This patch updates and fixes sys_tas() routine for m32r.
In the previous implementation, a lockup rarely caused at sys_tas()
routine in SMP environment.
> > The problem is that touching *addr will generate an oops if that page isn't
> > paged in. If we convert it to use get_user() then that's an improvement,
> > but we must not run get_user() under spinlock or local_irq_disable().
I rewrote sys_tas() routine by using "lock -> unlock" instructions, and
utilizing the m32r's interrupt handling characteristics; the m32r processor
can accept interrupts only at the 32-bit instruction boundary. So, the
"unlock" instruction can be executed continuously after the "lock"
instruction execution without any interruptions.
In addition, to solve such a page_fault problem, I use a fixup code like
get_user().
And, as for the kernel lockup problem, we found that a calling
do_page_fault() routine with disabling interrupts might cause a lockup at
flush_tlb_others(), because we checked a completion of IPI handler's
operations in a spin-locked critical section.
Therefore, by using "lock -> unlock" code, we can implement the sys_tas()
rouitine without disabling interrupts explicitly, then no lockups would
happen at flush_tlb_others(), I hope.
Compile check and some working test in SMP environment have done.
Signed-off-by: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch fixes a deadlock problem of the m32r SMP kernel.
In the m32r kernel, sys_tas() system call is provided as a test-and-set
function for userspace, for backward compatibility.
In some multi-threading application program, deadlocks were rarely caused
at sys_tas() funcion. Such a deadlock was caused due to a collision of
__pthread_lock() and __pthread_unlock() operations.
The "tas" syscall is repeatedly called by pthread_mutex_lock() to get a
lock, while a lock variable's value is not 0. On the other hand,
pthead_mutex_unlock() sets the lock variable to 0 for unlocking.
In the previous implementation of sys_tas() routine, there was a
possibility that a unlock operation was ignored in the following case:
- Assume a lock variable (*addr) was equal to 1 before sys_tas() execution.
- __pthread_unlock() operation is executed by the other processor
and the lock variable (*addr) is set to 0, between a read operation
("oldval = *addr;") and the following write operation ("*addr = 1;")
during a execution of sys_tas().
In this case, the following write operation ("*addr = 1;") overwrites the
__pthread_unlock() result, and sys_tas() fails to get a lock in the next
turn and after that.
According to the attatched patch, sys_tas() returns 0 value in the next
turn and deadlocks never happen.
Signed-off-by: Hitoshi Yamamoto <Yamamoto.Hitoshi@ap.MitsubishiElectric.co.jp>
Signed-off-by: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Jörn Engel