| Message ID | 1504198860-12951-22-git-send-email-Dave.Martin@arm.com |
|---|---|
| State | New |
| Headers | show |
| Series | ARM Scalable Vector Extension (SVE) | expand |
Dave Martin <Dave.Martin@arm.com> writes: > Because of the effect of SVE on the size of the signal frame, the > default vector length used for new processes involves a tradeoff > between performance of SVE-enabled software on the one hand, and > reliability of non-SVE-aware software on the other hand. > > For this reason, the best choice depends on the repertoire of > userspace software in use and is thus best left up to distro > maintainers, sysadmins and developers. > > If CONFIG_SYSCTL is enabled, this patch exposes the default vector > length in /proc/sys/abi/sve_default_vector_length, where boot > scripts or the adventurous can poke it. > > In common with other arm64 ABI sysctls, this control is currently > global: setting it requires CAP_SYS_ADMIN in the root user > namespace, but the value set is effective for subsequent execs in > all namespaces. The control only affects _new_ processes, however: > changing it does not affect the vector length of any existing > process. > > The intended usage model is that if userspace is known to be fully > SVE-tolerant (or a developer is curious to find out) then init > scripts can crank this up during startup. > > Signed-off-by: Dave Martin <Dave.Martin@arm.com> > Cc: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> > > --- > > Changes since v1 > ---------------- > > Changes requested by Alex Bennée: > > * Thin out BUG_ON()s: > Redundant BUG_ON()s and ones that just check invariants are removed. > Important sanity-checks are migrated to WARN_ON()s, with some > minimal best-effort patch-up code. > --- > arch/arm64/kernel/fpsimd.c | 62 +++++++++++++++++++++++++++++++++++++++++++++- > 1 file changed, 61 insertions(+), 1 deletion(-) > > diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c > index 42e8331..b430ee0 100644 > --- a/arch/arm64/kernel/fpsimd.c > +++ b/arch/arm64/kernel/fpsimd.c > @@ -34,6 +34,7 @@ > #include <linux/sched/signal.h> > #include <linux/signal.h> > #include <linux/slab.h> > +#include <linux/sysctl.h> > > #include <asm/fpsimd.h> > #include <asm/cputype.h> > @@ -244,6 +245,65 @@ static unsigned int find_supported_vector_length(unsigned int vl) > return sve_vl_from_vq(bit_to_vq(bit)); > } > > +#ifdef CONFIG_SYSCTL > + > +static int sve_proc_do_default_vl(struct ctl_table *table, int write, > + void __user *buffer, size_t *lenp, > + loff_t *ppos) > +{ > + int ret; > + int vl = sve_default_vl; > + struct ctl_table tmp_table = { > + .data = &vl, > + .maxlen = sizeof(vl), > + }; > + > + ret = proc_dointvec(&tmp_table, write, buffer, lenp, ppos); > + if (ret || !write) > + return ret; > + > + /* Writing -1 has the special meaning "set to max": */ > + if (vl == -1) { > + /* Fail safe if sve_max_vl wasn't initialised */ > + if (WARN_ON(!sve_vl_valid(sve_max_vl))) > + vl = SVE_VL_MIN; > + else > + vl = sve_max_vl; > + > + goto chosen; > + } > + > + if (!sve_vl_valid(vl)) > + return -EINVAL; > + > + vl = find_supported_vector_length(vl); > +chosen: > + sve_default_vl = vl; > + return 0; > +} > + > +static struct ctl_table sve_default_vl_table[] = { > + { > + .procname = "sve_default_vector_length", > + .mode = 0644, > + .proc_handler = sve_proc_do_default_vl, > + }, > + { } > +}; > + > +static int __init sve_sysctl_init(void) > +{ > + if (system_supports_sve()) > + if (!register_sysctl("abi", sve_default_vl_table)) > + return -EINVAL; > + > + return 0; > +} > + > +#else /* ! CONFIG_SYSCTL */ > +static int __init sve_sysctl_init(void) { return 0; } > +#endif /* ! CONFIG_SYSCTL */ > + > #define ZREG(sve_state, vq, n) ((char *)(sve_state) + \ > (SVE_SIG_ZREG_OFFSET(vq, n) - SVE_SIG_REGS_OFFSET)) > > @@ -1030,6 +1090,6 @@ static int __init fpsimd_init(void) > if (!(elf_hwcap & HWCAP_ASIMD)) > pr_notice("Advanced SIMD is not implemented\n"); > > - return 0; > + return sve_sysctl_init(); > } > late_initcall(fpsimd_init); -- Alex Bennée
diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c index 42e8331..b430ee0 100644 --- a/arch/arm64/kernel/fpsimd.c +++ b/arch/arm64/kernel/fpsimd.c @@ -34,6 +34,7 @@ #include <linux/sched/signal.h> #include <linux/signal.h> #include <linux/slab.h> +#include <linux/sysctl.h> #include <asm/fpsimd.h> #include <asm/cputype.h> @@ -244,6 +245,65 @@ static unsigned int find_supported_vector_length(unsigned int vl) return sve_vl_from_vq(bit_to_vq(bit)); } +#ifdef CONFIG_SYSCTL + +static int sve_proc_do_default_vl(struct ctl_table *table, int write, + void __user *buffer, size_t *lenp, + loff_t *ppos) +{ + int ret; + int vl = sve_default_vl; + struct ctl_table tmp_table = { + .data = &vl, + .maxlen = sizeof(vl), + }; + + ret = proc_dointvec(&tmp_table, write, buffer, lenp, ppos); + if (ret || !write) + return ret; + + /* Writing -1 has the special meaning "set to max": */ + if (vl == -1) { + /* Fail safe if sve_max_vl wasn't initialised */ + if (WARN_ON(!sve_vl_valid(sve_max_vl))) + vl = SVE_VL_MIN; + else + vl = sve_max_vl; + + goto chosen; + } + + if (!sve_vl_valid(vl)) + return -EINVAL; + + vl = find_supported_vector_length(vl); +chosen: + sve_default_vl = vl; + return 0; +} + +static struct ctl_table sve_default_vl_table[] = { + { + .procname = "sve_default_vector_length", + .mode = 0644, + .proc_handler = sve_proc_do_default_vl, + }, + { } +}; + +static int __init sve_sysctl_init(void) +{ + if (system_supports_sve()) + if (!register_sysctl("abi", sve_default_vl_table)) + return -EINVAL; + + return 0; +} + +#else /* ! CONFIG_SYSCTL */ +static int __init sve_sysctl_init(void) { return 0; } +#endif /* ! CONFIG_SYSCTL */ + #define ZREG(sve_state, vq, n) ((char *)(sve_state) + \ (SVE_SIG_ZREG_OFFSET(vq, n) - SVE_SIG_REGS_OFFSET)) @@ -1030,6 +1090,6 @@ static int __init fpsimd_init(void) if (!(elf_hwcap & HWCAP_ASIMD)) pr_notice("Advanced SIMD is not implemented\n"); - return 0; + return sve_sysctl_init(); } late_initcall(fpsimd_init);
Because of the effect of SVE on the size of the signal frame, the default vector length used for new processes involves a tradeoff between performance of SVE-enabled software on the one hand, and reliability of non-SVE-aware software on the other hand. For this reason, the best choice depends on the repertoire of userspace software in use and is thus best left up to distro maintainers, sysadmins and developers. If CONFIG_SYSCTL is enabled, this patch exposes the default vector length in /proc/sys/abi/sve_default_vector_length, where boot scripts or the adventurous can poke it. In common with other arm64 ABI sysctls, this control is currently global: setting it requires CAP_SYS_ADMIN in the root user namespace, but the value set is effective for subsequent execs in all namespaces. The control only affects _new_ processes, however: changing it does not affect the vector length of any existing process. The intended usage model is that if userspace is known to be fully SVE-tolerant (or a developer is curious to find out) then init scripts can crank this up during startup. Signed-off-by: Dave Martin <Dave.Martin@arm.com> Cc: Alex Bennée <alex.bennee@linaro.org> --- Changes since v1 ---------------- Changes requested by Alex Bennée: * Thin out BUG_ON()s: Redundant BUG_ON()s and ones that just check invariants are removed. Important sanity-checks are migrated to WARN_ON()s, with some minimal best-effort patch-up code. --- arch/arm64/kernel/fpsimd.c | 62 +++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 61 insertions(+), 1 deletion(-)