g_NCR5380 - 2.6.0 - problem with reloading module

g_NCR5380 - 2.6.0 - problem with reloading module

[Kevin Tappe]

Hi,

i hope this is the right emailadress for my problem...

The first time I load the module g_NCR5380 everything works fine:
scsi0 : interrupts not enabled. for better interactive performance,
scsi0 : please jumper the board for a free IRQ.
scsi0 : at port 0x280 interrupts disabled options CAN_QUEUE=16  
	CMD_PER_LUN=2 release=1 generic options
	AUTOPROBE_IRQ AUTOSENSE USLEEP, USLEEP_POLL=1 USLEEP_SLEEP=20 
	generic release=7
scsi0 : Generic NCR5380/53C400 Driver
	   Vendor: UMAX      Model: Astra 1200S
		Rev: V2.9   Type:   Scanner ANSI SCSI revision: 02
Attached scsi generic sg0 at scsi0, channel 0, id 5, lun 0,  type 6

I can't reload the module g_NCR5380 after an rmmod:

nihil:~# modprobe -v g_NCR5380
insmod /lib/modules/2.6.0/kernel/drivers/scsi/g_NCR5380.ko ncr_irq=255
ncr_addr=0x280 dtc_3181e=1
FATAL: Error inserting g_NCR5380
(/lib/modules/2.6.0/kernel/drivers/scsi/g_NCR53
80.ko): No such device

After a reboot everything is just fine again.
I use a vanilla 2.6.0 kernel. Please let me know if you need any further
information.

-Kevin
-- 
Think how much fear is going to be required to stop people from passing
along copies of things on their computers. I hope you don't want to live
in a world with that much fear. - Richard M. Stallman

Re: g_NCR5380 - 2.6.0 - problem with reloading module

[Randy.Dunlap]

On Mon, 5 Jan 2004 23:42:20 +0100 Kevin Tappe <kevin@zerodivide.de> wrote:

| Hi,
| 
| i hope this is the right emailadress for my problem...
| 
| The first time I load the module g_NCR5380 everything works fine:
| scsi0 : interrupts not enabled. for better interactive performance,
| scsi0 : please jumper the board for a free IRQ.
| scsi0 : at port 0x280 interrupts disabled options CAN_QUEUE=16  
| 	CMD_PER_LUN=2 release=1 generic options
| 	AUTOPROBE_IRQ AUTOSENSE USLEEP, USLEEP_POLL=1 USLEEP_SLEEP=20 
| 	generic release=7
| scsi0 : Generic NCR5380/53C400 Driver
| 	   Vendor: UMAX      Model: Astra 1200S
| 		Rev: V2.9   Type:   Scanner ANSI SCSI revision: 02
| Attached scsi generic sg0 at scsi0, channel 0, id 5, lun 0,  type 6
| 
| I can't reload the module g_NCR5380 after an rmmod:
| 
| nihil:~# modprobe -v g_NCR5380
| insmod /lib/modules/2.6.0/kernel/drivers/scsi/g_NCR5380.ko ncr_irq=255
| ncr_addr=0x280 dtc_3181e=1
| FATAL: Error inserting g_NCR5380
| (/lib/modules/2.6.0/kernel/drivers/scsi/g_NCR53
| 80.ko): No such device
| 
| After a reboot everything is just fine again.
| I use a vanilla 2.6.0 kernel. Please let me know if you need any further
| information.

Hi,

Can you show that 'modprobe -v g_NCR5380' output for the first time
that the driver is loaded also?

What's in /proc/scsi/g_NCR5380 while the driver is loaded the first time?
and in /proc/ioports and /proc/iomem?
Also, the latter 2 files after the driver is removed.
and in /proc/scsi after the driver is removed.

--
~Randy
MOTD:  Always include version info.

[PATCH] g_NCR5380 - 2.6.0 - problem with reloading module

[Randy.Dunlap]

[-- Attachment #1: Type: text/plain, Size: 2732 bytes --]

On Sat, 10 Jan 2004 20:38:15 +0100 Kevin Tappe <kevin@zerodivide.de> wrote:

| Hi,
| 
| On Fri, Jan 09, 2004 at 11:26:56AM -0800, Randy.Dunlap wrote:
| > Can you show that 'modprobe -v g_NCR5380' output for the first time
| > that the driver is loaded also?
| 
| nihil:~# modprobe -v g_NCR5380
| insmod /lib/modules/2.6.0/kernel/drivers/scsi/g_NCR5380.ko ncr_irq=255
| ncr_addr=0x280 dtc_3181e=1

no output from the driver??
That's OK though, I found the problem.

| > What's in /proc/scsi/g_NCR5380 while the driver is loaded the first time?
| > and in /proc/ioports and /proc/iomem?
| 
| nihil:~# cat /proc/scsi/g_NCR5380/0
| SCSI host number 0 : Generic NCR5380/NCR53C400 Scsi Driver
| Generic NCR5380 driver version 1
| NCR5380 core version 7
| NO NCR53C400 driver extensions
| Using port mapping at port 0x280, no interrupt
| REQ not asserted, phase unknown.
| No currently connected command
| issue_queue
| disconnected_queue
| 
| nihil:~# cat /proc/ioports
...
| 0213-0213 : ISAPnP
| 0220-022f : SoundBlaster
| 0260-0261 : pnp 00:11
| 0280-028f : ncr53c80
| 02f8-02ff : serial
...
| 0a79-0a79 : isapnp write
...
| f800-f81f : pnp 00:12
| f820-f83f : pnp 00:12
| fc00-fc0f : pnp 00:12
| 
| nihil:~# cat /proc/iomem
...
| 
| > Also, the latter 2 files after the driver is removed.
| > and in /proc/scsi after the driver is removed.
| 
| nihil:~# cat /proc/ioports
| Segmentation fault

Yes, I see that one also.

The problem is that the detect function requests an IO region
of 16 bytes (at least when a command line override parameter is
used) but the release function only tries to release 8 bytes,
and this request isn't done because it doesn't match any allocated
IO region.  [NCR53C400 extensions are not enabled, so
NCR5380_region_size is 8, not 16, but the request uses
NCR5380_region_size regardless.]

If enough kernel messages are enabled (like 'echo 9 > /proc/sysrq-trigger'),
you should have seen this message when the driver module was unloaded:
  Trying to free nonexistent resource <280-287>
because 280-28f was allocated.

Options to fix this:

1/ if the release_region() in the release function fails, have it try
   a second time with a different region size; (easy, sorta sloppy);
2/ save the allocated region size in instance->n_io_ports and release
   that size only;  <<<<< PATCH ATTACHED <<<<<
3/ if an override was used, release a region size of 16 in the
   release function; (this only looks clean for one adapter, not
   multiple)

Kevin, please test and report the results.  I think that the
module will be reloadable now.

Patch comments?  If no problems, James, please apply.


| nihil:~# cat /proc/iomem
...
| 
| The files in /proc/scsi are attached. Hope that helps...
| -Kevin
| -- 

--
~Randy

[-- Attachment #2: gNCR_regions.patch --]
[-- Type: application/octet-stream, Size: 2183 bytes --]

description:      release region of allocated (requested) size
product_versions: Linux 2.6.1

diffstat:=
 drivers/scsi/g_NCR5380.c |    9 +++++++--
  1 files changed, 7 insertions(+), 2 deletions(-)

diff -Naurp ./drivers/scsi/g_NCR5380.c~fakehw ./drivers/scsi/g_NCR5380.c
--- ./drivers/scsi/g_NCR5380.c~fakehw	2004-01-08 22:59:10.000000000 -0800
+++ ./drivers/scsi/g_NCR5380.c	2004-01-11 15:21:00.000000000 -0800
@@ -290,6 +290,7 @@ int __init generic_NCR5380_detect(Scsi_H
 	static int current_override = 0;
 	int count, i;
 	unsigned int *ports;
+	unsigned long region_size = 16;
 	static unsigned int __initdata ncr_53c400a_ports[] = {
 		0x280, 0x290, 0x300, 0x310, 0x330, 0x340, 0x348, 0x350, 0
 	};
@@ -420,6 +421,7 @@ int __init generic_NCR5380_detect(Scsi_H
 			/* Not a 53C400A style setup - just grab */
 			if(!(request_region(overrides[current_override].NCR5380_map_name, NCR5380_region_size, "ncr5380")))
 				continue;
+			region_size = NCR5380_region_size;
 		}
 #else
 		if(!request_mem_region(overrides[current_override].NCR5380_map_name, NCR5380_region_size, "ncr5380"))
@@ -428,7 +430,7 @@ int __init generic_NCR5380_detect(Scsi_H
 		instance = scsi_register(tpnt, sizeof(struct NCR5380_hostdata));
 		if (instance == NULL) {
 #ifndef CONFIG_SCSI_G_NCR5380_MEM
-			release_region(overrides[current_override].NCR5380_map_name, NCR5380_region_size);
+			release_region(overrides[current_override].NCR5380_map_name, region_size);
 #else
 			release_mem_region(overrides[current_override].NCR5380_map_name, NCR5380_region_size);
 #endif
@@ -436,6 +438,9 @@ int __init generic_NCR5380_detect(Scsi_H
 		}
 
 		instance->NCR5380_instance_name = overrides[current_override].NCR5380_map_name;
+#ifndef CONFIG_SCSI_G_NCR5380_MEM
+		instance->n_io_port = region_size;
+#endif
 
 		NCR5380_init(instance, flags);
 
@@ -498,7 +503,7 @@ int generic_NCR5380_release_resources(st
 	NCR5380_setup(instance);
 
 #ifndef CONFIG_SCSI_G_NCR5380_MEM
-	release_region(instance->NCR5380_instance_name, NCR5380_region_size);
+	release_region(instance->NCR5380_instance_name, instance->n_io_port);
 #else
 	release_mem_region(instance->NCR5380_instance_name, NCR5380_region_size);
 #endif