svn commit: r357728 - head/sys/sys

[Gleb Smirnoff]

On Mon, Feb 10, 2020 at 01:52:26PM +0000, Mateusz Guzik wrote:
M> Author: mjg
M> Date: Mon Feb 10 13:52:25 2020
M> New Revision: 357728
M> URL: https://svnweb.freebsd.org/changeset/base/357728
M> 
M> Log:
M>   Tidy up zpcpu_replace*
M>   
M>   - only compute the target address once
M>   - remove spurious type casting, zpcpu_get already return the correct type
M>   
M>   While here add missing newlines to other routines.
M> 
M> Modified:
M>   head/sys/sys/pcpu.h
M> 
M> Modified: head/sys/sys/pcpu.h
M> ==============================================================================
M> --- head/sys/sys/pcpu.h	Mon Feb 10 13:24:14 2020	(r357727)
M> +++ head/sys/sys/pcpu.h	Mon Feb 10 13:52:25 2020	(r357728)
M> @@ -245,32 +245,41 @@ extern struct pcpu *cpuid_to_pcpu[];
M>   * If you need atomicity use xchg.
M>   * */
M>  #define zpcpu_replace(base, val) ({					\
M> -	__typeof(val) _old = *(__typeof(base))zpcpu_get(base);		\
M> -	*(__typeof(val) *)zpcpu_get(base) = val;			\
M> +	__typeof(val) *_ptr = zpcpu_get(base);				\
M> +	__typeof(val) _old;						\
M> +									\
M> +	_old = *_ptr;							\
M> +	*_ptr = val;							\
M>  	_old;								\
M>  })

I think this function must have only protected variant that asserts
that curthread->td_critnest is on.

zpcpu_get() sometimes can be used without critical section, when we
are fine with getting a value from a different CPU.

However, can't imagine a situation where migrating during a replace
operation is acceptable.

-- 
Gleb Smirnoff