ide-scsi emulation hangs the kernel?

ide-scsi emulation hangs the kernel?

[Yuri Prushinsky]

Hey all! Could anybody please tell me what's wrong with the ide-scsi
emulation in 2.4.21? I have a cd-rw drive as a hdd, it works fine but
the system (redhat9) suddenly freezes at any time with the following
messages:


Dec  8 01:11:14 localhost kernel: scsi : aborting command due to timeout : pid 22251, scsi0, channel 0, id 0, lun 0 Request
Sense 00 00 00 40 00
Dec  8 01:11:14 localhost kernel: hdd: lost interrupt
Dec  8 01:11:14 localhost kernel: ide-scsi: CoD != 0 in idescsi_pc_intr
Dec  8 01:11:14 localhost kernel: hdd: DMA disabled
Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete
Dec  8 01:11:14 localhost kernel: hdd: irq timeout: status=0x80 { Busy }
Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete
Dec  8 01:11:14 localhost kernel: hdd: irq timeout: status=0x80 { Busy }
Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete

  

-- 
Best regards,
 Yura                          mailto:prushinsky@hotbox.ru

ide-scsi emulation hangs the kernel?

[Pat LaVarre]

> Could anybody please tell me what's wrong with
> the ide-scsi emulation in 2.4.21?

This question I'm too ignorant to answer.

Google linux-scsi tells me lk 2.6 deprecates ide-scsi in favour of
ide-cd, not that lk 2.4 drops support for ide-scsi.

> I have a cd-rw drive as a hdd, it works fine
> but the system (redhat9) suddenly freezes at
> any time with the following messages:

I'm writing because these messages do make some kind of sense to Me,
after having studied some of http://www.t13.org/ over the years.

Below you can see I conclude: "the host gave up entirely, declaring this
device broken".  Without more info I cannot know whether we should try
to fix the host and/or device and/or disc.

Please understand, I know nearly zero of lk, just enough of t13.org
ATA/PI to guess what these messages mean without being sure my guesses
are correct.

---

SCSI, including SCSI over IDE also known as ATAPI, works by the host and
the device taking turns to speak.

> Dec  8 01:11:14 localhost kernel: scsi : aborting command due to timeout : pid 22251, scsi0, channel 0, id 0, lun 0 Request Sense 00 00 00 40 00

"Timeout" often means the host lost patience while waiting for the
device to yield its turn to speak.

> Dec  8 01:11:14 localhost kernel: hdd: lost interrupt

Linux "hdd" often means the ID 1 "slave" device of the "secondary" IDE
bus.

Often the way an ATA/PI device yields its turn to speak is to interrupt
the host.  "Lost interrupt" could be a restatement of the "timeout"
idea, that 

> Dec  8 01:11:14 localhost kernel: ide-scsi: CoD != 0 in idescsi_pc_intr

ATA/PI sequences thru four states: x01 Command Out, x00 Data Out, x02
Data In, x03 Status In.

"CoD" may refer to that x 01 00 02 03 encoding.

> Dec  8 01:11:14 localhost kernel: hdd: DMA disabled

The port x1F2 SectorCount/InterruptReason mask x03
+Command/-Data:+In/-Out bits have no meaning unless DMA is disabled (in
favour of PIO).  The so-called "port x1F2" traditionally appeared as x86
i/o port address x1F2, also x172, etc.

> Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete
> Dec  8 01:11:14 localhost kernel: hdd: irq timeout: status=0x80 { Busy }

Automagic reset may be how this host choose to recover from its own loss
of patience.

In IDE PIO, ports x1F7/3F6 Status/AlternateStatus mask x88 BSY:DRQ
simulate a shared memory semaphore, in which the value x00 means host
talks, x08 means host copies data, and either of x80 or x88 means the
device talks and the host has to wait.  Here we have x80 i.e. host
waiting, again consistent with the theory that our host lost patience.

> Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete
> Dec  8 01:11:14 localhost kernel: hdd: irq timeout: status=0x80 { Busy }

This looks like a retry of a reset - this host is more passionately
losing patience.

> Dec  8 01:11:14 localhost kernel: hdd: ATAPI reset complete

This looks like the host gave up entirely, declaring this device broken.

---

Pat LaVarre

Re: ide-scsi emulation hangs the kernel?

[Philip R. Auld]

Hi,

Rumor has it that on Wed, Dec 10, 2003 at 04:36:26PM -0700 Pat LaVarre said:
> > Could anybody please tell me what's wrong with
> > the ide-scsi emulation in 2.4.21?
> 

I didn't see the original email, but have had this problem in the past. 
I hit this with a TOSHIBA DVD-ROM SD-C2612 and serverworks CSB5 IDE chip.

There is no timeout handling in ide-scsi. And once it times out the 
command gets re-issued with a timeout that can go off immediately. This 
patch fixes the problem on my system. It's for 2.4.9 but from the sound 
of it should be applicable to later kernels. Your milage may vary.

Cheers,

Phil


diff -ruN -X ../patches/dontdiff linux-clean/drivers/scsi/ide-scsi.c linux-cblade/drivers/scsi/ide-scsi.c
--- linux-clean/drivers/scsi/ide-scsi.c	Fri Aug 29 10:15:16 2003
+++ linux-cblade/drivers/scsi/ide-scsi.c	Fri Aug 29 10:23:25 2003
@@ -77,6 +77,7 @@
 #define PC_DMA_IN_PROGRESS		0	/* 1 while DMA in progress */
 #define PC_WRITING			1	/* Data direction */
 #define PC_TRANSFORM			2	/* transform SCSI commands */
+#define PC_TIMEDOUT                    3       /* command timed out */
 
 /*
  *	SCSI command transformation layer
@@ -113,6 +114,11 @@
 #define IDESCSI_IREASON_COD	0x1		/* Information transferred is command */
 #define IDESCSI_IREASON_IO	0x2		/* The device requests us to read */
 
+/* 
+ *      Timeout less than given scsi cmd timeout to use.
+ */
+#define IDESCSI_TIMEOUT_DELTA     (2 * HZ)
+
 static void idescsi_discard_data (ide_drive_t *drive, unsigned int bcount)
 {
 	while (bcount--)
@@ -279,7 +285,12 @@
 		pc->scsi_cmd->result = (CHECK_CONDITION << 1) | (DID_OK << 16);
 		if (log)
 			printk ("ide-scsi: %s: check condition for %lu\n", drive->name, pc->scsi_cmd->serial_number);
-	} else {
+	} else  if (test_bit(PC_TIMEDOUT, &pc->flags)){
+		pc->scsi_cmd->result = DID_TIME_OUT << 16;
+		if (log)
+			printk ("ide-scsi: %s: timed out  for %lu\n", drive->name, pc->scsi_cmd->serial_number);
+	}
+	else {
 		pc->scsi_cmd->result = DID_OK << 16;
 		idescsi_transform_pc2 (drive, pc);
 		if (log) {
@@ -301,7 +312,21 @@
 
 static inline unsigned long get_timeout(idescsi_pc_t *pc)
 {
-	return IDE_MAX(WAIT_CMD, pc->timeout - jiffies);
+	return IDE_MAX(WAIT_CMD, (signed long) (pc->timeout -jiffies));
+}
+
+/* timeout handler. Mark command as having timed out. 
+ */
+int idescsi_expiry (ide_drive_t *drive){
+	idescsi_scsi_t *scsi = drive->driver_data;
+	idescsi_pc_t *pc=scsi->pc;
+
+#if IDESCSI_DEBUG_LOG
+	printk(KERN_INFO "idescsi_expiry called for %lu at %lu\n", pc->scsi_cmd->serial_number, jiffies);
+#endif
+	set_bit(PC_TIMEDOUT, &pc->flags);
+
+	return 0;
 }
 
 /*
@@ -320,6 +345,15 @@
 	printk (KERN_INFO "ide-scsi: Reached idescsi_pc_intr interrupt handler\n");
 #endif /* IDESCSI_DEBUG_LOG */
 
+	if (test_bit(PC_TIMEDOUT, &pc->flags)){
+#if IDESCSI_DEBUG_LOG
+		printk("idescsi_pc_intr: got timed out packet  %lu at %lu\n", pc->scsi_cmd->serial_number, jiffies);
+#endif
+		/* end this request now - scsi should retry it*/
+		idescsi_end_request (0, HWGROUP(drive));
+		return ide_do_reset (drive);
+	}
+
 	if (test_and_clear_bit (PC_DMA_IN_PROGRESS, &pc->flags)) {
 #if IDESCSI_DEBUG_LOG
 		printk ("ide-scsi: %s: DMA complete\n", drive->name);
@@ -363,7 +397,7 @@
 				pc->actually_transferred += temp;
 				pc->current_position += temp;
 				idescsi_discard_data (drive,bcount - temp);
-				ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), NULL);
+				ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), idescsi_expiry);
 				return ide_started;
 			}
 #if IDESCSI_DEBUG_LOG
@@ -387,7 +421,7 @@
 	pc->actually_transferred+=bcount;				/* Update the current position */
 	pc->current_position+=bcount;
 
-	ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), NULL);	/* And set the interrupt handler again */
+	ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), idescsi_expiry);	/* And set the interrupt handler again */
 	return ide_started;
 }
 
@@ -407,7 +441,7 @@
 		printk (KERN_ERR "ide-scsi: (IO,CoD) != (0,1) while issuing a packet command\n");
 		return ide_do_reset (drive);
 	}
-	ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), NULL);	/* Set the interrupt routine */
+	ide_set_handler(drive, &idescsi_pc_intr, get_timeout(pc), idescsi_expiry);	/* Set the interrupt routine */
 	atapi_output_bytes (drive, scsi->pc->c, 12);			/* Send the actual packet */
 	return ide_started;
 }
@@ -442,7 +476,7 @@
 		(void) (HWIF(drive)->dmaproc(ide_dma_begin, drive));
 	}
 	if (test_bit (IDESCSI_DRQ_INTERRUPT, &scsi->flags)) {
-		ide_set_handler (drive, &idescsi_transfer_pc, get_timeout(pc), NULL);
+		ide_set_handler (drive, &idescsi_transfer_pc, get_timeout(pc), idescsi_expiry);
 		OUT_BYTE (WIN_PACKETCMD, IDE_COMMAND_REG);		/* Issue the packet command */
 		return ide_started;
 	} else {
@@ -776,7 +810,7 @@
 	pc->request_transfer = pc->buffer_size = cmd->request_bufflen;
 	pc->scsi_cmd = cmd;
 	pc->done = done;
-	pc->timeout = jiffies + cmd->timeout_per_command;
+	pc->timeout = jiffies + cmd->timeout_per_command - IDESCSI_TIMEOUT_DELTA;
 
 	if (should_transform(drive, cmd))
 		set_bit(PC_TRANSFORM, &pc->flags);


-- 
Philip R. Auld, Ph.D.  	        	       Egenera, Inc.    
Principal Software Engineer                   165 Forest St.
(508) 858-2628                            Marlboro, MA 01752

Re: ide-scsi emulation hangs the kernel?

[Pat LaVarre]

> > > Could anybody please tell me what's wrong with
> > > the ide-scsi emulation in 2.4.21?
> ...
> no timeout handling in ide-scsi. And once it
> times out the command gets re-issued with a
> timeout that can go off immediately. This patch
> fixes the problem on my system. It's for 2.4.9 ...

Patch!  Thank you.

> I didn't see the original email, ...

http://marc.theaimsgroup.com/?t=107100962500008
begins with:

http://marc.theaimsgroup.com/?l=linux-scsi&m=107100959521497
List:     linux-scsi
Subject:  ide-scsi emulation hangs the kernel?
From:     Yuri Prushinsky <prushinsky () hotbox ! ru>
Date:     2003-12-09 22:36:44

Pat LaVarre