[U-Boot-Users] [PATCH] fdt: Add simple alias support to fdt print command
If the path we are trying to print doesn't exist see if it matches an aliases. We don't do anything fancy at this point, but just strip the leading '/' if it exists and see if we have an exact match to an alias.
In the future we could try and prefix matching so the alias could be used as a shorter path reference.
Signed-off-by: Kumar Gala galak@kernel.crashing.org --- common/cmd_fdt.c | 20 +++++++++++++++++--- 1 files changed, 17 insertions(+), 3 deletions(-)
diff --git a/common/cmd_fdt.c b/common/cmd_fdt.c index 97b9dd7..65c5cdf 100644 --- a/common/cmd_fdt.c +++ b/common/cmd_fdt.c @@ -678,9 +678,23 @@ static int fdt_print(const char *pathp, char *prop, int depth) /* * Not found or something else bad happened. */ - printf ("libfdt fdt_path_offset() returned %s\n", - fdt_strerror(nodeoffset)); - return 1; + + /* see if we match an alias */ + int aliasoffset = fdt_path_offset(working_fdt, "/aliases"); + if (aliasoffset >= 0) { + const char *aliasp = pathp; + if (pathp[0] == '/') + aliasp++; + aliasp = fdt_getprop(working_fdt, + aliasoffset, aliasp, NULL); + nodeoffset = fdt_path_offset(working_fdt, aliasp); + } + + if (nodeoffset < 0) { + printf("libfdt fdt_path_offset() returned %s\n", + fdt_strerror(nodeoffset)); + return 1; + } } /* * The user passed in a property as well as node path.
Kumar Gala wrote:
If the path we are trying to print doesn't exist see if it matches an aliases. We don't do anything fancy at this point, but just strip the leading '/' if it exists and see if we have an exact match to an alias.
In the future we could try and prefix matching so the alias could be used as a shorter path reference.
Signed-off-by: Kumar Gala galak@kernel.crashing.org
Cool and useful too. Out of curiousity, does "real" Open Firmware do this sort of thing with aliases?
One reservation I have (which may disappear if the answer to the previous question is "yes"), it automatically and silently dereferences the /aliases/X node when asked to display /X or X (but only if /X doesn't exist in the dtb). This is not an obvious behavior since X isn't real.
Should we have a different display syntax to force the dereference of an alias X? Assuming "*" is a good choice, this would change the behavior fdt print *ethernet0 to dereference /aliases/ethernet0 and print out /soc8360@e0000000/.../enet0 (or whatever).
Thanks, gvb
On Jul 9, 2008, at 10:17 AM, Jerry Van Baren wrote:
Kumar Gala wrote:
If the path we are trying to print doesn't exist see if it matches an aliases. We don't do anything fancy at this point, but just strip the leading '/' if it exists and see if we have an exact match to an alias. In the future we could try and prefix matching so the alias could be used as a shorter path reference. Signed-off-by: Kumar Gala galak@kernel.crashing.org
Cool and useful too. Out of curiousity, does "real" Open Firmware do this sort of thing with aliases?
I'm pretty sure it the Apple OF implementation does. However I dont remember to what extent it does from my playing around with OF on Apple HW.
One reservation I have (which may disappear if the answer to the previous question is "yes"), it automatically and silently dereferences the /aliases/X node when asked to display /X or X (but only if /X doesn't exist in the dtb). This is not an obvious behavior since X isn't real.
we could print out something about using an alias so the user knows that its happening.
Should we have a different display syntax to force the dereference of an alias X? Assuming "*" is a good choice, this would change the behavior fdt print *ethernet0 to dereference /aliases/ethernet0 and print out /soc8360@e0000000/.../enet0 (or whatever).
Lets says I have an alias for 'soc' to 'soc8360@e000000'. I want to be able to in the future do print /soc/enet0 and have that work. Introducing some new syntax would make that difficult and more ugly.
- k
Kumar Gala wrote:
On Jul 9, 2008, at 10:17 AM, Jerry Van Baren wrote:
Kumar Gala wrote:
If the path we are trying to print doesn't exist see if it matches an aliases. We don't do anything fancy at this point, but just strip the leading '/' if it exists and see if we have an exact match to an alias. In the future we could try and prefix matching so the alias could be used as a shorter path reference. Signed-off-by: Kumar Gala galak@kernel.crashing.org
Cool and useful too. Out of curiousity, does "real" Open Firmware do this sort of thing with aliases?
I'm pretty sure it the Apple OF implementation does. However I dont remember to what extent it does from my playing around with OF on Apple HW.
Sounds like I'm going to have to preempt my kids on the olpc and see what OF does on it. :-)
One reservation I have (which may disappear if the answer to the previous question is "yes"), it automatically and silently dereferences the /aliases/X node when asked to display /X or X (but only if /X doesn't exist in the dtb). This is not an obvious behavior since X isn't real.
we could print out something about using an alias so the user knows that its happening.
Should we have a different display syntax to force the dereference of an alias X? Assuming "*" is a good choice, this would change the behavior fdt print *ethernet0 to dereference /aliases/ethernet0 and print out /soc8360@e0000000/.../enet0 (or whatever).
Lets says I have an alias for 'soc' to 'soc8360@e000000'. I want to be able to in the future do print /soc/enet0 and have that work. Introducing some new syntax would make that difficult and more ugly.
- k
Thinking out loud... we could define the syntax that a leading "*" indicates the first part of the path is a dereference of /aliases.
Assuming /aliases/soc = /soc8360@e000000 /aliases/ethernet0 = /soc8360@e0000000/.../enet0 then print *soc/enet0 and print *ethernet0 would both work and print the right thing. You *would* have to know that the first element of the path is an /aliases dereference. Your original patch did not require that piece of knowledge (but silently and automagically, which makes me nervous).
gvb
On Jul 9, 2008, at 12:02 PM, Jerry Van Baren wrote:
Thinking out loud... we could define the syntax that a leading "*" indicates the first part of the path is a dereference of /aliases.
Assuming /aliases/soc = /soc8360@e000000 /aliases/ethernet0 = /soc8360@e0000000/.../enet0 then print *soc/enet0 and print *ethernet0 would both work and print the right thing. You *would* have to know that the first element of the path is an /aliases dereference. Your original patch did not require that piece of knowledge (but silently and automagically, which makes me nervous).
did we come to resolution on this? I'd like to see this in 1.3.5.
- k
Kumar Gala wrote:
On Jul 9, 2008, at 12:02 PM, Jerry Van Baren wrote:
Thinking out loud... we could define the syntax that a leading "*" indicates the first part of the path is a dereference of /aliases.
Assuming /aliases/soc = /soc8360@e000000 /aliases/ethernet0 = /soc8360@e0000000/.../enet0 then print *soc/enet0 and print *ethernet0 would both work and print the right thing. You *would* have to know that the first element of the path is an /aliases dereference. Your original patch did not require that piece of knowledge (but silently and automagically, which makes me nervous).
did we come to resolution on this? I'd like to see this in 1.3.5.
- k
Hi Kumar,
I think we have basic resolution - I would like to see it in 1.3.5 too. I haven't pushed on this, waiting for 1.3.5 window to open (or some free time, whichever comes last).
I've CC:ed David Gibson in case he has some advice - the concept is to indicate a dereference of /aliases nodes so that us lazy engineers don't have to cut'n'paste the whole long path from the alias. Kumar originally proposed to do it automagically and I countered proposing using "*" to indicate the next path name should be looked up in /aliases and the result used instead (i.e. dereferenced). Discussion thread: http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/43575/focus=44941
Looking at the ieee1275 doc http://playground.sun.com/pub/1275/coredoc/1275-1994/1275.ps.gz it looks like "*" will work for a dereference delimiter as it is not listed as one of the permitted punctuation characters in a node name. Quoting 3.2.1.1 Node names: ---------------------------------------------------------------------- The driver name field is a sequence of between one and 31 letters, digits, and punctuation characters from the set ", . _ + - ". Uppercase and lowercase characters are distinct. ----------------------------------------------------------------------
We do have a problem with property names, where "*" _is_ a legal name component. Quoting 3.2.2.1.1 Property names: ---------------------------------------------------------------------- The property name is a human-readable text string consisting of one to thirty-one printable characters. Property names shall not contain uppercase characters or the characters "/", "", ":", "[", "]" and "@". ---------------------------------------------------------------------- Note that "*" is not proscribed, making it a legal character in a property name.
Having noted that, I'm willing to take the risk and use "*" for the "alias dereference" separator.
Looking back at the original patch, Kumar's original patch only did the /aliases dereference for the "fdt print" command. I'm thinking more general purpose: being able to dereference /aliases in all "fdt" commands. This seems helpful for the "fdt set" command, for instance. Whether this is reasonable to implement remains to be seen...
Best regards, gvb
On Sat, Aug 02, 2008 at 08:51:54PM -0400, Jerry Van Baren wrote:
Kumar Gala wrote:
On Jul 9, 2008, at 12:02 PM, Jerry Van Baren wrote:
Thinking out loud... we could define the syntax that a leading "*" indicates the first part of the path is a dereference of /aliases.
Assuming /aliases/soc = /soc8360@e000000 /aliases/ethernet0 = /soc8360@e0000000/.../enet0 then print *soc/enet0 and print *ethernet0 would both work and print the right thing. You *would* have to know that the first element of the path is an /aliases dereference. Your original patch did not require that piece of knowledge (but silently and automagically, which makes me nervous).
did we come to resolution on this? I'd like to see this in 1.3.5.
- k
Hi Kumar,
I think we have basic resolution - I would like to see it in 1.3.5 too. I haven't pushed on this, waiting for 1.3.5 window to open (or some free time, whichever comes last).
I've CC:ed David Gibson in case he has some advice - the concept is to indicate a dereference of /aliases nodes so that us lazy engineers don't have to cut'n'paste the whole long path from the alias. Kumar originally proposed to do it automagically and I countered proposing using "*" to indicate the next path name should be looked up in /aliases and the result used instead (i.e. dereferenced). Discussion thread: http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/43575/focus=44941
No, I really don't think using this "dereference" character is a good idea. If you're going to expand aliases, you should do it as real OF does - see section 4.3 of IEEE1275. Essentially it's the *lack* of a leading '/' character that triggers alias expansion. So you could use e.g. /soc8360@e0000000/ethernet@e0000400 or soc/ethernet@e0000400 or ethernet0
Looking at the ieee1275 doc http://playground.sun.com/pub/1275/coredoc/1275-1994/1275.ps.gz it looks like "*" will work for a dereference delimiter as it is not listed as one of the permitted punctuation characters in a node name. Quoting 3.2.1.1 Node names:
The driver name field is a sequence of between one and 31 letters, digits, and punctuation characters from the set ", . _ + - ". Uppercase and lowercase characters are distinct.
We do have a problem with property names, where "*" _is_ a legal name component. Quoting 3.2.2.1.1 Property names:
The property name is a human-readable text string consisting of one to thirty-one printable characters. Property names shall not contain uppercase characters or the characters "/", "", ":", "[", "]" and "@".
Note that "*" is not proscribed, making it a legal character in a property name.
I've observed a '*' in the device tree in the wild - some Apple tree, unfortunately I can't find it right at the moment. So I can't check if the '*' was in a node node, making it IEEE1275 illegal, or in a property name. I remember thinking it was illegal at the time, but I thought '*' was invalid in property names too, which seems to have been a misinterpretation.
Having noted that, I'm willing to take the risk and use "*" for the "alias dereference" separator.
Looking back at the original patch, Kumar's original patch only did the /aliases dereference for the "fdt print" command. I'm thinking more general purpose: being able to dereference /aliases in all "fdt" commands. This seems helpful for the "fdt set" command, for instance. Whether this is reasonable to implement remains to be seen...
If you're interepreting them in one place, you should probably interpret them everywhwere and have a single "resolve pathname" function. In fact, I should quite possibly put such a function into libfdt.
David Gibson wrote:
On Sat, Aug 02, 2008 at 08:51:54PM -0400, Jerry Van Baren wrote:
[snip]
I've CC:ed David Gibson in case he has some advice - the concept is to indicate a dereference of /aliases nodes so that us lazy engineers don't have to cut'n'paste the whole long path from the alias. Kumar originally proposed to do it automagically and I countered proposing using "*" to indicate the next path name should be looked up in /aliases and the result used instead (i.e. dereferenced). Discussion thread: http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/43575/focus=44941
No, I really don't think using this "dereference" character is a good idea. If you're going to expand aliases, you should do it as real OF does - see section 4.3 of IEEE1275. Essentially it's the *lack* of a leading '/' character that triggers alias expansion. So you could use e.g. /soc8360@e0000000/ethernet@e0000400 or soc/ethernet@e0000400 or ethernet0
Ahh, I didn't read far enough. The algorithm in section 4.3 is much better thought out than either of our proposals.
[snip]
If you're interepreting them in one place, you should probably interpret them everywhwere and have a single "resolve pathname" function.
Yes.
In fact, I should quite possibly put such a function into libfdt.
That would be very useful. :-)
Thanks, gvb
participants (4)
-
David Gibson -
Jerry Van Baren -
Jerry Van Baren -
Kumar Gala