Re: [RFC PATCH net-next 2/4] switchdev: add fwd_mark generator helper

[Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>]

Hello.

On 6/15/2015 4:52 PM, Scott Feldman wrote:

>>>>> From: Scott Feldman <sfeldma@gmail.com>

>>>>> skb->fwd_mark and dev->fwd_mark are 32-bit and should be unique for device
>>>>> and maybe even unique for a sub-set of ports within device, so add
>>>>> switchdev helper function to generate unique marks based on driver-supplied
>>>>> key.  Typically, the driver would use device switch ID for key, and maybe
>>>>> additional fields in key for grouped ports such as bridge ifindex.  The key
>>>>> can be of arbitrary length.

>>>>> The generator uses a global hash table to store fwd_marks hashed by key.

>>>>> Signed-off-by: Scott Feldman <sfeldma@gmail.com>

>>> <snip>

>>>>> +u32 switchdev_mark_get(void *key, size_t key_len)
>>>>> +{
>>>>> +      struct switchdev_mark_ht_entry {
>>>>> +              struct hlist_node entry;
>>>>> +              void *key;
>>>>> +              size_t key_len;
>>>>> +              u32 key_crc32;
>>>>> +              u32 mark;
>>>>> +      } *entry;
>>>>> +      u32 key_crc32 = crc32(~0, key, key_len);
>>>>> +      u32 mark = 0;
>>>>> +      unsigned long flags;
>>>>> +
>>>>> +      spin_lock_irqsave(&switchdev_mark_lock, flags);

>>>> I fail to see why _irqsave variant is needed here.

>>> I don't know what context caller is in, so using most conservative
>>> spinlock.  Is there a better way?

>> I don't see why would someone call this from irq.

> Ok, good point, I'll adjust to spin_lock.

     I guess spi_lock_irq() is what you meant. Disabling IRQs when called from 
the hardirq context made no sense since hardirq handlrs are executed with IRQs 
disabled anyway.

WBR, Sergei