summaryrefslogtreecommitdiff
path: root/Documentation/cli-sti-removal.txt
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /Documentation/cli-sti-removal.txt
downloadlwn-1da177e4c3f41524e886b7f1b8a0c1fc7321cac2.tar.gz
lwn-1da177e4c3f41524e886b7f1b8a0c1fc7321cac2.zip
Linux-2.6.12-rc2v2.6.12-rc2
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!
Diffstat (limited to 'Documentation/cli-sti-removal.txt')
-rw-r--r--Documentation/cli-sti-removal.txt133
1 files changed, 133 insertions, 0 deletions
diff --git a/Documentation/cli-sti-removal.txt b/Documentation/cli-sti-removal.txt
new file mode 100644
index 000000000000..0223c9d20331
--- /dev/null
+++ b/Documentation/cli-sti-removal.txt
@@ -0,0 +1,133 @@
+
+#### cli()/sti() removal guide, started by Ingo Molnar <mingo@redhat.com>
+
+
+as of 2.5.28, five popular macros have been removed on SMP, and
+are being phased out on UP:
+
+ cli(), sti(), save_flags(flags), save_flags_cli(flags), restore_flags(flags)
+
+until now it was possible to protect driver code against interrupt
+handlers via a cli(), but from now on other, more lightweight methods
+have to be used for synchronization, such as spinlocks or semaphores.
+
+for example, driver code that used to do something like:
+
+ struct driver_data;
+
+ irq_handler (...)
+ {
+ ....
+ driver_data.finish = 1;
+ driver_data.new_work = 0;
+ ....
+ }
+
+ ...
+
+ ioctl_func (...)
+ {
+ ...
+ cli();
+ ...
+ driver_data.finish = 0;
+ driver_data.new_work = 2;
+ ...
+ sti();
+ ...
+ }
+
+was SMP-correct because the cli() function ensured that no
+interrupt handler (amongst them the above irq_handler()) function
+would execute while the cli()-ed section is executing.
+
+but from now on a more direct method of locking has to be used:
+
+ spinlock_t driver_lock = SPIN_LOCK_UNLOCKED;
+ struct driver_data;
+
+ irq_handler (...)
+ {
+ unsigned long flags;
+ ....
+ spin_lock_irqsave(&driver_lock, flags);
+ ....
+ driver_data.finish = 1;
+ driver_data.new_work = 0;
+ ....
+ spin_unlock_irqrestore(&driver_lock, flags);
+ ....
+ }
+
+ ...
+
+ ioctl_func (...)
+ {
+ ...
+ spin_lock_irq(&driver_lock);
+ ...
+ driver_data.finish = 0;
+ driver_data.new_work = 2;
+ ...
+ spin_unlock_irq(&driver_lock);
+ ...
+ }
+
+the above code has a number of advantages:
+
+- the locking relation is easier to understand - actual lock usage
+ pinpoints the critical sections. cli() usage is too opaque.
+ Easier to understand means it's easier to debug.
+
+- it's faster, because spinlocks are faster to acquire than the
+ potentially heavily-used IRQ lock. Furthermore, your driver does
+ not have to wait eg. for a big heavy SCSI interrupt to finish,
+ because the driver_lock spinlock is only used by your driver.
+ cli() on the other hand was used by many drivers, and extended
+ the critical section to the whole IRQ handler function - creating
+ serious lock contention.
+
+
+to make the transition easier, we've still kept the cli(), sti(),
+save_flags(), save_flags_cli() and restore_flags() macros defined
+on UP systems - but their usage will be phased out until 2.6 is
+released.
+
+drivers that want to disable local interrupts (interrupts on the
+current CPU), can use the following five macros:
+
+ local_irq_disable(), local_irq_enable(), local_save_flags(flags),
+ local_irq_save(flags), local_irq_restore(flags)
+
+but beware, their meaning and semantics are much simpler, far from
+that of the old cli(), sti(), save_flags(flags) and restore_flags(flags)
+SMP meaning:
+
+ local_irq_disable() => turn local IRQs off
+
+ local_irq_enable() => turn local IRQs on
+
+ local_save_flags(flags) => save the current IRQ state into flags. The
+ state can be on or off. (on some
+ architectures there's even more bits in it.)
+
+ local_irq_save(flags) => save the current IRQ state into flags and
+ disable interrupts.
+
+ local_irq_restore(flags) => restore the IRQ state from flags.
+
+(local_irq_save can save both irqs on and irqs off state, and
+local_irq_restore can restore into both irqs on and irqs off state.)
+
+another related change is that synchronize_irq() now takes a parameter:
+synchronize_irq(irq). This change too has the purpose of making SMP
+synchronization more lightweight - this way you can wait for your own
+interrupt handler to finish, no need to wait for other IRQ sources.
+
+
+why were these changes done? The main reason was the architectural burden
+of maintaining the cli()/sti() interface - it became a real problem. The
+new interrupt system is much more streamlined, easier to understand, debug,
+and it's also a bit faster - the same happened to it that will happen to
+cli()/sti() using drivers once they convert to spinlocks :-)
+