Skip to contentSkip to conversation

Finishing off cpumask...

DISCUSSION Rusty Russell 2 participants 3 messages Oct 21, 2010 – Oct 22, 2010

lore
Share & tools
lore
#0 lore raw

I have had five final cpumask patches in my queue for over a year now, and I'd like to merge them *next* merge window (not this one, too late).

The first two finally convert mm_struct and task_struct to variable cpumasks, which are the last in the core kernel. The next two turn off and break old-style cpumask uses when CONFIG_CPUMASK_OFFSTACK is set, and the final one actually means we don't allocate all NR_CPUS bits for bitmasks.

Historically, this stuff has revealed issues when turned on, so I'd like it to have a long baking cycle in Ingo's tree and big-machine testers.

I'm rebasing and retesting now, will post git tree when I have it.

Thanks,
Rusty.

#1 lore raw
Quoted 2 lines · Rusty Russell

The next two turn off and break
old-style cpumask uses when CONFIG_CPUMASK_OFFSTACK is set,

Breaking old style for !CONFIG_CPUMASK_OFFSTAK too would be awesome, but I guess that's hard to do?

#2 lore raw
Quoted 6 lines · Peter Zijlstra
Quoted 2 lines · Rusty Russell

The next two turn off and break
old-style cpumask uses when CONFIG_CPUMASK_OFFSTACK is set,

Breaking old style for !CONFIG_CPUMASK_OFFSTAK too would be awesome, but I guess that's hard to do?

There are four things we could do.
(1) Get rid of the old cpus_ operators (=> cpumask_*). They're just confusing, and removing them would be a clarity win.
(2) Get rid of cpu_*_map (=> cpu_*_mask). The maps are just wrappers.
(3) Getting rid of cpumask_t (=> struct cpumask) and NR_CPUS (=> CONFIG_NR_CPUS). These are totally cosmetic.
(4) Getting rid of cpumask_t/struct cpumask (=> cpumask_var_t) is basically only useful if an arch is thinking about CPUMASK_OFFSTACK, and most painful.

So, I've changed the CONFIG option which disables (1) and (2); now anyone will be able to turn it on if they want to test. That might encourage the per-arch cleanup...

Thanks,
Rusty.