On 11/12/26 19:37, Konstantin Khorenko wrote:
I won't apply this version, please send v4 with the changes below.
On Thu, 29 Oct 2026 11:16:06 +0100, Vladimir Riabchun
<[email protected]> wrote:
The root cause is incorrect bitmap_fill usage - it sets whole
integers only, without handling bit tails, so the actual number of
set bits is 64 = bits in ULONG.
Items 1 and 2 of the cover letter reply: add the Fixes: and Feature:
trailers, write "whole unsigned longs", and say what a CT sees now.
Why: 87211cbff4b4 added the CT branch with bitmap_fill(). Since v3
the vCPU limit comes from the target task and the count follows the
cpuset, so nproc may differ from /proc/cpuinfo. A reader of the
message should not have to find that in the cover changelog.
Will fix in v4.
+ /*
+ * setaffinity is not available inside CT, so these bits
+ * can only be used for counting.
Item 3: sched_setaffinity() returns 0 in a CT and does nothing, it is
not "not available".
This is actually very misleading. Maybe we want to explicitly return
-EPERM from sched_setaffinity in !ve_is_super(get_exec_env()) case?
Or the userspace will break? I tried with -EPERM and CTs start
just fine.
--
Best regards, Riabchun Vladimir
Linux Kernel Developer, Virtuozzo
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel