[RFC 1/6] dtoc: add support to scan drivers

[Walter Lozano]

Hi Simon,

On 20/5/20 00:07, Simon Glass wrote:
> Hi Walter,
>
> On Wed, 13 May 2020 at 14:13, Walter Lozano<walter.lozano at collabora.com>  wrote:
>> Currently dtoc scans dtbs to convert them to struct platdata and
>> to generate U_BOOT_DEVICE entries. These entries need to be filled
>> with the driver name, but at this moment the information used is the
>> compatible name present in the dtb. This causes that only nodes with
>> a compatible name that matches a driver name generate a working
>> entry.
>>
>> In order to improve this behaviour, this patch adds to dtoc the
>> capability of scan drivers source code to generate a list of valid driver
>> names. This allows to rise a warning in the case that an U_BOOT_DEVICE
>> entry will try to use a name not valid.
>>
>> Additionally, in order to add more flexibility to the solution, adds the
>> U_BOOT_DRIVER_ALIAS macro, which generates no code at all, but allows an
>> easy way to declare driver name aliases. Thanks to this, dtoc can look
>> for the driver name based on its alias when it populates the U_BOOT_DEVICE entry.
>>
>> Signed-off-by: Walter Lozano<walter.lozano at collabora.com>
>> ---
>>   include/dm/device.h        |  6 +++++
>>   tools/dtoc/dtb_platdata.py | 53 +++++++++++++++++++++++++++++++++++---
>>   2 files changed, 55 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/dm/device.h b/include/dm/device.h
>> index 975eec5d0e..d5b3e732c3 100644
>> --- a/include/dm/device.h
>> +++ b/include/dm/device.h
>> @@ -282,6 +282,12 @@ struct driver {
>>   #define DM_GET_DRIVER(__name)                                          \
>>          ll_entry_get(struct driver, __name, driver)
>>
>> +/* Declare a macro to state a alias for a driver name. This macro will
> /*
>   * Declare ...
>
Sure.

>> + * produce no code but its information will be parsed by tools like
>> + * dtoc
>> + */
>> +#define U_BOOT_DRIVER_ALIAS(__name, __alias)
>> +
>>   /**
>>    * dev_get_platdata() - Get the platform data for a device
>>    *
>> diff --git a/tools/dtoc/dtb_platdata.py b/tools/dtoc/dtb_platdata.py
>> index ecfe0624d1..3f899a8bac 100644
>> --- a/tools/dtoc/dtb_platdata.py
>> +++ b/tools/dtoc/dtb_platdata.py
>> @@ -14,6 +14,8 @@ static data.
>>   import collections
>>   import copy
>>   import sys
>> +import os
>> +import re
> please keep sorted


Sure

>>   from dtoc import fdt
>>   from dtoc import fdt_util
>> @@ -149,6 +151,20 @@ class DtbPlatdata(object):
>>           self._outfile = None
>>           self._lines = []
>>           self._aliases = {}
>> +        self._drivers = []
>> +        self._driver_aliases = {}
> Please update comment above for what these new things are for.
OK.
>> +
>> +    def get_normalized_compat_name(self, node):
> Needs function comment (also for those below).
Sure, no problem.
>> +        compat_c, aliases_c = get_compat_name(node)
>> +        if compat_c not in self._drivers:
>> +            try: # pragma: no cover
> Instead of an exception can you use .get() and check for None?
Sure.
>> +                compat_c_old = compat_c
>> +                compat_c = self._driver_aliases[compat_c]
>> +                aliases_c = [compat_c_old] + aliases
>> +            except:
>> +                print('WARNING: the driver %s was not found in the driver list' % (compat_c))

This is something I would like to check with you. As you can see, a 
warning is raised when a driver name is not found, which I think is the 
right thing to do at this point. However, this warning will be trigger 
but almost all the tests as the drivers names are no valid (there is no 
U_BOOT_DRIVER). How do you think is best to fix this?

I see different possibilities here

- To add specific driver names to self._drivers when running a test

- To change driver names on tests to match a valid one

Maybe you have a more clean way to deal with this.

>> +
>> +        return compat_c, aliases_c
>>
>>       def setup_output(self, fname):
>>           """Set up the output destination
>> @@ -243,6 +259,34 @@ class DtbPlatdata(object):
>>               return PhandleInfo(max_args, args)
>>           return None
>>
>> +    def scan_driver(self, fn):
>> +        f = open(fn)
> Can you avoid single-char var names? Also how about either using
> tools.ReadFile() or:
>
> with open(fname) as fd:
>      ...rest of function
Agree.
>> +
>> +        b = f.read()
>> +
>> +        drivers = re.findall('U_BOOT_DRIVER\((.*)\)', b)
>> +
>> +        for d in drivers:
>> +            self._drivers.append(d)
>> +
>> +        driver_aliases = re.findall('U_BOOT_DRIVER_ALIAS\(\s*(\w+)\s*,\s*(\w+)\s*\)', b)
> Can you add a comment before this with an example of the line you are
> parsing? It helps make sense of the regex.
No problem.
>> +
>> +        for a in driver_aliases: # pragma: no cover
>> +            try:
>> +                self._driver_aliases[a[1]] = a[0]
>> +            except:
>> +                pass
> Again please avoid exceptions.
No problem at all.
>> +
>> +    def scan_drivers(self):
>> +        """Scan the driver folders to build a list of driver names and possible
>> +        aliases
>> +        """
>> +        for (dirpath, dirnames, filenames) in os.walk('./'):
>> +            for fn in filenames:
>> +                if not fn.endswith('.c'):
>> +                    continue
>> +                self.scan_driver(dirpath + '/' + fn)
>> +
>>       def scan_dtb(self):
>>           """Scan the device tree to obtain a tree of nodes and properties
>>
>> @@ -353,7 +397,7 @@ class DtbPlatdata(object):
>>           """
>>           structs = {}
>>           for node in self._valid_nodes:
>> -            node_name, _ = get_compat_name(node)
>> +            node_name, _ = self.get_normalized_compat_name(node)
>>               fields = {}
> I wonder how long this scan takes? Should we cache the results somewhere?

Probably you are worried because this task will be repeated every time 
we compile U-boot, and it could be a potential issue when testing 
several different configs. However, at least in my setup the scan takes 
less than 1 sec, which I think is promising. Also I think that this scan 
is only perform for those configs which enable OF_PLATDATA, which are 
15, should not cause much overhead.

What do you think?

>>               # Get a list of all the valid properties in this node.
>> @@ -377,14 +421,14 @@ class DtbPlatdata(object):
>>
>>           upto = 0
>>           for node in self._valid_nodes:
>> -            node_name, _ = get_compat_name(node)
>> +            node_name, _ = self.get_normalized_compat_name(node)
>>               struct = structs[node_name]
>>               for name, prop in node.props.items():
>>                   if name not in PROP_IGNORE_LIST and name[0] != '#':
>>                       prop.Widen(struct[name])
>>               upto += 1
>>
>> -            struct_name, aliases = get_compat_name(node)
>> +            struct_name, aliases = self.get_normalized_compat_name(node)
>>               for alias in aliases:
>>                   self._aliases[alias] = struct_name
>>
>> @@ -461,7 +505,7 @@ class DtbPlatdata(object):
>>           Args:
>>               node: node to output
>>           """
>> -        struct_name, _ = get_compat_name(node)
>> +        struct_name, _ = self.get_normalized_compat_name(node)
>>           var_name = conv_name_to_c(node.name)
>>           self.buf('static const struct %s%s %s%s = {\n' %
>>                    (STRUCT_PREFIX, struct_name, VAL_PREFIX, var_name))
>> @@ -562,6 +606,7 @@ def run_steps(args, dtb_file, include_disabled, output):
>>           raise ValueError('Please specify a command: struct, platdata')
>>
>>       plat = DtbPlatdata(dtb_file, include_disabled)
>> +    plat.scan_drivers()
>>       plat.scan_dtb()
>>       plat.scan_tree()
>>       plat.scan_reg_sizes()
>> --
>> 2.20.1
>>
Regards,

Walter