Zsh Mailing List Archive
Messages sorted by:
Reverse Date,
Date,
Thread,
Author
Re: ulimit -a: -r vs -N [was Re: pkgsrc patches for zsh]
On Wed, Jun 25, 2014 at 09:26:59AM +0100, Peter Stephenson wrote:
> That's supposed to be covered by -T for RLIMIT_PTHREAD, but the way
> NetBSD sets up its limits this doesn't work, and there appears to be a
> case missing in the handler. It was moved by zsh-workers/31936
> (http://www.zsh.org/mla/workers/2013/msg00970.html) to avoid the clash
> with other meanings of -r, which seems reasonable given the possible
> confusions, although we could provide backward compatibility where
> RTPRIO doesn't exist.
Thanks for the explanation and fix. I can confirm that with:
> diff --git a/Src/Builtins/rlimits.c b/Src/Builtins/rlimits.c
> index 0bcafda..a9eb328 100644
> --- a/Src/Builtins/rlimits.c
> +++ b/Src/Builtins/rlimits.c
> @@ -32,12 +32,14 @@
>
> #if defined(HAVE_GETRLIMIT) && defined(RLIM_INFINITY)
>
> -#ifdef RLIMIT_POSIXLOCKS
> +#if defined(HAVE_RLIMIT_POSIXLOCKS) && !defined(HAVE_RLIMIT_LOCKS)
> # define RLIMIT_LOCKS RLIMIT_POSIXLOCKS
> +# define HAVE_RLIMIT_LOCKS 1
> #endif
>
> -#ifdef RLIMIT_NTHR
> +#if defined(HAVE_RLIMIT_NTHR) && !defined(HAVE_RLIMIT_PTHREAD)
> # define RLIMIT_PTHREAD RLIMIT_NTHR
> +# define HAVE_RLIMIT_PTHREAD 1
> #endif
>
> enum {
> @@ -876,6 +878,11 @@ bin_ulimit(char *name, char **argv, UNUSED(Options ops), UNUSED(int func))
> res = RLIMIT_KQUEUES;
> break;
> # endif
> +# ifdef HAVE_RLIMIT_PTHREAD
> + case 'T':
> + res = RLIMIT_PTHREAD;
> + break;
> +# endif
> default:
> /* unrecognised limit */
> zwarnnam(name, "bad option: -%c", *options);
I see and can use -T:
# ulimit -a
-t: cpu time (seconds) unlimited
-f: file size (blocks) unlimited
-d: data seg size (kbytes) 262144
-s: stack size (kbytes) 4096
-c: core file size (blocks) unlimited
-m: resident set size (kbytes) 32485916
-l: locked-in-memory size (kbytes) 10828638
-u: processes 160
-n: file descriptors 128
-b: socket buffer size (bytes) unlimited
-v: virtual memory size (kbytes) unlimited
-T: threads per process 160
# ulimit -T 161
# ulimit -a
-t: cpu time (seconds) unlimited
-f: file size (blocks) unlimited
-d: data seg size (kbytes) 262144
-s: stack size (kbytes) 4096
-c: core file size (blocks) unlimited
-m: resident set size (kbytes) 32485916
-l: locked-in-memory size (kbytes) 10828638
-u: processes 160
-n: file descriptors 128
-b: socket buffer size (bytes) unlimited
-v: virtual memory size (kbytes) unlimited
-T: threads per process 161
The description for -T is not correct for NetBSD though. (I don't know
how this limit works on other operating systems.)
From NetBSD's sh(1)'s ulimit section:
-r show or set the limit on the number of threads this
user can have at one time
So the limit is not per-process, but for the user in total.
This part, however, is unneeded:
> diff --git a/configure.ac b/configure.ac
> index a2a6b9e..7c04c3a 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -1836,6 +1836,7 @@ zsh_LIMIT_PRESENT(RLIMIT_POSIXLOCKS)
> zsh_LIMIT_PRESENT(RLIMIT_NPTS)
> zsh_LIMIT_PRESENT(RLIMIT_SWAP)
> zsh_LIMIT_PRESENT(RLIMIT_KQUEUES)
> +zsh_LIMIT_PRESENT(RLIMIT_NTHR)
>
> AH_TEMPLATE([RLIMIT_VMEM_IS_RSS],
> [Define to 1 if RLIMIT_VMEM and RLIMIT_RSS both exist and are equal.])
The check for RLIMIT_NTHR is already in that file, about 10 lines
higher.
Cheers,
Thomas
Messages sorted by:
Reverse Date,
Date,
Thread,
Author