All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 00/10] x86: use gcc 6+ asm flag output feature
@ 2016-06-07 23:30 H. Peter Anvin
  2016-06-07 23:31 ` [PATCH 01/10] x86, bitops: remove use of "sbb" to return CF H. Peter Anvin
                   ` (11 more replies)
  0 siblings, 12 replies; 43+ messages in thread
From: H. Peter Anvin @ 2016-06-07 23:30 UTC (permalink / raw)
  To: Ingo Molnar, Thomas Gleixner, H. Peter Anvin,
	Linux Kernel Mailing List
  Cc: Peter Zijlstra, Andy Lutomirski, Borislav Petkov

From: "H. Peter Anvin" <hpa@zytor.com>

gcc 6+ has the ability to let flags (actually, conditions, which are
specific combinations of flags) to be used directly as asm() outputs.
The syntax for that is "=@cc<cc>" where <cc> is the same set of
letters that would be used in a j<cc> or set<cc> instruction
(e.g. "=@ccz" to test the ZF flag.)

This patchset by itself reduces the size of the x86-64 kernel by
0.12%, from a baseline of 4.7-rc2 built with gcc 6.1 (first line is
with the patchset, the second one is without):

     text       data       bss        dec       hex filename
     
 68245656   41004339  20533248  129783243   7bc55cb o.i386-allconfig/vmlinux
 68355716   41008499  20533248  129897463   7be13f7 o.i386-allconfig/vmlinux

127384005  129742359  38150144  295276508  11998fdc o.x86_64-allconfig/vmlinux
127538765  129742295  38150144  295431204  119bec24 o.x86_64-allconfig/vmlinux

^ permalink raw reply	[flat|nested] 43+ messages in thread
* [PATCH v2 10/10] x86, asm: Use CC_SET()/CC_OUT() and static_cpu_has() in archrandom.h
@ 2016-06-08 19:38 H. Peter Anvin
  2016-06-08 19:49 ` [tip:x86/asm] " tip-bot for H. Peter Anvin
  0 siblings, 1 reply; 43+ messages in thread
From: H. Peter Anvin @ 2016-06-08 19:38 UTC (permalink / raw)
  To: Ingo Molnar, Thomas Gleixner, H. Peter Anvin,
	Linux Kernel Mailing List
  Cc: Andy Lutomirski, Borislav Petkov, H. Peter Anvin

Use CC_SET()/CC_OUT() and static_cpu_has().  This produces code good
enough to eliminate ad hoc use of alternatives in <asm/archrandom.h>,
greatly simplifying the code.

While we are at it, make x86_init_rdrand() compile out completely if
we don't need it.

Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

v2: fix a conflict between <linux/random.h> and <asm/archrandom.h>
    discovered by Ingo Molnar.  There are a few places in x86-specific
    code where we need all of <arch/archrandom.h> even when
    CONFIG_ARCH_RANDOM is disabled, so <linux/random.h> does not
    suffice.
---
 arch/x86/include/asm/archrandom.h | 128 ++++++++++++++++++--------------------
 arch/x86/kernel/cpu/rdrand.c      |   4 +-
 2 files changed, 62 insertions(+), 70 deletions(-)

diff --git a/arch/x86/include/asm/archrandom.h b/arch/x86/include/asm/archrandom.h
index ab6f599..5b0579a 100644
--- a/arch/x86/include/asm/archrandom.h
+++ b/arch/x86/include/asm/archrandom.h
@@ -25,8 +25,6 @@
 
 #include <asm/processor.h>
 #include <asm/cpufeature.h>
-#include <asm/alternative.h>
-#include <asm/nops.h>
 
 #define RDRAND_RETRY_LOOPS	10
 
@@ -40,97 +38,91 @@
 # define RDSEED_LONG	RDSEED_INT
 #endif
 
-#ifdef CONFIG_ARCH_RANDOM
+/* Unconditional execution of RDRAND and RDSEED */
 
-/* Instead of arch_get_random_long() when alternatives haven't run. */
 static inline bool rdrand_long(unsigned long *v)
 {
-	int ok;
-	asm volatile("1: " RDRAND_LONG "\n\t"
-		     "jc 2f\n\t"
-		     "decl %0\n\t"
-		     "jnz 1b\n\t"
-		     "2:"
-		     : "=r" (ok), "=a" (*v)
-		     : "0" (RDRAND_RETRY_LOOPS));
-	return !!ok;
+	bool ok;
+	unsigned int retry = RDRAND_RETRY_LOOPS;
+	do {
+		asm volatile(RDRAND_LONG "\n\t"
+			     CC_SET(c)
+			     : CC_OUT(c) (ok), "=a" (*v));
+		if (ok)
+			return true;
+	} while (--retry);
+	return false;
+}
+
+static inline bool rdrand_int(unsigned int *v)
+{
+	bool ok;
+	unsigned int retry = RDRAND_RETRY_LOOPS;
+	do {
+		asm volatile(RDRAND_INT "\n\t"
+			     CC_SET(c)
+			     : CC_OUT(c) (ok), "=a" (*v));
+		if (ok)
+			return true;
+	} while (--retry);
+	return false;
 }
 
-/* A single attempt at RDSEED */
 static inline bool rdseed_long(unsigned long *v)
 {
 	bool ok;
 	asm volatile(RDSEED_LONG "\n\t"
-		     "setc %0"
-		     : "=qm" (ok), "=a" (*v));
+		     CC_SET(c)
+		     : CC_OUT(c) (ok), "=a" (*v));
 	return ok;
 }
 
-#define GET_RANDOM(name, type, rdrand, nop)			\
-static inline bool name(type *v)				\
-{								\
-	int ok;							\
-	alternative_io("movl $0, %0\n\t"			\
-		       nop,					\
-		       "\n1: " rdrand "\n\t"			\
-		       "jc 2f\n\t"				\
-		       "decl %0\n\t"                            \
-		       "jnz 1b\n\t"                             \
-		       "2:",                                    \
-		       X86_FEATURE_RDRAND,                      \
-		       ASM_OUTPUT2("=r" (ok), "=a" (*v)),       \
-		       "0" (RDRAND_RETRY_LOOPS));		\
-	return !!ok;						\
-}
-
-#define GET_SEED(name, type, rdseed, nop)			\
-static inline bool name(type *v)				\
-{								\
-	bool ok;						\
-	alternative_io("movb $0, %0\n\t"			\
-		       nop,					\
-		       rdseed "\n\t"				\
-		       "setc %0",				\
-		       X86_FEATURE_RDSEED,                      \
-		       ASM_OUTPUT2("=q" (ok), "=a" (*v)));	\
-	return ok;						\
+static inline bool rdseed_int(unsigned int *v)
+{
+	bool ok;
+	asm volatile(RDSEED_INT "\n\t"
+		     CC_SET(c)
+		     : CC_OUT(c) (ok), "=a" (*v));
+	return ok;
 }
 
-#ifdef CONFIG_X86_64
-
-GET_RANDOM(arch_get_random_long, unsigned long, RDRAND_LONG, ASM_NOP5);
-GET_RANDOM(arch_get_random_int, unsigned int, RDRAND_INT, ASM_NOP4);
-
-GET_SEED(arch_get_random_seed_long, unsigned long, RDSEED_LONG, ASM_NOP5);
-GET_SEED(arch_get_random_seed_int, unsigned int, RDSEED_INT, ASM_NOP4);
-
-#else
-
-GET_RANDOM(arch_get_random_long, unsigned long, RDRAND_LONG, ASM_NOP3);
-GET_RANDOM(arch_get_random_int, unsigned int, RDRAND_INT, ASM_NOP3);
-
-GET_SEED(arch_get_random_seed_long, unsigned long, RDSEED_LONG, ASM_NOP4);
-GET_SEED(arch_get_random_seed_int, unsigned int, RDSEED_INT, ASM_NOP4);
-
-#endif /* CONFIG_X86_64 */
-
+/* Conditional execution based on CPU type */
 #define arch_has_random()	static_cpu_has(X86_FEATURE_RDRAND)
 #define arch_has_random_seed()	static_cpu_has(X86_FEATURE_RDSEED)
 
-#else
+/*
+ * These are the generic interfaces; they must not be declared if the
+ * stubs in <linux/random.h> are to be invoked,
+ * i.e. CONFIG_ARCH_RANDOM is not defined.
+ */
+#ifdef CONFIG_ARCH_RANDOM
 
-static inline bool rdrand_long(unsigned long *v)
+static inline bool arch_get_random_long(unsigned long *v)
 {
-	return 0;
+	return arch_has_random() ? rdrand_long(v) : false;
 }
 
-static inline bool rdseed_long(unsigned long *v)
+static inline bool arch_get_random_int(unsigned int *v)
 {
-	return 0;
+	return arch_has_random() ? rdrand_int(v) : false;
 }
 
-#endif  /* CONFIG_ARCH_RANDOM */
+static inline bool arch_get_random_seed_long(unsigned long *v)
+{
+	return arch_has_random_seed() ? rdseed_long(v) : false;
+}
+
+static inline bool arch_get_random_seed_int(unsigned int *v)
+{
+	return arch_has_random_seed() ? rdseed_int(v) : false;
+}
 
 extern void x86_init_rdrand(struct cpuinfo_x86 *c);
 
+#else  /* !CONFIG_ARCH_RANDOM */
+
+static inline void x86_init_rdrand(struct cpuinfo_x86 *c) { }
+
+#endif  /* !CONFIG_ARCH_RANDOM */
+
 #endif /* ASM_X86_ARCHRANDOM_H */
diff --git a/arch/x86/kernel/cpu/rdrand.c b/arch/x86/kernel/cpu/rdrand.c
index f6f50c4..cfa97ff 100644
--- a/arch/x86/kernel/cpu/rdrand.c
+++ b/arch/x86/kernel/cpu/rdrand.c
@@ -39,9 +39,9 @@ __setup("nordrand", x86_rdrand_setup);
  */
 #define SANITY_CHECK_LOOPS 8
 
+#ifdef CONFIG_ARCH_RANDOM
 void x86_init_rdrand(struct cpuinfo_x86 *c)
 {
-#ifdef CONFIG_ARCH_RANDOM
 	unsigned long tmp;
 	int i;
 
@@ -55,5 +55,5 @@ void x86_init_rdrand(struct cpuinfo_x86 *c)
 			return;
 		}
 	}
-#endif
 }
+#endif
-- 
2.7.3.0.11.gd79db92

^ permalink raw reply related	[flat|nested] 43+ messages in thread
end of thread, other threads:[~2016-06-08 19:50 UTC | newest]

Thread overview: 43+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2016-06-07 23:30 [PATCH 00/10] x86: use gcc 6+ asm flag output feature H. Peter Anvin
2016-06-07 23:31 ` [PATCH 01/10] x86, bitops: remove use of "sbb" to return CF H. Peter Anvin
2016-06-07 23:39   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 02/10] x86, asm: use bool for bitops and other assembly outputs H. Peter Anvin
2016-06-07 23:40   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-08  7:49   ` [PATCH 02/10] " Peter Zijlstra
2016-06-08  8:28     ` Ingo Molnar
2016-06-08  8:33       ` Ingo Molnar
2016-06-08  8:55         ` H. Peter Anvin
2016-06-08  9:01           ` Ingo Molnar
2016-06-08  9:12             ` H. Peter Anvin
2016-06-08  9:20               ` Ingo Molnar
2016-06-08  9:31                 ` H. Peter Anvin
2016-06-08  9:39                   ` Peter Zijlstra
2016-06-08  9:43                     ` H. Peter Anvin
2016-06-08  9:40                 ` H. Peter Anvin
2016-06-08  8:58       ` H. Peter Anvin
2016-06-08  9:06         ` Ingo Molnar
2016-06-07 23:31 ` [PATCH 03/10] x86, asm: change the GEN_*_RMWcc() macros to not quote the condition H. Peter Anvin
2016-06-07 23:40   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 04/10] x86, asm: define CC_SET() and CC_OUT() macros H. Peter Anvin
2016-06-07 23:40   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-08  5:09   ` [PATCH 04/10] " Andy Lutomirski
2016-06-08 19:29     ` H. Peter Anvin
2016-06-08 19:34       ` H. Peter Anvin
2016-06-08 19:35         ` Borislav Petkov
2016-06-07 23:31 ` [PATCH 05/10] x86, asm: change GEN_*_RMWcc() to use CC_SET()/CC_OUT() H. Peter Anvin
2016-06-07 23:41   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 06/10] x86, asm: Use CC_SET()/CC_OUT() in <asm/bitops.h> H. Peter Anvin
2016-06-07 23:41   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 07/10] x86, asm: Use CC_SET()/CC_OUT() in <asm/percpu.h> H. Peter Anvin
2016-06-07 23:42   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 08/10] x86, asm: Use CC_SET()/CC_OUT() in <asm/rwsem.h> H. Peter Anvin
2016-06-07 23:42   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-07 23:31 ` [PATCH 09/10] x86, asm: Use CC_SET()/CC_OUT() and static_cpu_has() in archrandom.h H. Peter Anvin
2016-06-07 23:42   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-08  5:12   ` [PATCH 09/10] " Andy Lutomirski
2016-06-07 23:31 ` [PATCH 10/10] x86, asm, boot: Use CC_SET()/CC_OUT() in arch/x86/boot/boot.h H. Peter Anvin
2016-06-07 23:43   ` [tip:x86/asm] " tip-bot for H. Peter Anvin
2016-06-08  8:00 ` [PATCH 00/10] x86: use gcc 6+ asm flag output feature Peter Zijlstra
2016-06-08  8:56   ` H. Peter Anvin
2016-06-08  8:26 ` Ingo Molnar
  -- strict thread matches above, loose matches on Subject: below --
2016-06-08 19:38 [PATCH v2 10/10] x86, asm: Use CC_SET()/CC_OUT() and static_cpu_has() in archrandom.h H. Peter Anvin
2016-06-08 19:49 ` [tip:x86/asm] " tip-bot for H. Peter Anvin

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.