Re: [PATCH v2] riscv: Discard vector state on syscalls

["Rémi Denis-Courmont" <remi@remlab.net>]

[Resend with different MTA]

Le 26 juin 2023 19:45:20 GMT+02:00, "Björn Töpel" <bjorn@kernel.org> a écrit :
>Björn Töpel <bjorn@kernel.org> writes:
>
>> From: Björn Töpel <bjorn@rivosinc.com>
>>
>> The RISC-V vector specification states:
>>   Executing a system call causes all caller-saved vector registers
>>   (v0-v31, vl, vtype) and vstart to become unspecified.
>
>A bit of a corner case, but this will make sigreturn syscalls discard
>the vector state as well.
>
>Is that an issue? E.g. a user cannot build userspace context switching
>application. Does arm64 SVE handle sigreturn in a special way?

Isn't sigreturn() supposed to return the status from the arch-dependent 
machine state within the siginfo structure, rather than whatever was saved on 
sigreturn() syscall entry? That being the case, I think throwing the vector 
register bank away on *entry* of sigreturn() is fine as with any other syscall, 
but the state must *not* be cleared on syscall exit.

An example usecase would be emulating RVV extensions (on a CPU supporting 
baseline RVV 1.0) with a SIGILL handler.

-- 
Rémi



_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv