On Mon, May 15, 2023 at 7:38 PM Björn Töpel <bjorn@xxxxxxxxxx> wrote:
>
> Andy Chiu <andy.chiu@xxxxxxxxxx> writes:
>
> > This patch add two riscv-specific prctls, to allow usespace control the
> > use of vector unit:
>
> A more general question; I know that it's only x86 that implements
> arch_prctl(), and that arm64 added the SVE prctl kernel/sys.c -- but is
> there a reason not to have an arch-specific prctl for riscv?
I didn't notice that there is an arch-specific prctl for x86 when
implementing this. Maintaining a separate prctl out of the generic one
to do arch-specific configurations makes code elegant. But the role of
generic prctl has becoming more "arch-specific" due to porting of
architectures. For example, the generic prctl are used by arm64 for
SVE/SME configs, which apparently are arch-specific. And adding a
syscal for a similar interface might confuse users if the line between
the two is not clear.
I think the question would be more like "Is it worth adding a
arch_prctl when the generic prctl has already been used by other
architectures for arch-specific configurations?".
>
> > * PR_RISCV_V_SET_CONTROL: control the permission to use Vector at next,
> > or all following execve for a thread. Turning off a thread's Vector
> > live is not possible since libraries may have registered ifunc that
> > may execute Vector instructions.
> > * PR_RISCV_V_GET_CONTROL: get the same permission setting for the
> > current thread, and the setting for following execve(s).
> >
> > Signed-off-by: Andy Chiu <andy.chiu@xxxxxxxxxx>
> > Reviewed-by: Greentime Hu <greentime.hu@xxxxxxxxxx>
> > Reviewed-by: Vincent Chen <vincent.chen@xxxxxxxxxx>
> > ---
> > arch/riscv/include/asm/processor.h | 13 ++++
> > arch/riscv/include/asm/vector.h | 4 ++
> > arch/riscv/kernel/process.c | 1 +
> > arch/riscv/kernel/vector.c | 108 +++++++++++++++++++++++++++++
> > arch/riscv/kvm/vcpu.c | 2 +
> > include/uapi/linux/prctl.h | 11 +++
> > kernel/sys.c | 12 ++++
> > 7 files changed, 151 insertions(+)
> >
> > diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h
> > index 38ded8c5f207..79261da74cfd 100644
> > --- a/arch/riscv/include/asm/processor.h
> > +++ b/arch/riscv/include/asm/processor.h
> > @@ -40,6 +40,7 @@ struct thread_struct {
> > unsigned long s[12]; /* s[0]: frame pointer */
> > struct __riscv_d_ext_state fstate;
> > unsigned long bad_cause;
> > + unsigned long vstate_ctrl;
> > struct __riscv_v_ext_state vstate;
> > };
> >
> > @@ -83,6 +84,18 @@ extern void riscv_fill_hwcap(void);
> > extern int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src);
> >
> > extern unsigned long signal_minsigstksz __ro_after_init;
> > +
> > +#ifdef CONFIG_RISCV_ISA_V
> > +/* Userspace interface for PR_RISCV_V_{SET,GET}_VS prctl()s: */
> > +#define RISCV_V_SET_CONTROL(arg) riscv_v_vstate_ctrl_set_current(arg)
> > +#define RISCV_V_GET_CONTROL() riscv_v_vstate_ctrl_get_current()
> > +extern unsigned int riscv_v_vstate_ctrl_set_current(unsigned long arg);
> > +extern unsigned int riscv_v_vstate_ctrl_get_current(void);
> > +#else /* !CONFIG_RISCV_ISA_V */
> > +#define RISCV_V_SET_CONTROL(arg) (-EINVAL)
> > +#define RISCV_V_GET_CONTROL() (-EINVAL)
>
> The else-clause is not needed (see my comment below for kernel/sys.c),
> and can be removed.
>
> > +#endif /* CONFIG_RISCV_ISA_V */
> > +
> > #endif /* __ASSEMBLY__ */
> >
> > #endif /* _ASM_RISCV_PROCESSOR_H */
>
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index 339fee3eff6a..412d2c126060 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -140,6 +140,12 @@
> > #ifndef GET_TAGGED_ADDR_CTRL
> > # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
> > #endif
> > +#ifndef PR_RISCV_V_SET_CONTROL
> > +# define PR_RISCV_V_SET_CONTROL(a) (-EINVAL)
> > +#endif
> > +#ifndef PR_RISCV_V_GET_CONTROL
> > +# define PR_RISCV_V_GET_CONTROL() (-EINVAL)
>
> Both SET/GET above should be RISCV_V_{SET,GET}_CONTROL (without the
> prefix "PR_"), and nothing else, otherwise...
>
> > +#endif
> >
> > /*
> > * this is where the system-wide overflow UID and GID are defined, for
> > @@ -2708,6 +2714,12 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> > error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
> > break;
> > #endif
> > + case PR_RISCV_V_SET_CONTROL:
> > + error = RISCV_V_SET_CONTROL(arg2);
> > + break;
> > + case PR_RISCV_V_GET_CONTROL:
> > + error = RISCV_V_GET_CONTROL();
> > + break;
>
>
> ...the case here will be weird. ;-)
Yes... fixing that now
>
>
> Björn
Thanks,
Andy
