Date:	Wed, 25 Nov 2015 14:17:09 +1100 (AEDT)
From:	Finn Thain <fthain@...egraphics.com.au>
To:	Ondrej Zary <linux@...nbow-software.org>
cc:	"James E.J. Bottomley" <JBottomley@...n.com>,
	Michael Schmitz <schmitzmic@...il.com>,
	linux-m68k@...r.kernel.org, linux-scsi@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH 22/71] ncr5380: Eliminate selecting state


On Tue, 24 Nov 2015, Ondrej Zary wrote:

> On Wednesday 18 November 2015 09:35:17 Finn Thain wrote:
> > Linux v2.1.105 changed the algorithm for polling for the BSY signal
> > in NCR5380_select() and NCR5380_main().
> > 
> > Presently, this code has a bug. Back then, NCR5380_set_timer(hostdata, 1)
> > meant reschedule main() after sleeping for 10 ms. Repeated 25 times this
> > provided the recommended 250 ms selection time-out delay. This got broken
> > when HZ became configurable.
> > 
> > We could fix this but there's no need to reschedule the main loop. This
> > BSY polling presently happens when the NCR5380_main() work queue item
> > calls NCR5380_select(), which in turn schedules NCR5380_main(), which
> > calls NCR5380_select() again, and so on.
> > 
> > This algorithm is a deviation from the simpler one in atari_NCR5380.c.
> > The extra complexity and state is pointless. There's no reason to
> > stop selection half-way and return to to the main loop when the main
> > loop can do nothing useful until selection completes.
> > 
> > So just poll for BSY. We can sleep while polling now that we have a
> > suitable workqueue.
> 
> Bisecting slow module initialization pointed to this commit.

That's disappointing. This patch removed some nasty code. Anyway, thanks 
for taking the trouble to bisect.

> 
> Before this commit (2 seconds):
> [   60.317374] scsi host2: Generic NCR5380/NCR53C400 SCSI, io_port 0x0, n_io_port 0, base 0xd8000, irq 0, can_queue 16, cmd_per_lun 2, sg_tablesize 128, this_id 7, flags { NCR53C400 }, USLEEP_POLL 3, USLEEP_SLEEP 50, options { AUTOPROBE_IRQ PSEUDO_DMA }
> [   60.780715] scsi 2:0:1:0: Direct-Access     QUANTUM  LP240S GM240S01X 4.6  PQ: 0 ANSI: 2 CCS
> [   62.606260] sd 2:0:1:0: Attached scsi generic sg1 type 0
> 
> 
> After this commit (22 seconds):
> [  137.511711] scsi host2: Generic NCR5380/NCR53C400 SCSI, io_port 0x0, n_io_port 0, base 0xd8000, irq 0, can_queue 16, cmd_per_lun 2, sg_tablesize 128, this_id 7, flags { NCR53C400 }, USLEEP_POLL 3, USLEEP_SLEEP 50, options { AUTOPROBE_IRQ PSEUDO_DMA }
> [  145.028532] clocksource: timekeeping watchdog: Marking clocksource 'tsc' as unstable because the skew is too large:
> [  145.029767] clocksource:                       'acpi_pm' wd_now: a49738 wd_last: f4da04 mask: ffffff
> [  145.029828] clocksource:                       'tsc' cs_now: 2ea624698e cs_last: 2c710aa17f mask: ffffffffffffffff
> [  145.032733] clocksource: Switched to clocksource acpi_pm

I figured that it was okay to sleep from an unbound CPU-intensive 
workqueue but doing so seems to cause problems. (See also patch 66/71 
"Fix soft lockups".)

Perhaps a kthread is needed instead of a workqueue? (This workqueue 
already has it's own kthread, but top shows that it doesn't accrue CPU 
time.)

> [  145.236951] scsi 2:0:1:0: Direct-Access     QUANTUM  LP240S GM240S01X 4.6  PQ: 0 ANSI: 2 CCS
> [  159.959308] sd 2:0:1:0: Attached scsi generic sg1 type 0
> 
> 

This problem doesn't show up on my hardware, and I'd like to know where 
those 22 seconds are being spent. Would you please apply the entire series 
and add,
#define NDEBUG (NDEBUG_ARBITRATION | NDEBUG_SELECTION | NDEBUG_MAIN)
to the top of g_NCR5380.c and send me the messages logged during modprobe?

-- 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists