Re: [PATCH v3 2/3] platform/x86: think-lmi: Add possible_values for ThinkStation

["Mark Pearson" <mpearson-lenovo@squebb.ca>]

Thanks Thomas

On Sat, Mar 18, 2023, at 12:35 PM, Thomas Weißschuh wrote:
> Hi Mark,
>
> please also CC linux-kernel@vger.kernel.org and previous reviewers.
>
> On Fri, Mar 17, 2023 at 11:46:34AM -0400, Mark Pearson wrote:
>> ThinkStation platforms don't support the API to return possible_values
>> but instead embed it in the settings string.
>> 
>> Try and extract this information and set the possible_values attribute
>> appropriately.
>> 
>> If there aren't any values possible then don't display possible_values.
>> 
>> Signed-off-by: Mark Pearson <mpearson-lenovo@squebb.ca>
>> ---
>> Changes in V3:
>>  - Use is_visible attribute to determine if possible_values should be
>>    available
>>  - Code got refactored a bit to make compilation cleaner
>> Changes in V2:
>>  - Move no value for possible_values handling into show function
>>  - use kstrndup for allocating string
>> 
>>  drivers/platform/x86/think-lmi.c | 82 ++++++++++++++++++++++----------
>>  1 file changed, 56 insertions(+), 26 deletions(-)
>> 
>> diff --git a/drivers/platform/x86/think-lmi.c b/drivers/platform/x86/think-lmi.c
>> index 5fa5451c4802..d89a1c9bdbf1 100644
>> --- a/drivers/platform/x86/think-lmi.c
>> +++ b/drivers/platform/x86/think-lmi.c
>> @@ -917,6 +917,8 @@ static ssize_t display_name_show(struct kobject *kobj, struct kobj_attribute *at
>>  	return sysfs_emit(buf, "%s\n", setting->display_name);
>>  }
>>  
>> +static struct kobj_attribute attr_displ_name = __ATTR_RO(display_name);
>> +
>>  static ssize_t current_value_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
>>  {
>>  	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> @@ -937,30 +939,6 @@ static ssize_t current_value_show(struct kobject *kobj, struct kobj_attribute *a
>>  	return ret;
>>  }
>>  
>> -static ssize_t possible_values_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
>> -{
>> -	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> -
>> -	if (!tlmi_priv.can_get_bios_selections)
>> -		return -EOPNOTSUPP;
>> -
>> -	return sysfs_emit(buf, "%s\n", setting->possible_values);
>> -}
>> -
>> -static ssize_t type_show(struct kobject *kobj, struct kobj_attribute *attr,
>> -		char *buf)
>> -{
>> -	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> -
>> -	if (setting->possible_values) {
>> -		/* Figure out what setting type is as BIOS does not return this */
>> -		if (strchr(setting->possible_values, ','))
>> -			return sysfs_emit(buf, "enumeration\n");
>> -	}
>> -	/* Anything else is going to be a string */
>> -	return sysfs_emit(buf, "string\n");
>> -}
>> -
>>  static ssize_t current_value_store(struct kobject *kobj,
>>  		struct kobj_attribute *attr,
>>  		const char *buf, size_t count)
>> @@ -1044,14 +1022,46 @@ static ssize_t current_value_store(struct kobject *kobj,
>>  	return ret ?: count;
>>  }
>>  
>> -static struct kobj_attribute attr_displ_name = __ATTR_RO(display_name);
>> +static struct kobj_attribute attr_current_val = __ATTR_RW_MODE(current_value, 0600);
>> +
>> +static ssize_t possible_values_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
>> +{
>> +	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> +
>> +	return sysfs_emit(buf, "%s\n", setting->possible_values);
>> +}
>>  
>>  static struct kobj_attribute attr_possible_values = __ATTR_RO(possible_values);
>>  
>> -static struct kobj_attribute attr_current_val = __ATTR_RW_MODE(current_value, 0600);
>> +static ssize_t type_show(struct kobject *kobj, struct kobj_attribute *attr,
>> +		char *buf)
>> +{
>> +	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> +
>> +	if (setting->possible_values) {
>> +		/* Figure out what setting type is as BIOS does not return this */
>> +		if (strchr(setting->possible_values, ','))
>> +			return sysfs_emit(buf, "enumeration\n");
>> +	}
>> +	/* Anything else is going to be a string */
>> +	return sysfs_emit(buf, "string\n");
>> +}
>
> This patch seems to introduce a lot of churn, is it intentional?
Yes(ish). It got cleaned up as the functions were in a weird order when I introduced the is_visible. The actual changes are very small - but it did make it look messier than it really is.
Is this a big concern? I know it makes the review a bit more painful and my apologies for that.

>>  
>>  static struct kobj_attribute attr_type = __ATTR_RO(type);
>>  
>> +static umode_t attr_is_visible(struct kobject *kobj,
>> +					     struct attribute *attr, int n)
>> +{
>> +	struct tlmi_attr_setting *setting = to_tlmi_attr_setting(kobj);
>> +
>> +	/* We don't want to display possible_values attributes if not available */
>> +	if (attr == (struct attribute *)&attr_possible_values)
>
> This cast is unsafe, if the struct kobj_attribute order is randomised it
> will break.
>
> You can use
>
> 	if (attr == &attr_possible_values.attr)
>
Ack. Will change.

>> +		if (!setting->possible_values)
>> +			return 0;
>> +
>> +	return attr->mode;
>> +}
>> +
>>  static struct attribute *tlmi_attrs[] = {
>>  	&attr_displ_name.attr,
>>  	&attr_current_val.attr,
>> @@ -1061,6 +1071,7 @@ static struct attribute *tlmi_attrs[] = {
>>  };
>>  
>>  static const struct attribute_group tlmi_attr_group = {
>> +	.is_visible = attr_is_visible,
>>  	.attrs = tlmi_attrs,
>>  };
>>  
>> @@ -1440,6 +1451,25 @@ static int tlmi_analyze(void)
>>  			if (ret || !setting->possible_values)
>>  				pr_info("Error retrieving possible values for %d : %s\n",
>>  						i, setting->display_name);
>> +		} else {
>> +			/*
>> +			 * Older Thinkstations don't support the bios_selections API.
>> +			 * Instead they store this as a [Optional:Option1,Option2] section of the
>> +			 * name string.
>> +			 * Try and pull that out if it's available.
>> +			 */
>> +			char *item, *optstart, *optend;
>> +
>> +			if (!tlmi_setting(setting->index, &item, LENOVO_BIOS_SETTING_GUID)) {
>> +				optstart = strstr(item, "[Optional:");
>> +				if (optstart) {
>> +					optstart += strlen("[Optional:");
>> +					optend = strstr(optstart, "]");
>> +					if (optend)
>> +						setting->possible_values =
>> +							kstrndup(optstart, optend - optstart, GFP_KERNEL);
>> +				}
>> +			}
>
> The patch now does two things:
> 1) Hide the sysfs attributes if the value is not available
> 2) Extract the value from the description
>
> Maybe it could be split in two?
Sure. I did contemplate that and then ultimately decided it was all from the same intent so left it. But I can split.

>
> Another observation:
> Would it make sense to remove the part
> "[Optional:Option1,Option2]" from the name attribute?
>
I considered this previously and I was concerned about if this could have impacts that I couldn't foresee. The BIOS teams do strange things with this string so I was playing safe and leaving it alone (especially as it differs across the different portfolios)

I know it would be nice to have one standard for everything but sadly that's not the case, and not a battle I can win.

>>  		}
>>  		kobject_init(&setting->kobj, &tlmi_attr_setting_ktype);
>>  		tlmi_priv.setting[i] = setting;
>> -- 
>> 2.39.2
>>