Starting with kernel 2.6.13 we're going to have to change the syscall
numbers we use - the ssisys syscall number is now used by one of the
inotify calls in the base kernel.
To ease the pain of the transition here is a patch to libcluster to make
it try the new numbers, and if they don't work fall back to the old ones.
This would allow us to install a new kernel, and if necessary fall back
to an older one without being required to reinstall the system.
(It's actually a patch to CI code, but the relationship between CI and
OpenSSI is nothing if not incestuous).
(To make this work we would have something in unistd.h like:
/* BEGIN CLUSTER SYSTEM CALLS */
#define __NR_old_ssisys 290 /* conflict with __NR_ioprio_get */
#define __NR_old_rfork 291
#define __NR_old_rexecve 292
#define __NR_old_migrate 293
#define __NR_ssisys 400 /* Real number to be decided later */
#define __NR_rfork 401
#define __NR_rexecve 402
#define __NR_migrate 403
The actual numbers to used must be big enough to last some time. By
2.6.23 the base system has got up to 324!)
From: John Hughes <john@Calva.COM> - 2009-01-22 10:52:32
John Hughes wrote:
> Starting with kernel 2.6.13 we're going to have to change the syscall
> numbers we use - the ssisys syscall number is now used by one of the
> inotify calls in the base kernel.
Oh, yes, strace will also need some kind of fix.
Get latest updates about Open Source Projects, Conferences and News.