[TUHS] XID register

[TUHS] XID register

[ron minnich]

[-- Attachment #1: Type: text/plain, Size: 677 bytes --]

SunRPC, among other protocols, needs transaction IDs (XIDs) to distinguish
RPCs.For SunRPC, it's important that XIDs not be reused (not for all
protocols; 9p has no such requirement). Stateless protocols like NFS and
reused XIDs can get messy.

There is a vague, 30 year old memory, I have, that at some point SPARC got
a time register, or some other register, that always provided a different
answer each time it was read, even if read back to back, in part to enable
creation of non-reused XIDs. Note that things like the TSC or RISC-V MTIME
register make no such guarantee.

I am pretty sure someone here can fill me in, or tell me I'm wrong, about
my SPARC memory.

thanks

[-- Attachment #2: Type: text/html, Size: 780 bytes --]

[TUHS] Re: XID register

[Rico Pajarola]

[-- Attachment #1: Type: text/plain, Size: 1000 bytes --]

SPARCv9 has %tick register which just counts up on each clock cycle. Older
SPARCs don't have that. I'd still suspect you'd get non-unique results in a
multi CPU machine without additional effort.

On Sat, Dec 2, 2023 at 10:02 PM ron minnich <rminnich@gmail.com> wrote:

> SunRPC, among other protocols, needs transaction IDs (XIDs) to distinguish
> RPCs.For SunRPC, it's important that XIDs not be reused (not for all
> protocols; 9p has no such requirement). Stateless protocols like NFS and
> reused XIDs can get messy.
>
> There is a vague, 30 year old memory, I have, that at some point SPARC got
> a time register, or some other register, that always provided a different
> answer each time it was read, even if read back to back, in part to enable
> creation of non-reused XIDs. Note that things like the TSC or RISC-V MTIME
> register make no such guarantee.
>
> I am pretty sure someone here can fill me in, or tell me I'm wrong, about
> my SPARC memory.
>
> thanks
>

[-- Attachment #2: Type: text/html, Size: 1370 bytes --]