Skip to contentSkip to conversation

`uclass_find_device_by_name` fails to find the device

U-Boot Nandor Han 2 participants 4 messages Nov 6, 2019 – Mar 14, 2020

lore
Share & tools
lore

Nandor Han

author
#0 lore raw

Hi everybody,
I've noticed that commit 4213609cc7fb78f84b2ea63f4a5691b60d01c248
changes how uclass.c:uclass_find_device_by_name function compares the names for finding devices.

Because of this I have an issue when having a DTS and trying to find a device by name.

The lists.c:lists_bind_fdt function will bind a device to a device node tree and use the node name as device name.

```
lists.c:
int lists_bind_fdt(..)
...
name = ofnode_get_name(node);
...

```

Because of this devices will end up having names like: i2c at 021a0000, pfuze100 at 8...

When using uclass.c:uclass_find_device_by_name to get a device, before the change, this was not an issue since strncmp was used with the length of the name making a call like uclass_find_device_by_name(pfuze100) successfully even if the device name was pfuze100 at 8.

```
if (!strncmp(dev->name, name, strlen(name))) {
```

However after the change (which I think is correct), the check fails, resulting in not finding the expected device, since the device name has also the suffix @... included.

What's the proper fix here:

1. Changing the lists.c:lists_bind_fdt to drop the @ suffix?
Note: Could this be a problem when having the same node names but
different addressees in DTS {e.g.foo at 1, foo at 2}

2. When using uclass.c:uclass_find_device_by_name the user should take in consideration that @.. suffix needs to be added.
Note: This could be a problem when same code is used with different
DTSs.

3. Changing the comparation in uclass.c:uclass_find_device_by_name to ignore the part after @ for device names.

4. What else?

Regads,
Nandor

#1 lore raw

HI Nandor,

Quoted 47 lines · Nandor Han

Hi everybody,
I've noticed that commit 4213609cc7fb78f84b2ea63f4a5691b60d01c248
changes how uclass.c:uclass_find_device_by_name function compares the names for finding devices.

Because of this I have an issue when having a DTS and trying to find a device by name.

The lists.c:lists_bind_fdt function will bind a device to a device node tree and use the node name as device name.

```
lists.c:
int lists_bind_fdt(..)
...
name = ofnode_get_name(node);
...

```

Because of this devices will end up having names like: i2c at 021a0000, pfuze100 at 8...

When using uclass.c:uclass_find_device_by_name to get a device, before the change, this was not an issue since strncmp was used with the length of the name making a call like uclass_find_device_by_name(pfuze100) successfully even if the device name was pfuze100 at 8.

```
if (!strncmp(dev->name, name, strlen(name))) {
```

However after the change (which I think is correct), the check fails, resulting in not finding the expected device, since the device name has also the suffix @... included.

What's the proper fix here:

1. Changing the lists.c:lists_bind_fdt to drop the @ suffix?
Note: Could this be a problem when having the same node names but
different addressees in DTS {e.g.foo at 1, foo at 2}

2. When using uclass.c:uclass_find_device_by_name the user should take in consideration that @.. suffix needs to be added.
Note: This could be a problem when same code is used with different
DTSs.

Quoted 4 lines

3. Changing the comparation in uclass.c:uclass_find_device_by_name to ignore the part after @ for device names.

4. What else?

Can you use a phandle to refer to the device from another node, perhaps a UCLASS_BOARD node if needed? It is not a good idea to use this function.

Alternatively we could add a new function which finds by partial name.

Regards,
Simon

author
#2 lore raw

Hi Simon and thanks,

On 2019-12-24 17:58, Simon Glass wrote:

<snip>

Quoted 4 lines

Can you use a phandle to refer to the device from another node, perhaps a UCLASS_BOARD node if needed? It is not a good idea to use this function.

I'm not sure, I'll have to check that. Anyway uclass_find_device_by_name is used by uclass_get_device_by_name, which already seems to be used in plenty of places.

Quoted 2 lines

Alternatively we could add a new function which finds by partial name.

My fix was to ignore the suffix starting from "@", if that exist, since that's more like metadata...and compare entire string when "@" is missing.

Something like:

         uclass_foreach_dev(dev, uc) {
-               if (!strcmp(dev->name, name)) {
+               char *at_char = strchr(dev->name, '@');
+               int name_length;

+
+ if (at_char)
+ name_length = at_char - dev->name;
+ else
+ name_length = max(strlen(dev->name), strlen(name));
+

+               if (!strncmp(dev->name, name, name_length)) {
                         *devp = dev;
                         return 0;
                 }

I could probably to prepare an RFC with this change.

Regards,
Nandor

#3 lore raw

Hi Nandor,

Quoted 6 lines · Nandor Han

Hi Simon and thanks,

On 2019-12-24 17:58, Simon Glass wrote:

Quoted 8 lines

<snip>

Quoted 4 lines

Can you use a phandle to refer to the device from another node, perhaps a UCLASS_BOARD node if needed? It is not a good idea to use this function.

I'm not sure, I'll have to check that.

OK, that is the preferred method.

Quoted 27 lines

Anyway
uclass_find_device_by_name is used by uclass_get_device_by_name, which already seems to be used in plenty of places.

Quoted 2 lines

Alternatively we could add a new function which finds by partial name.

My fix was to ignore the suffix starting from "@", if that exist, since that's more like metadata...and compare entire string when "@" is missing.

Something like:

         uclass_foreach_dev(dev, uc) {
(quoted diff) +9 -1
-               if (!strcmp(dev->name, name)) {
+               char *at_char = strchr(dev->name, '@');
+               int name_length;
+
+               if (at_char)
+                       name_length = at_char - dev->name;
+               else
+                       name_length = max(strlen(dev->name), strlen(name));
+
+               if (!strncmp(dev->name, name, name_length)) {
                         *devp = dev;
                         return 0;
                 }
Quoted 1 line

I could probably to prepare an RFC with this change.

I think this is OK as a new function, but do add a comment in the header that it should not be used and phandles should be used instead. Also please remember to add a test.

Regards,
Simon