--- linux-2.6.0-test5/drivers/cdrom/sonycd535.c Tue Sep 9 14:42:52 2003
+++ linux-2.6.0-test5-fwd/drivers/cdrom/sonycd535.c Tue Sep 9 14:41:30 2003
@@ -812,14 +812,14 @@
block = req->sector;
nsect = req->nr_sectors;
- if (!(req->flags & REQ_CMD))
- continue; /* FIXME */
+ if (!blk_fs_request(req)) {
+ end_request(req, 0);
+ continue;
+ }
if (rq_data_dir(req) == WRITE) {
end_request(req, 0);
continue;
}
- if (rq_data_dir(req) != READ)
- panic("Unknown SONY CD cmd");
/*
* If the block address is invalid or the request goes beyond
* the end of the media, return an error.
@@ -896,8 +896,10 @@
}
if (readStatus == BAD_STATUS) {
/* Sleep for a while, then retry */
- current->state = TASK_INTERRUPTIBLE;
+ set_current_state(TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&sonycd535_lock);
schedule_timeout(RETRY_FOR_BAD_STATUS*HZ/10);
+ spin_lock_irq(&sonycd535_lock);
}
#if DEBUG > 0
printk(CDU535_MESSAGE_NAME
@@ -1481,7 +1483,7 @@
/* look for the CD-ROM, follows the procedure in the DOS driver */
inb(select_unit_reg);
/* wait for 40 18 Hz ticks (reverse-engineered from DOS driver) */
- current->state = TASK_INTERRUPTIBLE;
+ set_current_state(TASK_INTERRUPTIBLE);
schedule_timeout((HZ+17)*40/18);
inb(result_reg);
On Tue, Sep 09 2003, Felipe W Damasio wrote:
> Hi Jens,
>
> Jens Axboe wrote:
> >held! Ugh, and the request function do_cdu535_request calls
> >schedule_timeout with the queue lock held (that is held when that
> >function is entered), that is very buggy as well. Should also use
> >set_current_state() right above that call, not open code it (that also
> >misses a memory barrier). Same function also has problems with request
> >handling. You should kill:
> >
> > if (!(req->flags & REQ_CMD))
> > continue; /* FIXME */
> >
> >that is very broken, make that:
> >
> > if (!blk_fs_request(req)) {
> > end_request(req, 0);
> > continue;
> > }
> >
> >and kill these two lines:
> >
> > if (rq_data_dir(req) != READ)
> > panic("Unknown SONY CD cmd");
> >
> >they are screwy too.
> >
> >Care to fix the things I outlined above?
>
> This patch I think fixes all these, doesn't it?
>
> It applies on top of my latest cli-sti-removal patch that I sent you.
That needed changes too, as per my last mail. Please send one complete
patch with all the fixes in, thanks.
--
Jens Axboe
--- linux-2.6.0-test5/drivers/cdrom/sonycd535.c Tue Sep 9 17:01:08 2003
+++ linux-2.6.0-test5-fwd/drivers/cdrom/sonycd535.c Tue Sep 9 14:41:30 2003
@@ -36,6 +36,10 @@
* module_init & module_exit.
* Torben Mathiasen <[email protected]>
*
+ * September 2003 - Fix SMP support by removing cli/sti calls.
+ * Using spinlocks with a wait_queue instead.
+ * Felipe Damasio <[email protected]>
+ *
* Things to do:
* - handle errors and status better, put everything into a single word
* - use interrupts (code mostly there, but a big hole still missing)
@@ -340,10 +344,14 @@
if (sony535_irq_used <= 0) { /* poll */
yield();
} else { /* Interrupt driven */
- cli();
+ DEFINE_WAIT(wait);
+
+ spin_lock_irq(&sonycd535_lock);
enable_interrupts();
- interruptible_sleep_on(&cdu535_irq_wait);
- sti();
+ prepare_to_wait(&cdu535_irq_wait, &wait, TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&sonycd535_lock);
+ schedule();
+ finish_wait(&cdu535_irq_wait, &wait);
}
}
@@ -804,14 +812,14 @@
block = req->sector;
nsect = req->nr_sectors;
- if (!(req->flags & REQ_CMD))
- continue; /* FIXME */
+ if (!blk_fs_request(req)) {
+ end_request(req, 0);
+ continue;
+ }
if (rq_data_dir(req) == WRITE) {
end_request(req, 0);
continue;
}
- if (rq_data_dir(req) != READ)
- panic("Unknown SONY CD cmd");
/*
* If the block address is invalid or the request goes beyond
* the end of the media, return an error.
@@ -888,8 +896,10 @@
}
if (readStatus == BAD_STATUS) {
/* Sleep for a while, then retry */
- current->state = TASK_INTERRUPTIBLE;
+ set_current_state(TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&sonycd535_lock);
schedule_timeout(RETRY_FOR_BAD_STATUS*HZ/10);
+ spin_lock_irq(&sonycd535_lock);
}
#if DEBUG > 0
printk(CDU535_MESSAGE_NAME
@@ -1473,7 +1483,7 @@
/* look for the CD-ROM, follows the procedure in the DOS driver */
inb(select_unit_reg);
/* wait for 40 18 Hz ticks (reverse-engineered from DOS driver) */
- current->state = TASK_INTERRUPTIBLE;
+ set_current_state(TASK_INTERRUPTIBLE);
schedule_timeout((HZ+17)*40/18);
inb(result_reg);
This cli-sti removal seems exactly as broken as all the ones i've NAKed on
kernel-janitors. There's no evidence that I can see for locking in the
interrupt handler. Here's a race example:
CPU 1 CPU 2
sony_sleep();
spin_lock_irq(&sonycd535_lock);
enable_interrupts();
cdu535_interrupt();
disable_interrupts();
if (waitqueue_active(&cdu535_irq_wait)) {
prepare_to_wait(&cdu535_irq_wait, &wait, TASK_INTERRUPTIBLE);
spin_unlock_irq(&sonycd535_lock);
Either you need to move prepare_to_wait before enable_interrupts or
grab the sonycd535_lock in interrupt context.
Hang on a minute. This driver is always polled, and never interrupt
driven. There's no problem because this code path is never executed :-P
Nevertheless, it's probaby worth fixing so other more complex drivers
(eg cdu31a) don't copy it wrongly.
BTW, I bet sony_sleep() shouldn't be calling the new-and-improved yield().
--
"It's not Hollywood. War is real, war is primarily not about defeat or
victory, it is about death. I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk