This might be broken on several other architectures as well therefore
posting here:
We had a bug on s390 when preemption is enabled and preemption of the BKL
is enabled as well. If returning from an interrupt to kernel context we
check for TIF_NEED_RESCHED and if that is true we just call schedule().
However this is broken. Instead of schedule we must call
preempt_schedule_irq() otherwise the BKL will be unlocked and somebody else
might grab it.
This bug explains some strange list corruptions we've seen in fs/locks.c
where all the locking is done with the BKL. And probably a bunch of other
unexplainable bugs.
Patch below is more or less what is needed on s390.
---
arch/s390/kernel/entry.S | 8 +++-----
arch/s390/kernel/entry64.S | 7 +------
2 files changed, 4 insertions(+), 11 deletions(-)
Index: linux-2.6/arch/s390/kernel/entry64.S
===================================================================
--- linux-2.6.orig/arch/s390/kernel/entry64.S
+++ linux-2.6/arch/s390/kernel/entry64.S
@@ -583,12 +583,7 @@ io_preempt:
io_resume_loop:
tm __TI_flags+7(%r9),_TIF_NEED_RESCHED
jno io_leave
- larl %r1,.Lc_pactive
- mvc __TI_precount(4,%r9),0(%r1)
- stosm __SF_EMPTY(%r15),0x03 # reenable interrupts
- brasl %r14,schedule # call schedule
- stnsm __SF_EMPTY(%r15),0xfc # disable I/O and ext. interrupts
- xc __TI_precount(4,%r9),__TI_precount(%r9)
+ brasl %r14,preempt_schedule_irq # call schedule
j io_resume_loop
#endif
Index: linux-2.6/arch/s390/kernel/entry.S
===================================================================
--- linux-2.6.orig/arch/s390/kernel/entry.S
+++ linux-2.6/arch/s390/kernel/entry.S
@@ -597,12 +597,8 @@ io_preempt:
io_resume_loop:
tm __TI_flags+3(%r9),_TIF_NEED_RESCHED
bno BASED(io_leave)
- mvc __TI_precount(4,%r9),BASED(.Lc_pactive)
- stosm __SF_EMPTY(%r15),0x03 # reenable interrupts
- l %r1,BASED(.Lschedule)
+ l %r1,BASED(.Lpreempt_schedule_irq)
basr %r14,%r1 # call schedule
- stnsm __SF_EMPTY(%r15),0xfc # disable I/O and ext. interrupts
- xc __TI_precount(4,%r9),__TI_precount(%r9)
b BASED(io_resume_loop)
#endif
@@ -1043,6 +1039,8 @@ cleanup_io_leave_insn:
.Lexecve_tail: .long execve_tail
.Ljump_table: .long pgm_check_table
.Lschedule: .long schedule
+.Lpreempt_schedule_irq:
+ .long preempt_schedule_irq
.Ltrace: .long syscall_trace
.Lschedtail: .long schedule_tail
.Lsysc_table: .long sys_call_table
-
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
