[PATCH 0 of 7 v2] Make xen build with clang/LLVM again

[PATCH 0 of 7 v2] Make xen build with clang/LLVM again

[Tim Deegan]

This series makes the hypervisor build with clang/LLVM again,
after a certain amount of bit-rot.

Since v1:
 - Changed an xmalloc+memset pair to xzalloc in the memset patch
 - Reworked the spinlock patch not to touch gcc builds
 - Added a patch to indirect all __section__ directives through a macro.
 - Commented up the ugly __attribute__((used)) change and moved it
   into the definition of __section().

[PATCH 1 of 7 v2] xen: Add -Wno-unused-value to the clang CFLAGS

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID 7922a921d2034930780e41f1bc71d2c4ddb3b782
# Parent  d690c7e896a26c54a5ab85458824059de72d5cba
xen: Add -Wno-unused-value to the clang CFLAGS

clang complains about a lot of functions and macros whose return value
is unused.  I started on patches to drop some functions' return values
and scatter (void)s around callers, but it was getting too messy.
Just turn off the warning instead.

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r d690c7e896a2 -r 7922a921d203 Config.mk
--- a/Config.mk	Thu Apr 05 11:06:03 2012 +0100
+++ b/Config.mk	Thu Apr 05 16:49:15 2012 +0100
@@ -159,7 +159,8 @@ CFLAGS += -Wall -Wstrict-prototypes
 
 # Clang complains about macros that expand to 'if ( ( foo == bar ) ) ...'
 # and is over-zealous with the printf format lint
-CFLAGS-$(clang) += -Wno-parentheses -Wno-format
+# and is a bit too fierce about unused return values
+CFLAGS-$(clang) += -Wno-parentheses -Wno-format -Wno-unused-value
 
 $(call cc-option-add,HOSTCFLAGS,HOSTCC,-Wdeclaration-after-statement)
 $(call cc-option-add,CFLAGS,CC,-Wdeclaration-after-statement)

[PATCH 2 of 7 v2] x86/mm: Another couple of comparisons of unsigned vars with < 0

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID 08612e81926857363618746fd01f9b08b7c3ac73
# Parent  7922a921d2034930780e41f1bc71d2c4ddb3b782
x86/mm: Another couple of comparisons of unsigned vars with < 0.

Adding the explicit (unsigned) casts in case enums ever end up signed.

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r 7922a921d203 -r 08612e819268 xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
@@ -1305,7 +1305,7 @@ int p2m_set_mem_access(struct domain *d,
         p2m->default_access,
     };
 
-    if ( access >= HVMMEM_access_default || access < 0 )
+    if ( (unsigned) access >= HVMMEM_access_default )
         return -EINVAL;
 
     a = memaccess[access];
@@ -1367,7 +1367,7 @@ int p2m_get_mem_access(struct domain *d,
     if ( mfn_x(mfn) == INVALID_MFN )
         return -ESRCH;
     
-    if ( a >= ARRAY_SIZE(memaccess) || a < 0 )
+    if ( (unsigned) a >= ARRAY_SIZE(memaccess) )
         return -ERANGE;
 
     *access =  memaccess[a];

[PATCH 3 of 7 v2] x86: fix logical ANDs used to mask bitfields

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID 4674ce03c62a3e916954fd445b4510ffe72e64f4
# Parent  08612e81926857363618746fd01f9b08b7c3ac73
x86: fix logical ANDs used to mask bitfields.

Signed-off-by: Tim Deegan <tim@xen.org>
Acked-by: Jan Beulich <jbeulich@suse.com>

diff -r 08612e819268 -r 4674ce03c62a xen/arch/x86/hvm/svm/svm.c
--- a/xen/arch/x86/hvm/svm/svm.c	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/hvm/svm/svm.c	Thu Apr 05 16:49:15 2012 +0100
@@ -752,7 +752,7 @@ static void svm_lwp_interrupt(struct cpu
     ack_APIC_irq();
     vlapic_set_irq(
         vcpu_vlapic(curr),
-        (curr->arch.hvm_svm.guest_lwp_cfg >> 40) && 0xff,
+        (curr->arch.hvm_svm.guest_lwp_cfg >> 40) & 0xff,
         0);
 }
 
diff -r 08612e819268 -r 4674ce03c62a xen/arch/x86/hvm/vmx/vmx.c
--- a/xen/arch/x86/hvm/vmx/vmx.c	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/hvm/vmx/vmx.c	Thu Apr 05 16:49:15 2012 +0100
@@ -1382,7 +1382,7 @@ void vmx_inject_extint(int trap)
     if ( nestedhvm_vcpu_in_guestmode(v) ) {
         pin_based_cntrl = __get_vvmcs(vcpu_nestedhvm(v).nv_vvmcx, 
                                      PIN_BASED_VM_EXEC_CONTROL);
-        if ( pin_based_cntrl && PIN_BASED_EXT_INTR_MASK ) {
+        if ( pin_based_cntrl & PIN_BASED_EXT_INTR_MASK ) {
             nvmx_enqueue_n2_exceptions (v, 
                INTR_INFO_VALID_MASK | (X86_EVENTTYPE_EXT_INTR<<8) | trap,
                HVM_DELIVER_NO_ERROR_CODE);
@@ -1401,7 +1401,7 @@ void vmx_inject_nmi(void)
     if ( nestedhvm_vcpu_in_guestmode(v) ) {
         pin_based_cntrl = __get_vvmcs(vcpu_nestedhvm(v).nv_vvmcx, 
                                      PIN_BASED_VM_EXEC_CONTROL);
-        if ( pin_based_cntrl && PIN_BASED_NMI_EXITING ) {
+        if ( pin_based_cntrl & PIN_BASED_NMI_EXITING ) {
             nvmx_enqueue_n2_exceptions (v, 
                INTR_INFO_VALID_MASK | (X86_EVENTTYPE_NMI<<8) | TRAP_nmi,
                HVM_DELIVER_NO_ERROR_CODE);

[PATCH 4 of 7 v2] x86: fix memset(ptr, 0, sizeof ptr)

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID a93381049790e4f8a02f2322851f78175c254c5b
# Parent  4674ce03c62a3e916954fd445b4510ffe72e64f4
x86: fix memset(ptr, 0, sizeof ptr).

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/cpu/mcheck/amd_f10.c
--- a/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
@@ -73,9 +73,9 @@ amd_f10_handler(struct mc_info *mi, uint
 		return NULL;
 	}
 
-	memset(mc_ext, 0, sizeof(mc_ext));
+	memset(mc_ext, 0, sizeof(*mc_ext));
 	mc_ext->common.type = MC_TYPE_EXTENDED;
-	mc_ext->common.size = sizeof(mc_ext);
+	mc_ext->common.size = sizeof(*mc_ext);
 	mc_ext->mc_msrs = 3;
 
 	mc_ext->mc_msr[0].reg = MSR_F10_MC4_MISC1;
diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/mm/p2m.c
--- a/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
@@ -1232,11 +1232,10 @@ bool_t p2m_mem_access_check(unsigned lon
     }
 
     *req_ptr = NULL;
-    req = xmalloc(mem_event_request_t);
+    req = xzalloc(mem_event_request_t);
     if ( req )
     {
         *req_ptr = req;
-        memset(req, 0, sizeof(req));
         req->reason = MEM_EVENT_REASON_VIOLATION;
 
         /* Pause the current VCPU */

Re: [PATCH 4 of 7 v2] x86: fix memset(ptr, 0, sizeof ptr)

[Jan Beulich]

>>> On 05.04.12 at 17:51, Tim Deegan <tim@xen.org> wrote:
> # HG changeset patch
> # User Tim Deegan <tim@xen.org>
> # Date 1333640955 -3600
> # Node ID a93381049790e4f8a02f2322851f78175c254c5b
> # Parent  4674ce03c62a3e916954fd445b4510ffe72e64f4
> x86: fix memset(ptr, 0, sizeof ptr).
> 
> Signed-off-by: Tim Deegan <tim@xen.org>

Acked-by: Jan Beulich <jbeulich@suse.com>

> diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/cpu/mcheck/amd_f10.c
> --- a/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
> +++ b/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
> @@ -73,9 +73,9 @@ amd_f10_handler(struct mc_info *mi, uint
>  		return NULL;
>  	}
>  
> -	memset(mc_ext, 0, sizeof(mc_ext));
> +	memset(mc_ext, 0, sizeof(*mc_ext));
>  	mc_ext->common.type = MC_TYPE_EXTENDED;
> -	mc_ext->common.size = sizeof(mc_ext);
> +	mc_ext->common.size = sizeof(*mc_ext);
>  	mc_ext->mc_msrs = 3;
>  
>  	mc_ext->mc_msr[0].reg = MSR_F10_MC4_MISC1;
> diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/mm/p2m.c
> --- a/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
> +++ b/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
> @@ -1232,11 +1232,10 @@ bool_t p2m_mem_access_check(unsigned lon
>      }
>  
>      *req_ptr = NULL;
> -    req = xmalloc(mem_event_request_t);
> +    req = xzalloc(mem_event_request_t);
>      if ( req )
>      {
>          *req_ptr = req;
> -        memset(req, 0, sizeof(req));
>          req->reason = MEM_EVENT_REASON_VIOLATION;
>  
>          /* Pause the current VCPU */
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org 
> http://lists.xen.org/xen-devel 

Re: [PATCH 4 of 7 v2] x86: fix memset(ptr, 0, sizeof ptr)

[Christoph Egger]

On 04/05/12 17:51, Tim Deegan wrote:
> # HG changeset patch
> # User Tim Deegan<tim@xen.org>
> # Date 1333640955 -3600
> # Node ID a93381049790e4f8a02f2322851f78175c254c5b
> # Parent  4674ce03c62a3e916954fd445b4510ffe72e64f4
> x86: fix memset(ptr, 0, sizeof ptr).
>
> Signed-off-by: Tim Deegan<tim@xen.org>

Acked-by: Christoph Egger <Christoph.Egger@amd.com>

>
> diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/cpu/mcheck/amd_f10.c
> --- a/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
> +++ b/xen/arch/x86/cpu/mcheck/amd_f10.c	Thu Apr 05 16:49:15 2012 +0100
> @@ -73,9 +73,9 @@ amd_f10_handler(struct mc_info *mi, uint
>   		return NULL;
>   	}
>
> -	memset(mc_ext, 0, sizeof(mc_ext));
> +	memset(mc_ext, 0, sizeof(*mc_ext));
>   	mc_ext->common.type = MC_TYPE_EXTENDED;
> -	mc_ext->common.size = sizeof(mc_ext);
> +	mc_ext->common.size = sizeof(*mc_ext);
>   	mc_ext->mc_msrs = 3;
>
>   	mc_ext->mc_msr[0].reg = MSR_F10_MC4_MISC1;
> diff -r 4674ce03c62a -r a93381049790 xen/arch/x86/mm/p2m.c
> --- a/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
> +++ b/xen/arch/x86/mm/p2m.c	Thu Apr 05 16:49:15 2012 +0100
> @@ -1232,11 +1232,10 @@ bool_t p2m_mem_access_check(unsigned lon
>       }
>
>       *req_ptr = NULL;
> -    req = xmalloc(mem_event_request_t);
> +    req = xzalloc(mem_event_request_t);
>       if ( req )
>       {
>           *req_ptr = req;
> -        memset(req, 0, sizeof(req));
>           req->reason = MEM_EVENT_REASON_VIOLATION;
>
>           /* Pause the current VCPU */
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>


-- 
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Einsteinring 24, 85689 Dornach b. Muenchen
Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632

[PATCH 5 of 7 v2] x86: don't use .subsection when compiling with clang

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID 0908535327a5b01e49b69cd96db464be21ff3ee6
# Parent  a93381049790e4f8a02f2322851f78175c254c5b
x86: don't use .subsection when compiling with clang

LLVM's assembler doesn't support the .subsection directive, so put
the out-of-line failure path in .fixup instead.

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r a93381049790 -r 0908535327a5 xen/include/asm-x86/spinlock.h
--- a/xen/include/asm-x86/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/asm-x86/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
@@ -45,11 +45,19 @@ static always_inline int _raw_read_trylo
     asm volatile (
         "    lock; decl %0         \n"
         "    jns 2f                \n"
+#ifdef __clang__ /* clang's builtin assember can't do .subsection */
+        "1:  .pushsection .fixup,\"ax\"\n"
+#else
         "1:  .subsection 1         \n"
+#endif
         "2:  lock; incl %0         \n"
         "    decl %1               \n"
         "    jmp 1b                \n"
+#ifdef __clang__
+        "    .popsection           \n"
+#else
         "    .subsection 0         \n"
+#endif
         : "=m" (rw->lock), "=r" (acquired) : "1" (1) : "memory" );
 
     return acquired;

Re: [PATCH 5 of 7 v2] x86: don't use .subsection when compiling with clang

[Jan Beulich]

>>> On 05.04.12 at 17:51, Tim Deegan <tim@xen.org> wrote:
> # HG changeset patch
> # User Tim Deegan <tim@xen.org>
> # Date 1333640955 -3600
> # Node ID 0908535327a5b01e49b69cd96db464be21ff3ee6
> # Parent  a93381049790e4f8a02f2322851f78175c254c5b
> x86: don't use .subsection when compiling with clang
> 
> LLVM's assembler doesn't support the .subsection directive, so put
> the out-of-line failure path in .fixup instead.

Given that it's a single place only, addressing this inline rather than
creating a proper abstraction is probably fine, so ...

> Signed-off-by: Tim Deegan <tim@xen.org>

Acked-by: Jan Beulich <jbeulich@suse.com>

> diff -r a93381049790 -r 0908535327a5 xen/include/asm-x86/spinlock.h
> --- a/xen/include/asm-x86/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
> +++ b/xen/include/asm-x86/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
> @@ -45,11 +45,19 @@ static always_inline int _raw_read_trylo
>      asm volatile (
>          "    lock; decl %0         \n"
>          "    jns 2f                \n"
> +#ifdef __clang__ /* clang's builtin assember can't do .subsection */
> +        "1:  .pushsection .fixup,\"ax\"\n"
> +#else
>          "1:  .subsection 1         \n"
> +#endif
>          "2:  lock; incl %0         \n"
>          "    decl %1               \n"
>          "    jmp 1b                \n"
> +#ifdef __clang__
> +        "    .popsection           \n"
> +#else
>          "    .subsection 0         \n"
> +#endif
>          : "=m" (rw->lock), "=r" (acquired) : "1" (1) : "memory" );
>  
>      return acquired;
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org 
> http://lists.xen.org/xen-devel 

[PATCH 6 of 7 v2] xen: define __section() and friends and use them for section annotations

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID fba1917c638768c0a4e45c507c29f1020e08dfc1
# Parent  0908535327a5b01e49b69cd96db464be21ff3ee6
xen: define __section() and friends and use them for section annotations.

By itself this is just code-tidying, but it's also useful for the
following patch, which will adjust __section() for clang compiles.

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r 0908535327a5 -r fba1917c6387 xen/include/asm-arm/cache.h
--- a/xen/include/asm-arm/cache.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/asm-arm/cache.h	Thu Apr 05 16:49:15 2012 +0100
@@ -7,7 +7,7 @@
 #define L1_CACHE_SHIFT  (CONFIG_ARM_L1_CACHE_SHIFT)
 #define L1_CACHE_BYTES  (1 << L1_CACHE_SHIFT)
 
-#define __read_mostly __attribute__((__section__(".data.read_mostly")))
+#define __read_mostly __section(".data.read_mostly")
 
 #endif
 /*
diff -r 0908535327a5 -r fba1917c6387 xen/include/asm-arm/percpu.h
--- a/xen/include/asm-arm/percpu.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/asm-arm/percpu.h	Thu Apr 05 16:49:15 2012 +0100
@@ -8,7 +8,7 @@ void percpu_init_areas(void);
 
 /* Separate out the type, so (int[3], foo) works. */
 #define __DEFINE_PER_CPU(type, name, suffix)                    \
-    __attribute__((__section__(".bss.percpu" #suffix)))         \
+    __section(".bss.percpu" #suffix)                            \
     __typeof__(type) per_cpu_##name
 
 
diff -r 0908535327a5 -r fba1917c6387 xen/include/asm-x86/cache.h
--- a/xen/include/asm-x86/cache.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/asm-x86/cache.h	Thu Apr 05 16:49:15 2012 +0100
@@ -10,6 +10,6 @@
 #define L1_CACHE_SHIFT	(CONFIG_X86_L1_CACHE_SHIFT)
 #define L1_CACHE_BYTES	(1 << L1_CACHE_SHIFT)
 
-#define __read_mostly __attribute__((__section__(".data.read_mostly")))
+#define __read_mostly __section(".data.read_mostly")
 
 #endif
diff -r 0908535327a5 -r fba1917c6387 xen/include/asm-x86/percpu.h
--- a/xen/include/asm-x86/percpu.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/asm-x86/percpu.h	Thu Apr 05 16:49:15 2012 +0100
@@ -9,7 +9,7 @@ void percpu_init_areas(void);
 
 /* Separate out the type, so (int[3], foo) works. */
 #define __DEFINE_PER_CPU(type, name, suffix)                    \
-    __attribute__((__section__(".bss.percpu" #suffix)))         \
+    __section(".bss.percpu" #suffix)                            \
     __typeof__(type) per_cpu_##name
 
 /* var is in discarded region: offset to particular copy we want */
diff -r 0908535327a5 -r fba1917c6387 xen/include/xen/compiler.h
--- a/xen/include/xen/compiler.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/xen/compiler.h	Thu Apr 05 16:49:15 2012 +0100
@@ -14,6 +14,10 @@
 #define always_inline __inline__ __attribute__ ((always_inline))
 #define noinline      __attribute__((noinline))
 
+#define __section(s)      __attribute__((__section__(s)))
+#define __used_section(s) __attribute_used__ __attribute__((__section__(s)))
+#define __text_section(s) __attribute__((__section__(s)))
+
 #ifdef INIT_SECTIONS_ONLY
 /*
  * For sources indicated to have only init code, make sure even
diff -r 0908535327a5 -r fba1917c6387 xen/include/xen/init.h
--- a/xen/include/xen/init.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/xen/init.h	Thu Apr 05 16:49:15 2012 +0100
@@ -7,20 +7,13 @@
  * Mark functions and data as being only used at initialization
  * or exit time.
  */
-#define __init       \
-    __attribute__ ((__section__ (".init.text")))
-#define __exit       \
-    __attribute_used__ __attribute__ ((__section__(".exit.text")))
-#define __initdata   \
-    __attribute__ ((__section__ (".init.data")))
-#define __exitdata   \
-    __attribute_used__ __attribute__ ((__section__ (".exit.data")))
-#define __initsetup  \
-    __attribute_used__ __attribute__ ((__section__ (".init.setup")))
-#define __init_call(lvl)  \
-    __attribute_used__ __attribute__ ((__section__ (".initcall" lvl ".init")))
-#define __exit_call  \
-    __attribute_used__ __attribute__ ((__section__ (".exitcall.exit")))
+#define __init            __text_section(".init.text")
+#define __exit            __text_section(".exit.text")
+#define __initdata        __section(".init.data")
+#define __exitdata        __used_section(".exit.data")
+#define __initsetup       __used_section(".init.setup")
+#define __init_call(lvl)  __used_section(".initcall" lvl ".init")
+#define __exit_call       __used_section(".exitcall.exit")
 
 /* These macros are used to mark some functions or 
  * initialized data (doesn't apply to uninitialized data)
@@ -95,7 +88,7 @@ struct kernel_param {
 extern struct kernel_param __setup_start, __setup_end;
 
 #define __setup_str static __initdata __attribute__((__aligned__(1))) char
-#define __kparam static __attribute_used__ __initsetup struct kernel_param
+#define __kparam static __initsetup struct kernel_param
 
 #define custom_param(_name, _var) \
     __setup_str __setup_str_##_var[] = _name; \
diff -r 0908535327a5 -r fba1917c6387 xen/include/xen/spinlock.h
--- a/xen/include/xen/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/xen/spinlock.h	Thu Apr 05 16:49:15 2012 +0100
@@ -77,8 +77,8 @@ struct lock_profile_qhead {
 
 #define _LOCK_PROFILE(name) { 0, #name, &name, 0, 0, 0, 0, 0 }
 #define _LOCK_PROFILE_PTR(name)                                               \
-    static struct lock_profile *__lock_profile_##name __attribute_used__      \
-    __attribute__ ((__section__(".lockprofile.data"))) =                      \
+    static struct lock_profile *__lock_profile_##name                         \
+    __used_section(".lockprofile.data") =                                     \
     &__lock_profile_data_##name
 #define _SPIN_LOCK_UNLOCKED(x) { _RAW_SPIN_LOCK_UNLOCKED, 0xfffu, 0,          \
                                  _LOCK_DEBUG, x }
diff -r 0908535327a5 -r fba1917c6387 xen/include/xsm/xsm.h
--- a/xen/include/xsm/xsm.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/xsm/xsm.h	Thu Apr 05 16:49:15 2012 +0100
@@ -44,7 +44,7 @@ extern xsm_initcall_t __xsm_initcall_sta
 
 #define xsm_initcall(fn) \
     static xsm_initcall_t __initcall_##fn \
-    __attribute_used__ __attribute__((__section__(".xsm_initcall.init"))) = fn
+    __used_section(".xsm_initcall.init") = fn
 
 struct xsm_operations {
     void (*security_domaininfo) (struct domain *d,

[PATCH 7 of 7 v2] x86: explicitly mark __initdata variables as used when building with clang

[Tim Deegan]

# HG changeset patch
# User Tim Deegan <tim@xen.org>
# Date 1333640955 -3600
# Node ID 51174f350092962ad36e406adf7ae1f173330a37
# Parent  fba1917c638768c0a4e45c507c29f1020e08dfc1
x86: explicitly mark __initdata variables as used when building with clang.

This stops LLVM from replacing it with a different, auto-generated
variable as part of an optimization.  (The auto-generated variable
ends up in the normal data section.)

Remove stray __read_mostly annotations on declarations that this unmasked.

Signed-off-by: Tim Deegan <tim@xen.org>

diff -r fba1917c6387 -r 51174f350092 xen/arch/x86/oprofile/op_x86_model.h
--- a/xen/arch/x86/oprofile/op_x86_model.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/arch/x86/oprofile/op_x86_model.h	Thu Apr 05 16:49:15 2012 +0100
@@ -53,6 +53,6 @@ extern struct op_x86_model_spec const op
 void arch_perfmon_setup_counters(void);
 
 extern int ppro_has_global_ctrl;
-extern struct op_x86_model_spec const *__read_mostly model;
+extern struct op_x86_model_spec const *model;
 
 #endif /* OP_X86_MODEL_H */
diff -r fba1917c6387 -r 51174f350092 xen/include/acpi/cpufreq/cpufreq.h
--- a/xen/include/acpi/cpufreq/cpufreq.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/acpi/cpufreq/cpufreq.h	Thu Apr 05 16:49:15 2012 +0100
@@ -22,7 +22,7 @@
 
 DECLARE_PER_CPU(spinlock_t, cpufreq_statistic_lock);
 
-extern bool_t __read_mostly cpufreq_verbose;
+extern bool_t cpufreq_verbose;
 
 struct cpufreq_governor;
 
diff -r fba1917c6387 -r 51174f350092 xen/include/xen/compiler.h
--- a/xen/include/xen/compiler.h	Thu Apr 05 16:49:15 2012 +0100
+++ b/xen/include/xen/compiler.h	Thu Apr 05 16:49:15 2012 +0100
@@ -14,7 +14,13 @@
 #define always_inline __inline__ __attribute__ ((always_inline))
 #define noinline      __attribute__((noinline))
 
+#ifdef __clang__
+/* Clang can replace some vars with new automatic ones that go in .data;
+ * mark all explicit-segment vars 'used' to prevent that. */
+#define __section(s)      __attribute_used__ __attribute__((__section__(s)))
+#else
 #define __section(s)      __attribute__((__section__(s)))
+#endif
 #define __used_section(s) __attribute_used__ __attribute__((__section__(s)))
 #define __text_section(s) __attribute__((__section__(s)))

Re: [PATCH 0 of 7 v2] Make xen build with clang/LLVM again

[Jan Beulich]

>>> On 05.04.12 at 17:51, Tim Deegan <tim@xen.org> wrote:
> This series makes the hypervisor build with clang/LLVM again,
> after a certain amount of bit-rot.

I ack-ed the ones I uesfully can; the rest looks good to me too but
may need an ack from Keir.

Jan

> Since v1:
>  - Changed an xmalloc+memset pair to xzalloc in the memset patch
>  - Reworked the spinlock patch not to touch gcc builds
>  - Added a patch to indirect all __section__ directives through a macro.
>  - Commented up the ugly __attribute__((used)) change and moved it
>    into the definition of __section().
>  
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org 
> http://lists.xen.org/xen-devel 

Re: [PATCH 0 of 7 v2] Make xen build with clang/LLVM again

[Keir Fraser]

On 10/04/2012 08:46, "Jan Beulich" <JBeulich@suse.com> wrote:

>>>> On 05.04.12 at 17:51, Tim Deegan <tim@xen.org> wrote:
>> This series makes the hypervisor build with clang/LLVM again,
>> after a certain amount of bit-rot.
> 
> I ack-ed the ones I uesfully can; the rest looks good to me too but
> may need an ack from Keir.

I acked already, I think, but just in case
Acked-by: Keir Fraser <keir@xen.org>

> Jan
> 
>> Since v1:
>>  - Changed an xmalloc+memset pair to xzalloc in the memset patch
>>  - Reworked the spinlock patch not to touch gcc builds
>>  - Added a patch to indirect all __section__ directives through a macro.
>>  - Commented up the ugly __attribute__((used)) change and moved it
>>    into the definition of __section().
>>  
>> 
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xen.org
>> http://lists.xen.org/xen-devel
> 
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel