Skip to contentSkip to conversation

Signal-per-fd for RT signals

RFC PATCH In review Vitaly Luban 2 participants 2 messages May 19, 2001

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 Signal-per-fd for RT signals

Hi,

I'm sorry, the previous message slipped out w/o subj.

Attached patch is an implementation of "signal-per-fd" enhancement to kernel RT signal mechanism, AFAIK first proposed by A. Chandra and D. Mosberger :

http://www.hpl.hp.com/techreports/2000/HPL-2000-174.html

which should dramatically increase linux based network servers scalability.

Patch is made against 2.4.4 tree.

This patch allows to set signal-per-fd mode per each file descriptor by introducing new fcntls "F_SETAUXFL" and "F_GETAUXFL" with one possible argument "O_ONESIGFD" to F_SETAUXFL defined.

When set, no additional siginfo will be queued for an RT signal, generated by event on file descriptor, while there is already one queued, though event report field in already queued struct siginfo - "si_band" is updated.

I'd also like to hear an opinion on the signal filtering capability. I.e, it's relatively easy to filter signals upon an interest mask, supplied by the same F_SETAUXFL in the form of POLL_... This will bring functionality of RT signals event notification on the level with 'select' or 'poll' one, while more efficient and scalable. If there's an interest in such a feature, I'd be eager to publish a patch.

Thanks,
Vitaly.

#1 lore raw
Quoted 3 lines · Vitaly Luban

Hi,

<snip/>

Quoted 9 lines

the form of POLL_... This will bring functionality of RT signals event notification on the level with 'select' or 'poll' one, while more efficient and scalable. If there's an interest in such a feature, I'd be eager to publish a patch.

Thanks,
Vitaly.

I have been waiting for this patch since 2.4.0.

The SIGIO signal is a nightmare when it arrives :
The machine is already under high load and has to stop using the most efficient way to handle it.

The filter changes would be the cream on top of this patch. Do not hurry, but please not for long.

best Regards

Gerold