We use the presence of sched_setscheduler() and cpu_set_t as
witnesses for whether we can set process affinity and the
like.

This entire class of functionality is missing on Hurd, but
unfortunately it still gets advertised, so the end result is
that a number of warnings are produced during build:

  sched_get_priority_max is not implemented and will always fail
  sched_get_priority_min is not implemented and will always fail
  sched_getaffinity is not implemented and will always fail
  sched_setaffinity is not implemented and will always fail

Don't even attempt to detect the known-broken APIs, so that
we don't get compiler warnings and can report the missing
functionality correctly at runtime if attempts to use it are
made.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
---
 meson.build | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/meson.build b/meson.build
index 7845f60ff7..eef00ee32a 100644
--- a/meson.build
+++ b/meson.build
@@ -584,13 +584,19 @@ functions = [
   'posix_fallocate',
   'posix_memalign',
   'prlimit',
-  'sched_setscheduler',
   'setgroups',
   'setrlimit',
   'symlink',
   'sysctlbyname',
 ]
 
+# sched_* APIs are know to be broken on Hurd
+if host_machine.system() != 'gnu'
+  functions += [
+    'sched_setscheduler',
+  ]
+endif
+
 stat_functions = [
   '__lxstat',
   '__lxstat64',
@@ -669,8 +675,6 @@ symbols = [
 
   # Check for BSD approach for setting MAC addr
   [ 'net/if_dl.h', 'link_addr', '#include <sys/types.h>\n#include <sys/socket.h>' ],
-
-  [ 'sched.h', 'cpu_set_t' ],
 ]
 
 if host_machine.system() == 'linux'
@@ -680,6 +684,13 @@ if host_machine.system() == 'linux'
   ]
 endif
 
+# sched_* APIs are know to be broken on Hurd
+if host_machine.system() != 'gnu'
+  symbols += [
+    [ 'sched.h', 'cpu_set_t' ],
+  ]
+endif
+
 foreach symbol : symbols
   if cc.has_header_symbol(symbol[0], symbol[1], args: '-D_GNU_SOURCE', prefix: symbol.get(2, ''))
     conf.set('WITH_DECL_@0@'.format(symbol[1].to_upper()), 1)
-- 
2.43.2
_______________________________________________
Devel mailing list -- devel@lists.libvirt.org
To unsubscribe send an email to devel-leave@lists.libvirt.org

On 2/23/24 16:52, Andrea Bolognani wrote:
> We use the presence of sched_setscheduler() and cpu_set_t as
> witnesses for whether we can set process affinity and the
> like.
> 
> This entire class of functionality is missing on Hurd, but
> unfortunately it still gets advertised, so the end result is
> that a number of warnings are produced during build:
> 
>   sched_get_priority_max is not implemented and will always fail
>   sched_get_priority_min is not implemented and will always fail
>   sched_getaffinity is not implemented and will always fail
>   sched_setaffinity is not implemented and will always fail

This feels like a Hurd bug, sorry. Unless there is bug report against
Hurd (IOW, there's a chance this is fixed), I don't we should work
around it. Otherwise we'd be turning libvirt into a dumpster of workarounds.

Michal
_______________________________________________
Devel mailing list -- devel@lists.libvirt.org
To unsubscribe send an email to devel-leave@lists.libvirt.org

On Mon, Feb 26, 2024 at 11:25:12AM +0100, Michal Prívozník wrote:
> On 2/23/24 16:52, Andrea Bolognani wrote:
> > We use the presence of sched_setscheduler() and cpu_set_t as
> > witnesses for whether we can set process affinity and the
> > like.
> >
> > This entire class of functionality is missing on Hurd, but
> > unfortunately it still gets advertised, so the end result is
> > that a number of warnings are produced during build:
> >
> >   sched_get_priority_max is not implemented and will always fail
> >   sched_get_priority_min is not implemented and will always fail
> >   sched_getaffinity is not implemented and will always fail
> >   sched_setaffinity is not implemented and will always fail
>
> This feels like a Hurd bug, sorry. Unless there is bug report against
> Hurd (IOW, there's a chance this is fixed), I don't we should work
> around it. Otherwise we'd be turning libvirt into a dumpster of workarounds.

I've figured out a more sane approach, which also comes with free bug
fixes as a bonus :)

https://lists.libvirt.org/archives/list/devel@lists.libvirt.org/thread/MZZWENGR56QBABKQL2DTDSBRVM5XMI4W/

-- 
Andrea Bolognani / Red Hat / Virtualization
_______________________________________________
Devel mailing list -- devel@lists.libvirt.org
To unsubscribe send an email to devel-leave@lists.libvirt.org