sergiotarxz/nuttx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

xtensa_testset.c: Simplify the test-set function and remove some old

Browse Source 
comments.

Signed-off-by: Abdelatif Guettouche <abdelatif.guettouche@espressif.com>
This commit is contained in:
Abdelatif Guettouche 2021-10-13 10:32:24 +02:00 committed by Xiang Xiao
parent dfe1637864
commit eeb68bda3d
1 changed files with 4 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
arch/xtensa/src/common/xtensa_testset.c
View File

@ -56,8 +56,9 @@ static inline uint32_t xtensa_compareset(volatile uint32_t *addr,
__asm__ __volatile__
(
"WSR %2, SCOMPARE1\n" /* Initialize SCOMPARE1 */
"S32C1I %0, %1, 0\n" /* Store id into the lock, if the lock is the
* same as comparel. Otherwise, no write-access */
"S32C1I %0, %1, 0\n" /* Store the compare value into the lock,
* if the lock is the same as compare1.
* Otherwise, no write-access */
: "=r"(set) : "r"(addr), "r"(compare), "0"(set)
);
@ -90,36 +91,10 @@ static inline uint32_t xtensa_compareset(volatile uint32_t *addr,
spinlock_t up_testset(volatile spinlock_t *lock)
{
spinlock_t prev;
/* Perform the 32-bit compare and set operation */
prev = xtensa_compareset((volatile uint32_t *)lock,
return xtensa_compareset((volatile uint32_t *)lock,
SP_UNLOCKED, SP_LOCKED);
/* xtensa_compareset() should return either SP_UNLOCKED if the spinlock
* was locked or SP_LOCKED or possibly ~SP_UNLOCKED if the spinlock was
* not locked:
*
* "In the RE-2013.0 release and after, there is a slight change in the
* semantics of the S32C1I instruction. Nothing is changed about the
* operation on memory. In rare cases the resulting value in register
* at can be different in this and later releases. The rule still holds
* that memory has been written if and only if the register result
* equals SCOMPARE1.
*
* "The difference is that in some cases where memory has not been
* written, the instruction returns ~SCOMPARE1 instead of the current
* value of memory. Although this change can, in principle, affect
* the operation of code, scanning all internal Cadence code produced
* no examples where this change would change the operation of the
* code."
*
* In any case, the return value of SP_UNLOCKED can be trusted and will
* always mean that the spinlock was set.
*/
return (prev == SP_UNLOCKED) ? SP_UNLOCKED : SP_LOCKED;
}
#endif /* CONFIG_SPINLOCK */