[illumos-gate] 15278 libsmb: unused parameter -- Toomas Soome <tsoome⊙mc>

happy new year everyone, I have question regarding to the API to equate to sched_setaffinity on linux, or cpuset_getaffinity on FreeBSD, what’s the API for illumos? (seems Oracle Solaris had processor_affinity), is there anyone could answer? I’m going to bind postgres process on assigned cores, thanks in advance

tozhu pooladm to assign cpu pool for example

or use FSS and assign cpu shares

good idea with FSS

tsoome_ : thanks , we just use C code to bind the process,

I don’t know how to do by FSS

is there any C api ?

ye, there is api to create cpu set, but not the processor_affinity()

tsoome_ : would you like to show me an example? or is there any docs for the API ?

see psrset manual, it has further links

but the truth is, this feature set is from the solaris 10/opensolaris fork time...

tsoome_: do you say pset_create / pset_bind ?

yes.

and pset_assign

needs set projects in /etc/projects file

yes, thats the feature to allow admin to create cpu set, assign processes to use that set or create pooled resource and assign processes to use resource pools.

processor_affinity can’t be used on illumos?

we do not have that api

the world has moved on since the fork and we only have so many people to do things. This kind of development needs time and motivation....

got it, thank you all for the hard work

if you are working for some company, then company tends to have priorities of their own...

by the way, needs set this value in the project?; docs.oracle.com/cd/E37838_01/html/E61042/gpusj.html#scrolltoc

that could be. I haven't used processor_affinity API myself (the solaris installations are being replaced by exadata/cloud bits here and practically everything is run on linux or windows).

For illumos.org/issues/15271 andyf pointed out using `set viona:viona_force_copy_state = 2` probably also works, I just verified it did by booting to omnios-bloody package set from 20221228 without my change

→ FEATURE 15271: Want tunable to force viona to always copy tx buffers (New) | code.illumos.org/c/illumos-gate/+/2565

The dtrace result with this set for `dtrace -n 'fbt:viona:viona_tx_copy_needed:return { printf("%u", args[1]) }'` was the same a with my patch

However is it OK to modify the enum directly ?

(github.com/illumos/illumos-gate/blo…s/intel/io/viona/viona_tx.c#L60-L64)

There's the same question if you set a tuneable to boolean_t, since that's an enum. system(5) says `set` can modify an integer, and they're ints, so I expect it's ok but I don't think I've seen a boolean or enum tuneable before.

I based it of omniosorg/illumos-omnios ed9257a

I might just have chosen a bad example to copy though

Heh, just a short memory then!

I did set that one to get the guid change quicker on my bloody setup

So I know that worked :D

[illumos-gate] 15280 uts: remove pragma ident -- Toomas Soome <tsoome⊙mc>

tsoome: am I right in thinking that #1128 means we have a kb8042 instance that isn't 0?

how the heck?

OS image migrated to different system

ah

in my case, I was switching UTM system between Q35 and i440FX

I forgot how long builds take when you have a shadow compiler and smatch works

tsoome: do you envision clang (or even better clang-tidy) replacing smatch?

unless there's a way to get clang to output the same info clang-tidy would.

I've considered hacking make to write out the json that makes clang-tidy more convenient to use

or a lil tool to create it from .make.state

perhaps, but I think it is still too early to tell...

my only real input is that in my experience clang-tidy (unless you can tell clang to run the full analyzer passes), gives much more detailed output about subtle issues.

and perhaps v. clang _any_ output about subtle issues.

of course, I haven't really checked what is the difference with current smatch (assuming it has been developed;)

it gives the me the entire voluminous control flow of uninitialized uses for eg

yes, I like that with clang:)

can I tell the compiler to do that rather than running clang-tidy separately?

'cos if so ignore me about a clang-tidy pass, obviously :)

there are some interesting bits of smatch we don't use... though unsure how useful they'd be for us

like you can tag stuff with something that indiciates it's a ptr w/ a userland address vs. kernel address to catch some stuff

IIUC

(and IIRC)