[PATCH 2.6.21-rc7-mm2 2/2] sata_promise: move port reset from error intr to EH prereset

[PATCH 2.6.21-rc7-mm2 2/2] sata_promise: move port reset from error intr to EH prereset

[Mikael Pettersson]

When sata_promise detects an error, it resets the port and
triggers EH. This is how it has always done things. New-style
EH however appears to prefer (based on ahci and sata_sil24)
error detection to just freeze the port or otherwise render
it harmless, and for the reset to be delayed until EH runs.

This patch updates sata_promise EH according to this principle.
When an error is detected in pdc_error_intr, the port is frozen
but not reset. Later in the error handler, the prereset procedure
performs the necessary Promise-specific reset of the port.

Notes:
- Since pdc_error_intr no longer resets the port, ata_do_eh()
  can do its error analysis on the current register contents, so
  pdc_error_intr doesn't need to copy SCR_ERROR to ehi->serror.
- The reset is done in prereset since that is the one point that
  (a) comes after libata's error analysis, and (b) dominates
  (in a control-flow sense) all subsequent reset parts of EH.

Tested on SATA300 TX4, SATA300 TX2plus, SATAII-150 TX2plus,
SATA150 TX2plus, and TX4000.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
---
 drivers/ata/sata_promise.c |   19 ++++++++++---------
 1 files changed, 10 insertions(+), 9 deletions(-)

--- linux-2.6.21-rc7-mm2/drivers/ata/sata_promise.c.~1~	2007-04-28 23:16:57.000000000 +0200
+++ linux-2.6.21-rc7-mm2/drivers/ata/sata_promise.c	2007-04-28 23:55:49.000000000 +0200
@@ -45,7 +45,7 @@
 #include "sata_promise.h"
 
 #define DRV_NAME	"sata_promise"
-#define DRV_VERSION	"2.06"
+#define DRV_VERSION	"2.07"
 
 
 enum {
@@ -598,13 +598,16 @@ static void pdc_thaw(struct ata_port *ap
 	readl(mmio + PDC_CTLSTAT); /* flush */
 }
 
-static void pdc_common_error_handler(struct ata_port *ap, ata_reset_fn_t hardreset)
+static int pdc_prereset(struct ata_port *ap, unsigned long deadline)
 {
-	if (!(ap->pflags & ATA_PFLAG_FROZEN))
-		pdc_reset_port(ap);
+	pdc_reset_port(ap);
+	return ata_std_prereset(ap, deadline);
+}
 
+static void pdc_common_error_handler(struct ata_port *ap, ata_reset_fn_t hardreset)
+{
 	/* perform recovery */
-	ata_do_eh(ap, ata_std_prereset, ata_std_softreset, hardreset,
+	ata_do_eh(ap, pdc_prereset, ata_std_softreset, hardreset,
 		  ata_std_postreset);
 }
 
@@ -647,12 +650,10 @@ static void pdc_error_intr(struct ata_po
 			   | PDC_PCI_SYS_ERR | PDC1_PCI_PARITY_ERR))
 		ac_err_mask |= AC_ERR_HOST_BUS;
 
-	if (sata_scr_valid(ap))
-		ehi->serror |= pdc_sata_scr_read(ap, SCR_ERROR);
-
+	ehi->action |= ATA_EH_SOFTRESET;
 	qc->err_mask |= ac_err_mask;
 
-	pdc_reset_port(ap);
+	ata_port_freeze(ap);
 }
 
 static inline unsigned int pdc_host_intr( struct ata_port *ap,

Re: [PATCH 2.6.21-rc7-mm2 2/2] sata_promise: move port reset from error intr to EH prereset

[Tejun Heo]

Hello,

Mikael Pettersson wrote:
> @@ -647,12 +650,10 @@ static void pdc_error_intr(struct ata_po
>  			   | PDC_PCI_SYS_ERR | PDC1_PCI_PARITY_ERR))
>  		ac_err_mask |= AC_ERR_HOST_BUS;
>  
> -	if (sata_scr_valid(ap))
> -		ehi->serror |= pdc_sata_scr_read(ap, SCR_ERROR);
> -
> +	ehi->action |= ATA_EH_SOFTRESET;
>  	qc->err_mask |= ac_err_mask;
>  
> -	pdc_reset_port(ap);
> +	ata_port_freeze(ap);
>  }

You really don't wanna schedule ATA_EH_SOFTRESET and freeze the port on
every error.  If you do that, ATAPI devices will be reset after each
CHECK CONDITION and it probably won't work properly.  FWIW, libata EH
automatically issues softreset if the port is frozen (as that's the only
way to thaw the port), so setting EH_SOFTRESET is optional if you freeze
the port.

Another problem is that EH may issue other commands to while trying to
recover - e.g. PACKET - REQUEST SENSE or READ LOG PAGE.  This only
happens if the port is ready for commands, IOW, !frozen.  Before, the
pdc_reset_port() used to be called on entry to EH if it's not frozen but
it's not after this patch.  Is this safe?

I think adding a call to pdc_reset_port() to prereset() and freezing the
port if the port is in weird state should do it but I dunno much about
the controller.

Thanks.

-- 
tejun

Re: [PATCH 2.6.21-rc7-mm2 2/2] sata_promise: move port reset from error intr to EH prereset

[Mikael Pettersson]

On Mon, 30 Apr 2007 11:27:46 +0200, Tejun Heo wrote:
>Mikael Pettersson wrote:
>> @@ -647,12 +650,10 @@ static void pdc_error_intr(struct ata_po
>>  			   | PDC_PCI_SYS_ERR | PDC1_PCI_PARITY_ERR))
>>  		ac_err_mask |= AC_ERR_HOST_BUS;
>>  
>> -	if (sata_scr_valid(ap))
>> -		ehi->serror |= pdc_sata_scr_read(ap, SCR_ERROR);
>> -
>> +	ehi->action |= ATA_EH_SOFTRESET;
>>  	qc->err_mask |= ac_err_mask;
>>  
>> -	pdc_reset_port(ap);
>> +	ata_port_freeze(ap);
>>  }
>
>You really don't wanna schedule ATA_EH_SOFTRESET and freeze the port on
>every error.  If you do that, ATAPI devices will be reset after each
>CHECK CONDITION and it probably won't work properly.

Would that be hal polling to see if media has been inserted?

>  FWIW, libata EH
>automatically issues softreset if the port is frozen (as that's the only
>way to thaw the port), so setting EH_SOFTRESET is optional if you freeze
>the port.
>
>Another problem is that EH may issue other commands to while trying to
>recover - e.g. PACKET - REQUEST SENSE or READ LOG PAGE.  This only
>happens if the port is ready for commands, IOW, !frozen.  Before, the
>pdc_reset_port() used to be called on entry to EH if it's not frozen but
>it's not after this patch.  Is this safe?

Oh dear. I assumed EH would only want to inspect the standard
status registers, and that's why I wanted to delay the reset.
If EH wants to talk to the device then it's a very bad idea to
not have issued a pdc_reset_port() first.

Jeff, please toss this patch in the dust bin.

/Mikael

Re: [PATCH 2.6.21-rc7-mm2 2/2] sata_promise: move port reset from error intr to EH prereset

[Tejun Heo]

Mikael Pettersson wrote:
>> You really don't wanna schedule ATA_EH_SOFTRESET and freeze the port on
>> every error.  If you do that, ATAPI devices will be reset after each
>> CHECK CONDITION and it probably won't work properly.
> 
> Would that be hal polling to see if media has been inserted?

Yes, it can be hal.  It can be anything.  Raising CHECK CONDITION and
reporting the details via sense data is part of normal operations for
ATAPI devices.

>>  FWIW, libata EH
>> automatically issues softreset if the port is frozen (as that's the only
>> way to thaw the port), so setting EH_SOFTRESET is optional if you freeze
>> the port.
>>
>> Another problem is that EH may issue other commands to while trying to
>> recover - e.g. PACKET - REQUEST SENSE or READ LOG PAGE.  This only
>> happens if the port is ready for commands, IOW, !frozen.  Before, the
>> pdc_reset_port() used to be called on entry to EH if it's not frozen but
>> it's not after this patch.  Is this safe?
> 
> Oh dear. I assumed EH would only want to inspect the standard
> status registers, and that's why I wanted to delay the reset.
> If EH wants to talk to the device then it's a very bad idea to
> not have issued a pdc_reset_port() first.

Yeap, if port reset clears some registers, they need to be saved into
appropriate fields in ehi/c or qc.  If something isn't there, please let
me know.

Thanks.

-- 
tejun