Build failures in -next due to 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'

Hi Peter,

I am seeing build failures in -next when trying to build arc / arcv2 targets.

arch/arc/include/asm/atomic.h:74:2: error: ?SCOND_FAIL_RETRY_VAR_DEF? undeclared arch/arc/include/asm/atomic.h:87:2: error: expected ?:? or ?)? before ?SCOND_FAIL_RETRY_ASM?

Problems seem to be caused by 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'. Both SCOND_FAIL_RETRY_VAR_DEF and SCOND_FAIL_RETRY_ASM are undefined.

Quoted 7 lines · Guenter Roeck

Hi Peter,

I am seeing build failures in -next when trying to build arc / arcv2 targets.

arch/arc/include/asm/atomic.h:74:2: error: ?SCOND_FAIL_RETRY_VAR_DEF? undeclared arch/arc/include/asm/atomic.h:87:2: error: expected ?:? or ?)? before ?SCOND_FAIL_RETRY_ASM?

Quoted 3 lines

Problems seem to be caused by 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'. Both SCOND_FAIL_RETRY_VAR_DEF and SCOND_FAIL_RETRY_ASM are undefined.

Crud, I messed up the rebase against the backoff reverts. Lemme go do fixups.

Quoted 8 lines · Peter Zijlstra

Quoted 7 lines · Guenter Roeck

Hi Peter,

I am seeing build failures in -next when trying to build arc / arcv2 targets.

arch/arc/include/asm/atomic.h:74:2: error: ?SCOND_FAIL_RETRY_VAR_DEF? undeclared arch/arc/include/asm/atomic.h:87:2: error: expected ?:? or ?)? before ?SCOND_FAIL_RETRY_ASM?

Quoted 6 lines

Quoted 3 lines

Problems seem to be caused by 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'. Both SCOND_FAIL_RETRY_VAR_DEF and SCOND_FAIL_RETRY_ASM are undefined.

Crud, I messed up the rebase against the backoff reverts. Lemme go do fixups.

I've misplaced my arc compiler, but does this make it go again?

---
  arch/arc/include/asm/atomic.h | 8 --------
  1 file changed, 8 deletions(-)

arch/arc/include/asm/atomic.h +0 -8

index bd9c51cb2bfd..4e3c1b6b0806 100644
@@ -71,7 +71,6 @@ static inline int atomic_##op##_return(int i, atomic_t *v)		\
  static inline int atomic_fetch_##op(int i, atomic_t *v)			\
  {									\
  	unsigned int val, orig;						\
-	SCOND_FAIL_RETRY_VAR_DEF                                        \
  									\
  	/*								\
  	 * Explicit full memory barrier needed before/after as		\

@@ -84,11 +83,8 @@ static inline int atomic_fetch_##op(int i, atomic_t *v)			\
  	"	" #asm_op " %[val], %[orig], %[i]	\n"		\
  	"	scond   %[val], [%[ctr]]		\n"		\
  	"						\n"		\
-	SCOND_FAIL_RETRY_ASM						\
-									\
  	: [val]	"=&r"	(val),						\
  	  [orig] "=&r" (orig)						\
-	  SCOND_FAIL_RETRY_VARS						\
  	: [ctr]	"r"	(&v->counter),					\
  	  [i]	"ir"	(i)						\
  	: "cc");							\

@@ -199,10 +195,6 @@ ATOMIC_OPS(andnot, &= ~, bic)
  ATOMIC_OPS(or, |=, or)
  ATOMIC_OPS(xor, ^=, xor)
 
-#undef SCOND_FAIL_RETRY_VAR_DEF
-#undef SCOND_FAIL_RETRY_ASM
-#undef SCOND_FAIL_RETRY_VARS
-
  #else /* CONFIG_ARC_PLAT_EZNPS */
 
  static inline int atomic_read(const atomic_t *v)

Quoted 5 lines

Quoted 2 lines

Crud, I messed up the rebase against the backoff reverts. Lemme go do fixups.

I've misplaced my arc compiler, but does this make it go again?

Looks like it, yes. Build still fails for another reason, but this problem is gone. I can not do any runtime tests, though, only build tests.

Guenter

Question; why does atomic_read() have a full memory clobber on arc?

Will thinks its because you don't use a memory constraint for *v.

Quoted 8 lines · Peter Zijlstra

Quoted 7 lines · Guenter Roeck

Hi Peter,

I am seeing build failures in -next when trying to build arc / arcv2 targets.

arch/arc/include/asm/atomic.h:74:2: error: ?SCOND_FAIL_RETRY_VAR_DEF? undeclared arch/arc/include/asm/atomic.h:87:2: error: expected ?:? or ?)? before ?SCOND_FAIL_RETRY_ASM?

Quoted 6 lines

Quoted 3 lines

Problems seem to be caused by 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'. Both SCOND_FAIL_RETRY_VAR_DEF and SCOND_FAIL_RETRY_ASM are undefined.

Crud, I messed up the rebase against the backoff reverts. Lemme go do fixups.

I've misplaced my arc compiler, but does this make it go again?

There's no arc compiler on korg's cross-building directory.

Quoted 1 line

I've misplaced my arc compiler, but does this make it go again?

There's no arc compiler on korg's cross-building directory.

Quoted 2 lines · Ingo Molnar

Quoted 1 line

I've misplaced my arc compiler, but does this make it go again?

There's no arc compiler on korg's cross-building directory.

Yeah, I'd requested that a few times...

However we do have ARC kernels building at atleast 2 places
1. Guenther's build service which sent this error report
2. kisskb build by Michale Ellerman

Although I don't know if it is possible to download cross tools from either of above.

Quoted 2 lines · Ingo Molnar

Quoted 1 line

I've misplaced my arc compiler, but does this make it go again?

There's no arc compiler on korg's cross-building directory.

Yeah, I'd requested that a few times...

However we do have ARC kernels building at atleast 2 places
1. Guenther's build service which sent this error report
2. kisskb build by Michale Ellerman

Although I don't know if it is possible to download cross tools from either of above.

@Peter, upstream buildroot allows you to build toolchain for ARC ! Trust me it does build :-)

I use segher's buildall script.

The buildall script I put to work and after about an hour I have however many of the 20+ targets that build with gcc.git and binutils.git cross compilers.

As you can see, arc doesn't even build upstream binutils :/ (binutils-2_26-branch as of today).

Quoted 2 lines · Peter Zijlstra

As you can see, arc doesn't even build upstream binutils :/ (binutils-2_26-branch as of today).

Hmm - we are still in fight w.r.t. upstreaming gcc fully (ARCv2 gcc support) although ARCompact has been there for some time. binutils upstream should work - I'll check why the buildall stuff fails.

We currently have our tools on github but presume that won't work for you due to non canonical location / version !

Quoted 7 lines · Vineet Gupta

Quoted 7 lines · Peter Zijlstra

Quoted 2 lines

Quoted 2 lines

As you can see, arc doesn't even build upstream binutils :/ (binutils-2_26-branch as of today).

Hmm - we are still in fight w.r.t. upstreaming gcc fully (ARCv2 gcc support) although ARCompact has been there for some time. binutils upstream should work - I'll check why the buildall stuff fails.

I can run that build again and send you the resulting log files in private.

Quoted 8 lines · Peter Zijlstra

Quoted 7 lines · Vineet Gupta

Quoted 7 lines · Peter Zijlstra

Quoted 2 lines

Quoted 2 lines

As you can see, arc doesn't even build upstream binutils :/ (binutils-2_26-branch as of today).

Hmm - we are still in fight w.r.t. upstreaming gcc fully (ARCv2 gcc support) although ARCompact has been there for some time. binutils upstream should work - I'll check why the buildall stuff fails.

Quoted 2 lines

I can run that build again and send you the resulting log files in private.

Thx for the off-list log. The issue is gdb not yet upstream. So it seems we will have to wait for things on upstreaming to settle down before I can formally update segher's scripts to support ARC.

Quoted 9 lines · Vineet Gupta

Quoted 8 lines · Peter Zijlstra

Quoted 7 lines · Vineet Gupta

Quoted 7 lines · Peter Zijlstra

Quoted 2 lines

Quoted 2 lines

As you can see, arc doesn't even build upstream binutils :/ (binutils-2_26-branch as of today).

Hmm - we are still in fight w.r.t. upstreaming gcc fully (ARCv2 gcc support) although ARCompact has been there for some time. binutils upstream should work - I'll check why the buildall stuff fails.

Quoted 6 lines

Quoted 2 lines

I can run that build again and send you the resulting log files in private.

Thx for the off-list log. The issue is gdb not yet upstream. So it seems we will have to wait for things on upstreaming to settle down before I can formally update segher's scripts to support ARC.

Ah, lemme add --disable-gdb to the binutils build.

Should arc-*-* list gdb in their noconfigdirs like many other archs do too?

Should arc-*-* list gdb in their noconfigdirs like many other archs do too?

Quoted 8 lines · Peter Zijlstra

Quoted 7 lines · Guenter Roeck

Hi Peter,

I am seeing build failures in -next when trying to build arc / arcv2 targets.

arch/arc/include/asm/atomic.h:74:2: error: ?SCOND_FAIL_RETRY_VAR_DEF? undeclared arch/arc/include/asm/atomic.h:87:2: error: expected ?:? or ?)? before ?SCOND_FAIL_RETRY_ASM?

Quoted 6 lines

Quoted 3 lines

Problems seem to be caused by 'locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()'. Both SCOND_FAIL_RETRY_VAR_DEF and SCOND_FAIL_RETRY_ASM are undefined.

Crud, I messed up the rebase against the backoff reverts. Lemme go do fixups.

I've misplaced my arc compiler, but does this make it go again?

---
 arch/arc/include/asm/atomic.h | 8 --------
 1 file changed, 8 deletions(-)

arch/arc/include/asm/atomic.h +0 -8

index bd9c51cb2bfd..4e3c1b6b0806 100644
@@ -71,7 +71,6 @@ static inline int atomic_##op##_return(int i, atomic_t *v)		\
 static inline int atomic_fetch_##op(int i, atomic_t *v)			\
 {									\
 	unsigned int val, orig;						\
-	SCOND_FAIL_RETRY_VAR_DEF                                        \
 									\
 	/*								\
 	 * Explicit full memory barrier needed before/after as		\

@@ -84,11 +83,8 @@ static inline int atomic_fetch_##op(int i, atomic_t *v)			\
 	"	" #asm_op " %[val], %[orig], %[i]	\n"		\
 	"	scond   %[val], [%[ctr]]		\n"		\
 	"						\n"		\
-	SCOND_FAIL_RETRY_ASM						\
-									\
 	: [val]	"=&r"	(val),						\
 	  [orig] "=&r" (orig)						\
-	  SCOND_FAIL_RETRY_VARS						\
 	: [ctr]	"r"	(&v->counter),					\
 	  [i]	"ir"	(i)						\
 	: "cc");							\

@@ -199,10 +195,6 @@ ATOMIC_OPS(andnot, &= ~, bic)
 ATOMIC_OPS(or, |=, or)
 ATOMIC_OPS(xor, ^=, xor)
 
-#undef SCOND_FAIL_RETRY_VAR_DEF
-#undef SCOND_FAIL_RETRY_ASM
-#undef SCOND_FAIL_RETRY_VARS
-
 #else /* CONFIG_ARC_PLAT_EZNPS */
 
 static inline int atomic_read(const atomic_t *v)

_______________________________________________
linux-snps-arc mailing list
linux-snps-arc at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

Peter the fixlet below doesn't show up in linux-next yet. Could you please do that soon. Our internal build lords are yelling at me :-)

To keep things bisectable, it would be desirable to squash this with the orig commit !

Quoted 39 lines

---
 arch/arc/include/asm/atomic.h | 8 --------
 1 file changed, 8 deletions(-)

arch/arc/include/asm/atomic.h +0 -8

index bd9c51cb2bfd..4e3c1b6b0806 100644
@@ -71,7 +71,6 @@ static inline int atomic_##op##_return(int i, atomic_t *v)		\
 static inline int atomic_fetch_##op(int i, atomic_t *v)			\
 {									\
 	unsigned int val, orig;						\
-	SCOND_FAIL_RETRY_VAR_DEF                                        \
 									\
 	/*								\
 	 * Explicit full memory barrier needed before/after as		\

@@ -84,11 +83,8 @@ static inline int atomic_fetch_##op(int i, atomic_t *v)			\
 	"	" #asm_op " %[val], %[orig], %[i]	\n"		\
 	"	scond   %[val], [%[ctr]]		\n"		\
 	"						\n"		\
-	SCOND_FAIL_RETRY_ASM						\
-									\
 	: [val]	"=&r"	(val),						\
 	  [orig] "=&r" (orig)						\
-	  SCOND_FAIL_RETRY_VARS						\
 	: [ctr]	"r"	(&v->counter),					\
 	  [i]	"ir"	(i)						\
 	: "cc");							\

@@ -199,10 +195,6 @@ ATOMIC_OPS(andnot, &= ~, bic)
 ATOMIC_OPS(or, |=, or)
 ATOMIC_OPS(xor, ^=, xor)
 
-#undef SCOND_FAIL_RETRY_VAR_DEF
-#undef SCOND_FAIL_RETRY_ASM
-#undef SCOND_FAIL_RETRY_VARS
-
 #else /* CONFIG_ARC_PLAT_EZNPS */
 
 static inline int atomic_read(const atomic_t *v)