Differential D4964 Diff 12388 sys/libkern/explicit_bzero.c

Changeset View

Standalone View

sys/libkern/explicit_bzero.c

Context not available.
	#include <string.h>	#include <string.h>
	#endif /* _KERNEL */	#endif /* _KERNEL */

	__attribute__((weak)) void __explicit_bzero_hook(void *, size_t);

	__attribute__((weak)) void	#if __clang__
	__explicit_bzero_hook(void *buf, size_t len)	/*
	{	* http://clang.llvm.org/docs/LanguageExtensions.html#feature-checking-macros
	}	* http://lists.cs.uiuc.edu/pipermail/cfe-dev/2014-December/040627.html
	impUnsubmitted Not Done Inline Actions I'd move this to after the clang / gcc block that you aded below. I'd condition it on #ifndef NOOPT. I'd also include the explicit_bzero definition from below. This will support old compilers that don't do LTO as well as new that can / do. imp: I'd move this to after the clang / gcc block that you aded below. I'd condition it on #ifndef…
		*/
		#if __has_attribute( noinline ) /* && __has_attribute( optnone ) */
		#define NOOPT /* __attribute__ (( optnone )) */
		#define NOINLINE __attribute__ (( noinline ))
		#else
		#error "require clang with noinline and optnone attributes"
		impUnsubmitted Not Done Inline Actions I'd ditch the #error stuff here and below. imp: I'd ditch the #error stuff here and below.
		#endif
		theravenUnsubmitted Not Done Inline Actions Does this do anything? Can we compile the kernel with a compiler that doesn't support GNU C? theraven: Does this do anything? Can we compile the kernel with a compiler that doesn't support GNU C?
		#elif __GNUC__
		/*
		* http://gcc.gnu.org/onlinedocs/gcc/Function-Specific-Option-Pragmas.html
		* http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html
		*/
		#if __GNUC__ > 4 \|\| ( __GNUC__ == 4 && __GNUC_MINOR__ >= 4 )
		#define NOOPT __attribute__ (( optimize( 0 ) ))
		#define NOINLINE __attribute__ (( noinline ))
		#else
		#error "require gcc >= 4.4"
		#endif
		#else
		theravenUnsubmitted Not Done Inline Actions I'm not sure that NOOPT is needed here. Even turning this into a memset / bzero call should be fine for the compiler to do, if this function is not inlined. It should also be safe for LLVM to replace a call to explicit_bzero with an LLVM memset intrinsic with the volatile flag set, and so generate an optimised memset sequence that can't be removed. theraven: I'm not sure that NOOPT is needed here. Even turning this into a memset / bzero call should be…
		#error "unrecognised compiler"
		explode
		#endif

	void	NOOPT NOINLINE void
	explicit_bzero(void *buf, size_t len)	explicit_bzero(void *buf, size_t len)
	{	{
	memset(buf, 0, len);	size_t i;
	__explicit_bzero_hook(buf, len);	unsigned char *p = buf;

		for (i = 0; i < len; i++) {
		p[i] = 0;
		}
	}	}
		impUnsubmitted Not Done Inline Actions I'd be tempted to do: #ifdef NOOPT NOOPT NOLINE void explicit_bzero(void buf, size_t len) { <body> } #endif imp:* I'd be tempted to do: #ifdef NOOPT NOOPT NOLINE void explicit_bzero(void *buf, size_t len) {…

Context not available.