Re: [RFC PATCH 6/6] cpuset: Add cpuset.isolation_mask file

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Jul 15, 2021 at 01:13:38AM +0200, Frederic Weisbecker wrote:
> On Wed, Jul 14, 2021 at 06:52:43PM +0200, Peter Zijlstra wrote:

> > cpusets already has means to create paritions; why are you creating
> > something else?
> 
> I was about to answer that the semantics of isolcpus, which reference
> a NULL domain, are different from SD_LOAD_BALANCE implied by
> cpuset.sched_load_balance. But then I realize that SD_LOAD_BALANCE has
> been removed.
> 
> How cpuset.sched_load_balance is implemented then? Commit
> e669ac8ab952df2f07dee1e1efbf40647d6de332 ("sched: Remove checks against
> SD_LOAD_BALANCE") advertize that setting cpuset.sched_load_balance to 0
> ends up creating NULL domain but that's not what I get. For example if I
> mount a single cpuset root (no other cpuset mountpoints):

SD_LOAD_BALANCE was only for when you wanted to stop balancing inside a
domain tree. That no longer happens (and hasn't for a *long* time).
Cpusets simply creates multiple domain trees (or the empty one if its
just one CPU).

> $ mount -t cgroup none ./cpuset -o cpuset
> $ cd cpuset
> $ cat cpuset.cpus
> 0-7
> $ cat cpuset.sched_load_balance
> 1
> $ echo 0 > cpuset.sched_load_balance
> $ ls /sys/kernel/debug/domains/cpu1/
> domain0  domain1
> 
> I still get the domains on all CPUs...

(note, that's the cgroup-v1 interface, the cgroup-v2 interface is
significantly different)

I'd suggest doing: echo 1 > /debug/sched/verbose, if I do the above I
get:

[1290784.889705] CPU0 attaching NULL sched-domain.
[1290784.894830] CPU1 attaching NULL sched-domain.
[1290784.899947] CPU2 attaching NULL sched-domain.
[1290784.905056] CPU3 attaching NULL sched-domain.
[1290784.910153] CPU4 attaching NULL sched-domain.
[1290784.915252] CPU5 attaching NULL sched-domain.
[1290784.920338] CPU6 attaching NULL sched-domain.
[1290784.925439] CPU7 attaching NULL sched-domain.
[1290784.930535] CPU8 attaching NULL sched-domain.
[1290784.935660] CPU9 attaching NULL sched-domain.
[1290784.940911] CPU10 attaching NULL sched-domain.
[1290784.946117] CPU11 attaching NULL sched-domain.
[1290784.951317] CPU12 attaching NULL sched-domain.
[1290784.956507] CPU13 attaching NULL sched-domain.
[1290784.961688] CPU14 attaching NULL sched-domain.
[1290784.966876] CPU15 attaching NULL sched-domain.
[1290784.972047] CPU16 attaching NULL sched-domain.
[1290784.977218] CPU17 attaching NULL sched-domain.
[1290784.982383] CPU18 attaching NULL sched-domain.
[1290784.987552] CPU19 attaching NULL sched-domain.
[1290784.992724] CPU20 attaching NULL sched-domain.
[1290784.997893] CPU21 attaching NULL sched-domain.
[1290785.003063] CPU22 attaching NULL sched-domain.
[1290785.008230] CPU23 attaching NULL sched-domain.
[1290785.013400] CPU24 attaching NULL sched-domain.
[1290785.018568] CPU25 attaching NULL sched-domain.
[1290785.023736] CPU26 attaching NULL sched-domain.
[1290785.028905] CPU27 attaching NULL sched-domain.
[1290785.034074] CPU28 attaching NULL sched-domain.
[1290785.039241] CPU29 attaching NULL sched-domain.
[1290785.044409] CPU30 attaching NULL sched-domain.
[1290785.049579] CPU31 attaching NULL sched-domain.
[1290785.054816] CPU32 attaching NULL sched-domain.
[1290785.059986] CPU33 attaching NULL sched-domain.
[1290785.065154] CPU34 attaching NULL sched-domain.
[1290785.070323] CPU35 attaching NULL sched-domain.
[1290785.075492] CPU36 attaching NULL sched-domain.
[1290785.080662] CPU37 attaching NULL sched-domain.
[1290785.085832] CPU38 attaching NULL sched-domain.
[1290785.091001] CPU39 attaching NULL sched-domain.

Then when I do:

# mkdir /cgroup/A
# echo 0,20 > /cgroup/A/cpuset.cpus

I get:

[1291020.749036] CPU0 attaching sched-domain(s):
[1291020.754251]  domain-0: span=0,20 level=SMT
[1291020.759061]   groups: 0:{ span=0 }, 20:{ span=20 }
[1291020.765386] CPU20 attaching sched-domain(s):
[1291020.770399]  domain-0: span=0,20 level=SMT
[1291020.775210]   groups: 20:{ span=20 }, 0:{ span=0 }
[1291020.780831] root domain span: 0,20 (max cpu_capacity = 1024)

IOW, I've created a load-balance domain on just the first core of the
system.

# echo 0-1,20-21 > /cgroup/A/cpuset.cpus

Extends it to the first two cores:

[1291340.260699] CPU0 attaching NULL sched-domain.
[1291340.265820] CPU20 attaching NULL sched-domain.
[1291340.271403] CPU0 attaching sched-domain(s):
[1291340.276315]  domain-0: span=0,20 level=SMT
[1291340.281122]   groups: 0:{ span=0 }, 20:{ span=20 }
[1291340.286719]   domain-1: span=0-1,20-21 level=MC
[1291340.292011]    groups: 0:{ span=0,20 cap=2048 }, 1:{ span=1,21 cap=2048 }
[1291340.299855] CPU1 attaching sched-domain(s):
[1291340.304757]  domain-0: span=1,21 level=SMT
[1291340.309564]   groups: 1:{ span=1 }, 21:{ span=21 }
[1291340.315190]   domain-1: span=0-1,20-21 level=MC
[1291340.320474]    groups: 1:{ span=1,21 cap=2048 }, 0:{ span=0,20 cap=2048 }
[1291340.328307] CPU20 attaching sched-domain(s):
[1291340.333344]  domain-0: span=0,20 level=SMT
[1291340.338136]   groups: 20:{ span=20 }, 0:{ span=0 }
[1291340.343721]   domain-1: span=0-1,20-21 level=MC
[1291340.348980]    groups: 0:{ span=0,20 cap=2048 }, 1:{ span=1,21 cap=2048 }
[1291340.356783] CPU21 attaching sched-domain(s):
[1291340.361755]  domain-0: span=1,21 level=SMT
[1291340.366534]   groups: 21:{ span=21 }, 1:{ span=1 }
[1291340.372099]   domain-1: span=0-1,20-21 level=MC
[1291340.377364]    groups: 1:{ span=1,21 cap=2048 }, 0:{ span=0,20 cap=2048 }
[1291340.385216] root domain span: 0-1,20-21 (max cpu_capacity = 1024)