2.6.0-test1 gets corrupted data when loading init

2.6.0-test1 gets corrupted data when loading init

[Anders Gustafsson]

Hi,

with 2.6.0-test1 I get error like unresolved symbol in libs used by init or
bad elf-file. Sometimes init manages to start and there are same type of
errors in bash when it tries to run initscripts. This is on a dual xeon, no
highmem (512M), ServerWorks chipset, aic79xx scsi, / on scsi.

It breaks between 2.5.70 and 2.5.70-bk1, which contains a update in the
aic79xx-drivers, so my guess is related to that.

Scsi-related hardware in machine:

scsi0 : Adaptec AIC79XX PCI-X SCSI HBA DRIVER, Rev 1.3.0
        <Adaptec AIC7902 Ultra320 SCSI adapter>
        aic7902: Ultra320 Wide Channel A, SCSI Id=7, PCI 33 or 66Mhz, 512 SCBs

scsi1 : Adaptec AIC79XX PCI-X SCSI HBA DRIVER, Rev 1.3.0
        <Adaptec AIC7902 Ultra320 SCSI adapter>
        aic7902: Ultra320 Wide Channel B, SCSI Id=7, PCI 33 or 66Mhz, 512 SCBs

(scsi1:A:0): 320.000MB/s transfers (160.000MHz DT|IU|QAS, 16bit)
  Vendor: SEAGATE   Model: ST373307LW        Rev: 0002
  Type:   Direct-Access                      ANSI SCSI revision: 03
    

-- 
Anders Gustafsson - andersg@0x63.nu - http://0x63.nu/

Re: 2.6.0-test1 gets corrupted data when loading init

[Anders Gustafsson]

On Fri, Jul 18, 2003 at 10:34:58AM +0200, Anders Gustafsson wrote:
> It breaks between 2.5.70 and 2.5.70-bk1, which contains a update in the
> aic79xx-drivers, so my guess is related to that.

http://linux.bkbits.net:8080/linux-2.5/cset@1.1127.6.4 is the changeset that
makes it stop working.

(and if that cset-number isn't stable the comments for it are:

Aic79xx Driver Update
 o Change handling of the Rev. A packetized lun output bug
   to be more efficient by having the sequencer copy the
   single byte of valid lun data into the long lun field.

)

-- 
Anders Gustafsson - andersg@0x63.nu - http://0x63.nu/

Re: 2.6.0-test1 gets corrupted data when loading init

[Anders Gustafsson]

On Fri, Jul 18, 2003 at 11:51:08AM +0200, Anders Gustafsson wrote:
> On Fri, Jul 18, 2003 at 10:34:58AM +0200, Anders Gustafsson wrote:
> > It breaks between 2.5.70 and 2.5.70-bk1, which contains a update in the
> > aic79xx-drivers, so my guess is related to that.
> 
> http://linux.bkbits.net:8080/linux-2.5/cset@1.1127.6.4 is the changeset that
> makes it stop working.

Yeah, and reversing that on 2.6.0-test+bk with the attached patch makes it
work on 2.6.0-test1.

But I get these messages:

(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0): 160.000MB/s transfers (80.000MHz DT|IU|QAS, 16bit)
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0): 160.000MB/s transfers (80.000MHz DT, 16bit)
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error
(scsi1:A:0): 80.000MB/s transfers (40.000MHz DT, 16bit)
(scsi1:A:0:0): CDB: 0x12 0x1 0x80 0x0 0x60 0x0
(scsi1:A:0:0): Saw underflow (72 of 96 bytes). Treated as error


-- 
Anders Gustafsson - andersg@0x63.nu - http://0x63.nu/

--- a/drivers/scsi/aic7xxx/aic79xx.seq	Fri Jul 18 19:34:59 2003
+++ b/drivers/scsi/aic7xxx/aic79xx.seq	Fri Jul 18 19:34:59 2003
@@ -261,15 +261,6 @@
 	clr	A;
 	add	CMDS_PENDING, 1;
 	adc	CMDS_PENDING[1], A;
-	if ((ahd->bugs & AHD_PKT_LUN_BUG) != 0) {
-		/*
-		 * "Short Luns" are not placed into outgoing LQ
-		 * packets in the correct byte order.  Use a full
-		 * sized lun field instead and fill it with the
-		 * one byte of lun information we support.
-		 */
-		mov	SCB_PKT_LUN[6], SCB_LUN;
-	}
 	/*
 	 * The FIFO use count field is shared with the
 	 * tag set by the host so that our SCB dma engine
--- a/drivers/scsi/aic7xxx/aic79xx_core.c	Fri Jul 18 19:34:59 2003
+++ b/drivers/scsi/aic7xxx/aic79xx_core.c	Fri Jul 18 19:34:59 2003
@@ -8272,6 +8272,8 @@
 	download_consts[PKT_OVERRUN_BUFOFFSET] =
 		(ahd->overrun_buf - (uint8_t *)ahd->qoutfifo) / 256;
 	download_consts[SCB_TRANSFER_SIZE] = SCB_TRANSFER_SIZE_1BYTE_LUN;
+	if ((ahd->bugs & AHD_PKT_LUN_BUG) != 0)
+		download_consts[SCB_TRANSFER_SIZE] = SCB_TRANSFER_SIZE_FULL_LUN;
 	cur_patch = patches;
 	downloaded = 0;
 	skip_addr = 0;
--- a/drivers/scsi/aic7xxx/aic79xx_inline.h	Fri Jul 18 19:34:59 2003
+++ b/drivers/scsi/aic7xxx/aic79xx_inline.h	Fri Jul 18 19:34:59 2003
@@ -272,6 +272,10 @@
 	if ((scb->flags & SCB_PACKETIZED) != 0) {
 		/* XXX what about ACA??  It is type 4, but TAG_TYPE == 0x3. */
 		scb->hscb->task_attribute = scb->hscb->control & SCB_TAG_TYPE;
+		/*
+		 * For Rev A short lun workaround.
+		 */
+		scb->hscb->pkt_long_lun[6] = scb->hscb->lun;
 	} else {
 		if (ahd_get_transfer_length(scb) & 0x01)
 			scb->hscb->task_attribute = SCB_XFERLEN_ODD;

Re: 2.6.0-test1 gets corrupted data when loading init

[Justin T. Gibbs]

> On Fri, Jul 18, 2003 at 11:51:08AM +0200, Anders Gustafsson wrote:
>> On Fri, Jul 18, 2003 at 10:34:58AM +0200, Anders Gustafsson wrote:
>> > It breaks between 2.5.70 and 2.5.70-bk1, which contains a update in the
>> > aic79xx-drivers, so my guess is related to that.
>> 
>> http://linux.bkbits.net:8080/linux-2.5/cset@1.1127.6.4 is the changeset that
>> makes it stop working.
> 
> Yeah, and reversing that on 2.6.0-test+bk with the attached patch makes it
> work on 2.6.0-test1.

There are a whole slew of later changesets that haven't made it in yet.
The root cause of your particular problem is not the lun copy optimization,
but a problem with the layout of a data structure that is dma'ed to the
controller and a controller errata.  The fix for this is available in 
the 20030603 bksend file at my site:

http://people.FreeBSD.org/~gibbs/linux/SRC/

I will try to find some time later this week to review the code that
is now in 2.6 and generate updated changesets for that branch.

--
Justin

Re: 2.6.0-test1 gets corrupted data when loading init

[Anders Gustafsson]

On Tue, Jul 22, 2003 at 03:46:31PM -0600, Justin T. Gibbs wrote:
> There are a whole slew of later changesets that haven't made it in yet.
> The root cause of your particular problem is not the lun copy optimization,
> but a problem with the layout of a data structure that is dma'ed to the
> controller and a controller errata.  The fix for this is available in 
> the 20030603 bksend file at my site:
> 
> http://people.FreeBSD.org/~gibbs/linux/SRC/

Yes, thanks, that works just fine.

-- 
Anders Gustafsson - andersg@0x63.nu - http://0x63.nu/