Re: [PATCH] Fix deadlock in RPC scheduling code.

[Aurelien Degremont <aurelien.degremont@cea.fr>]

Trond Myklebust wrote:
>>Hmmm... With those constraints, it seems difficult to be able to modify 
>>the current rpc_wake_up_task() function...
> That is the price of optimisation in this case.

Optimisation against potential race (see below) ?

> Your deadlock problem resulted in __rpc_wake_up_task() iterating forever
> on the same task since the while() loop would not ever exit before it
> was empty. By changing the iteration scheme into one where we only try
> to wake up each task once, we allow rpc_wake_up()/rpc_wake_up_status()
> to complete.

IMO, the code will still hang because, the list_for_each_entry_safe() 
loop will never complete, even with the new scheme. Indeed, the loop 
call __rpc_wake_up_task() will try to set RPC_TASK_WAKEUP bit, but it is 
already set by rpc_wake_up_task(), so it hangs...

A: run rpc_wake_up_task(task foo)
A: set RPC_TASK_WAKEUP bit
--interrupt--
B: run rpc_wake_up()
B: grab queue->lock
B: enter the list_for_each_entry_safe() loop
B: run __rpc_wake_up_task(task foo)
B: wait for RPC_TASK_WAKEUP bit
A: wait for queue->lock

  -> deadlock ?

-- 
Aurelien Degremont