2.4.19-pre3 s390 ret_from_fork patch
Greetings:
The 2.4.19-pre3 works just dandy as it is without this patch, but it is needed for O(1) scheduler. I think Martin was going to forward this into 2.5.x tree, where O(1) is standard.
The problem is that schedule_tail() is supposed to unlock runqueue, but if interrupts are enabled before that happens, a timer tick may squeeze in and deadlock. I observed such situation myself.
So, I would like this to be in 2.4.19 for the sake of O(1) users. Any objections?
-- Pete
arch/s390/kernel/entry.S +4 -3
@@ -254,13 +254,14 @@ ret_from_fork: basr %r13,0 l %r13,.Lentry_base-.(%r13) # setup base pointer to &entry_base + # not saving R14 here because we go to sysc_return ultimately + l %r1,BASED(.Lschedtail) + basr %r14,%r1 # call schedule_tail (unlock stuff) GET_CURRENT # load pointer to task_struct to R9 stosm 24(%r15),0x03 # reenable interrupts sr %r0,%r0 # child returns 0 st %r0,SP_R2(%r15) # store return value (change R2 on stack) - l %r1,BASED(.Lschedtail) - la %r14,BASED(sysc_return) - br %r1 # call schedule_tail, return to sysc_return + b BASED(sysc_return) # # clone, fork, vfork, exec and sigreturn need glue,
arch/s390x/kernel/entry.S +2 -2
@@ -240,11 +240,11 @@ # .globl ret_from_fork ret_from_fork: + brasl %r14,schedule_tail GET_CURRENT # load pointer to task_struct to R9 stosm 48(%r15),0x03 # reenable interrupts xc SP_R2(8,%r15),SP_R2(%r15) # child returns 0 - larl %r14,sysc_return - jg schedule_tail # return to sysc_return + j sysc_return # # clone, fork, vfork, exec and sigreturn need glue,