Skip to contentSkip to conversation

hwmon: (pc87360) Fix device resource

PATCH In review Jean Delvare 2 participants 7 messages Jul 7, 2010 – Aug 12, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hwmon: (pc87360) Fix device resource

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually registered. The other I/O resources were leaked.

---
Jim, do you still have your system with a PC8736x device? If you do, can you please test this fix and report? Thanks.

 drivers/hwmon/pc87360.c |   30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)
drivers/hwmon/pc87360.c +16 -14
@@ -1610,11 +1610,8 @@ static struct pc87360_data *pc87360_upda
 
 static int __init pc87360_device_add(unsigned short address)
 {
-	struct resource res = {
-		.name	= "pc87360",
-		.flags	= IORESOURCE_IO,
-	};
-	int err, i;
+	struct resource res[3];
+	int err, i, res_count;
 
 	pdev = platform_device_alloc("pc87360", address);
 	if (!pdev) {
@@ -1623,22 +1620,27 @@ static int __init pc87360_device_add(uns
 		goto exit;
 	}
 
+	memset(res, 0, 3 * sizeof(struct resource));
 	for (i = 0; i < 3; i++) {
 		if (!extra_isa[i])
 			continue;
-		res.start = extra_isa[i];
-		res.end = extra_isa[i] + PC87360_EXTENT - 1;
+		res[res_count].start = extra_isa[i];
+		res[res_count].end = extra_isa[i] + PC87360_EXTENT - 1;
+		res[res_count].name = "pc87360",
+		res[res_count].flags = IORESOURCE_IO,
 
-		err = acpi_check_resource_conflict(&res);
+		err = acpi_check_resource_conflict(&res[res_count]);
 		if (err)
 			goto exit_device_put;
 
-		err = platform_device_add_resources(pdev, &res, 1);
-		if (err) {
-			printk(KERN_ERR "pc87360: Device resource[%d] "
-			       "addition failed (%d)\n", i, err);
-			goto exit_device_put;
-		}
+		res_count++;
+	}
+
+	err = platform_device_add_resources(pdev, res, res_count);
+	if (err) {
+		printk(KERN_ERR "pc87360: Device resources addition failed "
+		       "(%d)\n", err);
+		goto exit_device_put;
 	}
 
 	err = platform_device_add(pdev);
Signature · 6 lines
Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
maintainer
#1 lore raw

hi Jean

Quoted 8 lines · Jean Delvare

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually registered. The other I/O resources were leaked.

Quoted 6 lines

---
Jim, do you still have your system with a PC8736x device? If you do, can you please test this fix and report? Thanks.

Its not handy, but I see about arranging access.

Did you catch this by inspection?
was there another thread that triggered your recollection/association ? Any simple check for this ?
I dont suppose a /proc/<file> is gonna shout RESOURCE LEAK .. that said, I'll look before&after for something more subtle.

thanks Jean
Jim

Quoted 60 lines

 drivers/hwmon/pc87360.c |   30 ++++++++++++++++--------------

 1 file changed, 16 insertions(+), 14 deletions(-)
drivers/hwmon/pc87360.c    2010-07-07 14:11:03.000000000 +0200 renamed from drivers/hwmon/pc87360.c       2010-05-17 18:20:27.000000000 +0200 +16 -15
@@ -1610,11 +1610,8 @@ static struct pc87360_data *pc87360_upda
 
  static int __init pc87360_device_add(unsigned short address)
  {
-       struct resource res = {
-               .name   = "pc87360",
-               .flags  = IORESOURCE_IO,
-       };
-       int err, i;
+       struct resource res[3];
+       int err, i, res_count;
 
        pdev = platform_device_alloc("pc87360", address);
        if (!pdev) {
@@ -1623,22 +1620,27 @@ static int __init pc87360_device_add(uns
                goto exit;
        }
 
+       memset(res, 0, 3 * sizeof(struct resource));
        for (i = 0; i < 3; i++) {
                if (!extra_isa[i])
                        continue;
-               res.start = extra_isa[i];
-               res.end = extra_isa[i] + PC87360_EXTENT - 1;
+               res[res_count].start = extra_isa[i];
+               res[res_count].end = extra_isa[i] + PC87360_EXTENT - 1;
+               res[res_count].name = "pc87360",
+               res[res_count].flags = IORESOURCE_IO,
 
-               err = acpi_check_resource_conflict(&res);
+               err = acpi_check_resource_conflict(&res[res_count]);
                if (err)
                        goto exit_device_put;
 
-               err = platform_device_add_resources(pdev, &res, 1);
-               if (err) {
-                       printk(KERN_ERR "pc87360: Device resource[%d] "
-                              "addition failed (%d)\n", i, err);
-                       goto exit_device_put;
-               }
+               res_count++;
+       }
+
+       err = platform_device_add_resources(pdev, res, res_count);
+       if (err) {
+               printk(KERN_ERR "pc87360: Device resources addition failed "
+                      "(%d)\n", err);
+               goto exit_device_put;
        }
 
        err = platform_device_add(pdev);
 
 
--
Quoted 2 lines

Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

#2 lore raw
In reply to Jean Delvare

Hi Jim,

Quoted 16 lines · Jim Cromie
Quoted 8 lines · Jean Delvare

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually registered. The other I/O resources were leaked.

Quoted 6 lines

---
Jim, do you still have your system with a PC8736x device? If you do, can you please test this fix and report? Thanks.

Quoted 5 lines

Its not handy, but I see about arranging access.

Did you catch this by inspection?
was there another thread that triggered your recollection/association ? Any simple check for this ?

I copied this code to the pc87427 driver yesterday, as I am in the process of adding support for temperature monitoring to that driver (currently it only supports fans). The pc87427 driver code is slightly cleaner than the pc87360 driver code, in that it doesn't use a global array to carry the multiple I/O addresses of the logical devices. Instead, it retrieves them using platform_get_resource(). I found it worked OK for the first resource, and returned NULL for the second. Which made me realize that my two calls to platform_device_add_resources() had resulted in a single resource being registered. Further code inspection (of platform_device_add_resources) revealed that both had been allocated, but the first one got lost when the second was added.

I've fixed the pc87427 driver already, and now it's the pc87360 driver's turn.

Quoted 1 line

I dont suppose a /proc/<file> is gonna shout RESOURCE LEAK ..

Indeed not. I know it leaked memory by code inspection, I didn't witness it. Unless you cycle the pc87360 driver, the leak of a few dozen bytes is unnoticeable.

Quoted 1 line

that said, I'll look before&after for something more subtle.

You should definitely see a difference in /proc/ioports before and after my fix.

Signature · 6 lines
Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
maintainer
#3 lore raw
In reply to Jean Delvare
Quoted 8 lines · Jean Delvare

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually registered. The other I/O resources were leaked.

Quoted 8 lines

---
Jim, do you still have your system with a PC8736x device? If you do, can you please test this fix and report? Thanks.

hi Jean,

I applied this patch, and it built w/o errors.
Howver I have some distro (voyage 0.6.5) vs kernel issues
to resolve before I have a solid testing env.
It looks like 2.6.35 doesnt like udev 145, which shipped along with 2.6.30 kernel

[   11.252182] udev: starting version 145
[   11.252232] <3>udev: missing sysfs features; please update the kernel or

disable the kernel's CONFIG_SYSFS_DEPRECATED option; udev may fail to work correctly

and it does fail to work correctly;

voyage:~# modprobe pc87360
WARNING: Could not open 'kernel/drivers/hwmon/hwmon-vid.ko': No such file or
directory
FATAL: Could not open 'kernel/drivers/hwmon/pc87360.ko': No such file or
directory

I'll try to straighten this out, and follow up on hotplug ml if I need help.

In the meantime, it looks like res_count isnt initialized before 1st use.

jimc@harpo linux-2.6.git]$ grep -n res_count drivers/hwmon/pc87360.c 1614: int err, i, res_count;
1627: res[res_count].start = extra_isa[i];
1628: res[res_count].end = extra_isa[i] + PC87360_EXTENT - 1; 1629: res[res_count].name = "pc87360",
1630: res[res_count].flags = IORESOURCE_IO,
1632: err = acpi_check_resource_conflict(&res[res_count]); 1636: res_count++;
1639: err = platform_device_add_resources(pdev, res, res_count);

Oddly, gcc doesnt complain about this,
is this somehow hidden by zeroed memory due to __init on the function ??

Jim

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

#4 lore raw
In reply to Jean Delvare

Hi Jim,

Quoted 26 lines · Jim Cromie
Quoted 8 lines · Jean Delvare

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually registered. The other I/O resources were leaked.

Quoted 8 lines

---
Jim, do you still have your system with a PC8736x device? If you do, can you please test this fix and report? Thanks.

hi Jean,

I applied this patch, and it built w/o errors.
Howver I have some distro (voyage 0.6.5) vs kernel issues
to resolve before I have a solid testing env.
It looks like 2.6.35 doesnt like udev 145, which shipped along with 2.6.30 kernel

Quoted 4 lines
[   11.252182] udev: starting version 145
[   11.252232] <3>udev: missing sysfs features; please update the kernel or

disable the kernel's CONFIG_SYSFS_DEPRECATED option; udev may fail to work correctly

I can't help you much with this, I'm afraid. Just try what they say...

Quoted 8 lines

and it does fail to work correctly;

voyage:~# modprobe pc87360
WARNING: Could not open 'kernel/drivers/hwmon/hwmon-vid.ko': No such file or
directory
FATAL: Could not open 'kernel/drivers/hwmon/pc87360.ko': No such file or
directory

I'll try to straighten this out, and follow up on hotplug ml if I need help.

The errors above seem related to module dependencies and/or installation, so I'd look at how and where the kernel modules were installed on the system, if module-init-tools is up-to-date, try running "depmod -a", see if "modprobe -c" complains. If nothing helps, try running modprobe through strace, to see what exactly is failing.

Quoted 12 lines

In the meantime, it looks like res_count isnt initialized before 1st use.

jimc@harpo linux-2.6.git]$ grep -n res_count drivers/hwmon/pc87360.c 1614: int err, i, res_count;
1627: res[res_count].start = extra_isa[i];
1628: res[res_count].end = extra_isa[i] + PC87360_EXTENT - 1; 1629: res[res_count].name = "pc87360",
1630: res[res_count].flags = IORESOURCE_IO,
1632: err = acpi_check_resource_conflict(&res[res_count]); 1636: res_count++;
1639: err = platform_device_add_resources(pdev, res, res_count);

Quoted 2 lines

Oddly, gcc doesnt complain about this,
is this somehow hidden by zeroed memory due to __init on the function ??

This is indeed a bug, thanks a lot for finding and reporting it. It's fixed now.

Signature · 6 lines
Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
maintainer
#5 lore raw
In reply to Jean Delvare
Quoted 9 lines · Jean Delvare

Hi Jim,

Quoted 5 lines · Jim Cromie
Quoted 3 lines · Jean Delvare

It's not OK to call platform_device_add_resources() multiple times in a row. Despite its name, this functions sets the resources, it doesn't add them. So we have to prepare an array with all the

snip

Quoted 10 lines
Quoted 1 line

I applied this patch, and it built w/o errors.

Howver I have some distro (voyage 0.6.5) vs kernel issues

Quoted 2 lines

to resolve before I have a solid testing env.
It looks like 2.6.35 doesnt like udev 145, which shipped along with

2.6.30

Quoted 4 lines

kernel

[   11.252182] udev: starting version 145
[   11.252232] <3>udev: missing sysfs features; please update the kernel
Quoted 8 lines

or

Quoted 1 line

disable the kernel's CONFIG_SYSFS_DEPRECATED option; udev may fail to

work

Quoted 1 line

correctly

I can't help you much with this, I'm afraid. Just try what they say...

FTR - udev 145 is added into the initramfs file from my laptop,

not from voyage-linux. So I should be able to fix this and test soon.

Quoted 10 lines

In the meantime, it looks like res_count isnt initialized before 1st use.

Quoted 3 lines

Oddly, gcc doesnt complain about this,
is this somehow hidden by zeroed memory due to __init on the function ??

This is indeed a bug, thanks a lot for finding and reporting it. It's fixed now.

WRT lack of uninit-var warning,

[jimc@harpo sk1]$ make V=1 drivers/hwmon/pc87360.ko
...
gcc -Wp,-MD,drivers/hwmon/.pc87360.o.d -nostdinc -isystem
/usr/lib/gcc/i686-redhat-linux/4.4.4/include -I/home/jimc/lx/linux-2.6.git/arch/x86/include -Iinclude
-I/home/jimc/lx/linux-2.6.git/include -include include/generated/autoconf.h -I/home/jimc/lx/linux-2.6.git/drivers/hwmon -Idrivers/hwmon -D__KERNEL__
-Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration -Wno-format-security -fno-delete-null-pointer-checks -Os -m32 -msoft-float -mregparm=3 -freg-struct-return -mpreferred-stack-boundary=2 -march=pentium-mmx -mtune=generic -Wa,-mtune=generic32 -ffreestanding -DCONFIG_AS_CFI=1 -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -DCONFIG_AS_CFI_SECTIONS=1 -pipe -Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -Wframe-larger-than=1024 -fno-stack-protector -fno-omit-frame-pointer -fno-optimize-sibling-calls -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow -fconserve-stack -DMODULE -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(pc87360)"
-D"KBUILD_MODNAME=KBUILD_STR(pc87360)" -c -o drivers/hwmon/pc87360.o
/home/jimc/lx/linux-2.6.git/drivers/hwmon/pc87360.c
make -f /home/jimc/lx/linux-2.6.git/scripts/Makefile.modpost
scripts/mod/modpost -o /home/jimc/lx/sk1/Module.symvers -S -s
WARNING: modpost: Found 21 section mismatch(es).

-Wuninitialized is not in there (explaining lack of warnings)

is this a local config problem, or has it somehow dropped out of the Kconfig process ?
or was it never in there (seems unlikely) ?

also, regarding your patch, is there a real need for res_count at all ? IOW, why not this (abbreviated) ?

- res[res_count].start = extra_isa[i];
+ res[i].start = extra_isa[i];
...
- err = platform_device_add_resources(pdev, res, res_count);
+ err = platform_device_add_resources(pdev, res, --i);

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

#6 lore raw
In reply to Jean Delvare

Hi Jim,

Quoted 36 lines · Jim Cromie
Quoted 7 lines · Jean Delvare
Quoted 4 lines

In the meantime, it looks like res_count isnt initialized before 1st use. (...)
Oddly, gcc doesnt complain about this,
is this somehow hidden by zeroed memory due to __init on the function ??

This is indeed a bug, thanks a lot for finding and reporting it. It's fixed now.

WRT lack of uninit-var warning,

[jimc@harpo sk1]$ make V=1 drivers/hwmon/pc87360.ko
...
gcc -Wp,-MD,drivers/hwmon/.pc87360.o.d -nostdinc -isystem
/usr/lib/gcc/i686-redhat-linux/4.4.4/include

(quoted diff) +0 -13
-I/home/jimc/lx/linux-2.6.git/arch/x86/include -Iinclude
 -I/home/jimc/lx/linux-2.6.git/include -include include/generated/autoconf.h
 -I/home/jimc/lx/linux-2.6.git/drivers/hwmon -Idrivers/hwmon -D__KERNEL__
-Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing
-fno-common -Werror-implicit-function-declaration -Wno-format-security
-fno-delete-null-pointer-checks -Os -m32 -msoft-float -mregparm=3
-freg-struct-return -mpreferred-stack-boundary=2 -march=pentium-mmx
-mtune=generic -Wa,-mtune=generic32 -ffreestanding -DCONFIG_AS_CFI=1
-DCONFIG_AS_CFI_SIGNAL_FRAME=1 -DCONFIG_AS_CFI_SECTIONS=1 -pipe
-Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx
-mno-sse2 -mno-3dnow -Wframe-larger-than24 -fno-stack-protector
-fno-omit-frame-pointer -fno-optimize-sibling-calls
-Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow
-fconserve-stack  -DMODULE -D"KBUILD_STR(s)=#s"
-D"KBUILD_BASENAME=KBUILD_STR(pc87360)"
 -D"KBUILD_MODNAME=KBUILD_STR(pc87360)"  -c -o drivers/hwmon/pc87360.o

/home/jimc/lx/linux-2.6.git/drivers/hwmon/pc87360.c make -f /home/jimc/lx/linux-2.6.git/scripts/Makefile.modpost
scripts/mod/modpost -o /home/jimc/lx/sk1/Module.symvers -S -s
WARNING: modpost: Found 21 section mismatch(es).

Quoted 5 lines

-Wuninitialized is not in there (explaining lack of warnings)

is this a local config problem, or has it somehow dropped out of the Kconfig process ?
or was it never in there (seems unlikely) ?

-Wuninitialized is implied by -Wall (which is there) "only with -O1 and above" (quoting the man page). I'm not sure if -Os qualifies. But anyway I'm not using CONFIG_CC_OPTIMIZE_FOR_SIZE and didn't get the warning either...

Maybe this option is simply not 100% reliable. After all, it prints false positives every now and then, so maybe it also fails to catch some bugs.

Quoted 8 lines

also, regarding your patch, is there a real need for res_count at all ? IOW, why not this (abbreviated) ?

- res[res_count].start = extra_isa[i];
+ res[i].start = extra_isa[i];
...
- err = platform_device_add_resources(pdev, res, res_count);
+ err = platform_device_add_resources(pdev, res, --i);

We're iterating over 2 different arrays: res and extra_isa. res_count is the index for res[], i is the index for extra_isa[]. If all resources are present then both are in sync, but if any resource is missing then they will drift. So no, I can't see any way to get rid of res_count.

Signature · 6 lines
Jean Delvare

_______________________________________________
lm-sensors mailing list
[email protected]
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors