2.4-ac affinity syscalls and how to pin a task?

The attached patch implements the task cpu affinity interface on top of the O(1) scheduler in 2.4.19-pre7-ac3. The migration_thread is now in -ac3 so we can implement such an interface.

This is the same interface as in 2.5 and posted previously for 2.4 (sans O(1)):

sched_getaffinity(pid_t pid, unsigned long len, unsigned long *mask) sched_setaffinity(pid_t pid, unsigned long len, unsigned long *mask)

Problem:

We need to assure the task stays valid and does not exit, but due to the semantics of set_cpus_allowed and migration_thread we can not hold a lock across a call. In 2.5 we can pin a task via get_task_struct. This assures us, at least, that the task struct is not deallocated until a final call to put_task_struct.

What equivalent mechanism do we have in 2.4? How can we assure a task stays put without grabbing a spinlock?

This patch grabs the tasklist_lock before calling set_cpus_allowed. While this is working fine here and is safe enough to play with now, it is not safe.

Comments?

Robert Love

diff -urN linux-2.4.19-pre7-ac3/arch/i386/kernel/entry.S linux/arch/i386/kernel/entry.S
@@ -639,8 +639,8 @@
  	.long SYMBOL_NAME(sys_tkill)
 	.long SYMBOL_NAME(sys_ni_syscall)	/* reserved for sendfile64 */
 	.long SYMBOL_NAME(sys_ni_syscall)	/* 240 reserved for futex */
-	.long SYMBOL_NAME(sys_ni_syscall)	/* reserved for sched_setaffinity */
-	.long SYMBOL_NAME(sys_ni_syscall)	/* reserved for sched_getaffinity */
+	.long SYMBOL_NAME(sys_sched_setaffinity)
+	.long SYMBOL_NAME(sys_sched_getaffinity)
 
 	.rept NR_syscalls-(.-sys_call_table)/4
 		.long SYMBOL_NAME(sys_ni_syscall)

diff -urN linux-2.4.19-pre7-ac3/include/linux/capability.h linux/include/linux/capability.h
@@ -243,6 +243,7 @@
 /* Allow use of FIFO and round-robin (realtime) scheduling on own
    processes and setting the scheduling algorithm used by another
    process. */
+/* Allow setting cpu affinity on other processes */
 
 #define CAP_SYS_NICE         23