Re: [PATCH] of: cache phandle nodes to decrease cost of of_find_node_by_phandle()

[Frank Rowand]

On 01/31/18 22:45, Chintan Pandya wrote:
> 
> 
> On 2/1/2018 1:35 AM, frowand.list@xxxxxxxxx wrote:
>> From: Frank Rowand <frank.rowand@xxxxxxxx>
> 
>> +
>> +static void of_populate_phandle_cache(void)
>> +{
>> +ÂÂÂ unsigned long flags;
>> +ÂÂÂ phandle max_phandle;
>> +ÂÂÂ u32 nodes = 0;
>> +ÂÂÂ struct device_node *np;
>> +
>> +ÂÂÂ if (phandle_cache)
>> +ÂÂÂÂÂÂÂ return;
>> +
>> +ÂÂÂ max_phandle = live_tree_max_phandle();
>> +
>> +ÂÂÂ raw_spin_lock_irqsave(&devtree_lock, flags);
>> +
>> +ÂÂÂ for_each_of_allnodes(np)
>> +ÂÂÂÂÂÂÂ nodes++;
>> +
>> +ÂÂÂ /* sanity cap for malformed tree */
>> +ÂÂÂ if (max_phandle > nodes)
>> +ÂÂÂÂÂÂÂ max_phandle = nodes;
> Shouldn't we speak up about this in kernel log ? May be WARN_ON() ?

Probably not.  If we care enough about a hand coded phandle property
value we should add a check to checkpatch and/or dtc instead of adding
the warning here.


>> +
>> +ÂÂÂ phandle_cache = kzalloc((max_phandle + 1) * sizeof(*phandle_cache),
>> +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ GFP_KERNEL);
> kzalloc (might_sleep) in critical context will break.

Yes, thanks.

I also need to ensure memory ordering in of_free_phandle_cache()
to ensure that max_phandle_cache is zero before the cache memory
is freed.


> Anyways, will fix this locally and share test results.

Thanks, I look forward to the results.


> Thanks,
> Chintan
>