/arch/arm64/include/asm/cpufeature.h
> +++ b/arch/arm64/include/asm/cpufeature.h
> @@ -38,8 +38,9 @@
> #define ARM64_HAS_32BIT_EL0 13
> #define ARM64_HYP_OFFSET_LOW 14
> #define ARM64_MISMATCHED_CACHE_LINE_SIZE 15
> +#define ARM64_HAS_NO_FPSIMD 16
>
> -#define ARM64_NCAPS 16
> +#define ARM64_NCAPS 17
>
> #ifndef __ASSEMBLY__
>
> @@ -236,6 +237,11 @@ static inline bool system_supports_mixed_endian_el0(void)
> return id_aa64mmfr0_mixed_endian_el0(read_system_reg(SYS_ID_AA64MMFR0_EL1));
> }
>
> +static inline bool system_supports_fpsimd(void)
> +{
> + return !cpus_have_const_cap(ARM64_HAS_NO_FPSIMD);
> +}
> +
> #endif /* __ASSEMBLY__ */
>
> #endif
> diff --git a/arch/arm64/include/asm/neon.h b/arch/arm64/include/asm/neon.h
> index 13ce4cc..ad4cdc9 100644
> --- a/arch/arm64/include/asm/neon.h
> +++ b/arch/arm64/include/asm/neon.h
> @@ -9,8 +9,9 @@
> */
>
> #include <linux/types.h>
> +#include <asm/fpsimd.h>
>
> -#define cpu_has_neon() (1)
> +#define cpu_has_neon() system_supports_fpsimd()
>
> #define kernel_neon_begin() kernel_neon_begin_partial(32)
>
> diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> index d577f26..8f22544 100644
> --- a/arch/arm64/kernel/cpufeature.c
> +++ b/arch/arm64/kernel/cpufeature.c
> @@ -744,6 +744,14 @@ static bool hyp_offset_low(const struct arm64_cpu_capabilities *entry,
> return idmap_addr > GENMASK(VA_BITS - 2, 0) && !is_kernel_in_hyp_mode();
> }
>
> +static bool has_no_fpsimd(const struct arm64_cpu_capabilities *entry, int __unused)
> +{
> + u64 pfr0 = read_system_reg(SYS_ID_AA64PFR0_EL1);
> +
> + return cpuid_feature_extract_signed_field(pfr0,
> + ID_AA64PFR0_FP_SHIFT) < 0;
> +}
> +
> static const struct arm64_cpu_capabilities arm64_features[] = {
> {
> .desc = "GIC system register CPU interface",
> @@ -827,6 +835,13 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
> .def_scope = SCOPE_SYSTEM,
> .matches = hyp_offset_low,
> },
> + {
> + /* FP/SIMD is not implemented */
> + .capability = ARM64_HAS_NO_FPSIMD,
> + .def_scope = SCOPE_SYSTEM,
> + .min_field_value = 0,
> + .matches = has_no_fpsimd,
> + },
> {},
> };
>
> diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
> index 975b274..80da5aa 100644
> --- a/arch/arm64/kernel/fpsimd.c
> +++ b/arch/arm64/kernel/fpsimd.c
> @@ -127,6 +127,8 @@ void do_fpsimd_exc(unsigned int esr, struct pt_regs *regs)
>
> void fpsimd_thread_switch(struct task_struct *next)
> {
> + if (!system_supports_fpsimd())
> + return;
> /*
> * Save the current FPSIMD state to memory, but only if whatever is in
> * the registers is in fact the most recent userland FPSIMD state of
> @@ -157,6 +159,8 @@ void fpsimd_thread_switch(struct task_struct *next)
>
> void fpsimd_flush_thread(void)
> {
> + if (!system_supports_fpsimd())
> + return;
> memset(&current->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
> fpsimd_flush_task_state(current);
> set_thread_flag(TIF_FOREIGN_FPSTATE);
> @@ -168,6 +172,8 @@ void fpsimd_flush_thread(void)
> */
> void fpsimd_preserve_current_state(void)
> {
> + if (!system_supports_fpsimd())
> + return;
> preempt_disable();
> if (!test_thread_flag(TIF_FOREIGN_FPSTATE))
> fpsimd_save_state(&current->thread.fpsimd_state);
> @@ -181,6 +187,8 @@ void fpsimd_preserve_current_state(void)
> */
> void fpsimd_restore_current_state(void)
> {
> + if (!system_supports_fpsimd())
> + return;
> preempt_disable();
> if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) {
> struct fpsimd_state *st = &current->thread.fpsimd_state;
> @@ -199,6 +207,8 @@ void fpsimd_restore_current_state(void)
> */
> void fpsimd_update_current_state(struct fpsimd_state *state)
> {
> + if (!system_supports_fpsimd())
> + return;
> preempt_disable();
> fpsimd_load_state(state);
> if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) {
> @@ -228,6 +238,8 @@ static DEFINE_PER_CPU(struct fpsimd_partial_state, softirq_fpsimdstate);
> */
> void kernel_neon_begin_partial(u32 num_regs)
> {
> + if (WARN_ON(!system_supports_fpsimd()))
> + return;
> if (in_interrupt()) {
> struct fpsimd_partial_state *s = this_cpu_ptr(
> in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate);
> @@ -252,6 +264,8 @@ EXPORT_SYMBOL(kernel_neon_begin_partial);
>
> void kernel_neon_end(void)
> {
> + if (!system_supports_fpsimd())
> + return;
> if (in_interrupt()) {
> struct fpsimd_partial_state *s = this_cpu_ptr(
> in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate);
> diff --git a/arch/arm64/kvm/handle_exit.c b/arch/arm64/kvm/handle_exit.c
> index fa96fe2..39e055a 100644
> --- a/arch/arm64/kvm/handle_exit.c
> +++ b/arch/arm64/kvm/handle_exit.c
> @@ -57,6 +57,16 @@ static int handle_smc(struct kvm_vcpu *vcpu, struct kvm_run *run)
> return 1;
> }
>
> +/*
> + * Guest access to FP/ASIMD registers are routed to this handler only
> + * when the system doesn't support FP/ASIMD.
> + */
> +static int handle_no_fpsimd(struct kvm_vcpu *vcpu, struct kvm_run *run)
> +{
> + kvm_inject_undefined(vcpu);
> + return 1;
> +}
> +
> /**
> * kvm_handle_wfx - handle a wait-for-interrupts or wait-for-event
> * instruction executed by a guest
> @@ -144,6 +154,7 @@ static exit_handle_fn arm_exit_handlers[] = {
> [ESR_ELx_EC_BREAKPT_LOW]= kvm_handle_guest_debug,
> [ESR_ELx_EC_BKPT32] = kvm_handle_guest_debug,
> [ESR_ELx_EC_BRK64] = kvm_handle_guest_debug,
> + [ESR_ELx_EC_FP_ASIMD] = handle_no_fpsimd,
> };
>
> static exit_handle_fn kvm_get_exit_handler(struct kvm_vcpu *vcpu)
> diff --git a/arch/arm64/kvm/hyp/hyp-entry.S b/arch/arm64/kvm/hyp/hyp-entry.S
> index f6d9694..16167d7 100644
> --- a/arch/arm64/kvm/hyp/hyp-entry.S
> +++ b/arch/arm64/kvm/hyp/hyp-entry.S
> @@ -117,9 +117,16 @@ el1_trap:
> * x2: ESR_EC
> */
>
> - /* Guest accessed VFP/SIMD registers, save host, restore Guest */
> + /*
> + * We trap the first access to the FP/SIMD to save the host context
> + * and restore the guest context lazily.
> + * If FP/SIMD is not implemented, handle the trap and inject an
> + * undefined instruction exception to the guest.
> + */
> +alternative_if_not ARM64_HAS_NO_FPSIMD
> cmp x2, #ESR_ELx_EC_FP_ASIMD
> b.eq __fpsimd_guest_restore
> +alternative_else_nop_endif
>
> mrs x0, tpidr_el2
> mov x1, #ARM_EXCEPTION_TRAP
> diff --git a/arch/arm64/kvm/hyp/switch.c b/arch/arm64/kvm/hyp/switch.c
> index ae7855f..f2d0c4f 100644
> --- a/arch/arm64/kvm/hyp/switch.c
> +++ b/arch/arm64/kvm/hyp/switch.c
> @@ -18,6 +18,7 @@
> #include <linux/types.h>
> #include <asm/kvm_asm.h>
> #include <asm/kvm_hyp.h>
> +#include <asm/fpsimd.h>
>
> static bool __hyp_text __fpsimd_enabled_nvhe(void)
> {
> @@ -73,9 +74,11 @@ static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
> * traps are only taken to EL2 if the operation would not otherwise
> * trap to EL1. Therefore, always make sure that for 32-bit guests,
> * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
> + * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
> + * it will cause an exception.
> */
> val = vcpu->arch.hcr_el2;
> - if (!(val & HCR_RW)) {
> + if (!(val & HCR_RW) && system_supports_fpsimd()) {
> write_sysreg(1 << 30, fpexc32_el2);
> isb();
> }
> --
> 2.7.4
>