[RFC PATCH 1/9] powerpc/pseries: add cpu DLPAR support for drc-info property

[Tyrel Datwyler]

On 10/10/19 11:56 AM, Nathan Lynch wrote:
> Hi Tyrel,
> 
> Tyrel Datwyler <tyreld at linux.ibm.com> writes:
>> +static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
>> +{
>> +	const __be32 *indexes;
>> +	int i;
>> +
>> +	if (of_find_property(parent, "ibm,drc-info", NULL))
>> +		return drc_info_valid_index(parent, drc_index);
>> +
>> +	indexes = of_get_property(parent, "ibm,drc-indexes", NULL);
>> +	if (!indexes)
>> +		return false;
>> +
>> +	for (i = 0; i < indexes[0]; i++) {
> 
> should this be:
> 
>         for (i = 0; i < be32_to_cpu(indexes[0]); i++) {
> ?

Yes!

> 
> 
>> +		if (be32_to_cpu(indexes[i + 1]) == drc_index)
>> +			return true;
>> +	}
>> +
>> +	return false;
>>  }
> 
> It looks like this rewrites valid_cpu_drc_index()'s existing code for
> parsing ibm,drc-indexes but I don't see the need for this.
> 
> This patch would be easier to review if that were dropped or split out.

Yeah, I'll split it out. There are multiple places where we iterate over the
drc_indexes, and it is implemented several different ways. I basically picked an
implementation to use across the board. I think a better way would be just to
implement a for_each_drc_index(dn, drc_index) macro to abstract away iterator
implementation.

> 
>>  
>>  static ssize_t dlpar_cpu_add(u32 drc_index)
>> @@ -720,8 +756,11 @@ static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
>>  static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
>>  {
>>  	struct device_node *parent;
>> +	struct property *info;
>> +	const __be32 *indexes;
>>  	int cpus_found = 0;
>> -	int index, rc;
>> +	int i, j;
>> +	u32 drc_index;
>>  
>>  	parent = of_find_node_by_path("/cpus");
>>  	if (!parent) {
>> @@ -730,24 +769,46 @@ static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
>>  		return -1;
>>  	}
>>  
>> -	/* Search the ibm,drc-indexes array for possible CPU drcs to
>> -	 * add. Note that the format of the ibm,drc-indexes array is
>> -	 * the number of entries in the array followed by the array
>> -	 * of drc values so we start looking at index = 1.
>> -	 */
>> -	index = 1;
>> -	while (cpus_found < cpus_to_add) {
>> -		u32 drc;
>> +	info = of_find_property(parent, "ibm,drc-info", NULL);
>> +	if (info) {
>> +		struct of_drc_info drc;
>> +		const __be32 *value;
>> +		int count;
>>  
>> -		rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
>> -						index++, &drc);
>> -		if (rc)
>> -			break;
>> +		value = of_prop_next_u32(info, NULL, &count);
>> +		if (value)
>> +			value++;
>>  
>> -		if (dlpar_cpu_exists(parent, drc))
>> -			continue;
>> +		for (i = 0; i < count; i++) {
>> +			of_read_drc_info_cell(&info, &value, &drc);
>> +			if (strncmp(drc.drc_type, "CPU", 3))
>> +				break;
>> +
>> +			for (j = 0; j < drc.num_sequential_elems; j++) {
>> +				drc_index = drc.drc_index_start + (drc.sequential_inc * j);
>> +
>> +				if (dlpar_cpu_exists(parent, drc_index))
>> +					continue;
>>  
>> -		cpu_drcs[cpus_found++] = drc;
>> +				cpu_drcs[cpus_found++] = drc_index;
> 
> I am failing to see how this loop is limited by the cpus_to_add
> parameter as it was before this change. It looks like this will overflow
> the cpu_drcs array when cpus_to_add is less than the number of cpus
> found.

You are right. The code is picking every non-present drc_index which will
overflow the supplied buffer as you stated when there are more available indexes
than requested cpus. Will fix to bound the search.

> 
> As an aside I don't understand how the add_by_count()/dlpar_cpu_exists()
> algorithm could be correct as it currently stands. It seems to pick the
> first X indexes for which a corresponding cpu node is absent, but that
> set of indexes does not necessarily match the set that is available to
> configure. Something to address separately I suppose.

I'm not sure I follow?

> 
>> +			}
>> +		}
>> +	} else {
>> +		indexes = of_get_property(parent, "ibm,drc-indexes", NULL);
>> +
>> +		/* Search the ibm,drc-indexes array for possible CPU drcs to
>> +	 	* add. Note that the format of the ibm,drc-indexes array is
>> +	 	* the number of entries in the array followed by the array
>> +	 	* of drc values so we start looking at index = 1.
>> +	 	*/
>> +		for (i = 1; i < indexes[0]; i++) {
>> +			drc_index = be32_to_cpu(indexes[i]);
>> +
>> +			if (dlpar_cpu_exists(parent, drc_index))
>> +				continue;
>> +
>> +			cpu_drcs[cpus_found++] = drc_index;
>> +		}
>>  	}
> 
> As above, not sure why this was rewritten, and similar comments as
> before apply.
> 

Again, wanted to use a single implementation everywere. Obviously, as pointed
out in the previous comment missed a byte swap. Will split out into a separate
patch for consideration.

-Tyrel