Re: [RFC PATCH] x86: explicit call to mmiotrace in do_page_fault()

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Sat, 2008-02-09 at 19:52 +0200, Pekka Paalanen wrote:

> +int mmiotrace_register_pf(pf_handler_func new_pfh)
>  {
> +	int ret = 0;
>  	unsigned long flags;

> +	spin_lock_irqsave(&mmiotrace_handler_lock, flags);
> +	if (mmiotrace_pf_handler)
> +		ret = -EBUSY;
> +	else
> +		mmiotrace_pf_handler = new_pfh;
> +	spin_unlock_irqrestore(&mmiotrace_handler_lock, flags);
> +	return ret;
>  }
> +EXPORT_SYMBOL_GPL(mmiotrace_register_pf);
>  
>  /**
> + * mmiotrace_unregister_pf:
>   * The caller must ensure @old_pfh is not in use anymore before freeing it.
> + * This function does not guarantee it. The handler function pointer is
> + * protected by RCU, so you can do this by e.g. calling synchronize_rcu().
>   */
> +int mmiotrace_unregister_pf(pf_handler_func old_pfh)
>  {
> +	int ret = 0;
>  	unsigned long flags;

> +	spin_lock_irqsave(&mmiotrace_handler_lock, flags);
> +	if (mmiotrace_pf_handler != old_pfh)
> +		ret = -EPERM;
> +	else
> +		mmiotrace_pf_handler = NULL;
> +	spin_unlock_irqrestore(&mmiotrace_handler_lock, flags);
> +	return ret;
>  }

So here you set/unset mmiotrace_pf_handler under protection of a
spinlock.

 
>  /* returns non-zero if do_page_fault() should return */
> +static inline int call_mmiotrace(struct pt_regs *regs,
> +					unsigned long error_code,
> +					unsigned long address)
>  {
> +#ifdef CONFIG_MMIOTRACE_HOOKS
>  	int ret = 0;
>  	rcu_read_lock();
> +	if (mmiotrace_pf_handler)
> +		ret = mmiotrace_pf_handler(regs, error_code, address);
>  	rcu_read_unlock();
>  	return ret;
>  #else

But here you use it under rcu_read_lock(). This doesn't match.