[committed] hppa: Fix memory barrier patterns for pre PA8800 processors

Message ID ZRb2k9qUdH1lgaxh@mx3210.localdomain
State Accepted
Headers
Series [committed] hppa: Fix memory barrier patterns for pre PA8800 processors |

Checks

Context Check Description
snail/gcc-patch-check success Github commit url

Commit Message

John David Anglin Sept. 29, 2023, 4:08 p.m. UTC
  Since 2005, it was assumed in the Linux kernel that all PA 2.0 processors
supported the ldcw cacheable hint and that natural alignment could be used
for ldcw,co.  However, I recently fired up an old A500 machine with PA8600
processors and found that 16-byte alignment was needed for ldcw,co on it.
As far as I can tell, only PA8800 and PA8900 processors support the
cacheable hint.

This change revises the memory barrier patterns. We alway use ldcw,co
when comp[iling for PA 2.0 but we disable the -mcoherent-ldcw option
by default. As a result, the 16-byte aligned patterns are now the default.
This is safer but slightly less efficient.
 
Dave
---

Fix memory barrier patterns for pre PA8800 processors

2023-09-29  John David Anglin  <danglin@gcc.gnu.org>

	* config/pa/pa.md (memory_barrier): Revise comment.
	(memory_barrier_64, memory_barrier_32): Use ldcw,co on PA 2.0.
	* config/pa/pa.opt (coherent-ldcw): Change default to disabled.
  

Patch

diff --git a/gcc/config/pa/pa.md b/gcc/config/pa/pa.md
index f603591447d..4f85991e6bd 100644
--- a/gcc/config/pa/pa.md
+++ b/gcc/config/pa/pa.md
@@ -10739,10 +10739,10 @@  add,l %2,%3,%3\;bv,n %%r0(%3)"
 ;; generating PA 1.x code even though all PA 1.x systems are strongly ordered.
 
 ;; When barriers are needed, we use a strongly ordered ldcw instruction as
-;; the barrier.  Most PA 2.0 targets are cache coherent.  In that case, we
-;; can use the coherent cache control hint and avoid aligning the ldcw
-;; address.  In spite of its description, it is not clear that the sync
-;; instruction works as a barrier.
+;; the barrier.  All PA 2.0 targets accept the "co" cache control hint but
+;; only PA8800 and PA8900 processors implement the cacheable hint.  In
+;; that case, we can avoid aligning the ldcw address.  In spite of its
+;; description, it is not clear that the sync instruction works as a barrier.
 
 (define_expand "memory_barrier"
   [(parallel
@@ -10772,7 +10772,7 @@  add,l %2,%3,%3\;bv,n %%r0(%3)"
         (unspec:BLK [(match_dup 0)] UNSPEC_MEMORY_BARRIER))
     (clobber (match_operand 1 "pmode_register_operand" "=&r"))]
   "TARGET_64BIT"
-  "ldo 15(%%sp),%1\n\tdepd %%r0,63,3,%1\n\tldcw 0(%1),%1"
+  "ldo 15(%%sp),%1\n\tdepd %%r0,63,3,%1\n\tldcw,co 0(%1),%1"
   [(set_attr "type" "binary")
    (set_attr "length" "12")])
 
@@ -10781,6 +10781,6 @@  add,l %2,%3,%3\;bv,n %%r0(%3)"
         (unspec:BLK [(match_dup 0)] UNSPEC_MEMORY_BARRIER))
     (clobber (match_operand 1 "pmode_register_operand" "=&r"))]
   ""
-  "ldo 15(%%sp),%1\n\t{dep|depw} %%r0,31,3,%1\n\tldcw 0(%1),%1"
+  "ldo 15(%%sp),%1\n\t{dep|depw} %%r0,31,3,%1\n\t{ldcw|ldcw,co} 0(%1),%1"
   [(set_attr "type" "binary")
    (set_attr "length" "12")])
diff --git a/gcc/config/pa/pa.opt b/gcc/config/pa/pa.opt
index dd358f2f26a..573edcea338 100644
--- a/gcc/config/pa/pa.opt
+++ b/gcc/config/pa/pa.opt
@@ -50,7 +50,7 @@  Target Mask(CALLER_COPIES)
 Caller copies function arguments passed by hidden reference.
 
 mcoherent-ldcw
-Target Var(TARGET_COHERENT_LDCW) Init(1)
+Target Var(TARGET_COHERENT_LDCW) Init(0)
 Use ldcw/ldcd coherent cache-control hint.
 
 mdisable-fpregs