Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 09:03:57PM +, Russell King wrote: > > It would be trivial to treat them both as foobar0 and have the > > registration succeed for whoever gets it first, but I could see that this > > would be problematic in the serial8250 case. On the other hand, this is > > then serial8250's problem. > > Thank you for ignoring the other case of i82385 to justify your point > of view of it being just a single driver problem. > I didn't ignore it, I said that this was useful for anything that had device names ending in numbers. The above was just in reply to what you had pointed out about the serial8250 behaviour. Thank you for missing the point though. > Maybe you can work out a patch to fix up this mess? > Yes, I'll hack something together in the morning. pgpsXpFv90phM.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 10:56:03PM +0200, Paul Mundt wrote: > On Fri, Mar 25, 2005 at 08:25:08PM +, Russell King wrote: > > Eh? How do you end up with "/sys/devices/platform/foobar0.0" for the > > former case? It has an ID of "-1", and not zero. Your idea doesn't > > make any sense. > > > Yes, I missed the -1 part, so Kyle is correct. > > It would be trivial to treat them both as foobar0 and have the > registration succeed for whoever gets it first, but I could see that this > would be problematic in the serial8250 case. On the other hand, this is > then serial8250's problem. Thank you for ignoring the other case of i82385 to justify your point of view of it being just a single driver problem. Maybe you can work out a patch to fix up this mess? -- Russell King Linux kernel2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 08:25:08PM +, Russell King wrote: > Eh? How do you end up with "/sys/devices/platform/foobar0.0" for the > former case? It has an ID of "-1", and not zero. Your idea doesn't > make any sense. > Yes, I missed the -1 part, so Kyle is correct. It would be trivial to treat them both as foobar0 and have the registration succeed for whoever gets it first, but I could see that this would be problematic in the serial8250 case. On the other hand, this is then serial8250's problem. > > The first case is a corner case, and really shouldn't happen that much in > > practice outside of broken drivers. > > It does happen today. Firstly, the 8250 driver registers a device of > "serial8250" with id = -1 for the backwards-compatible devices. > Platforms can then register a platform device called "serial8250" > with zero or positive id numbers. > That's fine, but that still doesn't make it any less of a corner case. > > We don't go around changing /dev semantics everytime someone decides to > > call their device something silly, I don't see why platform devices > > should be treated differently, better to just fix the broken drivers.. > > It's not about something being called something silly. It's about > the original concept of how to generate the path being down right > stupid. > is a fairly common thing, if you have a problem with this, maybe you would like to audit /dev while you are at it. I don't disagree with you that this is useful for the devices that do end with numbers in their names, but breaking everything else as a result of this makes no sense either. What would you do if you needed to register a character device using the name of the device (which may end in a number, and there was a range of them)? This likely doesn't happen enough in practice for anyone to actually care, but you would have the same problem there otherwise. This should arguably be the problem of the corner case driver, it certainly shouldn't change convention for everyone else. While the original concept of how to generate the path may have been "down right stupid", it works for /dev, and I don't see how adding a superfluous . in the paths of devices that just don't care and subsequently breaking existing expectations of behaviour is any more inspired.. pgpSKqMDAHDAO.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 09:58:26PM +0200, Paul Mundt wrote: > On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote: > > So how would you tell the difference between the following? > > device = "foobar0" > > id = -1 > > path = "/sys/devices/platform/foobar0" > > versus > > device = "foobar" > > id = 0 > > path = "/sys/devices/platform/foobar0" > > > Easy, we use the delimiter on anything ending with a number at the end of > the device name.. so for device = "foobar0", this would end up as > /sys/devices/platform/foobar0.0, whereas in the latter case this would > end up as /sys/devices/platform/foobar0. Eh? How do you end up with "/sys/devices/platform/foobar0.0" for the former case? It has an ID of "-1", and not zero. Your idea doesn't make any sense. > The first case is a corner case, and really shouldn't happen that much in > practice outside of broken drivers. It does happen today. Firstly, the 8250 driver registers a device of "serial8250" with id = -1 for the backwards-compatible devices. Platforms can then register a platform device called "serial8250" with zero or positive id numbers. > > It's not as nice to add the extra period, but otherwise you end up with > > a lot of _extra_ special cases in both the kernel _and_ applications, > > which helps nobody. > > > No you don't, it's pretty easy to figure out that if the end of the > device name is a number that there will be a delimiter between that and > the id. This should be the exception, not the rule. Note that id = -1 means _no id_. So, Kyle is quite correct to ask about that case. device = "serial8250" id = -1 => /sys/devices/platform/serial8250 The "-1" means "do not add the ID". but, under the old naming scenario, the following comes out to the same sysfs path: device = "serial825" id = 0 => /sys/devices/platform/serial8250 and device = "serial8250" id = 0 => /sys/devices/platform/serial82500 is just too confusing. Same problem with i82365 platform devices, etc. > We don't go around changing /dev semantics everytime someone decides to > call their device something silly, I don't see why platform devices > should be treated differently, better to just fix the broken drivers.. It's not about something being called something silly. It's about the original concept of how to generate the path being down right stupid. -- Russell King Linux kernel2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Mar 25, 2005, at 14:58, Paul Mundt wrote:
On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote:
So how would you tell the difference between the following?
device = "foobar0"
id = -1
path = "/sys/devices/platform/foobar0"
versus
device = "foobar"
id = 0
path = "/sys/devices/platform/foobar0"
Easy, we use the delimiter on anything ending with a number at the end
of
the device name.. so for device = "foobar0", this would end up as
/sys/devices/platform/foobar0.0, whereas in the latter case this would
end up as /sys/devices/platform/foobar0.
But then you've just created yet another special case that will clutter
the
interface in both the kernel _and_ all the applications. Besides, the
ID
on the first entry isn't 0, it's -1, which indicates it's a singleton.
You'd break existing applications anyways, because you're renaming a
device
that existed and worked fine until it got renamed. We should try to fix
the interface properly where it's broken so that we don't have to live
with
the consequences for the next 3 years.
The first case is a corner case, and really shouldn't happen that much
in
practice outside of broken drivers.
But a driver ending in a number _isn't_ really a corner case. Most
devices
have model numbers, and so we shouldn't twist that case to do something
funny when we shouldn't have to.
No you don't, it's pretty easy to figure out that if the end of the
device name is a number that there will be a delimiter between that and
the id. This should be the exception, not the rule.
But on the first example, there _isn't_ an ID, it's -1, it's a
singleton.
We don't go around changing /dev semantics everytime someone decides to
call their device something silly, I don't see why platform devices
should be treated differently, better to just fix the broken drivers..
Fix the broken interface to have a unique naming scheme that doesn't
need
special cases and this problem will be less likely to occur again in the
future.
We all agree that if we were just creating this interface now, we would
use something like this, right?
if (id == -1) {
snprintf( path, path_len, "/sys/devices/platform/%s", name);
} else {
snprintf( path, path_len, "/sys/devices/platform/%s.%lu", name, id );
}
So why not do this now? It's a lot simpler and easier to get right,
with
no special cases other than the already existing singleton case. It
also
only requires a 1-character change to all existing code, the extra ".".
Cheers,
Kyle Moffett
-BEGIN GEEK CODE BLOCK-
Version: 3.12
GCM/CS/IT/U d- s++: a18 C>$ UB/L/X/*(+)>$ P+++()>$
L(+++) E W++(+) N+++(++) o? K? w--- O? M++ V? PS+() PE+(-) Y+
PGP+++ t+(+++) 5 X R? tv-(--) b(++) DI+ D+ G e->$ h!*()>++$ r
!y?(-)
--END GEEK CODE BLOCK--
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote: > So how would you tell the difference between the following? > device = "foobar0" > id = -1 > path = "/sys/devices/platform/foobar0" > versus > device = "foobar" > id = 0 > path = "/sys/devices/platform/foobar0" > Easy, we use the delimiter on anything ending with a number at the end of the device name.. so for device = "foobar0", this would end up as /sys/devices/platform/foobar0.0, whereas in the latter case this would end up as /sys/devices/platform/foobar0. The first case is a corner case, and really shouldn't happen that much in practice outside of broken drivers. > It's not as nice to add the extra period, but otherwise you end up with > a lot of _extra_ special cases in both the kernel _and_ applications, > which helps nobody. > No you don't, it's pretty easy to figure out that if the end of the device name is a number that there will be a delimiter between that and the id. This should be the exception, not the rule. We don't go around changing /dev semantics everytime someone decides to call their device something silly, I don't see why platform devices should be treated differently, better to just fix the broken drivers.. pgpXOhohVGxXC.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Mar 25, 2005, at 13:35, Paul Mundt wrote: Anything that expects that it can open a /sys/devices/platform/ path. I have a few applications like this, I have no reason to doubt that others do too. I don't see any reason to go out of the way to break this convention if the end of the device name is not a number. So how would you tell the difference between the following? device = "foobar0" id = -1 path = "/sys/devices/platform/foobar0" versus device = "foobar" id = 0 path = "/sys/devices/platform/foobar0" I'll agree that having two drivers named like this is bad, but how is a userspace application given a path like "/sys/devices/platform/foobar0" supposed to figure out which one it is. It's not as nice to add the extra period, but otherwise you end up with a lot of _extra_ special cases in both the kernel _and_ applications, which helps nobody. Cheers, Kyle Moffett -BEGIN GEEK CODE BLOCK- Version: 3.12 GCM/CS/IT/U d- s++: a18 C>$ UB/L/X/*(+)>$ P+++()>$ L(+++) E W++(+) N+++(++) o? K? w--- O? M++ V? PS+() PE+(-) Y+ PGP+++ t+(+++) 5 X R? tv-(--) b(++) DI+ D+ G e->$ h!*()>++$ r !y?(-) --END GEEK CODE BLOCK-- - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 10:10:14AM -0800, Greg KH wrote: > > This might make sense for devices that end in numbers, but does it really > > make sense for devices that don't? > > Then fix those drivers to not put the number in there if they don't have > one :) > But they do have non -1 ids, the device name itself just doesn't end in a number. In this event, the delimiter makes no sense. These drivers are expecting that you will have , and application code is expecting the same. /dev follows this convention too, I don't see this as being an unreasonable expectation. If anything, serial8250 is broken and should rename itself to something not ending in a number. It's not nice when one driver exhibits a corner case and decides to change the semantics for everyone else. > I don't see the serial8250 driver adding that .0 to it on my machines, > does this happen on yours? > Yes, we end up having /sys/devices/platform/serial8250 and serial8250.0. Where serial8250.0 ends up as: drwxr-xr-x3 00 0 Jan 1 00:00 . drwxr-xr-x 18 00 0 Jan 1 00:00 .. lrwxrwxrwx1 00 0 Jan 1 00:00 bus -> ../../../bus/platform -rw-r--r--1 004096 Jan 1 00:00 detach_state lrwxrwxrwx1 00 0 Jan 1 00:00 driver -> ../../../bus/platform/drivers/serial8250 drwxr-xr-x2 00 0 Jan 1 00:00 power That doesn't really bother me, having serial8250.0 is more sensible then serial82500. For this type of corner case the delimiter makes sense, but not in a blanket sense. > What userspace code are you referring to? > Anything that expects that it can open a /sys/devices/platform/ path. I have a few applications like this, I have no reason to doubt that others do too. I don't see any reason to go out of the way to break this convention if the end of the device name is not a number. pgpskYF7jwfiV.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 08:01:36PM +0200, Paul Mundt wrote: > On Wed, Mar 09, 2005 at 04:34:39PM -0800, Greg KH wrote: > > [PATCH] driver core: Separate platform device name from platform device > > number > > > > Separate platform device name from platform device number such that > > names ending with numbers aren't confusing. > > > This might make sense for devices that end in numbers, but does it really > make sense for devices that don't? Then fix those drivers to not put the number in there if they don't have one :) > I don't really see how having something like randomfb.0 is intuitive, > this may make sense for things like serial8250 where another 0 would > be misleading without some form of delimiter, but those are the corner > cases and should be treated as such. I don't see the serial8250 driver adding that .0 to it on my machines, does this happen on yours? > It's a bit irritating to have to constantly update userspace code that is > acting under the false pretense that there is some sort of consistent > naming scheme in place that won't change every time some new corner case > crops up. What userspace code are you referring to? thanks, greg k-h - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Wed, Mar 09, 2005 at 04:34:39PM -0800, Greg KH wrote: > [PATCH] driver core: Separate platform device name from platform device number > > Separate platform device name from platform device number such that > names ending with numbers aren't confusing. > This might make sense for devices that end in numbers, but does it really make sense for devices that don't? I don't really see how having something like randomfb.0 is intuitive, this may make sense for things like serial8250 where another 0 would be misleading without some form of delimiter, but those are the corner cases and should be treated as such. It's a bit irritating to have to constantly update userspace code that is acting under the false pretense that there is some sort of consistent naming scheme in place that won't change every time some new corner case crops up. (And yes, I should have brought this up when the patch was posted, but I didn't see it until _after_ being bit by this change). pgpkDvoOIRkMI.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Wed, Mar 09, 2005 at 04:34:39PM -0800, Greg KH wrote: [PATCH] driver core: Separate platform device name from platform device number Separate platform device name from platform device number such that names ending with numbers aren't confusing. This might make sense for devices that end in numbers, but does it really make sense for devices that don't? I don't really see how having something like randomfb.0 is intuitive, this may make sense for things like serial8250 where another 0 would be misleading without some form of delimiter, but those are the corner cases and should be treated as such. It's a bit irritating to have to constantly update userspace code that is acting under the false pretense that there is some sort of consistent naming scheme in place that won't change every time some new corner case crops up. (And yes, I should have brought this up when the patch was posted, but I didn't see it until _after_ being bit by this change). pgpkDvoOIRkMI.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 08:01:36PM +0200, Paul Mundt wrote: On Wed, Mar 09, 2005 at 04:34:39PM -0800, Greg KH wrote: [PATCH] driver core: Separate platform device name from platform device number Separate platform device name from platform device number such that names ending with numbers aren't confusing. This might make sense for devices that end in numbers, but does it really make sense for devices that don't? Then fix those drivers to not put the number in there if they don't have one :) I don't really see how having something like randomfb.0 is intuitive, this may make sense for things like serial8250 where another 0 would be misleading without some form of delimiter, but those are the corner cases and should be treated as such. I don't see the serial8250 driver adding that .0 to it on my machines, does this happen on yours? It's a bit irritating to have to constantly update userspace code that is acting under the false pretense that there is some sort of consistent naming scheme in place that won't change every time some new corner case crops up. What userspace code are you referring to? thanks, greg k-h - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 10:10:14AM -0800, Greg KH wrote: This might make sense for devices that end in numbers, but does it really make sense for devices that don't? Then fix those drivers to not put the number in there if they don't have one :) But they do have non -1 ids, the device name itself just doesn't end in a number. In this event, the delimiter makes no sense. These drivers are expecting that you will have devicenameid, and application code is expecting the same. /dev follows this convention too, I don't see this as being an unreasonable expectation. If anything, serial8250 is broken and should rename itself to something not ending in a number. It's not nice when one driver exhibits a corner case and decides to change the semantics for everyone else. I don't see the serial8250 driver adding that .0 to it on my machines, does this happen on yours? Yes, we end up having /sys/devices/platform/serial8250 and serial8250.0. Where serial8250.0 ends up as: drwxr-xr-x3 00 0 Jan 1 00:00 . drwxr-xr-x 18 00 0 Jan 1 00:00 .. lrwxrwxrwx1 00 0 Jan 1 00:00 bus - ../../../bus/platform -rw-r--r--1 004096 Jan 1 00:00 detach_state lrwxrwxrwx1 00 0 Jan 1 00:00 driver - ../../../bus/platform/drivers/serial8250 drwxr-xr-x2 00 0 Jan 1 00:00 power That doesn't really bother me, having serial8250.0 is more sensible then serial82500. For this type of corner case the delimiter makes sense, but not in a blanket sense. What userspace code are you referring to? Anything that expects that it can open a /sys/devices/platform/deviceid path. I have a few applications like this, I have no reason to doubt that others do too. I don't see any reason to go out of the way to break this convention if the end of the device name is not a number. pgpskYF7jwfiV.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Mar 25, 2005, at 13:35, Paul Mundt wrote: Anything that expects that it can open a /sys/devices/platform/deviceid path. I have a few applications like this, I have no reason to doubt that others do too. I don't see any reason to go out of the way to break this convention if the end of the device name is not a number. So how would you tell the difference between the following? device = foobar0 id = -1 path = /sys/devices/platform/foobar0 versus device = foobar id = 0 path = /sys/devices/platform/foobar0 I'll agree that having two drivers named like this is bad, but how is a userspace application given a path like /sys/devices/platform/foobar0 supposed to figure out which one it is. It's not as nice to add the extra period, but otherwise you end up with a lot of _extra_ special cases in both the kernel _and_ applications, which helps nobody. Cheers, Kyle Moffett -BEGIN GEEK CODE BLOCK- Version: 3.12 GCM/CS/IT/U d- s++: a18 C$ UB/L/X/*(+)$ P+++()$ L(+++) E W++(+) N+++(++) o? K? w--- O? M++ V? PS+() PE+(-) Y+ PGP+++ t+(+++) 5 X R? tv-(--) b(++) DI+ D+ G e-$ h!*()++$ r !y?(-) --END GEEK CODE BLOCK-- - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote: So how would you tell the difference between the following? device = foobar0 id = -1 path = /sys/devices/platform/foobar0 versus device = foobar id = 0 path = /sys/devices/platform/foobar0 Easy, we use the delimiter on anything ending with a number at the end of the device name.. so for device = foobar0, this would end up as /sys/devices/platform/foobar0.0, whereas in the latter case this would end up as /sys/devices/platform/foobar0. The first case is a corner case, and really shouldn't happen that much in practice outside of broken drivers. It's not as nice to add the extra period, but otherwise you end up with a lot of _extra_ special cases in both the kernel _and_ applications, which helps nobody. No you don't, it's pretty easy to figure out that if the end of the device name is a number that there will be a delimiter between that and the id. This should be the exception, not the rule. We don't go around changing /dev semantics everytime someone decides to call their device something silly, I don't see why platform devices should be treated differently, better to just fix the broken drivers.. pgpXOhohVGxXC.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Mar 25, 2005, at 14:58, Paul Mundt wrote:
On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote:
So how would you tell the difference between the following?
device = foobar0
id = -1
path = /sys/devices/platform/foobar0
versus
device = foobar
id = 0
path = /sys/devices/platform/foobar0
Easy, we use the delimiter on anything ending with a number at the end
of
the device name.. so for device = foobar0, this would end up as
/sys/devices/platform/foobar0.0, whereas in the latter case this would
end up as /sys/devices/platform/foobar0.
But then you've just created yet another special case that will clutter
the
interface in both the kernel _and_ all the applications. Besides, the
ID
on the first entry isn't 0, it's -1, which indicates it's a singleton.
You'd break existing applications anyways, because you're renaming a
device
that existed and worked fine until it got renamed. We should try to fix
the interface properly where it's broken so that we don't have to live
with
the consequences for the next 3 years.
The first case is a corner case, and really shouldn't happen that much
in
practice outside of broken drivers.
But a driver ending in a number _isn't_ really a corner case. Most
devices
have model numbers, and so we shouldn't twist that case to do something
funny when we shouldn't have to.
No you don't, it's pretty easy to figure out that if the end of the
device name is a number that there will be a delimiter between that and
the id. This should be the exception, not the rule.
But on the first example, there _isn't_ an ID, it's -1, it's a
singleton.
We don't go around changing /dev semantics everytime someone decides to
call their device something silly, I don't see why platform devices
should be treated differently, better to just fix the broken drivers..
Fix the broken interface to have a unique naming scheme that doesn't
need
special cases and this problem will be less likely to occur again in the
future.
We all agree that if we were just creating this interface now, we would
use something like this, right?
if (id == -1) {
snprintf( path, path_len, /sys/devices/platform/%s, name);
} else {
snprintf( path, path_len, /sys/devices/platform/%s.%lu, name, id );
}
So why not do this now? It's a lot simpler and easier to get right,
with
no special cases other than the already existing singleton case. It
also
only requires a 1-character change to all existing code, the extra ..
Cheers,
Kyle Moffett
-BEGIN GEEK CODE BLOCK-
Version: 3.12
GCM/CS/IT/U d- s++: a18 C$ UB/L/X/*(+)$ P+++()$
L(+++) E W++(+) N+++(++) o? K? w--- O? M++ V? PS+() PE+(-) Y+
PGP+++ t+(+++) 5 X R? tv-(--) b(++) DI+ D+ G e-$ h!*()++$ r
!y?(-)
--END GEEK CODE BLOCK--
-
To unsubscribe from this list: send the line unsubscribe linux-kernel in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 09:58:26PM +0200, Paul Mundt wrote: On Fri, Mar 25, 2005 at 02:38:22PM -0500, Kyle Moffett wrote: So how would you tell the difference between the following? device = foobar0 id = -1 path = /sys/devices/platform/foobar0 versus device = foobar id = 0 path = /sys/devices/platform/foobar0 Easy, we use the delimiter on anything ending with a number at the end of the device name.. so for device = foobar0, this would end up as /sys/devices/platform/foobar0.0, whereas in the latter case this would end up as /sys/devices/platform/foobar0. Eh? How do you end up with /sys/devices/platform/foobar0.0 for the former case? It has an ID of -1, and not zero. Your idea doesn't make any sense. The first case is a corner case, and really shouldn't happen that much in practice outside of broken drivers. It does happen today. Firstly, the 8250 driver registers a device of serial8250 with id = -1 for the backwards-compatible devices. Platforms can then register a platform device called serial8250 with zero or positive id numbers. It's not as nice to add the extra period, but otherwise you end up with a lot of _extra_ special cases in both the kernel _and_ applications, which helps nobody. No you don't, it's pretty easy to figure out that if the end of the device name is a number that there will be a delimiter between that and the id. This should be the exception, not the rule. Note that id = -1 means _no id_. So, Kyle is quite correct to ask about that case. device = serial8250 id = -1 = /sys/devices/platform/serial8250 The -1 means do not add the ID. but, under the old naming scenario, the following comes out to the same sysfs path: device = serial825 id = 0 = /sys/devices/platform/serial8250 and device = serial8250 id = 0 = /sys/devices/platform/serial82500 is just too confusing. Same problem with i82365 platform devices, etc. We don't go around changing /dev semantics everytime someone decides to call their device something silly, I don't see why platform devices should be treated differently, better to just fix the broken drivers.. It's not about something being called something silly. It's about the original concept of how to generate the path being down right stupid. -- Russell King Linux kernel2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 08:25:08PM +, Russell King wrote: Eh? How do you end up with /sys/devices/platform/foobar0.0 for the former case? It has an ID of -1, and not zero. Your idea doesn't make any sense. Yes, I missed the -1 part, so Kyle is correct. It would be trivial to treat them both as foobar0 and have the registration succeed for whoever gets it first, but I could see that this would be problematic in the serial8250 case. On the other hand, this is then serial8250's problem. The first case is a corner case, and really shouldn't happen that much in practice outside of broken drivers. It does happen today. Firstly, the 8250 driver registers a device of serial8250 with id = -1 for the backwards-compatible devices. Platforms can then register a platform device called serial8250 with zero or positive id numbers. That's fine, but that still doesn't make it any less of a corner case. We don't go around changing /dev semantics everytime someone decides to call their device something silly, I don't see why platform devices should be treated differently, better to just fix the broken drivers.. It's not about something being called something silly. It's about the original concept of how to generate the path being down right stupid. devid is a fairly common thing, if you have a problem with this, maybe you would like to audit /dev while you are at it. I don't disagree with you that this is useful for the devices that do end with numbers in their names, but breaking everything else as a result of this makes no sense either. What would you do if you needed to register a character device using the name of the device (which may end in a number, and there was a range of them)? This likely doesn't happen enough in practice for anyone to actually care, but you would have the same problem there otherwise. This should arguably be the problem of the corner case driver, it certainly shouldn't change convention for everyone else. While the original concept of how to generate the path may have been down right stupid, it works for /dev, and I don't see how adding a superfluous . in the paths of devices that just don't care and subsequently breaking existing expectations of behaviour is any more inspired.. pgpSKqMDAHDAO.pgp Description: PGP signature
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 10:56:03PM +0200, Paul Mundt wrote: On Fri, Mar 25, 2005 at 08:25:08PM +, Russell King wrote: Eh? How do you end up with /sys/devices/platform/foobar0.0 for the former case? It has an ID of -1, and not zero. Your idea doesn't make any sense. Yes, I missed the -1 part, so Kyle is correct. It would be trivial to treat them both as foobar0 and have the registration succeed for whoever gets it first, but I could see that this would be problematic in the serial8250 case. On the other hand, this is then serial8250's problem. Thank you for ignoring the other case of i82385 to justify your point of view of it being just a single driver problem. Maybe you can work out a patch to fix up this mess? -- Russell King Linux kernel2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] driver core: Separate platform device name from platform device number
On Fri, Mar 25, 2005 at 09:03:57PM +, Russell King wrote: It would be trivial to treat them both as foobar0 and have the registration succeed for whoever gets it first, but I could see that this would be problematic in the serial8250 case. On the other hand, this is then serial8250's problem. Thank you for ignoring the other case of i82385 to justify your point of view of it being just a single driver problem. I didn't ignore it, I said that this was useful for anything that had device names ending in numbers. The above was just in reply to what you had pointed out about the serial8250 behaviour. Thank you for missing the point though. Maybe you can work out a patch to fix up this mess? Yes, I'll hack something together in the morning. pgpsXpFv90phM.pgp Description: PGP signature
