Skip to contentSkip to conversation

Introduce cmpxchg128() -- aka. the demise of cmpxchg_double()

PATCH Applied Peter Zijlstra 11 participants 45 messages May 31, 2023 – Jun 9, 2023

Subsystems (15)
lore
Share & tools
lore
author maintainer
#42 lore raw

PATCH 12/12 s390/cpum_sf: Convert to cmpxchg128()

In reply to Peter Zijlstra

Now that there is a cross arch u128 and cmpxchg128(), use those instead of the custom CDSG helper.

---
 arch/s390/include/asm/cpu_mf.h  |    2 +-
 arch/s390/kernel/perf_cpum_sf.c |   16 +++-------------
 2 files changed, 4 insertions(+), 14 deletions(-)
arch/s390/include/asm/cpu_mf.h +1 -1
@@ -140,7 +140,7 @@ union hws_trailer_header {
 		unsigned int dsdes:16;	/* 48-63: size of diagnostic SDE */
 		unsigned long long overflow; /* 64 - Overflow Count   */
 	};
-	__uint128_t val;
+	u128 val;
 };
 
 struct hws_trailer_entry {
arch/s390/kernel/perf_cpum_sf.c +3 -13
@@ -1271,16 +1271,6 @@ static void hw_collect_samples(struct pe
 	}
 }
 
-static inline __uint128_t __cdsg(__uint128_t *ptr, __uint128_t old, __uint128_t new)
-{
-	asm volatile(
-		"	cdsg	%[old],%[new],%[ptr]\n"
-		: [old] "+d" (old), [ptr] "+QS" (*ptr)
-		: [new] "d" (new)
-		: "memory", "cc");
-	return old;
-}
-
 /* hw_perf_event_update() - Process sampling buffer
  * @event:	The perf event
  * @flush_all:	Flag to also flush partially filled sample-data-blocks
@@ -1352,7 +1342,7 @@ static void hw_perf_event_update(struct
 			new.f = 0;
 			new.a = 1;
 			new.overflow = 0;
-			prev.val = __cdsg(&te->header.val, old.val, new.val);
+			prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 		} while (prev.val != old.val);
 
 		/* Advance to next sample-data-block */
@@ -1562,7 +1552,7 @@ static bool aux_set_alert(struct aux_buf
 		}
 		new.a = 1;
 		new.overflow = 0;
-		prev.val = __cdsg(&te->header.val, old.val, new.val);
+		prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 	} while (prev.val != old.val);
 	return true;
 }
@@ -1636,7 +1626,7 @@ static bool aux_reset_buffer(struct aux_
 				new.a = 1;
 			else
 				new.a = 0;
-			prev.val = __cdsg(&te->header.val, old.val, new.val);
+			prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 		} while (prev.val != old.val);
 		*overflow += orig_overflow;
 	}

Peter Zijlstra

author maintainer
#0 lore raw

cover Introduce cmpxchg128() -- aka. the demise of cmpxchg_double()

Hi!

After much breaking of things, find here the improved version.

Since v3:

- unbreak everything that does *NOT* have cmpxchg128()

Notably this_cpu_cmpxchg_double() is used unconditionally by SLUB which means that this_cpu_try_cmpxchg128() needs to be unconditionally available on all 64bit architectures.

- fixed up x86/x86_64 cmpxchg{8,16}b emulation for this_cpu_cmpxchg{64,128}()

- introduce {raw,this}_cpu_try_cmpxchg*()

- add fallback for !__SIZEOF_INT128__ 64bit architectures

Sadly there are supported 64bit architecture/compiler combinations that do not have __SIZEOF_INT128__, specifically it was found that HPPA64 only added this with GCC-11.

this is yuck, and ideally we'd simply raise compiler requirements, but this 'works'.

My plan is to re-add this to tip/locking/core and thus -next later this week.

Also available at:

git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git locking/core

---
 Documentation/core-api/this_cpu_ops.rst     |   2 -
 arch/arm64/include/asm/atomic_ll_sc.h       |  56 +++---
 arch/arm64/include/asm/atomic_lse.h         |  39 ++---
 arch/arm64/include/asm/cmpxchg.h            |  48 ++----
 arch/arm64/include/asm/percpu.h             |  30 ++--
 arch/s390/include/asm/cmpxchg.h             |  32 +---
 arch/s390/include/asm/cpu_mf.h              |   2 +-
 arch/s390/include/asm/percpu.h              |  34 ++--
 arch/s390/kernel/perf_cpum_sf.c             |  16 +-
 arch/x86/include/asm/cmpxchg.h              |  25 ---
 arch/x86/include/asm/cmpxchg_32.h           |   2 +-
 arch/x86/include/asm/cmpxchg_64.h           |  63 ++++++-
 arch/x86/include/asm/percpu.h               | 102 ++++++-----
 arch/x86/lib/Makefile                       |   3 +-
 arch/x86/lib/cmpxchg16b_emu.S               |  43 +++--
 arch/x86/lib/cmpxchg8b_emu.S                |  67 ++++++--
 drivers/iommu/amd/amd_iommu_types.h         |   9 +-
 drivers/iommu/amd/iommu.c                   |  10 +-
 drivers/iommu/intel/irq_remapping.c         |   8 +-
 include/asm-generic/percpu.h                | 257 ++++++++++++++++++++++------
 include/crypto/b128ops.h                    |  14 +-
 include/linux/atomic/atomic-arch-fallback.h |  95 +++++++++-
 include/linux/atomic/atomic-instrumented.h  |  93 ++++++++--
 include/linux/dmar.h                        | 125 +++++++-------
 include/linux/percpu-defs.h                 |  45 ++---
 include/linux/slub_def.h                    |  12 +-
 include/linux/types.h                       |  12 ++
 include/uapi/linux/types.h                  |   4 +
 lib/crypto/curve25519-hacl64.c              |   2 -
 lib/crypto/poly1305-donna64.c               |   2 -
 mm/slab.h                                   |  53 +++++-
 mm/slub.c                                   | 139 +++++++++------
 scripts/atomic/gen-atomic-fallback.sh       |   4 +-
 scripts/atomic/gen-atomic-instrumented.sh   |  19 +-
 34 files changed, 952 insertions(+), 515 deletions(-)
author maintainer
#16 lore raw

PATCH 7/12 percpu: #ifndef __SIZEOF_INT128__

Some 64bit architectures do not advertise __SIZEOF_INT128__ on all supported compiler versions. Notably the HPPA64 only started doing with GCC-11.

Since the per-cpu ops are universally availably, and this_cpu_{,try_}cmpxchg128() is expected to be available on all 64bit architectures a wee bodge is in order.

Sadly, while C reverts to memcpy() for assignment of POD types, it does not revert to memcmp() for for equality. Therefore frob that manually.

---
 include/asm-generic/percpu.h |   56 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/types.h        |    7 +++++
 2 files changed, 63 insertions(+)
include/asm-generic/percpu.h +56 -0
@@ -313,6 +313,35 @@ do {									\
 #define raw_cpu_xchg_8(pcp, nval)	raw_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef __SIZEOF_INT128__
+#define raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (!__builtin_memcmp(&__val, &__old, sizeof(pcp))) {		\
+		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
+	__ret;								\
+})
+
+#define raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg_memcpy(pcp, &__old, nval);		\
+	__old;								\
+})
+
+#define raw_cpu_cmpxchg128(pcp, oval, nval) \
+	raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+#endif
+
 #ifndef raw_cpu_try_cmpxchg_1
 #ifdef raw_cpu_cmpxchg_1
 #define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
@@ -503,6 +532,33 @@ do {									\
 #define this_cpu_xchg_8(pcp, nval)	this_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef __SIZEOF_INT128__
+#define this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)		\
+({									\
+	typeof(pcp) __ret;						\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_cmpxchg128(pcp, oval, nval) \
+	this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define this_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+#endif
+
 #ifndef this_cpu_try_cmpxchg_1
 #ifdef this_cpu_cmpxchg_1
 #define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
include/linux/types.h +7 -0
@@ -13,6 +13,13 @@
 #ifdef __SIZEOF_INT128__
 typedef __s128 s128;
 typedef __u128 u128;
+#else
+#ifdef CONFIG_64BIT
+/* hack for this_cpu_cmpxchg128 */
+typedef struct {
+	u64 a, b;
+} u128 __attribute__((aligned(16)));
+#endif
 #endif
 
 typedef u32 __kernel_dev_t;
author maintainer
#1 lore raw

PATCH 1/12 cyrpto/b128ops: Remove struct u128

In reply to Peter Zijlstra

Per git-grep u128_xor() and its related struct u128 are unused except to implement {be,le}128_xor(). Remove them to free up the namespace.

---
 include/crypto/b128ops.h |   14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)
include/crypto/b128ops.h +3 -11
@@ -50,10 +50,6 @@
 #include <linux/types.h>
 
 typedef struct {
-	u64 a, b;
-} u128;
-
-typedef struct {
 	__be64 a, b;
 } be128;
 
@@ -61,20 +57,16 @@ typedef struct {
 	__le64 b, a;
 } le128;
 
-static inline void u128_xor(u128 *r, const u128 *p, const u128 *q)
+static inline void be128_xor(be128 *r, const be128 *p, const be128 *q)
 {
 	r->a = p->a ^ q->a;
 	r->b = p->b ^ q->b;
 }
 
-static inline void be128_xor(be128 *r, const be128 *p, const be128 *q)
-{
-	u128_xor((u128 *)r, (u128 *)p, (u128 *)q);
-}
-
 static inline void le128_xor(le128 *r, const le128 *p, const le128 *q)
 {
-	u128_xor((u128 *)r, (u128 *)p, (u128 *)q);
+	r->a = p->a ^ q->a;
+	r->b = p->b ^ q->b;
 }
 
 #endif /* _CRYPTO_B128OPS_H */
author maintainer
#9 lore raw

PATCH 5/12 percpu: Add {raw,this}_cpu_try_cmpxchg()

In reply to Peter Zijlstra

Add the try_cmpxchg() form to the per-cpu ops.

---
 include/asm-generic/percpu.h |  113 +++++++++++++++++++++++++++++++++++++++++--
 include/linux/percpu-defs.h  |   19 +++++++
 2 files changed, 128 insertions(+), 4 deletions(-)
include/asm-generic/percpu.h +109 -4
@@ -89,16 +89,37 @@ do {									\
 	__ret;								\
 })
 
-#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+#define __cpu_fallback_try_cmpxchg(pcp, ovalp, nval, _cmpxchg)		\
+({									\
+	typeof(pcp) __val, __old = *(ovalp);				\
+ 	__val = _cmpxchg(pcp, __old, nval);				\
+	if (__val != __old)						\
+		*(ovalp) = __val;					\
+	__val == __old;							\
+})
+
+#define raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
 ({									\
 	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
-	typeof(pcp) __ret;						\
-	__ret = *__p;							\
-	if (__ret == (oval))						\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (__val == __old) {						\
 		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
 	__ret;								\
 })
 
+#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg(pcp, &__old, nval);			\
+	__old;								\
+})
+
 #define raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 ({									\
 	typeof(pcp1) *__p1 = raw_cpu_ptr(&(pcp1));			\
@@ -170,6 +191,16 @@ do {									\
 	__ret;								\
 })
 
+#define this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval);		\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
 #define this_cpu_generic_cmpxchg(pcp, oval, nval)			\
 ({									\
 	typeof(pcp) __ret;						\
@@ -282,6 +313,43 @@ do {									\
 #define raw_cpu_xchg_8(pcp, nval)	raw_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef raw_cpu_try_cmpxchg_1
+#ifdef raw_cpu_cmpxchg_1
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_1)
+#else
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_2
+#ifdef raw_cpu_cmpxchg_2
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_2)
+#else
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_4
+#ifdef raw_cpu_cmpxchg_4
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_4)
+#else
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_8
+#ifdef raw_cpu_cmpxchg_8
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_8)
+#else
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef raw_cpu_cmpxchg_1
 #define raw_cpu_cmpxchg_1(pcp, oval, nval) \
 	raw_cpu_generic_cmpxchg(pcp, oval, nval)
@@ -407,6 +475,43 @@ do {									\
 #define this_cpu_xchg_8(pcp, nval)	this_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef this_cpu_try_cmpxchg_1
+#ifdef this_cpu_cmpxchg_1
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_1)
+#else
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_2
+#ifdef this_cpu_cmpxchg_2
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_2)
+#else
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_4
+#ifdef this_cpu_cmpxchg_4
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_4)
+#else
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_8
+#ifdef this_cpu_cmpxchg_8
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_8)
+#else
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef this_cpu_cmpxchg_1
 #define this_cpu_cmpxchg_1(pcp, oval, nval) \
 	this_cpu_generic_cmpxchg(pcp, oval, nval)
include/linux/percpu-defs.h +19 -0
@@ -343,6 +343,21 @@ static __always_inline void __this_cpu_p
 	pscr2_ret__;							\
 })
 
+#define __pcpu_size_call_return2bool(stem, variable, ...)		\
+({									\
+	bool pscr2_ret__;						\
+	__verify_pcpu_ptr(&(variable));					\
+	switch(sizeof(variable)) {					\
+	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
+	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
+	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
+	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
+	default:							\
+		__bad_size_call_parameter(); break;			\
+	}								\
+	pscr2_ret__;							\
+})
+
 /*
  * Special handling for cmpxchg_double.  cmpxchg_double is passed two
  * percpu variables.  The first has to be aligned to a double word
@@ -426,6 +441,8 @@ do {									\
 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
 #define raw_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
+#define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
@@ -513,6 +530,8 @@ do {									\
 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
 #define this_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
+#define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
author maintainer
#14 lore raw

PATCH 6/12 percpu: Wire up cmpxchg128

In reply to Peter Zijlstra

In order to replace cmpxchg_double() with the newly minted cmpxchg128() family of functions, wire it up in this_cpu_cmpxchg().

---
 arch/arm64/include/asm/percpu.h |   20 ++++++++++
 arch/s390/include/asm/percpu.h  |   16 ++++++++
 arch/x86/include/asm/percpu.h   |   74 ++++++++++++++++++++++++++++++++++++----
 arch/x86/lib/Makefile           |    3 +
 arch/x86/lib/cmpxchg16b_emu.S   |   43 +++++++++++++----------
 arch/x86/lib/cmpxchg8b_emu.S    |   67 ++++++++++++++++++++++++++++--------
 include/asm-generic/percpu.h    |   56 ++++++++++++++++++++++++++++++
 7 files changed, 240 insertions(+), 39 deletions(-)
arch/arm64/include/asm/percpu.h +20 -0
@@ -140,6 +140,10 @@ PERCPU_RET_OP(add, add, ldadd)
  * re-enabling preemption for preemptible kernels, but doing that in a way
  * which builds inside a module would mean messing directly with the preempt
  * count. If you do this, peterz and tglx will hunt you down.
+ *
+ * Not to mention it'll break the actual preemption model for missing a
+ * preemption point when TIF_NEED_RESCHED gets set while preemption is
+ * disabled.
  */
 #define this_cpu_cmpxchg_double_8(ptr1, ptr2, o1, o2, n1, n2)		\
 ({									\
@@ -240,6 +244,22 @@ PERCPU_RET_OP(add, add, ldadd)
 #define this_cpu_cmpxchg_8(pcp, o, n)	\
 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
 
+#define this_cpu_cmpxchg64(pcp, o, n)	this_cpu_cmpxchg_8(pcp, o, n)
+
+#define this_cpu_cmpxchg128(pcp, o, n)					\
+({									\
+	typedef typeof(pcp) pcp_op_T__;					\
+	u128 old__, new__, ret__;					\
+	pcp_op_T__ *ptr__;						\
+	old__ = o;							\
+	new__ = n;							\
+	preempt_disable_notrace();					\
+	ptr__ = raw_cpu_ptr(&(pcp));					\
+	ret__ = cmpxchg128_local((void *)ptr__, old__, new__);		\
+	preempt_enable_notrace();					\
+	ret__;								\
+})
+
 #ifdef __KVM_NVHE_HYPERVISOR__
 extern unsigned long __hyp_per_cpu_offset(unsigned int cpu);
 #define __per_cpu_offset
arch/s390/include/asm/percpu.h +16 -0
@@ -148,6 +148,22 @@
 #define this_cpu_cmpxchg_4(pcp, oval, nval) arch_this_cpu_cmpxchg(pcp, oval, nval)
 #define this_cpu_cmpxchg_8(pcp, oval, nval) arch_this_cpu_cmpxchg(pcp, oval, nval)
 
+#define this_cpu_cmpxchg64(pcp, o, n)	this_cpu_cmpxchg_8(pcp, o, n)
+
+#define this_cpu_cmpxchg128(pcp, oval, nval)				\
+({									\
+	typedef typeof(pcp) pcp_op_T__;					\
+	u128 old__, new__, ret__;					\
+	pcp_op_T__ *ptr__;						\
+	old__ = oval;							\
+	new__ = nval;							\
+	preempt_disable_notrace();					\
+	ptr__ = raw_cpu_ptr(&(pcp));					\
+	ret__ = cmpxchg128((void *)ptr__, old__, new__);		\
+	preempt_enable_notrace();					\
+	ret__;								\
+})
+
 #define arch_this_cpu_xchg(pcp, nval)					\
 ({									\
 	typeof(pcp) *ptr__;						\
arch/x86/include/asm/percpu.h +68 -6

86 lines not on this page · raw message

arch/x86/lib/Makefile +2 -1
@@ -61,8 +61,9 @@ ifeq ($(CONFIG_X86_32),y)
         lib-y += strstr_32.o
         lib-y += string_32.o
         lib-y += memmove_32.o
+        lib-y += cmpxchg8b_emu.o
 ifneq ($(CONFIG_X86_CMPXCHG64),y)
-        lib-y += cmpxchg8b_emu.o atomic64_386_32.o
+        lib-y += atomic64_386_32.o
 endif
 else
         obj-y += iomap_copy_64.o
arch/x86/lib/cmpxchg16b_emu.S +25 -18

72 lines not on this page · raw message

arch/x86/lib/cmpxchg8b_emu.S +53 -14

95 lines not on this page · raw message

include/asm-generic/percpu.h +56 -0

80 lines not on this page · raw message

author maintainer
#7 lore raw

PATCH 4/12 instrumentation: Wire up cmpxchg128()

In reply to Peter Zijlstra

Wire up the cmpxchg128 family in the atomic wrapper scripts.

These provide the generic cmpxchg128 family of functions from the arch_ prefixed version, adding explicit instrumentation where needed.

---
 include/linux/atomic/atomic-arch-fallback.h |   95 +++++++++++++++++++++++++++-
 include/linux/atomic/atomic-instrumented.h  |   86 +++++++++++++++++++++++++
 scripts/atomic/gen-atomic-fallback.sh       |    4 -
 scripts/atomic/gen-atomic-instrumented.sh   |    4 -
 4 files changed, 183 insertions(+), 6 deletions(-)
include/linux/atomic/atomic-arch-fallback.h +94 -1
@@ -77,6 +77,29 @@
 
 #endif /* arch_cmpxchg64_relaxed */
 
+#ifndef arch_cmpxchg128_relaxed
+#define arch_cmpxchg128_acquire arch_cmpxchg128
+#define arch_cmpxchg128_release arch_cmpxchg128
+#define arch_cmpxchg128_relaxed arch_cmpxchg128
+#else /* arch_cmpxchg128_relaxed */
+
+#ifndef arch_cmpxchg128_acquire
+#define arch_cmpxchg128_acquire(...) \
+	__atomic_op_acquire(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_cmpxchg128_release
+#define arch_cmpxchg128_release(...) \
+	__atomic_op_release(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_cmpxchg128
+#define arch_cmpxchg128(...) \
+	__atomic_op_fence(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#endif /* arch_cmpxchg128_relaxed */
+
 #ifndef arch_try_cmpxchg_relaxed
 #ifdef arch_try_cmpxchg
 #define arch_try_cmpxchg_acquire arch_try_cmpxchg
@@ -217,6 +240,76 @@
 
 #endif /* arch_try_cmpxchg64_relaxed */
 
+#ifndef arch_try_cmpxchg128_relaxed
+#ifdef arch_try_cmpxchg128
+#define arch_try_cmpxchg128_acquire arch_try_cmpxchg128
+#define arch_try_cmpxchg128_release arch_try_cmpxchg128
+#define arch_try_cmpxchg128_relaxed arch_try_cmpxchg128
+#endif /* arch_try_cmpxchg128 */
+
+#ifndef arch_try_cmpxchg128
+#define arch_try_cmpxchg128(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128 */
+
+#ifndef arch_try_cmpxchg128_acquire
+#define arch_try_cmpxchg128_acquire(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_acquire((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_acquire */
+
+#ifndef arch_try_cmpxchg128_release
+#define arch_try_cmpxchg128_release(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_release((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_release */
+
+#ifndef arch_try_cmpxchg128_relaxed
+#define arch_try_cmpxchg128_relaxed(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_relaxed((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_relaxed */
+
+#else /* arch_try_cmpxchg128_relaxed */
+
+#ifndef arch_try_cmpxchg128_acquire
+#define arch_try_cmpxchg128_acquire(...) \
+	__atomic_op_acquire(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg128_release
+#define arch_try_cmpxchg128_release(...) \
+	__atomic_op_release(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg128
+#define arch_try_cmpxchg128(...) \
+	__atomic_op_fence(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#endif /* arch_try_cmpxchg128_relaxed */
+
 #ifndef arch_try_cmpxchg_local
 #define arch_try_cmpxchg_local(_ptr, _oldp, _new) \
 ({ \
@@ -2668,4 +2761,4 @@ arch_atomic64_dec_if_positive(atomic64_t
 #endif
 
 #endif /* _LINUX_ATOMIC_FALLBACK_H */
-// ad2e2b4d168dbc60a73922616047a9bfa446af36
+// 52dfc6fe4a2e7234bbd2aa3e16a377c1db793a53
include/linux/atomic/atomic-instrumented.h +85 -1

113 lines not on this page · raw message

scripts/atomic/gen-atomic-fallback.sh +2 -2
@@ -217,11 +217,11 @@ cat << EOF
 
 EOF
 
-for xchg in "arch_xchg" "arch_cmpxchg" "arch_cmpxchg64"; do
+for xchg in "arch_xchg" "arch_cmpxchg" "arch_cmpxchg64" "arch_cmpxchg128"; do
 	gen_xchg_fallbacks "${xchg}"
 done
 
-for cmpxchg in "cmpxchg" "cmpxchg64"; do
+for cmpxchg in "cmpxchg" "cmpxchg64" "cmpxchg128"; do
 	gen_try_cmpxchg_fallbacks "${cmpxchg}"
 done
 
scripts/atomic/gen-atomic-instrumented.sh +2 -2
@@ -166,14 +166,14 @@ grep '^[a-z]' "$1" | while read name met
 done
 
 
-for xchg in "xchg" "cmpxchg" "cmpxchg64" "try_cmpxchg" "try_cmpxchg64"; do
+for xchg in "xchg" "cmpxchg" "cmpxchg64" "cmpxchg128" "try_cmpxchg" "try_cmpxchg64" "try_cmpxchg128"; do
 	for order in "" "_acquire" "_release" "_relaxed"; do
 		gen_xchg "${xchg}" "${order}" ""
 		printf "\n"
 	done
 done
 
-for xchg in "cmpxchg_local" "cmpxchg64_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" ; do
+for xchg in "cmpxchg_local" "cmpxchg64_local" "cmpxchg128_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" "try_cmpxchg128_local"; do
 	gen_xchg "${xchg}" "" ""
 	printf "\n"
 done
author maintainer
#38 lore raw

PATCH 10/12 slub: Replace cmpxchg_double()

In reply to Peter Zijlstra
---
 include/linux/slub_def.h |   12 +++-
 mm/slab.h                |   53 +++++++++++++++--
 mm/slub.c                |  139 +++++++++++++++++++++++++++--------------------
 3 files changed, 137 insertions(+), 67 deletions(-)
include/linux/slub_def.h +9 -3
@@ -39,7 +39,8 @@ enum stat_item {
 	CPU_PARTIAL_FREE,	/* Refill cpu partial on free */
 	CPU_PARTIAL_NODE,	/* Refill cpu partial from node partial */
 	CPU_PARTIAL_DRAIN,	/* Drain cpu partial to node partial */
-	NR_SLUB_STAT_ITEMS };
+	NR_SLUB_STAT_ITEMS
+};
 
 #ifndef CONFIG_SLUB_TINY
 /*
@@ -47,8 +48,13 @@ enum stat_item {
  * with this_cpu_cmpxchg_double() alignment requirements.
  */
 struct kmem_cache_cpu {
-	void **freelist;	/* Pointer to next available object */
-	unsigned long tid;	/* Globally unique transaction id */
+	union {
+		struct {
+			void **freelist;	/* Pointer to next available object */
+			unsigned long tid;	/* Globally unique transaction id */
+		};
+		freelist_aba_t freelist_tid;
+	};
 	struct slab *slab;	/* The slab from which we are allocating */
 #ifdef CONFIG_SLUB_CPU_PARTIAL
 	struct slab *partial;	/* Partially allocated frozen slabs */
mm/slab.h +46 -7

74 lines not on this page · raw message

mm/slub.c +82 -57

238 lines not on this page · raw message

author maintainer
#40 lore raw

PATCH 11/12 arch: Remove cmpxchg_double

In reply to Peter Zijlstra

No moar users, remove the monster.

---
 Documentation/core-api/this_cpu_ops.rst    |    2 -
 arch/arm64/include/asm/atomic_ll_sc.h      |   33 ----------------
 arch/arm64/include/asm/atomic_lse.h        |   36 ------------------
 arch/arm64/include/asm/cmpxchg.h           |   46 -----------------------
 arch/arm64/include/asm/percpu.h            |   10 -----
 arch/s390/include/asm/cmpxchg.h            |   34 -----------------
 arch/s390/include/asm/percpu.h             |   18 ---------
 arch/x86/include/asm/cmpxchg.h             |   25 ------------
 arch/x86/include/asm/cmpxchg_32.h          |    1 
 arch/x86/include/asm/cmpxchg_64.h          |    1 
 arch/x86/include/asm/percpu.h              |   42 ---------------------
 include/asm-generic/percpu.h               |   58 -----------------------------
 include/linux/atomic/atomic-instrumented.h |   17 --------
 include/linux/percpu-defs.h                |   38 -------------------
 scripts/atomic/gen-atomic-instrumented.sh  |   15 ++-----
 15 files changed, 5 insertions(+), 371 deletions(-)
Documentation/core-api/this_cpu_ops.rst +0 -2
@@ -53,7 +53,6 @@ are defined. These operations can be use
 	this_cpu_add_return(pcp, val)
 	this_cpu_xchg(pcp, nval)
 	this_cpu_cmpxchg(pcp, oval, nval)
-	this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 	this_cpu_sub(pcp, val)
 	this_cpu_inc(pcp)
 	this_cpu_dec(pcp)
@@ -242,7 +241,6 @@ modifies the variable, then RMW actions
 	__this_cpu_add_return(pcp, val)
 	__this_cpu_xchg(pcp, nval)
 	__this_cpu_cmpxchg(pcp, oval, nval)
-	__this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 	__this_cpu_sub(pcp, val)
 	__this_cpu_inc(pcp)
 	__this_cpu_dec(pcp)
arch/arm64/include/asm/atomic_ll_sc.h +0 -33
@@ -294,39 +294,6 @@ __CMPXCHG_CASE( ,  ,  mb_, 64, dmb ish,
 
 #undef __CMPXCHG_CASE
 
-#define __CMPXCHG_DBL(name, mb, rel, cl)				\
-static __always_inline long						\
-__ll_sc__cmpxchg_double##name(unsigned long old1,			\
-				      unsigned long old2,		\
-				      unsigned long new1,		\
-				      unsigned long new2,		\
-				      volatile void *ptr)		\
-{									\
-	unsigned long tmp, ret;						\
-									\
-	asm volatile("// __cmpxchg_double" #name "\n"			\
-	"	prfm	pstl1strm, %2\n"				\
-	"1:	ldxp	%0, %1, %2\n"					\
-	"	eor	%0, %0, %3\n"					\
-	"	eor	%1, %1, %4\n"					\
-	"	orr	%1, %0, %1\n"					\
-	"	cbnz	%1, 2f\n"					\
-	"	st" #rel "xp	%w0, %5, %6, %2\n"			\
-	"	cbnz	%w0, 1b\n"					\
-	"	" #mb "\n"						\
-	"2:"								\
-	: "=&r" (tmp), "=&r" (ret), "+Q" (*(__uint128_t *)ptr)		\
-	: "r" (old1), "r" (old2), "r" (new1), "r" (new2)		\
-	: cl);								\
-									\
-	return ret;							\
-}
-
-__CMPXCHG_DBL(   ,        ,  ,         )
-__CMPXCHG_DBL(_mb, dmb ish, l, "memory")
-
-#undef __CMPXCHG_DBL
-
 union __u128_halves {
 	u128 full;
 	struct {
arch/arm64/include/asm/atomic_lse.h +0 -36
@@ -281,42 +281,6 @@ __CMPXCHG_CASE(x,  ,  mb_, 64, al, "memo
 
 #undef __CMPXCHG_CASE
 
-#define __CMPXCHG_DBL(name, mb, cl...)					\
-static __always_inline long						\
-__lse__cmpxchg_double##name(unsigned long old1,				\
-					 unsigned long old2,		\
-					 unsigned long new1,		\
-					 unsigned long new2,		\
-					 volatile void *ptr)		\
-{									\
-	unsigned long oldval1 = old1;					\
-	unsigned long oldval2 = old2;					\
-	register unsigned long x0 asm ("x0") = old1;			\
-	register unsigned long x1 asm ("x1") = old2;			\
-	register unsigned long x2 asm ("x2") = new1;			\
-	register unsigned long x3 asm ("x3") = new2;			\
-	register unsigned long x4 asm ("x4") = (unsigned long)ptr;	\
-									\
-	asm volatile(							\
-	__LSE_PREAMBLE							\
-	"	casp" #mb "\t%[old1], %[old2], %[new1], %[new2], %[v]\n"\
-	"	eor	%[old1], %[old1], %[oldval1]\n"			\
-	"	eor	%[old2], %[old2], %[oldval2]\n"			\
-	"	orr	%[old1], %[old1], %[old2]"			\
-	: [old1] "+&r" (x0), [old2] "+&r" (x1),				\
-	  [v] "+Q" (*(__uint128_t *)ptr)				\
-	: [new1] "r" (x2), [new2] "r" (x3), [ptr] "r" (x4),		\
-	  [oldval1] "r" (oldval1), [oldval2] "r" (oldval2)		\
-	: cl);								\
-									\
-	return x0;							\
-}
-
-__CMPXCHG_DBL(   ,   )
-__CMPXCHG_DBL(_mb, al, "memory")
-
-#undef __CMPXCHG_DBL
-
 #define __CMPXCHG128(name, mb, cl...)					\
 static __always_inline u128						\
 __lse__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)		\
arch/arm64/include/asm/cmpxchg.h +0 -46

58 lines not on this page · raw message

arch/arm64/include/asm/percpu.h +0 -10
@@ -145,16 +145,6 @@ PERCPU_RET_OP(add, add, ldadd)
  * preemption point when TIF_NEED_RESCHED gets set while preemption is
  * disabled.
  */
-#define this_cpu_cmpxchg_double_8(ptr1, ptr2, o1, o2, n1, n2)		\
-({									\
-	int __ret;							\
-	preempt_disable_notrace();					\
-	__ret = cmpxchg_double_local(	raw_cpu_ptr(&(ptr1)),		\
-					raw_cpu_ptr(&(ptr2)),		\
-					o1, o2, n1, n2);		\
-	preempt_enable_notrace();					\
-	__ret;								\
-})
 
 #define _pcp_protect(op, pcp, ...)					\
 ({									\
arch/s390/include/asm/cmpxchg.h +0 -34
@@ -190,40 +190,6 @@ static __always_inline unsigned long __c
 #define arch_cmpxchg_local	arch_cmpxchg
 #define arch_cmpxchg64_local	arch_cmpxchg
 
-#define system_has_cmpxchg_double()	1
-
-static __always_inline int __cmpxchg_double(unsigned long p1, unsigned long p2,
-					    unsigned long o1, unsigned long o2,
-					    unsigned long n1, unsigned long n2)
-{
-	union register_pair old = { .even = o1, .odd = o2, };
-	union register_pair new = { .even = n1, .odd = n2, };
-	int cc;
-
-	asm volatile(
-		"	cdsg	%[old],%[new],%[ptr]\n"
-		"	ipm	%[cc]\n"
-		"	srl	%[cc],28\n"
-		: [cc] "=&d" (cc), [old] "+&d" (old.pair)
-		: [new] "d" (new.pair),
-		  [ptr] "QS" (*(unsigned long *)p1), "Q" (*(unsigned long *)p2)
-		: "memory", "cc");
-	return !cc;
-}
-
-#define arch_cmpxchg_double(p1, p2, o1, o2, n1, n2)			\
-({									\
-	typeof(p1) __p1 = (p1);						\
-	typeof(p2) __p2 = (p2);						\
-									\
-	BUILD_BUG_ON(sizeof(*(p1)) != sizeof(long));			\
-	BUILD_BUG_ON(sizeof(*(p2)) != sizeof(long));			\
-	VM_BUG_ON((unsigned long)((__p1) + 1) != (unsigned long)(__p2));\
-	__cmpxchg_double((unsigned long)__p1, (unsigned long)__p2,	\
-			 (unsigned long)(o1), (unsigned long)(o2),	\
-			 (unsigned long)(n1), (unsigned long)(n2));	\
-})
-
 #define system_has_cmpxchg128()		1
 
 static __always_inline u128 arch_cmpxchg128(volatile u128 *ptr, u128 old, u128 new)
arch/s390/include/asm/percpu.h +0 -18
@@ -180,24 +180,6 @@
 #define this_cpu_xchg_4(pcp, nval) arch_this_cpu_xchg(pcp, nval)
 #define this_cpu_xchg_8(pcp, nval) arch_this_cpu_xchg(pcp, nval)
 
-#define arch_this_cpu_cmpxchg_double(pcp1, pcp2, o1, o2, n1, n2)	    \
-({									    \
-	typeof(pcp1) *p1__;						    \
-	typeof(pcp2) *p2__;						    \
-	int ret__;							    \
-									    \
-	preempt_disable_notrace();					    \
-	p1__ = raw_cpu_ptr(&(pcp1));					    \
-	p2__ = raw_cpu_ptr(&(pcp2));					    \
-	ret__ = __cmpxchg_double((unsigned long)p1__, (unsigned long)p2__,  \
-				 (unsigned long)(o1), (unsigned long)(o2),  \
-				 (unsigned long)(n1), (unsigned long)(n2)); \
-	preempt_enable_notrace();					    \
-	ret__;								    \
-})
-
-#define this_cpu_cmpxchg_double_8 arch_this_cpu_cmpxchg_double
-
 #include <asm-generic/percpu.h>
 
 #endif /* __ARCH_S390_PERCPU__ */
arch/x86/include/asm/cmpxchg.h +0 -25
@@ -239,29 +239,4 @@ extern void __add_wrong_size(void)
 #define __xadd(ptr, inc, lock)	__xchg_op((ptr), (inc), xadd, lock)
 #define xadd(ptr, inc)		__xadd((ptr), (inc), LOCK_PREFIX)
 
-#define __cmpxchg_double(pfx, p1, p2, o1, o2, n1, n2)			\
-({									\
-	bool __ret;							\
-	__typeof__(*(p1)) __old1 = (o1), __new1 = (n1);			\
-	__typeof__(*(p2)) __old2 = (o2), __new2 = (n2);			\
-	BUILD_BUG_ON(sizeof(*(p1)) != sizeof(long));			\
-	BUILD_BUG_ON(sizeof(*(p2)) != sizeof(long));			\
-	VM_BUG_ON((unsigned long)(p1) % (2 * sizeof(long)));		\
-	VM_BUG_ON((unsigned long)((p1) + 1) != (unsigned long)(p2));	\
-	asm volatile(pfx "cmpxchg%c5b %1"				\
-		     CC_SET(e)						\
-		     : CC_OUT(e) (__ret),				\
-		       "+m" (*(p1)), "+m" (*(p2)),			\
-		       "+a" (__old1), "+d" (__old2)			\
-		     : "i" (2 * sizeof(long)),				\
-		       "b" (__new1), "c" (__new2));			\
-	__ret;								\
-})
-
-#define arch_cmpxchg_double(p1, p2, o1, o2, n1, n2) \
-	__cmpxchg_double(LOCK_PREFIX, p1, p2, o1, o2, n1, n2)
-
-#define arch_cmpxchg_double_local(p1, p2, o1, o2, n1, n2) \
-	__cmpxchg_double(, p1, p2, o1, o2, n1, n2)
-
 #endif	/* ASM_X86_CMPXCHG_H */
arch/x86/include/asm/cmpxchg_32.h +0 -1
@@ -103,7 +103,6 @@ static inline bool __try_cmpxchg64(volat
 
 #endif
 
-#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX8)
 #define system_has_cmpxchg64()		boot_cpu_has(X86_FEATURE_CX8)
 
 #endif /* _ASM_X86_CMPXCHG_32_H */
arch/x86/include/asm/cmpxchg_64.h +0 -1
@@ -81,7 +81,6 @@ static __always_inline bool arch_try_cmp
 	return __arch_try_cmpxchg128(ptr, oldp, new,);
 }
 
-#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX16)
 #define system_has_cmpxchg128()		boot_cpu_has(X86_FEATURE_CX16)
 
 #endif /* _ASM_X86_CMPXCHG_64_H */
arch/x86/include/asm/percpu.h +0 -42

54 lines not on this page · raw message

include/asm-generic/percpu.h +0 -58

80 lines not on this page · raw message

include/linux/atomic/atomic-instrumented.h +1 -16
@@ -2234,21 +2234,6 @@ atomic_long_dec_if_positive(atomic_long_
 	arch_try_cmpxchg128_local(__ai_ptr, __ai_oldp, __VA_ARGS__); \
 })
 
-#define cmpxchg_double(ptr, ...) \
-({ \
-	typeof(ptr) __ai_ptr = (ptr); \
-	kcsan_mb(); \
-	instrument_atomic_read_write(__ai_ptr, 2 * sizeof(*__ai_ptr)); \
-	arch_cmpxchg_double(__ai_ptr, __VA_ARGS__); \
-})
-
-
-#define cmpxchg_double_local(ptr, ...) \
-({ \
-	typeof(ptr) __ai_ptr = (ptr); \
-	instrument_atomic_read_write(__ai_ptr, 2 * sizeof(*__ai_ptr)); \
-	arch_cmpxchg_double_local(__ai_ptr, __VA_ARGS__); \
-})
 
 #endif /* _LINUX_ATOMIC_INSTRUMENTED_H */
-// 82d1be694fab30414527d0877c29fa75ed5a0b74
+// 3611991b015450e119bcd7417a9431af7f3ba13c
include/linux/percpu-defs.h +0 -38
@@ -358,33 +358,6 @@ static __always_inline void __this_cpu_p
 	pscr2_ret__;							\
 })
 
-/*
- * Special handling for cmpxchg_double.  cmpxchg_double is passed two
- * percpu variables.  The first has to be aligned to a double word
- * boundary and the second has to follow directly thereafter.
- * We enforce this on all architectures even if they don't support
- * a double cmpxchg instruction, since it's a cheap requirement, and it
- * avoids breaking the requirement for architectures with the instruction.
- */
-#define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...)		\
-({									\
-	bool pdcrb_ret__;						\
-	__verify_pcpu_ptr(&(pcp1));					\
-	BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2));			\
-	VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1)));	\
-	VM_BUG_ON((unsigned long)(&(pcp2)) !=				\
-		  (unsigned long)(&(pcp1)) + sizeof(pcp1));		\
-	switch(sizeof(pcp1)) {						\
-	case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break;	\
-	default:							\
-		__bad_size_call_parameter(); break;			\
-	}								\
-	pdcrb_ret__;							\
-})
-
 #define __pcpu_size_call(stem, variable, ...)				\
 do {									\
 	__verify_pcpu_ptr(&(variable));					\
@@ -443,9 +416,6 @@ do {									\
 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
-#define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-	__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
-
 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
@@ -505,11 +475,6 @@ do {									\
 	raw_cpu_cmpxchg(pcp, oval, nval);				\
 })
 
-#define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-({	__this_cpu_preempt_check("cmpxchg_double");			\
-	raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2);	\
-})
-
 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
@@ -532,9 +497,6 @@ do {									\
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
-#define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
-
 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
scripts/atomic/gen-atomic-instrumented.sh +4 -11
@@ -84,7 +84,6 @@ gen_xchg()
 {
 	local xchg="$1"; shift
 	local order="$1"; shift
-	local mult="$1"; shift
 
 	kcsan_barrier=""
 	if [ "${xchg%_local}" = "${xchg}" ]; then
@@ -104,8 +103,8 @@ cat <<EOF
 EOF
 [ -n "$kcsan_barrier" ] && printf "\t${kcsan_barrier}; \\\\\n"
 cat <<EOF
-	instrument_atomic_read_write(__ai_ptr, ${mult}sizeof(*__ai_ptr)); \\
-	instrument_read_write(__ai_oldp, ${mult}sizeof(*__ai_oldp)); \\
+	instrument_atomic_read_write(__ai_ptr, sizeof(*__ai_ptr)); \\
+	instrument_read_write(__ai_oldp, sizeof(*__ai_oldp)); \\
 	arch_${xchg}${order}(__ai_ptr, __ai_oldp, __VA_ARGS__); \\
 })
 EOF
@@ -168,22 +167,16 @@ done
 
 for xchg in "xchg" "cmpxchg" "cmpxchg64" "cmpxchg128" "try_cmpxchg" "try_cmpxchg64" "try_cmpxchg128"; do
 	for order in "" "_acquire" "_release" "_relaxed"; do
-		gen_xchg "${xchg}" "${order}" ""
+		gen_xchg "${xchg}" "${order}"
 		printf "\n"
 	done
 done
 
 for xchg in "cmpxchg_local" "cmpxchg64_local" "cmpxchg128_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" "try_cmpxchg128_local"; do
-	gen_xchg "${xchg}" "" ""
+	gen_xchg "${xchg}" ""
 	printf "\n"
 done
 
-gen_xchg "cmpxchg_double" "" "2 * "
-
-printf "\n\n"
-
-gen_xchg "cmpxchg_double_local" "" "2 * "
-
 cat <<EOF
 
 #endif /* _LINUX_ATOMIC_INSTRUMENTED_H */
author maintainer
#3 lore raw

PATCH 2/12 types: Introduce [us]128

In reply to Peter Zijlstra

Introduce [us]128 (when available). Unlike [us]64, ensure they are always naturally aligned.

This also enables 128bit wide atomics (which require natural alignment) such as cmpxchg128().

---
 include/linux/types.h          |    5 +++++
 include/uapi/linux/types.h     |    4 ++++
 lib/crypto/curve25519-hacl64.c |    2 --
 lib/crypto/poly1305-donna64.c  |    2 --
 4 files changed, 9 insertions(+), 4 deletions(-)
include/linux/types.h +5 -0
@@ -10,6 +10,11 @@
 #define DECLARE_BITMAP(name,bits) \
 	unsigned long name[BITS_TO_LONGS(bits)]
 
+#ifdef __SIZEOF_INT128__
+typedef __s128 s128;
+typedef __u128 u128;
+#endif
+
 typedef u32 __kernel_dev_t;
 
 typedef __kernel_fd_set		fd_set;
include/uapi/linux/types.h +4 -0
@@ -13,6 +13,10 @@
 
 #include <linux/posix_types.h>
 
+#ifdef __SIZEOF_INT128__
+typedef __signed__ __int128 __s128 __attribute__((aligned(16)));
+typedef unsigned __int128 __u128 __attribute__((aligned(16)));
+#endif
 
 /*
  * Below are truly Linux-specific types that should never collide with
lib/crypto/curve25519-hacl64.c +0 -2
@@ -14,8 +14,6 @@
 #include <crypto/curve25519.h>
 #include <linux/string.h>
 
-typedef __uint128_t u128;
-
 static __always_inline u64 u64_eq_mask(u64 a, u64 b)
 {
 	u64 x = a ^ b;
lib/crypto/poly1305-donna64.c +0 -2
@@ -10,8 +10,6 @@
 #include <asm/unaligned.h>
 #include <crypto/internal/poly1305.h>
 
-typedef __uint128_t u128;
-
 void poly1305_core_setkey(struct poly1305_core_key *key,
 			  const u8 raw_key[POLY1305_BLOCK_SIZE])
 {
author maintainer
#36 lore raw

PATCH 9/12 x86,intel_iommu: Replace cmpxchg_double()

In reply to Peter Zijlstra
---
 drivers/iommu/intel/irq_remapping.c |    8 --
 include/linux/dmar.h                |  125 +++++++++++++++++++-----------------
 2 files changed, 68 insertions(+), 65 deletions(-)
drivers/iommu/intel/irq_remapping.c +2 -6
@@ -175,18 +175,14 @@ static int modify_irte(struct irq_2_iomm
 	irte = &iommu->ir_table->base[index];
 
 	if ((irte->pst == 1) || (irte_modified->pst == 1)) {
-		bool ret;
-
-		ret = cmpxchg_double(&irte->low, &irte->high,
-				     irte->low, irte->high,
-				     irte_modified->low, irte_modified->high);
 		/*
 		 * We use cmpxchg16 to atomically update the 128-bit IRTE,
 		 * and it cannot be updated by the hardware or other processors
 		 * behind us, so the return value of cmpxchg16 should be the
 		 * same as the old value.
 		 */
-		WARN_ON(!ret);
+		u128 old = irte->irte;
+		WARN_ON(!try_cmpxchg128(&irte->irte, &old, irte_modified->irte));
 	} else {
 		WRITE_ONCE(irte->low, irte_modified->low);
 		WRITE_ONCE(irte->high, irte_modified->high);
include/linux/dmar.h +66 -59

133 lines not on this page · raw message

author maintainer
#5 lore raw

PATCH 3/12 arch: Introduce arch_{,try_}_cmpxchg128{,_local}()

In reply to Peter Zijlstra

For all architectures that currently support cmpxchg_double() implement the cmpxchg128() family of functions that is basically the same but with a saner interface.

---
 arch/arm64/include/asm/atomic_ll_sc.h |   41 +++++++++++++++++++++
 arch/arm64/include/asm/atomic_lse.h   |   31 ++++++++++++++++
 arch/arm64/include/asm/cmpxchg.h      |   26 +++++++++++++
 arch/s390/include/asm/cmpxchg.h       |   14 +++++++
 arch/x86/include/asm/cmpxchg_32.h     |    3 +
 arch/x86/include/asm/cmpxchg_64.h     |   64 +++++++++++++++++++++++++++++++++-
 6 files changed, 177 insertions(+), 2 deletions(-)
arch/arm64/include/asm/atomic_ll_sc.h +41 -0
@@ -326,6 +326,47 @@ __CMPXCHG_DBL(   ,        ,  ,         )
 __CMPXCHG_DBL(_mb, dmb ish, l, "memory")
 
 #undef __CMPXCHG_DBL
+
+union __u128_halves {
+	u128 full;
+	struct {
+		u64 low, high;
+	};
+};
+
+#define __CMPXCHG128(name, mb, rel, cl...)                             \
+static __always_inline u128						\
+__ll_sc__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)	\
+{									\
+	union __u128_halves r, o = { .full = (old) },			\
+			       n = { .full = (new) };			\
+       unsigned int tmp;                                               \
+									\
+	asm volatile("// __cmpxchg128" #name "\n"			\
+       "       prfm    pstl1strm, %[v]\n"                              \
+       "1:     ldxp    %[rl], %[rh], %[v]\n"                           \
+       "       cmp     %[rl], %[ol]\n"                                 \
+       "       ccmp    %[rh], %[oh], 0, eq\n"                          \
+       "       b.ne    2f\n"                                           \
+       "       st" #rel "xp    %w[tmp], %[nl], %[nh], %[v]\n"          \
+       "       cbnz    %w[tmp], 1b\n"                                  \
+	"	" #mb "\n"						\
+	"2:"								\
+       : [v] "+Q" (*(u128 *)ptr),                                      \
+         [rl] "=&r" (r.low), [rh] "=&r" (r.high),                      \
+         [tmp] "=&r" (tmp)                                             \
+       : [ol] "r" (o.low), [oh] "r" (o.high),                          \
+         [nl] "r" (n.low), [nh] "r" (n.high)                           \
+       : "cc", ##cl);                                                  \
+									\
+	return r.full;							\
+}
+
+__CMPXCHG128(   ,        ,  )
+__CMPXCHG128(_mb, dmb ish, l, "memory")
+
+#undef __CMPXCHG128
+
 #undef K
 
 #endif	/* __ASM_ATOMIC_LL_SC_H */
arch/arm64/include/asm/atomic_lse.h +31 -0
@@ -317,4 +317,35 @@ __CMPXCHG_DBL(_mb, al, "memory")
 
 #undef __CMPXCHG_DBL
 
+#define __CMPXCHG128(name, mb, cl...)					\
+static __always_inline u128						\
+__lse__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)		\
+{									\
+	union __u128_halves r, o = { .full = (old) },			\
+			       n = { .full = (new) };			\
+	register unsigned long x0 asm ("x0") = o.low;			\
+	register unsigned long x1 asm ("x1") = o.high;			\
+	register unsigned long x2 asm ("x2") = n.low;			\
+	register unsigned long x3 asm ("x3") = n.high;			\
+	register unsigned long x4 asm ("x4") = (unsigned long)ptr;	\
+									\
+	asm volatile(							\
+	__LSE_PREAMBLE							\
+	"	casp" #mb "\t%[old1], %[old2], %[new1], %[new2], %[v]\n"\
+	: [old1] "+&r" (x0), [old2] "+&r" (x1),				\
+	  [v] "+Q" (*(u128 *)ptr)					\
+	: [new1] "r" (x2), [new2] "r" (x3), [ptr] "r" (x4),		\
+	  [oldval1] "r" (o.low), [oldval2] "r" (o.high)			\
+	: cl);								\
+									\
+	r.low = x0; r.high = x1;					\
+									\
+	return r.full;							\
+}
+
+__CMPXCHG128(   ,   )
+__CMPXCHG128(_mb, al, "memory")
+
+#undef __CMPXCHG128
+
 #endif	/* __ASM_ATOMIC_LSE_H */
arch/arm64/include/asm/cmpxchg.h +26 -0
@@ -146,6 +146,19 @@ __CMPXCHG_DBL(_mb)
 
 #undef __CMPXCHG_DBL
 
+#define __CMPXCHG128(name)						\
+static inline u128 __cmpxchg128##name(volatile u128 *ptr,		\
+				      u128 old, u128 new)		\
+{									\
+	return __lse_ll_sc_body(_cmpxchg128##name,			\
+				ptr, old, new);				\
+}
+
+__CMPXCHG128(   )
+__CMPXCHG128(_mb)
+
+#undef __CMPXCHG128
+
 #define __CMPXCHG_GEN(sfx)						\
 static __always_inline unsigned long __cmpxchg##sfx(volatile void *ptr,	\
 					   unsigned long old,		\
@@ -228,6 +241,19 @@ __CMPXCHG_GEN(_mb)
 	__ret;									\
 })
 
+/* cmpxchg128 */
+#define system_has_cmpxchg128()		1
+
+#define arch_cmpxchg128(ptr, o, n)						\
+({										\
+	__cmpxchg128_mb((ptr), (o), (n));					\
+})
+
+#define arch_cmpxchg128_local(ptr, o, n)					\
+({										\
+	__cmpxchg128((ptr), (o), (n));						\
+})
+
 #define __CMPWAIT_CASE(w, sfx, sz)					\
 static inline void __cmpwait_case_##sz(volatile void *ptr,		\
 				       unsigned long val)		\
arch/s390/include/asm/cmpxchg.h +14 -0
@@ -224,4 +224,18 @@ static __always_inline int __cmpxchg_dou
 			 (unsigned long)(n1), (unsigned long)(n2));	\
 })
 
+#define system_has_cmpxchg128()		1
+
+static __always_inline u128 arch_cmpxchg128(volatile u128 *ptr, u128 old, u128 new)
+{
+	asm volatile(
+		"	cdsg	%[old],%[new],%[ptr]\n"
+		: [old] "+d" (old), [ptr] "+QS" (*ptr)
+		: [new] "d" (new)
+		: "memory", "cc");
+	return old;
+}
+
+#define arch_cmpxchg128		arch_cmpxchg128
+
 #endif /* __ASM_CMPXCHG_H */
arch/x86/include/asm/cmpxchg_32.h +2 -1
@@ -103,6 +103,7 @@ static inline bool __try_cmpxchg64(volat
 
 #endif
 
-#define system_has_cmpxchg_double() boot_cpu_has(X86_FEATURE_CX8)
+#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX8)
+#define system_has_cmpxchg64()		boot_cpu_has(X86_FEATURE_CX8)
 
 #endif /* _ASM_X86_CMPXCHG_32_H */
arch/x86/include/asm/cmpxchg_64.h +63 -1

69 lines not on this page · raw message

author maintainer
#34 lore raw

PATCH 8/12 x86,amd_iommu: Replace cmpxchg_double()

In reply to Peter Zijlstra
---
 drivers/iommu/amd/amd_iommu_types.h |    9 +++++++--
 drivers/iommu/amd/iommu.c           |   10 ++++------
 2 files changed, 11 insertions(+), 8 deletions(-)
drivers/iommu/amd/amd_iommu_types.h +7 -2
@@ -986,8 +986,13 @@ union irte_ga_hi {
 };
 
 struct irte_ga {
-	union irte_ga_lo lo;
-	union irte_ga_hi hi;
+	union {
+		struct {
+			union irte_ga_lo lo;
+			union irte_ga_hi hi;
+		};
+		u128 irte;
+	};
 };
 
 struct irq_2_irte {
drivers/iommu/amd/iommu.c +4 -6
@@ -3003,10 +3003,10 @@ static int alloc_irq_index(struct amd_io
 static int modify_irte_ga(struct amd_iommu *iommu, u16 devid, int index,
 			  struct irte_ga *irte, struct amd_ir_data *data)
 {
-	bool ret;
 	struct irq_remap_table *table;
-	unsigned long flags;
 	struct irte_ga *entry;
+	unsigned long flags;
+	u128 old;
 
 	table = get_irq_table(iommu, devid);
 	if (!table)
@@ -3017,16 +3017,14 @@ static int modify_irte_ga(struct amd_iom
 	entry = (struct irte_ga *)table->table;
 	entry = &entry[index];
 
-	ret = cmpxchg_double(&entry->lo.val, &entry->hi.val,
-			     entry->lo.val, entry->hi.val,
-			     irte->lo.val, irte->hi.val);
 	/*
 	 * We use cmpxchg16 to atomically update the 128-bit IRTE,
 	 * and it cannot be updated by the hardware or other processors
 	 * behind us, so the return value of cmpxchg16 should be the
 	 * same as the old value.
 	 */
-	WARN_ON(!ret);
+	old = entry->irte;
+	WARN_ON(!try_cmpxchg128(&entry->irte, &old, irte->irte));
 
 	if (data)
 		data->ref = entry;
maintainer
#17 lore raw

Re: [PATCH 07/12] percpu: #ifndef __SIZEOF_INT128__

In reply to Peter Zijlstra
Quoted 3 lines · Peter Zijlstra

Some 64bit architectures do not advertise __SIZEOF_INT128__ on all supported compiler versions. Notably the HPPA64 only started doing with GCC-11.

I checked the other compilers to be sure that anything else we support (gcc-5.1 and up) across all 64-bit architectures does support int128.

It would be nice to have the hack more localized to parisc and guarded with a CONFIG_GCC_VERSION check so we can kill it off in the future, once we drop either gcc-10 or parisc support.

Quoted 12 lines
include/asm-generic/percpu.h +12 -0
+#ifndef __SIZEOF_INT128__
+#define raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (!__builtin_memcmp(&__val, &__old, sizeof(pcp))) {		\
+		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\

include/asm-generic/percpu.h do { \ · line 337

Quoted 10 lines
include/asm-generic/percpu.h +10 -0
+	}								\
+	__ret;								\
+})
+
+#define raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg_memcpy(pcp, &__old, nval);		\
+	__old;								\
+})

Instead of having this in include/asm-generic under !__SIZEOF_INT128__, could you just move this into the parisc files with a compiler version check?

Arnd

reviewer Reviewed-by Tested-by on the v1 series
#44 lore raw
In reply to Peter Zijlstra
Quoted 16 lines · Peter Zijlstra

Hi!

After much breaking of things, find here the improved version.

Since v3:

- unbreak everything that does *NOT* have cmpxchg128()

Notably this_cpu_cmpxchg_double() is used unconditionally by SLUB which means that this_cpu_try_cmpxchg128() needs to be unconditionally available on all 64bit architectures.

- fixed up x86/x86_64 cmpxchg{8,16}b emulation for this_cpu_cmpxchg{64,128}()

- introduce {raw,this}_cpu_try_cmpxchg*()

Quoted 8 lines

- add fallback for !__SIZEOF_INT128__ 64bit architectures

Sadly there are supported 64bit architecture/compiler combinations that do not have __SIZEOF_INT128__, specifically it was found that HPPA64 only added this with GCC-11.

this is yuck, and ideally we'd simply raise compiler requirements, but this 'works'.

The patches look good to me, and I used my local cross-build script to build test this with the kernel.org GCC 10.3.0 cross toolchain for all of the following arch/triplet/config combinations:

  alpha           alpha-linux             defconfig
  arc             arc-linux               defconfig
  arm             arm-linux-gnueabi       multi_v4t_defconfig
  arm             arm-linux-gnueabi       multi_v5_defconfig
  arm             arm-linux-gnueabi       multi_v7_defconfig
  arm             arm-linux-gnueabi       omap1_defconfig
  arm64           aarch64-linux           defconfig
  csky            csky-linux              defconfig
  i386            i386-linux              defconfig
  ia64            ia64-linux              defconfig
  m68k            m68k-linux              defconfig
  microblaze      microblaze-linux        defconfig
  mips            mips-linux              32r1_defconfig
  mips            mips-linux              32r2_defconfig
  mips            mips-linux              32r6_defconfig
  mips            mips64-linux            64r1_defconfig
  mips            mips64-linux            64r2_defconfig
  mips            mips64-linux            64r6_defconfig
  nios2           nios2-linux             defconfig
  openrisc        or1k-linux              defconfig
  parisc          hppa-linux              generic-32bit_defconfig
  parisc          hppa64-linux            generic-64bit_defconfig
  powerpc         powerpc-linux           ppc40x_defconfig
  powerpc         powerpc64-linux         ppc64_defconfig
  powerpc         powerpc64-linux         ppc64e_defconfig
  riscv           riscv32-linux           rv32_defconfig
  riscv           riscv64-linux           defconfig
  s390            s390-linux              defconfig
  sh              sh4-linux               defconfig
  sparc           sparc-linux             sparc32_defconfig
  sparc           sparc64-linux           sparc64_defconfig
  x86_64          x86_64-linux            defconfig
  xtensa          xtensa-linux            defconfig

... and everything seemed happy.

I've also boot-tested arm64 defconfig.

So FWIW, for the series:

Quoted 1 line

My plan is to re-add this to tip/locking/core and thus -next later this week.

I'll need to rebase my kerneldoc series atop this, so getting this into a stable branch soon would be great!

Thanks,
Mark.

Quoted 42 lines

Also available at:

git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git locking/core

---
 Documentation/core-api/this_cpu_ops.rst     |   2 -
 arch/arm64/include/asm/atomic_ll_sc.h       |  56 +++---
 arch/arm64/include/asm/atomic_lse.h         |  39 ++---
 arch/arm64/include/asm/cmpxchg.h            |  48 ++----
 arch/arm64/include/asm/percpu.h             |  30 ++--
 arch/s390/include/asm/cmpxchg.h             |  32 +---
 arch/s390/include/asm/cpu_mf.h              |   2 +-
 arch/s390/include/asm/percpu.h              |  34 ++--
 arch/s390/kernel/perf_cpum_sf.c             |  16 +-
 arch/x86/include/asm/cmpxchg.h              |  25 ---
 arch/x86/include/asm/cmpxchg_32.h           |   2 +-
 arch/x86/include/asm/cmpxchg_64.h           |  63 ++++++-
 arch/x86/include/asm/percpu.h               | 102 ++++++-----
 arch/x86/lib/Makefile                       |   3 +-
 arch/x86/lib/cmpxchg16b_emu.S               |  43 +++--
 arch/x86/lib/cmpxchg8b_emu.S                |  67 ++++++--
 drivers/iommu/amd/amd_iommu_types.h         |   9 +-
 drivers/iommu/amd/iommu.c                   |  10 +-
 drivers/iommu/intel/irq_remapping.c         |   8 +-
 include/asm-generic/percpu.h                | 257 ++++++++++++++++++++++------
 include/crypto/b128ops.h                    |  14 +-
 include/linux/atomic/atomic-arch-fallback.h |  95 +++++++++-
 include/linux/atomic/atomic-instrumented.h  |  93 ++++++++--
 include/linux/dmar.h                        | 125 +++++++-------
 include/linux/percpu-defs.h                 |  45 ++---
 include/linux/slub_def.h                    |  12 +-
 include/linux/types.h                       |  12 ++
 include/uapi/linux/types.h                  |   4 +
 lib/crypto/curve25519-hacl64.c              |   2 -
 lib/crypto/poly1305-donna64.c               |   2 -
 mm/slab.h                                   |  53 +++++-
 mm/slub.c                                   | 139 +++++++++------
 scripts/atomic/gen-atomic-fallback.sh       |   4 +-
 scripts/atomic/gen-atomic-instrumented.sh   |  19 +-
 34 files changed, 952 insertions(+), 515 deletions(-)
author maintainer
#18 lore raw

Re: [PATCH 07/12] percpu: #ifndef __SIZEOF_INT128__

In reply to Arnd Bergmann
Quoted 8 lines · Arnd Bergmann
Quoted 3 lines · Peter Zijlstra

Some 64bit architectures do not advertise __SIZEOF_INT128__ on all supported compiler versions. Notably the HPPA64 only started doing with GCC-11.

I checked the other compilers to be sure that anything else we support (gcc-5.1 and up) across all 64-bit architectures does support int128.

Oh excellent -- I didn't have sufficient old cross compilers to verify, hence me not doing what you suggest below.

If HPPA64 really is the only one so affected, then yes, I can move this hack into arch/parisc.

author maintainer
#19 lore raw

PATCH v2 7/12 parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Arnd Bergmann
Quoted 4 lines · Arnd Bergmann

It would be nice to have the hack more localized to parisc and guarded with a CONFIG_GCC_VERSION check so we can kill it off in the future, once we drop either gcc-10 or parisc support.

I vote for dropping parisc -- it's the only 64bit arch that doesn't have sane atomics.

Anyway, the below seems to work -- build tested with GCC-10.1

---
Subject: parisc/percpu: Work around the lack of __SIZEOF_INT128__
From: Peter Zijlstra <[email protected]>
Date: Tue May 30 22:27:40 CEST 2023

HPPA64 is unique in not providing __SIZEOF_INT128__ across all supported compilers, specifically it only started doing this with GCC-11.

Since the per-cpu ops are universally availably, and this_cpu_{,try_}cmpxchg128() is expected to be available on all 64bit architectures a wee bodge is in order.

Sadly, while C reverts to memcpy() for assignment of POD types, it does not revert to memcmp() for for equality. Therefore frob that manually.

---
 arch/parisc/include/asm/percpu.h |   77 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 77 insertions(+)
arch/parisc/include/asm/percpu.h new file +77 -0
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_PARISC_PERCPU_H
+#define _ASM_PARISC_PERCPU_H
+
+#include <linux/types.h>
+
+#if defined(CONFIG_64BIT) && CONFIG_GCC_VERSION < 1100000
+
+/*
+ * GCC prior to 11 does not provide __SIZEOF_INT128__ on HPPA64
+ * as such we need to provide an alternative implementation of
+ * {raw,this}_cpu_{,try_}cmpxchg128().
+ *
+ * This obviously doesn't function as u128 should, but for the purpose
+ * of per-cpu cmpxchg128 it might just do.
+ */
+typedef struct {
+	u64 a, b;
+} u128 __attribute__((aligned(16)));
+
+#define raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (!__builtin_memcmp(&__val, &__old, sizeof(pcp))) {		\
+		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
+	__ret;								\
+})
+
+#define raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg_memcpy(pcp, &__old, nval);		\
+	__old;								\
+})
+
+#define raw_cpu_cmpxchg128(pcp, oval, nval) \
+	raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+
+#define this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)		\
+({									\
+	typeof(pcp) __ret;						\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_cmpxchg128(pcp, oval, nval) \
+	this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define this_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+
+#endif /* !__SIZEOF_INT128__ */
+
+#include <asm-generic/percpu.h>
+
+#endif /* _ASM_PARISC_PERCPU_H */
maintainer
#20 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 9 lines · Peter Zijlstra
Quoted 4 lines · Arnd Bergmann

It would be nice to have the hack more localized to parisc and guarded with a CONFIG_GCC_VERSION check so we can kill it off in the future, once we drop either gcc-10 or parisc support.

I vote for dropping parisc -- it's the only 64bit arch that doesn't have sane atomics.

Of course I'm against dropping parisc.

Quoted 1 line

Anyway, the below seems to work -- build tested with GCC-10.1

I don't think we need to care about gcc-10 on parisc.
Debian and Gentoo are the only supported distributions, while Debian requires gcc-12 to build > 6.x kernels, and I assume Gentoo uses at least gcc-12 as well.

So raising the gcc limit for parisc only (at least temporarily for now) should be fine and your workaround below wouldn't be necessary, right?

Helge

Quoted 101 lines

---
Subject: parisc/percpu: Work around the lack of __SIZEOF_INT128__
From: Peter Zijlstra <[email protected]>
Date: Tue May 30 22:27:40 CEST 2023

HPPA64 is unique in not providing __SIZEOF_INT128__ across all supported compilers, specifically it only started doing this with GCC-11.

Since the per-cpu ops are universally availably, and this_cpu_{,try_}cmpxchg128() is expected to be available on all 64bit architectures a wee bodge is in order.

Sadly, while C reverts to memcpy() for assignment of POD types, it does not revert to memcmp() for for equality. Therefore frob that manually.

---
  arch/parisc/include/asm/percpu.h |   77 +++++++++++++++++++++++++++++++++++++++
  1 file changed, 77 insertions(+)
arch/parisc/include/asm/percpu.h new file +77 -0
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_PARISC_PERCPU_H
+#define _ASM_PARISC_PERCPU_H
+
+#include <linux/types.h>
+
+#if defined(CONFIG_64BIT) && CONFIG_GCC_VERSION < 1100000
+
+/*
+ * GCC prior to 11 does not provide __SIZEOF_INT128__ on HPPA64
+ * as such we need to provide an alternative implementation of
+ * {raw,this}_cpu_{,try_}cmpxchg128().
+ *
+ * This obviously doesn't function as u128 should, but for the purpose
+ * of per-cpu cmpxchg128 it might just do.
+ */
+typedef struct {
+	u64 a, b;
+} u128 __attribute__((aligned(16)));
+
+#define raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (!__builtin_memcmp(&__val, &__old, sizeof(pcp))) {		\
+		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
+	__ret;								\
+})
+
+#define raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg_memcpy(pcp, &__old, nval);		\
+	__old;								\
+})
+
+#define raw_cpu_cmpxchg128(pcp, oval, nval) \
+	raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+
+#define this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)		\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)		\
+({									\
+	typeof(pcp) __ret;						\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval);	\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
+#define this_cpu_cmpxchg128(pcp, oval, nval) \
+	this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval)
+#define this_cpu_try_cmpxchg128(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval)
+
+#endif /* !__SIZEOF_INT128__ */
+
+#include <asm-generic/percpu.h>
+
+#endif /* _ASM_PARISC_PERCPU_H */
author maintainer
#21 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 8 lines · Helge Deller
Quoted 7 lines · Peter Zijlstra
Quoted 4 lines · Arnd Bergmann

It would be nice to have the hack more localized to parisc and guarded with a CONFIG_GCC_VERSION check so we can kill it off in the future, once we drop either gcc-10 or parisc support.

Quoted 4 lines
Quoted 2 lines

I vote for dropping parisc -- it's the only 64bit arch that doesn't have sane atomics.

Of course I'm against dropping parisc.

:-)

Quoted 9 lines
Quoted 1 line

Anyway, the below seems to work -- build tested with GCC-10.1

I don't think we need to care about gcc-10 on parisc.
Debian and Gentoo are the only supported distributions, while Debian requires gcc-12 to build > 6.x kernels, and I assume Gentoo uses at least gcc-12 as well.

So raising the gcc limit for parisc only (at least temporarily for now) should be fine and your workaround below wouldn't be necessary, right?

Correct, if you're willing to set minimum GCC version to 11 for parisc all is well and this patch can go play in the bit bucket.

#23 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Helge Deller
Quoted 8 lines · Helge Deller

I don't think we need to care about gcc-10 on parisc.
Debian and Gentoo are the only supported distributions, while Debian requires gcc-12 to build > 6.x kernels, and I assume Gentoo uses at least gcc-12 as well.

So raising the gcc limit for parisc only (at least temporarily for now) should be fine and your workaround below wouldn't be necessary, right?

This absolutely sounds like the right option. Let's simplify the problem space by just saying that parisc needs the newer compiler.

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

I wonder if we could move the gcc minimum version check into the Kconfig file instead, and make it easier to let architectures override the minimum version.

I don't quite know how to do that sanely, though. I don't think we have a sane way to error out at Kconfig time (except by forcing some syntax error inside an 'if' statement or something horrendously hacky like that).

Added Masahiro to the (already overlong) participants list.

Linus

#24 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 14 lines · Linus Torvalds
Quoted 8 lines · Helge Deller

I don't think we need to care about gcc-10 on parisc.
Debian and Gentoo are the only supported distributions, while Debian requires gcc-12 to build > 6.x kernels, and I assume Gentoo uses at least gcc-12 as well.

So raising the gcc limit for parisc only (at least temporarily for now) should be fine and your workaround below wouldn't be necessary, right?

This absolutely sounds like the right option. Let's simplify the problem space by just saying that parisc needs the newer compiler.

Quoted 7 lines

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

I wonder if we could move the gcc minimum version check into the Kconfig file instead, and make it easier to let architectures override the minimum version.

Currently, it is invoked in the Kconfig time,
but not directly in Kconfig files.

scripts/Kconfig.include
-> scripts/cc-version.sh

    -> scripts/min-tool-version.sh

It would be ugly if we wrote the equivalent code
directly in Kconfig files.

Quoted 5 lines

I don't quite know how to do that sanely, though. I don't think we have a sane way to error out at Kconfig time (except by forcing some syntax error inside an 'if' statement or something horrendously hacky like that).

The parse stage can fail by $(error-if ) macro, but the evaluation stage never fails. I think it is a design.

I think checking the compiler version during the parse stage makes sense given the current situation. The compiler version is fixed when Kconfig starts. If the compiler is found to be too old, there is no meaning to proceed.

You suggested to choose a compiler in the Kconfig time:
https://lore.kernel.org/lkml/CAHk-=whdrvCkSWh=BRrwZwNo3=yLBXXM88NGx8VEpP1VTgmkyQ@mail.gmail.com/

When we achieve that, moving the min version to Kconfig files will be the right thing to do. Then, everything will be evaluated dynamically.

Quoted 4 lines

Added Masahiro to the (already overlong) participants list.

Linus

Signature · 2 lines
Best Regards
Masahiro Yamada
#22 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Peter Zijlstra
Quoted 23 lines · Peter Zijlstra
Quoted 8 lines · Helge Deller
Quoted 7 lines · Peter Zijlstra
Quoted 4 lines · Arnd Bergmann

It would be nice to have the hack more localized to parisc and guarded with a CONFIG_GCC_VERSION check so we can kill it off in the future, once we drop either gcc-10 or parisc support.

Quoted 4 lines
Quoted 2 lines

I vote for dropping parisc -- it's the only 64bit arch that doesn't have sane atomics.

Of course I'm against dropping parisc.

:-)

Quoted 7 lines
Quoted 1 line

Anyway, the below seems to work -- build tested with GCC-10.1

I don't think we need to care about gcc-10 on parisc.
Debian and Gentoo are the only supported distributions, while Debian requires gcc-12 to build > 6.x kernels, and I assume Gentoo uses at least gcc-12 as well.

Quoted 5 lines
Quoted 2 lines

So raising the gcc limit for parisc only (at least temporarily for now) should be fine and your workaround below wouldn't be necessary, right?

Correct, if you're willing to set minimum GCC version to 11 for parisc all is well and this patch can go play in the bit bucket.

It's fine for us in Gentoo, thanks!

author maintainer
#25 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Linus Torvalds
Quoted 3 lines · Linus Torvalds

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

Something like so then?

---
Subject: parisc: Raise minimal GCC version
From: Peter Zijlstra <[email protected]>
Date: Fri Jun 2 16:33:54 CEST 2023

With 64bit builds depending on __SIZEOF_INT128__ raise the parisc minimum compiler version to gcc-11.0.0.

All other 64bit architectures provide this from GCC-5.1.0 (and probably before), except hppa64 which only started advertising this with GCC-11.

---
 scripts/min-tool-version.sh |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
scripts/min-tool-version.sh +5 -1
@@ -17,7 +17,11 @@ binutils)
 	echo 2.25.0
 	;;
 gcc)
-	echo 5.1.0
+	if [ "$SRCARCH" = parisc ]; then
+		echo 11.0.0
+	else
+		echo 5.1.0
+	fi
 	;;
 llvm)
 	if [ "$SRCARCH" = s390 ]; then
reviewer Tested-by
#26 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 30 lines · Peter Zijlstra
Quoted 3 lines · Linus Torvalds

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

Something like so then?

---
Subject: parisc: Raise minimal GCC version
From: Peter Zijlstra <[email protected]>
Date: Fri Jun 2 16:33:54 CEST 2023

With 64bit builds depending on __SIZEOF_INT128__ raise the parisc minimum compiler version to gcc-11.0.0.

All other 64bit architectures provide this from GCC-5.1.0 (and probably before), except hppa64 which only started advertising this with GCC-11.

---
 scripts/min-tool-version.sh |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
scripts/min-tool-version.sh +0 -0
@@ -17,7 +17,11 @@ binutils)
 	echo 2.25.0
 	;;

scripts/min-tool-version.sh binutils) · line 27

Quoted 10 lines
scripts/min-tool-version.sh +5 -1
 gcc)
-	echo 5.1.0
+	if [ "$SRCARCH" = parisc ]; then
+		echo 11.0.0
+	else
+		echo 5.1.0
+	fi
 	;;
 llvm)
 	if [ "$SRCARCH" = s390 ]; then

I gave this a spin and it looks good to me:

[mark@lakrids:~/src/linux]% usekorg 10.3.0 make ARCH=arm64 CROSS_COMPILE=aarch64-linux- defconfig

  HOSTCC  scripts/basic/fixdep
  HOSTCC  scripts/kconfig/conf.o
  HOSTCC  scripts/kconfig/confdata.o
  HOSTCC  scripts/kconfig/expr.o
  LEX     scripts/kconfig/lexer.lex.c
  YACC    scripts/kconfig/parser.tab.[ch]
  HOSTCC  scripts/kconfig/lexer.lex.o
  HOSTCC  scripts/kconfig/menu.o
  HOSTCC  scripts/kconfig/parser.tab.o
  HOSTCC  scripts/kconfig/preprocess.o
  HOSTCC  scripts/kconfig/symbol.o
  HOSTCC  scripts/kconfig/util.o
  HOSTLD  scripts/kconfig/conf
*** Default configuration is based on 'defconfig'
#
# configuration written to .config
#
[mark@lakrids:~/src/linux]% usekorg 10.3.0 make ARCH=parisc CROSS_COMPILE=hppa64-linux- generic-64bit_defconfig
***
*** C compiler is too old.
***   Your GCC version:    10.3.0
***   Minimum GCC version: 11.0.0
***
scripts/Kconfig.include:44: Sorry, this C compiler is not supported.
make[1]: *** [scripts/kconfig/Makefile:94: generic-64bit_defconfig] Error 1
make: *** [Makefile:692: generic-64bit_defconfig] Error 2
[mark@lakrids:~/src/linux]% usekorg 11.3.0 make ARCH=parisc CROSS_COMPILE=hppa64-linux- generic-64bit_defconfig
#
# configuration written to .config
#

FWIW:

Mark.

#27 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Peter Zijlstra
Quoted 2 lines · Peter Zijlstra

Something like so then?

Ack. I think it would be much cleaner if we would have it as part of the Kconfig file and architectures could just override some GCC_MIN_VERSION value, but that's not the universe we currently have, so your patch looks like the best thing to do.

Linus

#28 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Peter Zijlstra
Quoted 30 lines · Peter Zijlstra
Quoted 3 lines · Linus Torvalds

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

Something like so then?

---
Subject: parisc: Raise minimal GCC version
From: Peter Zijlstra <[email protected]>
Date: Fri Jun 2 16:33:54 CEST 2023

With 64bit builds depending on __SIZEOF_INT128__ raise the parisc minimum compiler version to gcc-11.0.0.

All other 64bit architectures provide this from GCC-5.1.0 (and probably before), except hppa64 which only started advertising this with GCC-11.

---
scripts/min-tool-version.sh | 6 +++++-

1 file changed, 5 insertions(+), 1 deletion(-)
scripts/min-tool-version.sh +0 -0
@@ -17,7 +17,11 @@ binutils)
 	echo 2.25.0
 	;;

scripts/min-tool-version.sh binutils) · line 27

Quoted 10 lines
scripts/min-tool-version.sh +5 -1
 gcc)
-	echo 5.1.0
+	if [ "$SRCARCH" = parisc ]; then
+		echo 11.0.0
+	else
+		echo 5.1.0
+	fi
 	;;
 llvm)
 	if [ "$SRCARCH" = s390 ]; then

Dumb question: is this only about the cpp macro or is it about __int128 existing at all?

author maintainer
#29 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 1 line · H. Peter Anvin

Dumb question: is this only about the cpp macro or is it about __int128 existing at all?

It's mostly about __int128 being there, __SIZEOF_INT128__ is just the way we go about detecting if it's there or not.

#30 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Ok. So the patch description needs to be fixed. Otherwise the solution would be far simpler :)

author maintainer
#31 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 1 line · H. Peter Anvin

Ok. So the patch description needs to be fixed. Otherwise the solution would be far simpler :)

"With 64bit builds depending on __SIZEOF_INT128__ to detect the presence of __int128 raise the parisc minimum compiler version to gcc-11.0.0."

better?

#32 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

Quoted 6 lines · Peter Zijlstra

"With 64bit builds depending on __SIZEOF_INT128__ to detect the presence of __int128 raise the parisc minimum compiler version to gcc-11.0.0."

better?

I'd just say "64-bit targets need the __int128 type, which for pa-risc means raising the minimum gcc version to 11".

The __SIZEOF_INT128__ part isn't the important part. That's just the symptom.

Linus

maintainer Acked-by
#33 lore raw

Re: [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__

In reply to Peter Zijlstra
Quoted 13 lines · Peter Zijlstra
Quoted 3 lines · Linus Torvalds

Right now we have that "minimum gcc version" in a somewhat annoying
place: it's in the ./scripts/min-tool-version.sh file as a shell script.

Something like so then?

---
Subject: parisc: Raise minimal GCC version
From: Peter Zijlstra <[email protected]>
Date: Fri Jun 2 16:33:54 CEST 2023

Quoted 8 lines

With 64bit builds depending on __SIZEOF_INT128__ raise the parisc minimum compiler version to gcc-11.0.0.

All other 64bit architectures provide this from GCC-5.1.0 (and probably before), except hppa64 which only started advertising this with GCC-11.

The patch raises the compiler for 32- and 64-bit parisc builds, but that's OK.

So:

Thank you!
Helge

Quoted 19 lines
---
  scripts/min-tool-version.sh |    6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)
scripts/min-tool-version.sh +5 -1
@@ -17,7 +17,11 @@ binutils)
  	echo 2.25.0
  	;;
  gcc)
-	echo 5.1.0
+	if [ "$SRCARCH" = parisc ]; then
+		echo 11.0.0
+	else
+		echo 5.1.0
+	fi
  	;;
  llvm)
  	if [ "$SRCARCH" = s390 ]; then
merge bot Applied
#43 lore raw

[tip: locking/core] s390/cpum_sf: Convert to cmpxchg128()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     497cc42bf53b55185ab3d39c634fbf09eb6681ae
Gitweb:        https://git.kernel.org/tip/497cc42bf53b55185ab3d39c634fbf09eb6681ae
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:45 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:40 +02:00

s390/cpum_sf: Convert to cmpxchg128()

Now that there is a cross arch u128 and cmpxchg128(), use those instead of the custom CDSG helper.

---
 arch/s390/include/asm/cpu_mf.h  |  2 +-
 arch/s390/kernel/perf_cpum_sf.c | 16 +++-------------
 2 files changed, 4 insertions(+), 14 deletions(-)
arch/s390/include/asm/cpu_mf.h +1 -1
index 7e417d7..a0de5b9 100644
@@ -140,7 +140,7 @@ union hws_trailer_header {
 		unsigned int dsdes:16;	/* 48-63: size of diagnostic SDE */
 		unsigned long long overflow; /* 64 - Overflow Count   */
 	};
-	__uint128_t val;
+	u128 val;
 };
 
 struct hws_trailer_entry {
arch/s390/kernel/perf_cpum_sf.c +3 -13
index 7ef72f5..8ecfbce 100644
@@ -1271,16 +1271,6 @@ static void hw_collect_samples(struct perf_event *event, unsigned long *sdbt,
 	}
 }
 
-static inline __uint128_t __cdsg(__uint128_t *ptr, __uint128_t old, __uint128_t new)
-{
-	asm volatile(
-		"	cdsg	%[old],%[new],%[ptr]\n"
-		: [old] "+d" (old), [ptr] "+QS" (*ptr)
-		: [new] "d" (new)
-		: "memory", "cc");
-	return old;
-}
-
 /* hw_perf_event_update() - Process sampling buffer
  * @event:	The perf event
  * @flush_all:	Flag to also flush partially filled sample-data-blocks
@@ -1352,7 +1342,7 @@ static void hw_perf_event_update(struct perf_event *event, int flush_all)
 			new.f = 0;
 			new.a = 1;
 			new.overflow = 0;
-			prev.val = __cdsg(&te->header.val, old.val, new.val);
+			prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 		} while (prev.val != old.val);
 
 		/* Advance to next sample-data-block */
@@ -1562,7 +1552,7 @@ static bool aux_set_alert(struct aux_buffer *aux, unsigned long alert_index,
 		}
 		new.a = 1;
 		new.overflow = 0;
-		prev.val = __cdsg(&te->header.val, old.val, new.val);
+		prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 	} while (prev.val != old.val);
 	return true;
 }
@@ -1636,7 +1626,7 @@ static bool aux_reset_buffer(struct aux_buffer *aux, unsigned long range,
 				new.a = 1;
 			else
 				new.a = 0;
-			prev.val = __cdsg(&te->header.val, old.val, new.val);
+			prev.val = cmpxchg128(&te->header.val, old.val, new.val);
 		} while (prev.val != old.val);
 		*overflow += orig_overflow;
 	}
merge bot Applied
#35 lore raw

[tip: locking/core] x86,amd_iommu: Replace cmpxchg_double()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     0a0a6800b02489c1288f963ad290b6a4876a2cc8
Gitweb:        https://git.kernel.org/tip/0a0a6800b02489c1288f963ad290b6a4876a2cc8
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:41 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:38 +02:00

x86,amd_iommu: Replace cmpxchg_double()

---
 drivers/iommu/amd/amd_iommu_types.h |  9 +++++++--
 drivers/iommu/amd/iommu.c           | 10 ++++------
 2 files changed, 11 insertions(+), 8 deletions(-)
drivers/iommu/amd/amd_iommu_types.h +7 -2
index 2ddbda3..ab8aa8f 100644
@@ -986,8 +986,13 @@ union irte_ga_hi {
 };
 
 struct irte_ga {
-	union irte_ga_lo lo;
-	union irte_ga_hi hi;
+	union {
+		struct {
+			union irte_ga_lo lo;
+			union irte_ga_hi hi;
+		};
+		u128 irte;
+	};
 };
 
 struct irq_2_irte {
drivers/iommu/amd/iommu.c +4 -6
index 4a31464..1e9f85e 100644
@@ -3003,10 +3003,10 @@ out:
 static int modify_irte_ga(struct amd_iommu *iommu, u16 devid, int index,
 			  struct irte_ga *irte, struct amd_ir_data *data)
 {
-	bool ret;
 	struct irq_remap_table *table;
-	unsigned long flags;
 	struct irte_ga *entry;
+	unsigned long flags;
+	u128 old;
 
 	table = get_irq_table(iommu, devid);
 	if (!table)
@@ -3017,16 +3017,14 @@ static int modify_irte_ga(struct amd_iommu *iommu, u16 devid, int index,
 	entry = (struct irte_ga *)table->table;
 	entry = &entry[index];
 
-	ret = cmpxchg_double(&entry->lo.val, &entry->hi.val,
-			     entry->lo.val, entry->hi.val,
-			     irte->lo.val, irte->hi.val);
 	/*
 	 * We use cmpxchg16 to atomically update the 128-bit IRTE,
 	 * and it cannot be updated by the hardware or other processors
 	 * behind us, so the return value of cmpxchg16 should be the
 	 * same as the old value.
 	 */
-	WARN_ON(!ret);
+	old = entry->irte;
+	WARN_ON(!try_cmpxchg128(&entry->irte, &old, irte->irte));
 
 	if (data)
 		data->ref = entry;
merge bot Applied
#37 lore raw

[tip: locking/core] x86,intel_iommu: Replace cmpxchg_double()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     b1fe7f2cda2a003afe316ce8dfe8d3645694a67e
Gitweb:        https://git.kernel.org/tip/b1fe7f2cda2a003afe316ce8dfe8d3645694a67e
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:42 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:38 +02:00

x86,intel_iommu: Replace cmpxchg_double()

---
 drivers/iommu/intel/irq_remapping.c |   8 +--
 include/linux/dmar.h                | 125 ++++++++++++++-------------
 2 files changed, 68 insertions(+), 65 deletions(-)
drivers/iommu/intel/irq_remapping.c +2 -6
index a1b9873..08f5632 100644
@@ -175,18 +175,14 @@ static int modify_irte(struct irq_2_iommu *irq_iommu,
 	irte = &iommu->ir_table->base[index];
 
 	if ((irte->pst == 1) || (irte_modified->pst == 1)) {
-		bool ret;
-
-		ret = cmpxchg_double(&irte->low, &irte->high,
-				     irte->low, irte->high,
-				     irte_modified->low, irte_modified->high);
 		/*
 		 * We use cmpxchg16 to atomically update the 128-bit IRTE,
 		 * and it cannot be updated by the hardware or other processors
 		 * behind us, so the return value of cmpxchg16 should be the
 		 * same as the old value.
 		 */
-		WARN_ON(!ret);
+		u128 old = irte->irte;
+		WARN_ON(!try_cmpxchg128(&irte->irte, &old, irte_modified->irte));
 	} else {
 		WRITE_ONCE(irte->low, irte_modified->low);
 		WRITE_ONCE(irte->high, irte_modified->high);
include/linux/dmar.h +66 -59

133 lines not on this page · raw message

merge bot Applied
#39 lore raw

[tip: locking/core] slub: Replace cmpxchg_double()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     6801be4f2653e5fdddca73b527cf0728284ba8a3
Gitweb:        https://git.kernel.org/tip/6801be4f2653e5fdddca73b527cf0728284ba8a3
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:43 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:39 +02:00

slub: Replace cmpxchg_double()

---
 include/linux/slub_def.h |  12 ++-
 mm/slab.h                |  53 ++++++++++++--
 mm/slub.c                | 139 ++++++++++++++++++++++----------------
 3 files changed, 137 insertions(+), 67 deletions(-)
include/linux/slub_def.h +9 -3
index f6df03f..deb90cf 100644
@@ -39,7 +39,8 @@ enum stat_item {
 	CPU_PARTIAL_FREE,	/* Refill cpu partial on free */
 	CPU_PARTIAL_NODE,	/* Refill cpu partial from node partial */
 	CPU_PARTIAL_DRAIN,	/* Drain cpu partial to node partial */
-	NR_SLUB_STAT_ITEMS };
+	NR_SLUB_STAT_ITEMS
+};
 
 #ifndef CONFIG_SLUB_TINY
 /*
@@ -47,8 +48,13 @@ enum stat_item {
  * with this_cpu_cmpxchg_double() alignment requirements.
  */
 struct kmem_cache_cpu {
-	void **freelist;	/* Pointer to next available object */
-	unsigned long tid;	/* Globally unique transaction id */
+	union {
+		struct {
+			void **freelist;	/* Pointer to next available object */
+			unsigned long tid;	/* Globally unique transaction id */
+		};
+		freelist_aba_t freelist_tid;
+	};
 	struct slab *slab;	/* The slab from which we are allocating */
 #ifdef CONFIG_SLUB_CPU_PARTIAL
 	struct slab *partial;	/* Partially allocated frozen slabs */
mm/slab.h +46 -7

74 lines not on this page · raw message

mm/slub.c +82 -57

238 lines not on this page · raw message

merge bot Applied
#41 lore raw

[tip: locking/core] arch: Remove cmpxchg_double

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     febe950dbfb464799beb0339cc6fb10699f4a5da
Gitweb:        https://git.kernel.org/tip/febe950dbfb464799beb0339cc6fb10699f4a5da
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:44 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:39 +02:00

arch: Remove cmpxchg_double

No moar users, remove the monster.

---
 Documentation/core-api/this_cpu_ops.rst    |  2 +-
 arch/arm64/include/asm/atomic_ll_sc.h      | 33 +------------
 arch/arm64/include/asm/atomic_lse.h        | 36 +-------------
 arch/arm64/include/asm/cmpxchg.h           | 46 +-----------------
 arch/arm64/include/asm/percpu.h            | 10 +----
 arch/s390/include/asm/cmpxchg.h            | 34 +------------
 arch/s390/include/asm/percpu.h             | 18 +-------
 arch/x86/include/asm/cmpxchg.h             | 25 +---------
 arch/x86/include/asm/cmpxchg_32.h          |  1 +-
 arch/x86/include/asm/cmpxchg_64.h          |  1 +-
 arch/x86/include/asm/percpu.h              | 42 +---------------
 include/asm-generic/percpu.h               | 58 +---------------------
 include/linux/atomic/atomic-instrumented.h | 17 +------
 include/linux/percpu-defs.h                | 38 +--------------
 scripts/atomic/gen-atomic-instrumented.sh  | 15 +----
 15 files changed, 5 insertions(+), 371 deletions(-)
Documentation/core-api/this_cpu_ops.rst +0 -2
index 5cb8b88..91acbcf 100644
@@ -53,7 +53,6 @@ preemption and interrupts::
 	this_cpu_add_return(pcp, val)
 	this_cpu_xchg(pcp, nval)
 	this_cpu_cmpxchg(pcp, oval, nval)
-	this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 	this_cpu_sub(pcp, val)
 	this_cpu_inc(pcp)
 	this_cpu_dec(pcp)
@@ -242,7 +241,6 @@ safe::
 	__this_cpu_add_return(pcp, val)
 	__this_cpu_xchg(pcp, nval)
 	__this_cpu_cmpxchg(pcp, oval, nval)
-	__this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 	__this_cpu_sub(pcp, val)
 	__this_cpu_inc(pcp)
 	__this_cpu_dec(pcp)
arch/arm64/include/asm/atomic_ll_sc.h +0 -33
index b100a49..89d2ba2 100644
@@ -294,39 +294,6 @@ __CMPXCHG_CASE( ,  ,  mb_, 64, dmb ish,  , l, "memory", L)
 
 #undef __CMPXCHG_CASE
 
-#define __CMPXCHG_DBL(name, mb, rel, cl)				\
-static __always_inline long						\
-__ll_sc__cmpxchg_double##name(unsigned long old1,			\
-				      unsigned long old2,		\
-				      unsigned long new1,		\
-				      unsigned long new2,		\
-				      volatile void *ptr)		\
-{									\
-	unsigned long tmp, ret;						\
-									\
-	asm volatile("// __cmpxchg_double" #name "\n"			\
-	"	prfm	pstl1strm, %2\n"				\
-	"1:	ldxp	%0, %1, %2\n"					\
-	"	eor	%0, %0, %3\n"					\
-	"	eor	%1, %1, %4\n"					\
-	"	orr	%1, %0, %1\n"					\
-	"	cbnz	%1, 2f\n"					\
-	"	st" #rel "xp	%w0, %5, %6, %2\n"			\
-	"	cbnz	%w0, 1b\n"					\
-	"	" #mb "\n"						\
-	"2:"								\
-	: "=&r" (tmp), "=&r" (ret), "+Q" (*(__uint128_t *)ptr)		\
-	: "r" (old1), "r" (old2), "r" (new1), "r" (new2)		\
-	: cl);								\
-									\
-	return ret;							\
-}
-
-__CMPXCHG_DBL(   ,        ,  ,         )
-__CMPXCHG_DBL(_mb, dmb ish, l, "memory")
-
-#undef __CMPXCHG_DBL
-
 union __u128_halves {
 	u128 full;
 	struct {
arch/arm64/include/asm/atomic_lse.h +0 -36
index c781281..87f568a 100644
@@ -281,42 +281,6 @@ __CMPXCHG_CASE(x,  ,  mb_, 64, al, "memory")
 
 #undef __CMPXCHG_CASE
 
-#define __CMPXCHG_DBL(name, mb, cl...)					\
-static __always_inline long						\
-__lse__cmpxchg_double##name(unsigned long old1,				\
-					 unsigned long old2,		\
-					 unsigned long new1,		\
-					 unsigned long new2,		\
-					 volatile void *ptr)		\
-{									\
-	unsigned long oldval1 = old1;					\
-	unsigned long oldval2 = old2;					\
-	register unsigned long x0 asm ("x0") = old1;			\
-	register unsigned long x1 asm ("x1") = old2;			\
-	register unsigned long x2 asm ("x2") = new1;			\
-	register unsigned long x3 asm ("x3") = new2;			\
-	register unsigned long x4 asm ("x4") = (unsigned long)ptr;	\
-									\
-	asm volatile(							\
-	__LSE_PREAMBLE							\
-	"	casp" #mb "\t%[old1], %[old2], %[new1], %[new2], %[v]\n"\
-	"	eor	%[old1], %[old1], %[oldval1]\n"			\
-	"	eor	%[old2], %[old2], %[oldval2]\n"			\
-	"	orr	%[old1], %[old1], %[old2]"			\
-	: [old1] "+&r" (x0), [old2] "+&r" (x1),				\
-	  [v] "+Q" (*(__uint128_t *)ptr)				\
-	: [new1] "r" (x2), [new2] "r" (x3), [ptr] "r" (x4),		\
-	  [oldval1] "r" (oldval1), [oldval2] "r" (oldval2)		\
-	: cl);								\
-									\
-	return x0;							\
-}
-
-__CMPXCHG_DBL(   ,   )
-__CMPXCHG_DBL(_mb, al, "memory")
-
-#undef __CMPXCHG_DBL
-
 #define __CMPXCHG128(name, mb, cl...)					\
 static __always_inline u128						\
 __lse__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)		\
arch/arm64/include/asm/cmpxchg.h +0 -46

58 lines not on this page · raw message

arch/arm64/include/asm/percpu.h +0 -10
index 2549829..9abcc8e 100644
@@ -145,16 +145,6 @@ PERCPU_RET_OP(add, add, ldadd)
  * preemption point when TIF_NEED_RESCHED gets set while preemption is
  * disabled.
  */
-#define this_cpu_cmpxchg_double_8(ptr1, ptr2, o1, o2, n1, n2)		\
-({									\
-	int __ret;							\
-	preempt_disable_notrace();					\
-	__ret = cmpxchg_double_local(	raw_cpu_ptr(&(ptr1)),		\
-					raw_cpu_ptr(&(ptr2)),		\
-					o1, o2, n1, n2);		\
-	preempt_enable_notrace();					\
-	__ret;								\
-})
 
 #define _pcp_protect(op, pcp, ...)					\
 ({									\
arch/s390/include/asm/cmpxchg.h +0 -34
index 7517376..aae0315 100644
@@ -190,40 +190,6 @@ static __always_inline unsigned long __cmpxchg(unsigned long address,
 #define arch_cmpxchg_local	arch_cmpxchg
 #define arch_cmpxchg64_local	arch_cmpxchg
 
-#define system_has_cmpxchg_double()	1
-
-static __always_inline int __cmpxchg_double(unsigned long p1, unsigned long p2,
-					    unsigned long o1, unsigned long o2,
-					    unsigned long n1, unsigned long n2)
-{
-	union register_pair old = { .even = o1, .odd = o2, };
-	union register_pair new = { .even = n1, .odd = n2, };
-	int cc;
-
-	asm volatile(
-		"	cdsg	%[old],%[new],%[ptr]\n"
-		"	ipm	%[cc]\n"
-		"	srl	%[cc],28\n"
-		: [cc] "=&d" (cc), [old] "+&d" (old.pair)
-		: [new] "d" (new.pair),
-		  [ptr] "QS" (*(unsigned long *)p1), "Q" (*(unsigned long *)p2)
-		: "memory", "cc");
-	return !cc;
-}
-
-#define arch_cmpxchg_double(p1, p2, o1, o2, n1, n2)			\
-({									\
-	typeof(p1) __p1 = (p1);						\
-	typeof(p2) __p2 = (p2);						\
-									\
-	BUILD_BUG_ON(sizeof(*(p1)) != sizeof(long));			\
-	BUILD_BUG_ON(sizeof(*(p2)) != sizeof(long));			\
-	VM_BUG_ON((unsigned long)((__p1) + 1) != (unsigned long)(__p2));\
-	__cmpxchg_double((unsigned long)__p1, (unsigned long)__p2,	\
-			 (unsigned long)(o1), (unsigned long)(o2),	\
-			 (unsigned long)(n1), (unsigned long)(n2));	\
-})
-
 #define system_has_cmpxchg128()		1
 
 static __always_inline u128 arch_cmpxchg128(volatile u128 *ptr, u128 old, u128 new)
arch/s390/include/asm/percpu.h +0 -18
index 5603402..264095d 100644
@@ -180,24 +180,6 @@
 #define this_cpu_xchg_4(pcp, nval) arch_this_cpu_xchg(pcp, nval)
 #define this_cpu_xchg_8(pcp, nval) arch_this_cpu_xchg(pcp, nval)
 
-#define arch_this_cpu_cmpxchg_double(pcp1, pcp2, o1, o2, n1, n2)	    \
-({									    \
-	typeof(pcp1) *p1__;						    \
-	typeof(pcp2) *p2__;						    \
-	int ret__;							    \
-									    \
-	preempt_disable_notrace();					    \
-	p1__ = raw_cpu_ptr(&(pcp1));					    \
-	p2__ = raw_cpu_ptr(&(pcp2));					    \
-	ret__ = __cmpxchg_double((unsigned long)p1__, (unsigned long)p2__,  \
-				 (unsigned long)(o1), (unsigned long)(o2),  \
-				 (unsigned long)(n1), (unsigned long)(n2)); \
-	preempt_enable_notrace();					    \
-	ret__;								    \
-})
-
-#define this_cpu_cmpxchg_double_8 arch_this_cpu_cmpxchg_double
-
 #include <asm-generic/percpu.h>
 
 #endif /* __ARCH_S390_PERCPU__ */
arch/x86/include/asm/cmpxchg.h +0 -25
index 540573f..d536365 100644
@@ -239,29 +239,4 @@ extern void __add_wrong_size(void)
 #define __xadd(ptr, inc, lock)	__xchg_op((ptr), (inc), xadd, lock)
 #define xadd(ptr, inc)		__xadd((ptr), (inc), LOCK_PREFIX)
 
-#define __cmpxchg_double(pfx, p1, p2, o1, o2, n1, n2)			\
-({									\
-	bool __ret;							\
-	__typeof__(*(p1)) __old1 = (o1), __new1 = (n1);			\
-	__typeof__(*(p2)) __old2 = (o2), __new2 = (n2);			\
-	BUILD_BUG_ON(sizeof(*(p1)) != sizeof(long));			\
-	BUILD_BUG_ON(sizeof(*(p2)) != sizeof(long));			\
-	VM_BUG_ON((unsigned long)(p1) % (2 * sizeof(long)));		\
-	VM_BUG_ON((unsigned long)((p1) + 1) != (unsigned long)(p2));	\
-	asm volatile(pfx "cmpxchg%c5b %1"				\
-		     CC_SET(e)						\
-		     : CC_OUT(e) (__ret),				\
-		       "+m" (*(p1)), "+m" (*(p2)),			\
-		       "+a" (__old1), "+d" (__old2)			\
-		     : "i" (2 * sizeof(long)),				\
-		       "b" (__new1), "c" (__new2));			\
-	__ret;								\
-})
-
-#define arch_cmpxchg_double(p1, p2, o1, o2, n1, n2) \
-	__cmpxchg_double(LOCK_PREFIX, p1, p2, o1, o2, n1, n2)
-
-#define arch_cmpxchg_double_local(p1, p2, o1, o2, n1, n2) \
-	__cmpxchg_double(, p1, p2, o1, o2, n1, n2)
-
 #endif	/* ASM_X86_CMPXCHG_H */
arch/x86/include/asm/cmpxchg_32.h +0 -1
index d12f9ed..b5731c5 100644
@@ -103,7 +103,6 @@ static inline bool __try_cmpxchg64(volatile u64 *ptr, u64 *pold, u64 new)
 
 #endif
 
-#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX8)
 #define system_has_cmpxchg64()		boot_cpu_has(X86_FEATURE_CX8)
 
 #endif /* _ASM_X86_CMPXCHG_32_H */
arch/x86/include/asm/cmpxchg_64.h +0 -1
index fe29107..3e6e3ee 100644
@@ -81,7 +81,6 @@ static __always_inline bool arch_try_cmpxchg128_local(volatile u128 *ptr, u128 *
 	return __arch_try_cmpxchg128(ptr, oldp, new,);
 }
 
-#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX16)
 #define system_has_cmpxchg128()		boot_cpu_has(X86_FEATURE_CX16)
 
 #endif /* _ASM_X86_CMPXCHG_64_H */
arch/x86/include/asm/percpu.h +0 -42

54 lines not on this page · raw message

include/asm-generic/percpu.h +0 -58

80 lines not on this page · raw message

include/linux/atomic/atomic-instrumented.h +1 -16
index 8583720..a55b5b7 100644
@@ -2234,21 +2234,6 @@ atomic_long_dec_if_positive(atomic_long_t *v)
 	arch_try_cmpxchg128_local(__ai_ptr, __ai_oldp, __VA_ARGS__); \
 })
 
-#define cmpxchg_double(ptr, ...) \
-({ \
-	typeof(ptr) __ai_ptr = (ptr); \
-	kcsan_mb(); \
-	instrument_atomic_read_write(__ai_ptr, 2 * sizeof(*__ai_ptr)); \
-	arch_cmpxchg_double(__ai_ptr, __VA_ARGS__); \
-})
-
-
-#define cmpxchg_double_local(ptr, ...) \
-({ \
-	typeof(ptr) __ai_ptr = (ptr); \
-	instrument_atomic_read_write(__ai_ptr, 2 * sizeof(*__ai_ptr)); \
-	arch_cmpxchg_double_local(__ai_ptr, __VA_ARGS__); \
-})
 
 #endif /* _LINUX_ATOMIC_INSTRUMENTED_H */
-// 82d1be694fab30414527d0877c29fa75ed5a0b74
+// 3611991b015450e119bcd7417a9431af7f3ba13c
include/linux/percpu-defs.h +0 -38
index cbbf6d1..ec35731 100644
@@ -358,33 +358,6 @@ static __always_inline void __this_cpu_preempt_check(const char *op) { }
 	pscr2_ret__;							\
 })
 
-/*
- * Special handling for cmpxchg_double.  cmpxchg_double is passed two
- * percpu variables.  The first has to be aligned to a double word
- * boundary and the second has to follow directly thereafter.
- * We enforce this on all architectures even if they don't support
- * a double cmpxchg instruction, since it's a cheap requirement, and it
- * avoids breaking the requirement for architectures with the instruction.
- */
-#define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...)		\
-({									\
-	bool pdcrb_ret__;						\
-	__verify_pcpu_ptr(&(pcp1));					\
-	BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2));			\
-	VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1)));	\
-	VM_BUG_ON((unsigned long)(&(pcp2)) !=				\
-		  (unsigned long)(&(pcp1)) + sizeof(pcp1));		\
-	switch(sizeof(pcp1)) {						\
-	case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break;	\
-	case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break;	\
-	default:							\
-		__bad_size_call_parameter(); break;			\
-	}								\
-	pdcrb_ret__;							\
-})
-
 #define __pcpu_size_call(stem, variable, ...)				\
 do {									\
 	__verify_pcpu_ptr(&(variable));					\
@@ -443,9 +416,6 @@ do {									\
 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
-#define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-	__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
-
 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
@@ -505,11 +475,6 @@ do {									\
 	raw_cpu_cmpxchg(pcp, oval, nval);				\
 })
 
-#define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-({	__this_cpu_preempt_check("cmpxchg_double");			\
-	raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2);	\
-})
-
 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
@@ -532,9 +497,6 @@ do {									\
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
-#define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
-	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
-
 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
scripts/atomic/gen-atomic-instrumented.sh +4 -11
index 4f80347..a2ef735 100755
@@ -84,7 +84,6 @@ gen_xchg()
 {
 	local xchg="$1"; shift
 	local order="$1"; shift
-	local mult="$1"; shift
 
 	kcsan_barrier=""
 	if [ "${xchg%_local}" = "${xchg}" ]; then
@@ -104,8 +103,8 @@ cat <<EOF
 EOF
 [ -n "$kcsan_barrier" ] && printf "\t${kcsan_barrier}; \\\\\n"
 cat <<EOF
-	instrument_atomic_read_write(__ai_ptr, ${mult}sizeof(*__ai_ptr)); \\
-	instrument_read_write(__ai_oldp, ${mult}sizeof(*__ai_oldp)); \\
+	instrument_atomic_read_write(__ai_ptr, sizeof(*__ai_ptr)); \\
+	instrument_read_write(__ai_oldp, sizeof(*__ai_oldp)); \\
 	arch_${xchg}${order}(__ai_ptr, __ai_oldp, __VA_ARGS__); \\
 })
 EOF
@@ -168,22 +167,16 @@ done
 
 for xchg in "xchg" "cmpxchg" "cmpxchg64" "cmpxchg128" "try_cmpxchg" "try_cmpxchg64" "try_cmpxchg128"; do
 	for order in "" "_acquire" "_release" "_relaxed"; do
-		gen_xchg "${xchg}" "${order}" ""
+		gen_xchg "${xchg}" "${order}"
 		printf "\n"
 	done
 done
 
 for xchg in "cmpxchg_local" "cmpxchg64_local" "cmpxchg128_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" "try_cmpxchg128_local"; do
-	gen_xchg "${xchg}" "" ""
+	gen_xchg "${xchg}" ""
 	printf "\n"
 done
 
-gen_xchg "cmpxchg_double" "" "2 * "
-
-printf "\n\n"
-
-gen_xchg "cmpxchg_double_local" "" "2 * "
-
 cat <<EOF
 
 #endif /* _LINUX_ATOMIC_INSTRUMENTED_H */
merge bot Applied
#4 lore raw

[tip: locking/core] types: Introduce [us]128

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     224d80c584d3016cb8d83d1c33914fdd3508aa8c
Gitweb:        https://git.kernel.org/tip/224d80c584d3016cb8d83d1c33914fdd3508aa8c
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:35 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:35 +02:00

types: Introduce [us]128

Introduce [us]128 (when available). Unlike [us]64, ensure they are always naturally aligned.

This also enables 128bit wide atomics (which require natural alignment) such as cmpxchg128().

---
 include/linux/types.h          | 5 +++++
 include/uapi/linux/types.h     | 4 ++++
 lib/crypto/curve25519-hacl64.c | 2 --
 lib/crypto/poly1305-donna64.c  | 2 --
 4 files changed, 9 insertions(+), 4 deletions(-)
include/linux/types.h +5 -0
index 688fb94..becb8cd 100644
@@ -10,6 +10,11 @@
 #define DECLARE_BITMAP(name,bits) \
 	unsigned long name[BITS_TO_LONGS(bits)]
 
+#ifdef __SIZEOF_INT128__
+typedef __s128 s128;
+typedef __u128 u128;
+#endif
+
 typedef u32 __kernel_dev_t;
 
 typedef __kernel_fd_set		fd_set;
include/uapi/linux/types.h +4 -0
index 308433b..6375a06 100644
@@ -13,6 +13,10 @@
 
 #include <linux/posix_types.h>
 
+#ifdef __SIZEOF_INT128__
+typedef __signed__ __int128 __s128 __attribute__((aligned(16)));
+typedef unsigned __int128 __u128 __attribute__((aligned(16)));
+#endif
 
 /*
  * Below are truly Linux-specific types that should never collide with
lib/crypto/curve25519-hacl64.c +0 -2
index 771d82d..c40e5d9 100644
@@ -14,8 +14,6 @@
 #include <crypto/curve25519.h>
 #include <linux/string.h>
 
-typedef __uint128_t u128;
-
 static __always_inline u64 u64_eq_mask(u64 a, u64 b)
 {
 	u64 x = a ^ b;
lib/crypto/poly1305-donna64.c +0 -2
index d34cf40..988702c 100644
@@ -10,8 +10,6 @@
 #include <asm/unaligned.h>
 #include <crypto/internal/poly1305.h>
 
-typedef __uint128_t u128;
-
 void poly1305_core_setkey(struct poly1305_core_key *key,
 			  const u8 raw_key[POLY1305_BLOCK_SIZE])
 {
merge bot Applied
#8 lore raw

[tip: locking/core] instrumentation: Wire up cmpxchg128()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     8c8b096a23d12fedf3c0f50524f30113ef97aa8c
Gitweb:        https://git.kernel.org/tip/8c8b096a23d12fedf3c0f50524f30113ef97aa8c
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:37 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:36 +02:00

instrumentation: Wire up cmpxchg128()

Wire up the cmpxchg128 family in the atomic wrapper scripts.

These provide the generic cmpxchg128 family of functions from the arch_ prefixed version, adding explicit instrumentation where needed.

---
 include/linux/atomic/atomic-arch-fallback.h | 95 +++++++++++++++++++-
 include/linux/atomic/atomic-instrumented.h  | 86 +++++++++++++++++-
 scripts/atomic/gen-atomic-fallback.sh       |  4 +-
 scripts/atomic/gen-atomic-instrumented.sh   |  4 +-
 4 files changed, 183 insertions(+), 6 deletions(-)
include/linux/atomic/atomic-arch-fallback.h +94 -1
index a6e4437..1722ddb 100644
@@ -77,6 +77,29 @@
 
 #endif /* arch_cmpxchg64_relaxed */
 
+#ifndef arch_cmpxchg128_relaxed
+#define arch_cmpxchg128_acquire arch_cmpxchg128
+#define arch_cmpxchg128_release arch_cmpxchg128
+#define arch_cmpxchg128_relaxed arch_cmpxchg128
+#else /* arch_cmpxchg128_relaxed */
+
+#ifndef arch_cmpxchg128_acquire
+#define arch_cmpxchg128_acquire(...) \
+	__atomic_op_acquire(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_cmpxchg128_release
+#define arch_cmpxchg128_release(...) \
+	__atomic_op_release(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_cmpxchg128
+#define arch_cmpxchg128(...) \
+	__atomic_op_fence(arch_cmpxchg128, __VA_ARGS__)
+#endif
+
+#endif /* arch_cmpxchg128_relaxed */
+
 #ifndef arch_try_cmpxchg_relaxed
 #ifdef arch_try_cmpxchg
 #define arch_try_cmpxchg_acquire arch_try_cmpxchg
@@ -217,6 +240,76 @@
 
 #endif /* arch_try_cmpxchg64_relaxed */
 
+#ifndef arch_try_cmpxchg128_relaxed
+#ifdef arch_try_cmpxchg128
+#define arch_try_cmpxchg128_acquire arch_try_cmpxchg128
+#define arch_try_cmpxchg128_release arch_try_cmpxchg128
+#define arch_try_cmpxchg128_relaxed arch_try_cmpxchg128
+#endif /* arch_try_cmpxchg128 */
+
+#ifndef arch_try_cmpxchg128
+#define arch_try_cmpxchg128(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128 */
+
+#ifndef arch_try_cmpxchg128_acquire
+#define arch_try_cmpxchg128_acquire(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_acquire((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_acquire */
+
+#ifndef arch_try_cmpxchg128_release
+#define arch_try_cmpxchg128_release(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_release((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_release */
+
+#ifndef arch_try_cmpxchg128_relaxed
+#define arch_try_cmpxchg128_relaxed(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg128_relaxed((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg128_relaxed */
+
+#else /* arch_try_cmpxchg128_relaxed */
+
+#ifndef arch_try_cmpxchg128_acquire
+#define arch_try_cmpxchg128_acquire(...) \
+	__atomic_op_acquire(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg128_release
+#define arch_try_cmpxchg128_release(...) \
+	__atomic_op_release(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg128
+#define arch_try_cmpxchg128(...) \
+	__atomic_op_fence(arch_try_cmpxchg128, __VA_ARGS__)
+#endif
+
+#endif /* arch_try_cmpxchg128_relaxed */
+
 #ifndef arch_try_cmpxchg_local
 #define arch_try_cmpxchg_local(_ptr, _oldp, _new) \
 ({ \
@@ -2668,4 +2761,4 @@ arch_atomic64_dec_if_positive(atomic64_t *v)
 #endif
 
 #endif /* _LINUX_ATOMIC_FALLBACK_H */
-// ad2e2b4d168dbc60a73922616047a9bfa446af36
+// 52dfc6fe4a2e7234bbd2aa3e16a377c1db793a53
include/linux/atomic/atomic-instrumented.h +85 -1

113 lines not on this page · raw message

scripts/atomic/gen-atomic-fallback.sh +2 -2
index 6e853f0..a70acd5 100755
@@ -217,11 +217,11 @@ cat << EOF
 
 EOF
 
-for xchg in "arch_xchg" "arch_cmpxchg" "arch_cmpxchg64"; do
+for xchg in "arch_xchg" "arch_cmpxchg" "arch_cmpxchg64" "arch_cmpxchg128"; do
 	gen_xchg_fallbacks "${xchg}"
 done
 
-for cmpxchg in "cmpxchg" "cmpxchg64"; do
+for cmpxchg in "cmpxchg" "cmpxchg64" "cmpxchg128"; do
 	gen_try_cmpxchg_fallbacks "${cmpxchg}"
 done
 
scripts/atomic/gen-atomic-instrumented.sh +2 -2
index d9ffd74..4f80347 100755
@@ -166,14 +166,14 @@ grep '^[a-z]' "$1" | while read name meta args; do
 done
 
 
-for xchg in "xchg" "cmpxchg" "cmpxchg64" "try_cmpxchg" "try_cmpxchg64"; do
+for xchg in "xchg" "cmpxchg" "cmpxchg64" "cmpxchg128" "try_cmpxchg" "try_cmpxchg64" "try_cmpxchg128"; do
 	for order in "" "_acquire" "_release" "_relaxed"; do
 		gen_xchg "${xchg}" "${order}" ""
 		printf "\n"
 	done
 done
 
-for xchg in "cmpxchg_local" "cmpxchg64_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" ; do
+for xchg in "cmpxchg_local" "cmpxchg64_local" "cmpxchg128_local" "sync_cmpxchg" "try_cmpxchg_local" "try_cmpxchg64_local" "try_cmpxchg128_local"; do
 	gen_xchg "${xchg}" "" ""
 	printf "\n"
 done
merge bot Applied
#10 lore raw

[tip: locking/core] percpu: Add {raw,this}_cpu_try_cmpxchg()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     c5c0ba953b8c969c5d51bf1c57f239866a97c47c
Gitweb:        https://git.kernel.org/tip/c5c0ba953b8c969c5d51bf1c57f239866a97c47c
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:38 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:36 +02:00

percpu: Add {raw,this}_cpu_try_cmpxchg()

Add the try_cmpxchg() form to the per-cpu ops.

---
 include/asm-generic/percpu.h | 113 ++++++++++++++++++++++++++++++++--
 include/linux/percpu-defs.h  |  19 ++++++-
 2 files changed, 128 insertions(+), 4 deletions(-)
include/asm-generic/percpu.h +109 -4
index 6432a7f..96af32c 100644
@@ -89,16 +89,37 @@ do {									\
 	__ret;								\
 })
 
-#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+#define __cpu_fallback_try_cmpxchg(pcp, ovalp, nval, _cmpxchg)		\
+({									\
+	typeof(pcp) __val, __old = *(ovalp);				\
+	__val = _cmpxchg(pcp, __old, nval);				\
+	if (__val != __old)						\
+		*(ovalp) = __val;					\
+	__val == __old;							\
+})
+
+#define raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
 ({									\
 	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
-	typeof(pcp) __ret;						\
-	__ret = *__p;							\
-	if (__ret == (oval))						\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (__val == __old) {						\
 		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
 	__ret;								\
 })
 
+#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg(pcp, &__old, nval);			\
+	__old;								\
+})
+
 #define raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 ({									\
 	typeof(pcp1) *__p1 = raw_cpu_ptr(&(pcp1));			\
@@ -170,6 +191,16 @@ do {									\
 	__ret;								\
 })
 
+#define this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval);		\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
 #define this_cpu_generic_cmpxchg(pcp, oval, nval)			\
 ({									\
 	typeof(pcp) __ret;						\
@@ -282,6 +313,43 @@ do {									\
 #define raw_cpu_xchg_8(pcp, nval)	raw_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef raw_cpu_try_cmpxchg_1
+#ifdef raw_cpu_cmpxchg_1
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_1)
+#else
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_2
+#ifdef raw_cpu_cmpxchg_2
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_2)
+#else
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_4
+#ifdef raw_cpu_cmpxchg_4
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_4)
+#else
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_8
+#ifdef raw_cpu_cmpxchg_8
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_8)
+#else
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef raw_cpu_cmpxchg_1
 #define raw_cpu_cmpxchg_1(pcp, oval, nval) \
 	raw_cpu_generic_cmpxchg(pcp, oval, nval)
@@ -407,6 +475,43 @@ do {									\
 #define this_cpu_xchg_8(pcp, nval)	this_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef this_cpu_try_cmpxchg_1
+#ifdef this_cpu_cmpxchg_1
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_1)
+#else
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_2
+#ifdef this_cpu_cmpxchg_2
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_2)
+#else
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_4
+#ifdef this_cpu_cmpxchg_4
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_4)
+#else
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_8
+#ifdef this_cpu_cmpxchg_8
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_8)
+#else
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef this_cpu_cmpxchg_1
 #define this_cpu_cmpxchg_1(pcp, oval, nval) \
 	this_cpu_generic_cmpxchg(pcp, oval, nval)
include/linux/percpu-defs.h +19 -0
index e60727b..cbbf6d1 100644
@@ -343,6 +343,21 @@ static __always_inline void __this_cpu_preempt_check(const char *op) { }
 	pscr2_ret__;							\
 })
 
+#define __pcpu_size_call_return2bool(stem, variable, ...)		\
+({									\
+	bool pscr2_ret__;						\
+	__verify_pcpu_ptr(&(variable));					\
+	switch(sizeof(variable)) {					\
+	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
+	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
+	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
+	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
+	default:							\
+		__bad_size_call_parameter(); break;			\
+	}								\
+	pscr2_ret__;							\
+})
+
 /*
  * Special handling for cmpxchg_double.  cmpxchg_double is passed two
  * percpu variables.  The first has to be aligned to a double word
@@ -426,6 +441,8 @@ do {									\
 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
 #define raw_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
+#define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
@@ -513,6 +530,8 @@ do {									\
 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
 #define this_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
+#define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
merge bot Applied
#15 lore raw

[tip: locking/core] percpu: Wire up cmpxchg128

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     6d12c8d308e68b9b0fa98ca2df4f83db4b4c965d
Gitweb:        https://git.kernel.org/tip/6d12c8d308e68b9b0fa98ca2df4f83db4b4c965d
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:39 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:37 +02:00

percpu: Wire up cmpxchg128

In order to replace cmpxchg_double() with the newly minted cmpxchg128() family of functions, wire it up in this_cpu_cmpxchg().

---
 arch/arm64/include/asm/percpu.h | 20 +++++++++-
 arch/s390/include/asm/percpu.h  | 16 +++++++-
 arch/x86/include/asm/percpu.h   | 74 +++++++++++++++++++++++++++++---
 arch/x86/lib/Makefile           |  3 +-
 arch/x86/lib/cmpxchg16b_emu.S   | 43 +++++++++++--------
 arch/x86/lib/cmpxchg8b_emu.S    | 67 ++++++++++++++++++++++-------
 include/asm-generic/percpu.h    | 56 ++++++++++++++++++++++++-
 7 files changed, 240 insertions(+), 39 deletions(-)
arch/arm64/include/asm/percpu.h +20 -0
index b9ba19d..2549829 100644
@@ -140,6 +140,10 @@ PERCPU_RET_OP(add, add, ldadd)
  * re-enabling preemption for preemptible kernels, but doing that in a way
  * which builds inside a module would mean messing directly with the preempt
  * count. If you do this, peterz and tglx will hunt you down.
+ *
+ * Not to mention it'll break the actual preemption model for missing a
+ * preemption point when TIF_NEED_RESCHED gets set while preemption is
+ * disabled.
  */
 #define this_cpu_cmpxchg_double_8(ptr1, ptr2, o1, o2, n1, n2)		\
 ({									\
@@ -240,6 +244,22 @@ PERCPU_RET_OP(add, add, ldadd)
 #define this_cpu_cmpxchg_8(pcp, o, n)	\
 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
 
+#define this_cpu_cmpxchg64(pcp, o, n)	this_cpu_cmpxchg_8(pcp, o, n)
+
+#define this_cpu_cmpxchg128(pcp, o, n)					\
+({									\
+	typedef typeof(pcp) pcp_op_T__;					\
+	u128 old__, new__, ret__;					\
+	pcp_op_T__ *ptr__;						\
+	old__ = o;							\
+	new__ = n;							\
+	preempt_disable_notrace();					\
+	ptr__ = raw_cpu_ptr(&(pcp));					\
+	ret__ = cmpxchg128_local((void *)ptr__, old__, new__);		\
+	preempt_enable_notrace();					\
+	ret__;								\
+})
+
 #ifdef __KVM_NVHE_HYPERVISOR__
 extern unsigned long __hyp_per_cpu_offset(unsigned int cpu);
 #define __per_cpu_offset
arch/s390/include/asm/percpu.h +16 -0
index 081837b..5603402 100644
@@ -148,6 +148,22 @@
 #define this_cpu_cmpxchg_4(pcp, oval, nval) arch_this_cpu_cmpxchg(pcp, oval, nval)
 #define this_cpu_cmpxchg_8(pcp, oval, nval) arch_this_cpu_cmpxchg(pcp, oval, nval)
 
+#define this_cpu_cmpxchg64(pcp, o, n)	this_cpu_cmpxchg_8(pcp, o, n)
+
+#define this_cpu_cmpxchg128(pcp, oval, nval)				\
+({									\
+	typedef typeof(pcp) pcp_op_T__;					\
+	u128 old__, new__, ret__;					\
+	pcp_op_T__ *ptr__;						\
+	old__ = oval;							\
+	new__ = nval;							\
+	preempt_disable_notrace();					\
+	ptr__ = raw_cpu_ptr(&(pcp));					\
+	ret__ = cmpxchg128((void *)ptr__, old__, new__);		\
+	preempt_enable_notrace();					\
+	ret__;								\
+})
+
 #define arch_this_cpu_xchg(pcp, nval)					\
 ({									\
 	typeof(pcp) *ptr__;						\
arch/x86/include/asm/percpu.h +68 -6

86 lines not on this page · raw message

arch/x86/lib/Makefile +2 -1
index 01932af..ea3a28e 100644
@@ -61,8 +61,9 @@ ifeq ($(CONFIG_X86_32),y)
         lib-y += strstr_32.o
         lib-y += string_32.o
         lib-y += memmove_32.o
+        lib-y += cmpxchg8b_emu.o
 ifneq ($(CONFIG_X86_CMPXCHG64),y)
-        lib-y += cmpxchg8b_emu.o atomic64_386_32.o
+        lib-y += atomic64_386_32.o
 endif
 else
         obj-y += iomap_copy_64.o
arch/x86/lib/cmpxchg16b_emu.S +25 -18

72 lines not on this page · raw message

arch/x86/lib/cmpxchg8b_emu.S +53 -14

95 lines not on this page · raw message

include/asm-generic/percpu.h +56 -0

80 lines not on this page · raw message

merge bot Applied
#2 lore raw

[tip: locking/core] cyrpto/b128ops: Remove struct u128

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     f413e724818c6482146218b3bcaf3d75b1317fc4
Gitweb:        https://git.kernel.org/tip/f413e724818c6482146218b3bcaf3d75b1317fc4
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:34 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:35 +02:00

cyrpto/b128ops: Remove struct u128

Per git-grep u128_xor() and its related struct u128 are unused except to implement {be,le}128_xor(). Remove them to free up the namespace.

---
 include/crypto/b128ops.h | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)
include/crypto/b128ops.h +3 -11
index 0b8e6bc..f3b37cb 100644
@@ -50,10 +50,6 @@
 #include <linux/types.h>
 
 typedef struct {
-	u64 a, b;
-} u128;
-
-typedef struct {
 	__be64 a, b;
 } be128;
 
@@ -61,20 +57,16 @@ typedef struct {
 	__le64 b, a;
 } le128;
 
-static inline void u128_xor(u128 *r, const u128 *p, const u128 *q)
+static inline void be128_xor(be128 *r, const be128 *p, const be128 *q)
 {
 	r->a = p->a ^ q->a;
 	r->b = p->b ^ q->b;
 }
 
-static inline void be128_xor(be128 *r, const be128 *p, const be128 *q)
-{
-	u128_xor((u128 *)r, (u128 *)p, (u128 *)q);
-}
-
 static inline void le128_xor(le128 *r, const le128 *p, const le128 *q)
 {
-	u128_xor((u128 *)r, (u128 *)p, (u128 *)q);
+	r->a = p->a ^ q->a;
+	r->b = p->b ^ q->b;
 }
 
 #endif /* _CRYPTO_B128OPS_H */
merge bot Applied
#6 lore raw

[tip: locking/core] arch: Introduce arch_{,try_}_cmpxchg128{,_local}()

In reply to Peter Zijlstra
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     b23e139d0b66c0216e7e9361a5021290395f504c
Gitweb:        https://git.kernel.org/tip/b23e139d0b66c0216e7e9361a5021290395f504c
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 31 May 2023 15:08:36 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Mon, 05 Jun 2023 09:36:35 +02:00

arch: Introduce arch_{,try_}_cmpxchg128{,_local}()

For all architectures that currently support cmpxchg_double() implement the cmpxchg128() family of functions that is basically the same but with a saner interface.

---
 arch/arm64/include/asm/atomic_ll_sc.h | 41 +++++++++++++++++-
 arch/arm64/include/asm/atomic_lse.h   | 31 +++++++++++++-
 arch/arm64/include/asm/cmpxchg.h      | 26 +++++++++++-
 arch/s390/include/asm/cmpxchg.h       | 14 ++++++-
 arch/x86/include/asm/cmpxchg_32.h     |  3 +-
 arch/x86/include/asm/cmpxchg_64.h     | 64 +++++++++++++++++++++++++-
 6 files changed, 177 insertions(+), 2 deletions(-)
arch/arm64/include/asm/atomic_ll_sc.h +41 -0
index cbb3d96..b100a49 100644
@@ -326,6 +326,47 @@ __CMPXCHG_DBL(   ,        ,  ,         )
 __CMPXCHG_DBL(_mb, dmb ish, l, "memory")
 
 #undef __CMPXCHG_DBL
+
+union __u128_halves {
+	u128 full;
+	struct {
+		u64 low, high;
+	};
+};
+
+#define __CMPXCHG128(name, mb, rel, cl...)                             \
+static __always_inline u128						\
+__ll_sc__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)	\
+{									\
+	union __u128_halves r, o = { .full = (old) },			\
+			       n = { .full = (new) };			\
+       unsigned int tmp;                                               \
+									\
+	asm volatile("// __cmpxchg128" #name "\n"			\
+       "       prfm    pstl1strm, %[v]\n"                              \
+       "1:     ldxp    %[rl], %[rh], %[v]\n"                           \
+       "       cmp     %[rl], %[ol]\n"                                 \
+       "       ccmp    %[rh], %[oh], 0, eq\n"                          \
+       "       b.ne    2f\n"                                           \
+       "       st" #rel "xp    %w[tmp], %[nl], %[nh], %[v]\n"          \
+       "       cbnz    %w[tmp], 1b\n"                                  \
+	"	" #mb "\n"						\
+	"2:"								\
+       : [v] "+Q" (*(u128 *)ptr),                                      \
+         [rl] "=&r" (r.low), [rh] "=&r" (r.high),                      \
+         [tmp] "=&r" (tmp)                                             \
+       : [ol] "r" (o.low), [oh] "r" (o.high),                          \
+         [nl] "r" (n.low), [nh] "r" (n.high)                           \
+       : "cc", ##cl);                                                  \
+									\
+	return r.full;							\
+}
+
+__CMPXCHG128(   ,        ,  )
+__CMPXCHG128(_mb, dmb ish, l, "memory")
+
+#undef __CMPXCHG128
+
 #undef K
 
 #endif	/* __ASM_ATOMIC_LL_SC_H */
arch/arm64/include/asm/atomic_lse.h +31 -0
index 319958b..c781281 100644
@@ -317,4 +317,35 @@ __CMPXCHG_DBL(_mb, al, "memory")
 
 #undef __CMPXCHG_DBL
 
+#define __CMPXCHG128(name, mb, cl...)					\
+static __always_inline u128						\
+__lse__cmpxchg128##name(volatile u128 *ptr, u128 old, u128 new)		\
+{									\
+	union __u128_halves r, o = { .full = (old) },			\
+			       n = { .full = (new) };			\
+	register unsigned long x0 asm ("x0") = o.low;			\
+	register unsigned long x1 asm ("x1") = o.high;			\
+	register unsigned long x2 asm ("x2") = n.low;			\
+	register unsigned long x3 asm ("x3") = n.high;			\
+	register unsigned long x4 asm ("x4") = (unsigned long)ptr;	\
+									\
+	asm volatile(							\
+	__LSE_PREAMBLE							\
+	"	casp" #mb "\t%[old1], %[old2], %[new1], %[new2], %[v]\n"\
+	: [old1] "+&r" (x0), [old2] "+&r" (x1),				\
+	  [v] "+Q" (*(u128 *)ptr)					\
+	: [new1] "r" (x2), [new2] "r" (x3), [ptr] "r" (x4),		\
+	  [oldval1] "r" (o.low), [oldval2] "r" (o.high)			\
+	: cl);								\
+									\
+	r.low = x0; r.high = x1;					\
+									\
+	return r.full;							\
+}
+
+__CMPXCHG128(   ,   )
+__CMPXCHG128(_mb, al, "memory")
+
+#undef __CMPXCHG128
+
 #endif	/* __ASM_ATOMIC_LSE_H */
arch/arm64/include/asm/cmpxchg.h +26 -0
index c6bc5d8..097b832 100644
@@ -146,6 +146,19 @@ __CMPXCHG_DBL(_mb)
 
 #undef __CMPXCHG_DBL
 
+#define __CMPXCHG128(name)						\
+static inline u128 __cmpxchg128##name(volatile u128 *ptr,		\
+				      u128 old, u128 new)		\
+{									\
+	return __lse_ll_sc_body(_cmpxchg128##name,			\
+				ptr, old, new);				\
+}
+
+__CMPXCHG128(   )
+__CMPXCHG128(_mb)
+
+#undef __CMPXCHG128
+
 #define __CMPXCHG_GEN(sfx)						\
 static __always_inline unsigned long __cmpxchg##sfx(volatile void *ptr,	\
 					   unsigned long old,		\
@@ -228,6 +241,19 @@ __CMPXCHG_GEN(_mb)
 	__ret;									\
 })
 
+/* cmpxchg128 */
+#define system_has_cmpxchg128()		1
+
+#define arch_cmpxchg128(ptr, o, n)						\
+({										\
+	__cmpxchg128_mb((ptr), (o), (n));					\
+})
+
+#define arch_cmpxchg128_local(ptr, o, n)					\
+({										\
+	__cmpxchg128((ptr), (o), (n));						\
+})
+
 #define __CMPWAIT_CASE(w, sfx, sz)					\
 static inline void __cmpwait_case_##sz(volatile void *ptr,		\
 				       unsigned long val)		\
arch/s390/include/asm/cmpxchg.h +14 -0
index 06e0e42..7517376 100644
@@ -224,4 +224,18 @@ static __always_inline int __cmpxchg_double(unsigned long p1, unsigned long p2,
 			 (unsigned long)(n1), (unsigned long)(n2));	\
 })
 
+#define system_has_cmpxchg128()		1
+
+static __always_inline u128 arch_cmpxchg128(volatile u128 *ptr, u128 old, u128 new)
+{
+	asm volatile(
+		"	cdsg	%[old],%[new],%[ptr]\n"
+		: [old] "+d" (old), [ptr] "+QS" (*ptr)
+		: [new] "d" (new)
+		: "memory", "cc");
+	return old;
+}
+
+#define arch_cmpxchg128		arch_cmpxchg128
+
 #endif /* __ASM_CMPXCHG_H */
arch/x86/include/asm/cmpxchg_32.h +2 -1
index 6ba80ce..d12f9ed 100644
@@ -103,6 +103,7 @@ static inline bool __try_cmpxchg64(volatile u64 *ptr, u64 *pold, u64 new)
 
 #endif
 
-#define system_has_cmpxchg_double() boot_cpu_has(X86_FEATURE_CX8)
+#define system_has_cmpxchg_double()	boot_cpu_has(X86_FEATURE_CX8)
+#define system_has_cmpxchg64()		boot_cpu_has(X86_FEATURE_CX8)
 
 #endif /* _ASM_X86_CMPXCHG_32_H */
arch/x86/include/asm/cmpxchg_64.h +63 -1

69 lines not on this page · raw message

#11 lore raw

Re: [PATCH v2 5/12] percpu: Add {raw,this}_cpu_try_cmpxchg()

In reply to Peter Zijlstra
Quoted 4 lines · Peter Zijlstra

Add the try_cmpxchg() form to the per-cpu ops.

---

+CC Nathan, llvm list

Hi all, this patch seems to break booting on Qualcomm ARM64 platforms when compiled with clang (GCC works fine) for some reason..:

next-20230605 - works
next-20230606 - doesn't

grev -m 1 dc4e51fd9846 on next-20230606 - works again b4 shazam <this_msgid> -P 1-4 - still works b4 shazam <this_msgid> -P 5 - breaks

Confirmed on at least Qualcomm QCM2290, SM8250.

Checking the serial console, it hits a BUG_ON:

[    0.000000] ------------[ cut here ]------------
[    0.000000] kernel BUG at mm/vmalloc.c:1638!
[    0.000000] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
[    0.000000] Modules linked in:
[    0.000000] CPU: 0 PID: 0 Comm: swapper/0 Not tainted [snip]
[    0.000000] Hardware name: Qualcomm Technologies, Inc. Robotics RB1 (DT)
[    0.000000] pstate: 000000c5 (nzcv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[    0.000000] pc : alloc_vmap_area+0xafc/0xb08
[    0.000000] lr : alloc_vmap_area+0x9e4/0xb08
[    0.000000] sp : ffffa50137f53c20
[    0.000000] x29: ffffa50137f53c60 x28: ffffa50137f30c18 x27: 0000000000000000
[    0.000000] x26: 0000000000007fff x25: ffff800080000000 x24: 000000000000cfff
[    0.000000] x23: ffffffffffff8000 x22: ffffa50137fef970 x21: fffffbfff0000000
[    0.000000] x20: ffff022982003208 x19: ffff0229820031f8 x18: ffffa50137f64f70
[    0.000000] x17: ffffa50137fef980 x16: ffffa501375e6d08 x15: 0000000000000001
[    0.000000] x14: ffffa5013831e1a0 x13: ffffa50137f30c18 x12: 0000000000402dc2
[    0.000000] x11: 0000000000000000 x10: ffff022982003018 x9 : ffffa5013831e188
[    0.000000] x8 : ffffcb55ff003228 x7 : 0000000000000000 x6 : 0000000000000048
[    0.000000] x5 : 0000000000000000 x4 : ffffa50137f53bd0 x3 : ffffa50136490000
[    0.000000] x2 : 0000000000000001 x1 : ffffa5013831e190 x0 : ffff022982003208
[    0.000000] Call trace:
[    0.000000]  alloc_vmap_area+0xafc/0xb08
[    0.000000]  __get_vm_area_node+0x108/0x1e8
[    0.000000]  __vmalloc_node_range+0x1fc/0x728
[    0.000000]  __vmalloc_node+0x5c/0x70
[    0.000000]  init_IRQ+0x90/0x11c
[    0.000000]  start_kernel+0x1ac/0x3bc
[    0.000000]  __primary_switched+0xc4/0xcc
[    0.000000] Code: f000e300 91062000 943bd9ba 17ffff8f (d4210000)
[    0.000000] ---[ end trace 0000000000000000 ]---
[    0.000000] Kernel panic - not syncing: Attempted to kill the idle task!

Compiled with clang 15.0.7 from Arch repos, with
make ARCH=arm64 LLVM=1

Konrad

Quoted 195 lines
 include/asm-generic/percpu.h |  113 +++++++++++++++++++++++++++++++++++++++++--
 include/linux/percpu-defs.h  |   19 +++++++
 2 files changed, 128 insertions(+), 4 deletions(-)
include/asm-generic/percpu.h +109 -4
@@ -89,16 +89,37 @@ do {									\
 	__ret;								\
 })
 
-#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+#define __cpu_fallback_try_cmpxchg(pcp, ovalp, nval, _cmpxchg)		\
+({									\
+	typeof(pcp) __val, __old = *(ovalp);				\
+ 	__val = _cmpxchg(pcp, __old, nval);				\
+	if (__val != __old)						\
+		*(ovalp) = __val;					\
+	__val == __old;							\
+})
+
+#define raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
 ({									\
 	typeof(pcp) *__p = raw_cpu_ptr(&(pcp));				\
-	typeof(pcp) __ret;						\
-	__ret = *__p;							\
-	if (__ret == (oval))						\
+	typeof(pcp) __val = *__p, __old = *(ovalp);			\
+	bool __ret;							\
+	if (__val == __old) {						\
 		*__p = nval;						\
+		__ret = true;						\
+	} else {							\
+		*(ovalp) = __val;					\
+		__ret = false;						\
+	}								\
 	__ret;								\
 })
 
+#define raw_cpu_generic_cmpxchg(pcp, oval, nval)			\
+({									\
+	typeof(pcp) __old = (oval);					\
+	raw_cpu_generic_try_cmpxchg(pcp, &__old, nval);			\
+	__old;								\
+})
+
 #define raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 ({									\
 	typeof(pcp1) *__p1 = raw_cpu_ptr(&(pcp1));			\
@@ -170,6 +191,16 @@ do {									\
 	__ret;								\
 })
 
+#define this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)			\
+({									\
+	bool __ret;							\
+	unsigned long __flags;						\
+	raw_local_irq_save(__flags);					\
+	__ret = raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval);		\
+	raw_local_irq_restore(__flags);					\
+	__ret;								\
+})
+
 #define this_cpu_generic_cmpxchg(pcp, oval, nval)			\
 ({									\
 	typeof(pcp) __ret;						\
@@ -282,6 +313,43 @@ do {									\
 #define raw_cpu_xchg_8(pcp, nval)	raw_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef raw_cpu_try_cmpxchg_1
+#ifdef raw_cpu_cmpxchg_1
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_1)
+#else
+#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_2
+#ifdef raw_cpu_cmpxchg_2
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_2)
+#else
+#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_4
+#ifdef raw_cpu_cmpxchg_4
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_4)
+#else
+#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef raw_cpu_try_cmpxchg_8
+#ifdef raw_cpu_cmpxchg_8
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, raw_cpu_cmpxchg_8)
+#else
+#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef raw_cpu_cmpxchg_1
 #define raw_cpu_cmpxchg_1(pcp, oval, nval) \
 	raw_cpu_generic_cmpxchg(pcp, oval, nval)
@@ -407,6 +475,43 @@ do {									\
 #define this_cpu_xchg_8(pcp, nval)	this_cpu_generic_xchg(pcp, nval)
 #endif
 
+#ifndef this_cpu_try_cmpxchg_1
+#ifdef this_cpu_cmpxchg_1
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_1)
+#else
+#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_2
+#ifdef this_cpu_cmpxchg_2
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_2)
+#else
+#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_4
+#ifdef this_cpu_cmpxchg_4
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_4)
+#else
+#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+#ifndef this_cpu_try_cmpxchg_8
+#ifdef this_cpu_cmpxchg_8
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	__cpu_fallback_try_cmpxchg(pcp, ovalp, nval, this_cpu_cmpxchg_8)
+#else
+#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \
+	this_cpu_generic_try_cmpxchg(pcp, ovalp, nval)
+#endif
+#endif
+
 #ifndef this_cpu_cmpxchg_1
 #define this_cpu_cmpxchg_1(pcp, oval, nval) \
 	this_cpu_generic_cmpxchg(pcp, oval, nval)
include/linux/percpu-defs.h +19 -0
@@ -343,6 +343,21 @@ static __always_inline void __this_cpu_p
 	pscr2_ret__;							\
 })
 
+#define __pcpu_size_call_return2bool(stem, variable, ...)		\
+({									\
+	bool pscr2_ret__;						\
+	__verify_pcpu_ptr(&(variable));					\
+	switch(sizeof(variable)) {					\
+	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
+	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
+	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
+	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
+	default:							\
+		__bad_size_call_parameter(); break;			\
+	}								\
+	pscr2_ret__;							\
+})
+
 /*
  * Special handling for cmpxchg_double.  cmpxchg_double is passed two
  * percpu variables.  The first has to be aligned to a double word
@@ -426,6 +441,8 @@ do {									\
 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
 #define raw_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
+#define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
@@ -513,6 +530,8 @@ do {									\
 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
 #define this_cpu_cmpxchg(pcp, oval, nval) \
 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
+#define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
+	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
 #define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
 	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
 
#12 lore raw

Re: [PATCH v2 5/12] percpu: Add {raw,this}_cpu_try_cmpxchg()

Hi Konrad,

Quoted 55 lines · Konrad Dybcio
Quoted 4 lines · Peter Zijlstra

Add the try_cmpxchg() form to the per-cpu ops.

---

+CC Nathan, llvm list

Hi all, this patch seems to break booting on Qualcomm ARM64 platforms when compiled with clang (GCC works fine) for some reason..:

next-20230605 - works
next-20230606 - doesn't

grev -m 1 dc4e51fd9846 on next-20230606 - works again b4 shazam <this_msgid> -P 1-4 - still works b4 shazam <this_msgid> -P 5 - breaks

Confirmed on at least Qualcomm QCM2290, SM8250.

Checking the serial console, it hits a BUG_ON:

[    0.000000] ------------[ cut here ]------------
[    0.000000] kernel BUG at mm/vmalloc.c:1638!
[    0.000000] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
[    0.000000] Modules linked in:
[    0.000000] CPU: 0 PID: 0 Comm: swapper/0 Not tainted [snip]
[    0.000000] Hardware name: Qualcomm Technologies, Inc. Robotics RB1 (DT)
[    0.000000] pstate: 000000c5 (nzcv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[    0.000000] pc : alloc_vmap_area+0xafc/0xb08
[    0.000000] lr : alloc_vmap_area+0x9e4/0xb08
[    0.000000] sp : ffffa50137f53c20
[    0.000000] x29: ffffa50137f53c60 x28: ffffa50137f30c18 x27: 0000000000000000
[    0.000000] x26: 0000000000007fff x25: ffff800080000000 x24: 000000000000cfff
[    0.000000] x23: ffffffffffff8000 x22: ffffa50137fef970 x21: fffffbfff0000000
[    0.000000] x20: ffff022982003208 x19: ffff0229820031f8 x18: ffffa50137f64f70
[    0.000000] x17: ffffa50137fef980 x16: ffffa501375e6d08 x15: 0000000000000001
[    0.000000] x14: ffffa5013831e1a0 x13: ffffa50137f30c18 x12: 0000000000402dc2
[    0.000000] x11: 0000000000000000 x10: ffff022982003018 x9 : ffffa5013831e188
[    0.000000] x8 : ffffcb55ff003228 x7 : 0000000000000000 x6 : 0000000000000048
[    0.000000] x5 : 0000000000000000 x4 : ffffa50137f53bd0 x3 : ffffa50136490000
[    0.000000] x2 : 0000000000000001 x1 : ffffa5013831e190 x0 : ffff022982003208
[    0.000000] Call trace:
[    0.000000]  alloc_vmap_area+0xafc/0xb08
[    0.000000]  __get_vm_area_node+0x108/0x1e8
[    0.000000]  __vmalloc_node_range+0x1fc/0x728
[    0.000000]  __vmalloc_node+0x5c/0x70
[    0.000000]  init_IRQ+0x90/0x11c
[    0.000000]  start_kernel+0x1ac/0x3bc
[    0.000000]  __primary_switched+0xc4/0xcc
[    0.000000] Code: f000e300 91062000 943bd9ba 17ffff8f (d4210000)
[    0.000000] ---[ end trace 0000000000000000 ]---
[    0.000000] Kernel panic - not syncing: Attempted to kill the idle task!
Quoted 2 lines

Compiled with clang 15.0.7 from Arch repos, with
make ARCH=arm64 LLVM=1

Thanks a lot for testing with LLVM, submitting this report, and doing a bisect. I sent a patch to fix this a couple of days ago and Peter pushed it to -tip today, so it should be in the next -next release:

https://git.kernel.org/tip/093d9b240a1fa261ff8aeb7c7cc484dedacfda53

Cheers,
Nathan

#13 lore raw

Re: [PATCH v2 5/12] percpu: Add {raw,this}_cpu_try_cmpxchg()

Quoted 58 lines · Nathan Chancellor

Hi Konrad,

Quoted 55 lines · Konrad Dybcio
Quoted 4 lines · Peter Zijlstra

Add the try_cmpxchg() form to the per-cpu ops.

---

+CC Nathan, llvm list

Hi all, this patch seems to break booting on Qualcomm ARM64 platforms when compiled with clang (GCC works fine) for some reason..:

next-20230605 - works
next-20230606 - doesn't

grev -m 1 dc4e51fd9846 on next-20230606 - works again b4 shazam <this_msgid> -P 1-4 - still works b4 shazam <this_msgid> -P 5 - breaks

Confirmed on at least Qualcomm QCM2290, SM8250.

Checking the serial console, it hits a BUG_ON:

[    0.000000] ------------[ cut here ]------------
[    0.000000] kernel BUG at mm/vmalloc.c:1638!
[    0.000000] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
[    0.000000] Modules linked in:
[    0.000000] CPU: 0 PID: 0 Comm: swapper/0 Not tainted [snip]
[    0.000000] Hardware name: Qualcomm Technologies, Inc. Robotics RB1 (DT)
[    0.000000] pstate: 000000c5 (nzcv daIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[    0.000000] pc : alloc_vmap_area+0xafc/0xb08
[    0.000000] lr : alloc_vmap_area+0x9e4/0xb08
[    0.000000] sp : ffffa50137f53c20
[    0.000000] x29: ffffa50137f53c60 x28: ffffa50137f30c18 x27: 0000000000000000
[    0.000000] x26: 0000000000007fff x25: ffff800080000000 x24: 000000000000cfff
[    0.000000] x23: ffffffffffff8000 x22: ffffa50137fef970 x21: fffffbfff0000000
[    0.000000] x20: ffff022982003208 x19: ffff0229820031f8 x18: ffffa50137f64f70
[    0.000000] x17: ffffa50137fef980 x16: ffffa501375e6d08 x15: 0000000000000001
[    0.000000] x14: ffffa5013831e1a0 x13: ffffa50137f30c18 x12: 0000000000402dc2
[    0.000000] x11: 0000000000000000 x10: ffff022982003018 x9 : ffffa5013831e188
[    0.000000] x8 : ffffcb55ff003228 x7 : 0000000000000000 x6 : 0000000000000048
[    0.000000] x5 : 0000000000000000 x4 : ffffa50137f53bd0 x3 : ffffa50136490000
[    0.000000] x2 : 0000000000000001 x1 : ffffa5013831e190 x0 : ffff022982003208
[    0.000000] Call trace:
[    0.000000]  alloc_vmap_area+0xafc/0xb08
[    0.000000]  __get_vm_area_node+0x108/0x1e8
[    0.000000]  __vmalloc_node_range+0x1fc/0x728
[    0.000000]  __vmalloc_node+0x5c/0x70
[    0.000000]  init_IRQ+0x90/0x11c
[    0.000000]  start_kernel+0x1ac/0x3bc
[    0.000000]  __primary_switched+0xc4/0xcc
[    0.000000] Code: f000e300 91062000 943bd9ba 17ffff8f (d4210000)
[    0.000000] ---[ end trace 0000000000000000 ]---
[    0.000000] Kernel panic - not syncing: Attempted to kill the idle task!
Quoted 5 lines
Quoted 2 lines

Compiled with clang 15.0.7 from Arch repos, with
make ARCH=arm64 LLVM=1

Thanks a lot for testing with LLVM, submitting this report, and doing a bisect.

No, thank *you* for making it even possible ;)

I sent a patch to fix this a couple of days ago and Peter pushed

Quoted 3 lines

it to -tip today, so it should be in the next -next release:

https://git.kernel.org/tip/093d9b240a1fa261ff8aeb7c7cc484dedacfda53

Amazing, I can boot the most recent next-20230609 with it again!

Konrad

Quoted 3 lines

Cheers,
Nathan