Skip to contentSkip to conversation

Platform device matching, & weird strncmp usage

DISCUSSION Benjamin Herrenschmidt 4 participants 7 messages Jan 6, 2006 – Jan 7, 2006 Driver core

lore
Share & tools
lore
#0 lore raw

Hi !

In 2.6.15, platform device matching works according to this comment in the code, or rather are supposed to:

* Platform device IDs are assumed to be encoded like this:
* "<name><instance>", where <name> is a short description of the
* type of device, like "pci" or "floppy", and <instance> is the
* enumerated instance of the device, like '0' or '42'.

However, looking a few lines below, I see the actual implemetation:

static int platform_match(struct device * dev, struct device_driver * drv)

{
	struct platform_device *pdev = container_of(dev, struct platform_device, dev);

	return (strncmp(pdev->name, drv->name, BUS_ID_SIZE) == 0);
}

As far as I know, strncmp() is _NOT_ supposed to return 0 if one string is shorter than the other and they match until that point. Thus the above will never match unless the <name> portion of pdev->name is exactly of size BUS_ID_SIZE which is obviously not the case...

Did I miss something or do we expect a "special" semantic for strncmp in the kernel ?

Ben.

#1 lore raw
Quoted 4 lines · Benjamin Herrenschmidt

Hi !

In 2.6.15, platform device matching works according to this comment in the code, or rather are supposed to:

.../...

Just in case my analysis is correct, here's an untested fix:

---

Platform device matching doesn't work when an "id" is used.

drivers/base/platform.c +2 -1
@@ -447,7 +447,8 @@ static int platform_match(struct device 
 {
 	struct platform_device *pdev = container_of(dev, struct platform_device, dev);
 
-	return (strncmp(pdev->name, drv->name, BUS_ID_SIZE) == 0);
+	return strncmp(pdev->name, drv->name,
+			min(BUS_ID_SIZE, strlen(drv->name))) == 0;
 }
 
 static int platform_suspend(struct device * dev, pm_message_t state)
#2 lore raw
In reply to Benjamin Herrenschmidt
Quoted 11 lines · Benjamin Herrenschmidt

static int platform_match(struct device * dev, struct device_driver * drv)

{
	struct platform_device *pdev = container_of(dev, struct platform_device, dev);

	return (strncmp(pdev->name, drv->name, BUS_ID_SIZE) == 0);
}

As far as I know, strncmp() is _NOT_ supposed to return 0 if one string is shorter than the other and they match until that point. Thus the above will never match unless the <name> portion of pdev->name is exactly of size BUS_ID_SIZE which is obviously not the case...

pdev->name is just the <name> part - it's pdev->dev.name which has both the <name> and <instance>. I think the strncmp is unnecessary, and it can be replaced by a plain strcmp.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#3 lore raw
Quoted 8 lines · Russell King
Quoted 7 lines · Benjamin Herrenschmidt

static int platform_match(struct device * dev, struct device_driver * drv)

{
	struct platform_device *pdev = container_of(dev, struct platform_device, dev);

	return (strncmp(pdev->name, drv->name, BUS_ID_SIZE) == 0);
}
Quoted 8 lines
Quoted 4 lines

As far as I know, strncmp() is _NOT_ supposed to return 0 if one string is shorter than the other and they match until that point. Thus the above will never match unless the <name> portion of pdev->name is exactly of size BUS_ID_SIZE which is obviously not the case...

pdev->name is just the <name> part - it's pdev->dev.name which has both the <name> and <instance>. I think the strncmp is unnecessary, and it can be replaced by a plain strcmp.

Ah indeed, I got confused by having both strings.

Ben.

#4 lore raw
In reply to Benjamin Herrenschmidt
Quoted 17 lines · Benjamin Herrenschmidt

In 2.6.15, platform device matching works according to this comment in the code, or rather are supposed to:

* Platform device IDs are assumed to be encoded like this:
* "<name><instance>", where <name> is a short description of the
* type of device, like "pci" or "floppy", and <instance> is the
* enumerated instance of the device, like '0' or '42'.

However, looking a few lines below, I see the actual implemetation:

static int platform_match(struct device * dev, struct device_driver * drv)

{
	struct platform_device *pdev = container_of(dev, struct platform_device, dev);

	return (strncmp(pdev->name, drv->name, BUS_ID_SIZE) == 0);
}
Quoted 7 lines

As far as I know, strncmp() is _NOT_ supposed to return 0 if one string is shorter than the other and they match until that point. Thus the above will never match unless the <name> portion of pdev->name is exactly of size BUS_ID_SIZE which is obviously not the case...

Did I miss something or do we expect a "special" semantic for strncmp in the kernel ?

I can't speak to the correctness of that code but your understanding of strncmp() is incorrect. From "GNU C Library Application Fundamentals":

This function is the [sic] similar to strcmp, except that no more than size wide characters are compared. In other words, if the two strings are the same in their first size wide characters, the return value is zero.

And this has been may experience for the past 20 years and is confirmed by this trivial program which prints zero in both cases:

#include <string.h>
#include <stdio.h>

int main() {
    printf("%d\n", strncmp("abc","abcd",3));
    printf("%d\n", strncmp("abcd","abc",3));
}
#5 lore raw
Quoted 6 lines · Kurtis D. Rader
Quoted 2 lines · Benjamin Herrenschmidt

As far as I know, strncmp() is _NOT_ supposed to return 0 if one string is shorter than the other and they match until that point

I can't speak to the correctness of that code but your understanding of strncmp() is incorrect. From "GNU C Library Application Fundamentals":

I believe the original poster was asserting that 'strncmp("abc","abcd",100) should never return zero,' and not 'strncmp("abc","abcd",3) should never return zero.' Though I also found the statement confusing at first.

-Bob

#6 lore raw
In reply to Kurtis D. Rader
Quoted 12 lines

I can't speak to the correctness of that code but your understanding of strncmp() is incorrect. From "GNU C Library Application Fundamentals":

This function is the [sic] similar to strcmp, except that no more than size wide characters are compared. In other words, if the two strings are the same in their first size wide characters, the return value is zero.

And this has been may experience for the past 20 years and is confirmed by this trivial program which prints zero in both cases:

Quoted 6 lines

#include <string.h>
#include <stdio.h>

int main() {
    printf("%d\n", strncmp("abc","abcd",3));
    printf("%d\n", strncmp("abcd","abc",3));
}

Except that in my example, the count was larger than both strings...

Anyway, it happens to not be a problem as the string I was comparing was not in fact different.

Ben.