Skip to contentSkip to conversation

pdc202xx_old: fix resetproc() method

PATCH Applied Sergei Shtylyov 3 participants 5 messages May 29, 2009 – Jun 7, 2009 libata

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pdc202xx_old: fix resetproc() method

pdc202xx_reset() calls pdc202xx_reset_host() twice, for both channels, while that function actually twiddles the single, shared software reset bit -- the net effect is a duplicated reset and horrendous 4 second delay happening not only on a channel reset but also when dma_lost_irq() and dma_clear() methods are called. Fold pdc202xx_reset_host() into pdc202xx_reset(), fix printk(), and move it before the actual reset...

---
The patch is against the ide-2.6.git master branch...

Bart, I know you have the docs... from the now removed source code I was able to figure out that the bit in question most probably drives RST- signal on both channels (it seems to require re-tuning drives on both channels which we are currently not doing). If I'm right, why the hell we're twiddling it on a normal SRST reset, and what's worse on an interrupt timeout conditions? Anyway, it's hardly a good idea to reset both channels when you only have problem with only one of them, and I don't see any justification to doing that... So maybe this patch should've actually wiped out that whole reset insanity for good?..

 drivers/ide/pdc202xx_old.c |   19 +++++--------------
 1 files changed, 5 insertions(+), 14 deletions(-)
drivers/ide/pdc202xx_old.c +5 -14
Index: ide-2.6/drivers/ide/pdc202xx_old.c
===================================================================
@@ -1,6 +1,6 @@
 /*
  *  Copyright (C) 1998-2002		Andre Hedrick <[email protected]>
- *  Copyright (C) 2006-2007		MontaVista Software, Inc.
+ *  Copyright (C) 2006-2007, 2009	MontaVista Software, Inc.
  *  Copyright (C) 2007			Bartlomiej Zolnierkiewicz
  *
  *  Portions Copyright (C) 1999 Promise Technology, Inc.
@@ -227,28 +227,19 @@ somebody_else:
 	return (dma_stat & 4) == 4;	/* return 1 if INTR asserted */
 }
 
-static void pdc202xx_reset_host (ide_hwif_t *hwif)
+static void pdc202xx_reset(ide_drive_t *drive)
 {
+	ide_hwif_t *hwif	= drive->hwif;
 	unsigned long high_16	= hwif->extra_base - 16;
 	u8 udma_speed_flag	= inb(high_16 | 0x001f);
 
+	printk(KERN_WARNING "PDC202xx: software reset...\n");
+
 	outb(udma_speed_flag | 0x10, high_16 | 0x001f);
 	mdelay(100);
 	outb(udma_speed_flag & ~0x10, high_16 | 0x001f);
 	mdelay(2000);	/* 2 seconds ?! */
 
-	printk(KERN_WARNING "PDC202XX: %s channel reset.\n",
-		hwif->channel ? "Secondary" : "Primary");
-}
-
-static void pdc202xx_reset (ide_drive_t *drive)
-{
-	ide_hwif_t *hwif	= drive->hwif;
-	ide_hwif_t *mate	= hwif->mate;
-
-	pdc202xx_reset_host(hwif);
-	pdc202xx_reset_host(mate);
-
 	ide_set_max_pio(drive);
 }
 
#1 lore raw
Quoted 2 lines · On Sat, 30 May 2009 00:07:12 +0400 Sergei Shtylyov <[email protected]>…

pdc202xx_reset() calls pdc202xx_reset_host() twice, for both channels, while that function actually twiddles the single, shared software reset bit

Including while the other channel is active it appears - that doesn't look good if so

#2 lore raw
In reply to Sergei Shtylyov
Quoted 9 lines · On Friday 29 May 2009 22:07:12 Sergei Shtylyov wrote:

pdc202xx_reset() calls pdc202xx_reset_host() twice, for both channels, while that function actually twiddles the single, shared software reset bit -- the net effect is a duplicated reset and horrendous 4 second delay happening not only on a channel reset but also when dma_lost_irq() and dma_clear() methods are called. Fold pdc202xx_reset_host() into pdc202xx_reset(), fix printk(), and move it before the actual reset...

Quoted 8 lines

---
The patch is against the ide-2.6.git master branch...

Bart, I know you have the docs... from the now removed source code I was able to figure out that the bit in question most probably drives RST- signal on both channels (it seems to require re-tuning drives on both channels which we are currently not doing). If I'm right, why the hell we're twiddling it on a normal SRST reset, and what's worse on an interrupt timeout conditions? Anyway, it's

Unfortunately the documentation doesn't say anything more than that the changing of the bit value results in SRST on both channels...

IIRC from some old discussions this is required by the chipset sometimes..

Quoted 3 lines

hardly a good idea to reset both channels when you only have problem with only one of them, and I don't see any justification to doing that... So maybe this patch should've actually wiped out that whole reset insanity for good?..

I'm pretty sure that there are valid reasons behind some of this insanity, OTOH (per mail from Alan) it completely ignores the fact that the other port may be active..

We may try removing it in incremental patch so it is easier to track/revert it if things go wrong..

#3 lore raw

Hello.

Quoted 12 lines · Bartlomiej Zolnierkiewicz
Quoted 11 lines

pdc202xx_reset() calls pdc202xx_reset_host() twice, for both channels, while that function actually twiddles the single, shared software reset bit -- the net effect is a duplicated reset and horrendous 4 second delay happening not only on a channel reset but also when dma_lost_irq() and dma_clear() methods are called. Fold pdc202xx_reset_host() into pdc202xx_reset(), fix printk(), and move it before the actual reset...

---
The patch is against the ide-2.6.git master branch...

Quoted 8 lines
Quoted 5 lines

Bart, I know you have the docs... from the now removed source code I was able to figure out that the bit in question most probably drives RST- signal on both channels (it seems to require re-tuning drives on both channels which we are currently not doing). If I'm right, why the hell we're twiddling it on a normal SRST reset, and what's worse on an interrupt timeout conditions? Anyway, it's

Unfortunately the documentation doesn't say anything more than that the changing of the bit value results in SRST on both channels...

I find it somewhat hard to believe that this bit controls SRST bits in
the device control registers -- we have them both under our control anyway, why would we need such a "shortcut". I'd rather believe that this bit controls RESET- signal on both channels, with the neigboring bit 3 setting it to tristate mode the same way as it works on HPT37x (where BTW the datasheets call RESET- "SRST_") -- all this probably having to do with the hotswap support (busproc() method that manipulated bit 3 was removed from that driver by me 3 years ago)...

Quoted 1 line

IIRC from some old discussions this is required by the chipset sometimes..

If that bit indeed controls SRST bits, I don't see how setting it in
resetproc() helps (it can only harm, unwittingly resetting another channel). If this bit controls RESET-, I again don't see how/why it helps to hard-reset the channels in addition to the soft reset just performed. When pdc202xx_reset() is called as the dma_clear() method, the channel is probably going be reset soon due to BSY=1, and when pdc202xx_reset() is called from dma_lost_irq() method, it seems a clear overkill...

Quoted 10 lines
Quoted 3 lines

hardly a good idea to reset both channels when you only have problem with only one of them, and I don't see any justification to doing that... So maybe this patch should've actually wiped out that whole reset insanity for good?..

I'm pretty sure that there are valid reasons behind some of this insanity, OTOH (per mail from Alan) it completely ignores the fact that the other port may be active..

We may try removing it in incremental patch so it is easier to track/revert it if things go wrong..

OK, sounds like a plan...

MBR, Sergei

#4 lore raw
In reply to Sergei Shtylyov
Quoted 8 lines · On Friday 29 May 2009 22:07:12 Sergei Shtylyov wrote:

pdc202xx_reset() calls pdc202xx_reset_host() twice, for both channels, while that function actually twiddles the single, shared software reset bit -- the net effect is a duplicated reset and horrendous 4 second delay happening not only on a channel reset but also when dma_lost_irq() and dma_clear() methods are called. Fold pdc202xx_reset_host() into pdc202xx_reset(), fix printk(), and move it before the actual reset...

applied