Bogus interrupts: test fix

Bogus interrupts: test fix

[Benjamin Herrenschmidt]

I've updated the 2.2-pre4 test kernel on my web page, it contains a test
fix for the bogus interrupt problem. Please try it and give me feedback.
Also, do not worry about the "lost interrupts" counter I added to
/proc/interrupts, they are not really "lost" but I would be interested in
knowing how much of them you get with ordinary use of the system.

The "fix" looks like this:

void __pmac pmac_mask_and_ack_irq(int irq_nr)
{
	unsigned long bit = 1UL << (irq_nr & 0x1f);
	int i = irq_nr >> 5;

	if ((unsigned)irq_nr >= max_irqs)
		return;
	/*spin_lock(&irq_controller_lock);*/

	clear_bit(irq_nr, cached_irq_mask);
	if (test_and_clear_bit(irq_nr, lost_interrupts))
		atomic_dec(&n_lost_interrupts);
	out_le32(&pmac_irq_hw[i]->ack, bit);
	out_le32(&pmac_irq_hw[i]->enable, cached_irq_mask[i]);
	out_le32(&pmac_irq_hw[i]->ack, bit);
	do {
		/* make sure ack gets to controller before we enable interrupts */
		sync();
	} while(in_le32(&pmac_irq_hw[i]->flag) & bit);

	/*spin_unlock(&irq_controller_lock);*/
	/*if ( irq_controller_lock.lock )
	  panic("irq controller lock still held in mask and ack\n");*/
}

Note that the sync() may now be useless. I don't have anymore bogus
interrupts on the PowerBook for now but I usually have very few of them
anyway. I'm more interested by results from people with a real problem
(io slowdown due to lots of them).

I'll post this kernel on an ftp site later today too.

-- 
           E-Mail: <mailto:bh40@calva.net>
BenH.      Web   : <http://calvaweb.calvacom.fr/bh40/>





[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: Bogus interrupts: test fix

[Albrecht Dreß]

Benjamin Herrenschmidt wrote:
> 
> I've updated the 2.2-pre4 test kernel on my web page, it contains a test
> fix for the bogus interrupt problem. Please try it and give me feedback.

Really good! The improvement of ide performance is impressive. The slowdown
effect has gone during ide _reads_ [= almost "normal" response times compared to
Intel/Linux systems], but is still present during ide _writes_ (I tested it with
a program which writes or reads 200 MB with simple read/write calls, so I have
enough time for a login on a 2nd console). BTW, the read transfer speed given by
hdparm (PB G3/300, IBM DYLA 28100) is 7.08 MB/s.

The bogus interrupts are gone _completely_; still 0 after ~1 hour with a lot of
io. LST interrupts are == 1 (one) after the same time.

Yours, and THANKS for your work,

	Albrecht.

-- 
+-----------------------------------------------------------------------------+
| Dr.-Ing. Albrecht Dre\3                                      ----           |
| Max-Planck-Institut f"ur Radioastronomie    |\       /      /o  o\          |
| Technische Abt. Optische Interferometrie    |  \    /      |  /   |         |
| Auf dem H"ugel 69                           |    \ |        \ ---/          |
| D-53121 Bonn (Germany)          ------------+------+-------------------     |
|                                             |    / |                        |
| Phone (+49) 228 525 319                     |  /  /                         |
| Fax   (+49) 228 525 411                     |/   /                          |
| Mail  ad@mpifr-bonn.mpg.de                                                  |
+-------------- electrical engineers do it with less resistance --------------+

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]