git: e0363a08dc45 - stable/12 - atomic: Add atomic_cmpset_masked to powerpc and use it

From: Kyle Evans <kevans_at_FreeBSD.org>
Date: Thu, 07 Oct 2021 03:28:13 UTC
The branch stable/12 has been updated by kevans:

URL: https://cgit.FreeBSD.org/src/commit/?id=e0363a08dc453be7b67f38d8a9663ff41dde94ee

commit e0363a08dc453be7b67f38d8a9663ff41dde94ee
Author:     Justin Hibbits <jhibbits@FreeBSD.org>
AuthorDate: 2019-11-15 04:33:07 +0000
Commit:     Kyle Evans <kevans@FreeBSD.org>
CommitDate: 2021-10-07 03:27:27 +0000

    atomic: Add atomic_cmpset_masked to powerpc and use it
    
    Summary:
    This is a more optimal way of doing atomic_compset_masked() than the
    fallback in sys/_atomic_subword.h.  There's also an override for
    _atomic_fcmpset_masked_word(), which may or may not be necessary, and is
    unused for powerpc.
    
    (cherry picked from commit d0bdb11139424b9dcfe3b73cd5a003055382de03)
---
 sys/powerpc/include/atomic.h | 32 ++++++++++++++++++++++++++++++++
 sys/sys/_atomic_subword.h    |  4 ++++
 2 files changed, 36 insertions(+)

diff --git a/sys/powerpc/include/atomic.h b/sys/powerpc/include/atomic.h
index 788e7520d7bc..2510160a402a 100644
--- a/sys/powerpc/include/atomic.h
+++ b/sys/powerpc/include/atomic.h
@@ -612,6 +612,38 @@ atomic_cmpset_short(volatile u_short *p, u_short cmpval, u_short newval)
 
 	return (ret);
 }
+#else
+static __inline int
+atomic_cmpset_masked(uint32_t *p, uint32_t cmpval, uint32_t newval,
+    uint32_t mask)
+{
+	int		ret;
+	uint32_t	tmp;
+
+	__asm __volatile (
+		"1:\tlwarx %2, 0, %2\n\t"	/* load old value */
+		"and %0, %2, %7\n\t"
+		"cmplw %4, %0\n\t"		/* compare */
+		"bne- 2f\n\t"			/* exit if not equal */
+		"andc %2, %2, %7\n\t"
+		"or %2, %2, %5\n\t"
+		"stwcx. %2, 0, %3\n\t"      	/* attempt to store */
+		"bne- 1b\n\t"			/* spin if failed */
+		"li %0, 1\n\t"			/* success - retval = 1 */
+		"b 3f\n\t"			/* we've succeeded */
+		"2:\n\t"
+		"stwcx. %2, 0, %3\n\t"       	/* clear reservation (74xx) */
+		"li %0, 0\n\t"			/* failure - retval = 0 */
+		"3:\n\t"
+		: "=&r" (ret), "=m" (*p), "+&r" (tmp)
+		: "r" (p), "r" (cmpval), "r" (newval), "m" (*p),
+		  "r" (mask)
+		: "cr0", "memory");
+
+	return (ret);
+}
+
+#define	_atomic_cmpset_masked_word(a,o,v,m) atomic_cmpset_masked(a, o, v, m)
 #endif
 
 static __inline int
diff --git a/sys/sys/_atomic_subword.h b/sys/sys/_atomic_subword.h
index fd03ea6a6aed..08e54bf519a9 100644
--- a/sys/sys/_atomic_subword.h
+++ b/sys/sys/_atomic_subword.h
@@ -63,6 +63,7 @@
     ((((__uintptr_t)(p) % 4)) * NBBY)
 #endif
 
+#ifndef	_atomic_cmpset_masked_word
 /*
  * Pass these bad boys a couple words and a mask of the bits you care about,
  * they'll loop until we either succeed or fail because of those bits rather
@@ -92,7 +93,9 @@ _atomic_cmpset_masked_word(uint32_t *addr, uint32_t old, uint32_t val,
 
 	return (ret);
 }
+#endif
 
+#ifndef	_atomic_fcmpset_masked_word
 static __inline int
 _atomic_fcmpset_masked_word(uint32_t *addr, uint32_t *old, uint32_t val,
     uint32_t mask)
@@ -108,6 +111,7 @@ _atomic_fcmpset_masked_word(uint32_t *addr, uint32_t *old, uint32_t val,
 	*old = (*addr & ~mask) | *old;
 	return (atomic_fcmpset_32(addr, old, (*old & ~mask) | val));
 }
+#endif
 
 static __inline int
 atomic_cmpset_8(__volatile uint8_t *addr, uint8_t old, uint8_t val)