coreboot-kgpe-d16/payloads/libpayload/include/compiler.h

54 lines
1.6 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only */
#ifndef _COMMONLIB_BSD_COMPILER_H_
#define _COMMONLIB_BSD_COMPILER_H_
#ifndef __packed
#if defined(__WIN32) || defined(__WIN64)
#define __packed __attribute__((gcc_struct, packed))
#else
#define __packed __attribute__((packed))
#endif
#endif
#ifndef __aligned
#define __aligned(x) __attribute__((aligned(x)))
#endif
#ifndef __always_unused
#define __always_unused __attribute__((unused))
#endif
#ifndef __must_check
#define __must_check __attribute__((warn_unused_result))
#endif
#ifndef __weak
#define __weak __attribute__((weak))
#endif
#ifndef __noreturn
#define __noreturn __attribute__((noreturn))
#endif
#ifndef __always_inline
#define __always_inline inline __attribute__((always_inline))
#endif
Make common macros double-evaluation safe I just got hit by a double-evaluation bug again, it's time to attempt to fix this once more. Unfortunately there are several issues that don't make this easy: - bitfield variables don't support typeof() - local macro variables that shadow others trigger -Werror=shadow - sign warnings with integer literal and unsigned var in typeof-MIN() - ({ statement expressions }) can not be used outside functions - romcc doesn't support any of the fancy GCC/clang extensions This patch tries to address all of them as far as possible with macro magic. We don't have the technology to solve the bitfield and non-function context issues yet (__builtin_choose_expr() still throws a "no statement expression outside a function" error if it's only in the branch that's not chosen, unfortunately), so we'll have to provide alternative macros for use in those cases (and we'll avoid making __ALIGN_MASK() double-evaluation safe for now, since it would be annoying to do that there and having an alignment mask with side effects seems very unlikely). romcc can continue using unsafe versions since we're hopefully not writing a lot of new code for it. Sign warnings can be avoided in literal/variable comparisons by always using the type of the variable there. Shadowing is avoided by picking very explicit local variable names and using a special __COUNTER__ solution for MIN() and MAX() (the only ones of these you're likely to nest). Also add DIV_ROUND_UP() to libpayload since it's a generally quite useful thing to have. Change-Id: Iea35156c9aa9f6f2c7b8f00991418b746f44315d Signed-off-by: Julius Werner <jwerner@chromium.org> Reviewed-on: https://review.coreboot.org/c/coreboot/+/32027 Tested-by: build bot (Jenkins) <no-reply@coreboot.org> Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net> Reviewed-by: Patrick Georgi <pgeorgi@google.com>
2015-05-23 03:09:48 +02:00
/* This evaluates to the type of the first expression, unless that is constant
in which case it evalutates to the type of the second. This is useful when
assigning macro parameters to temporary variables, because that would
normally circumvent the special loosened type promotion rules for integer
literals. By using this macro, the promotion can happen at the time the
literal is assigned to the temporary variable. If the literal doesn't fit in
the chosen type, -Werror=overflow will catch it, so this should be safe. */
#define __TYPEOF_UNLESS_CONST(expr, fallback_expr) __typeof__( \
Make common macros double-evaluation safe I just got hit by a double-evaluation bug again, it's time to attempt to fix this once more. Unfortunately there are several issues that don't make this easy: - bitfield variables don't support typeof() - local macro variables that shadow others trigger -Werror=shadow - sign warnings with integer literal and unsigned var in typeof-MIN() - ({ statement expressions }) can not be used outside functions - romcc doesn't support any of the fancy GCC/clang extensions This patch tries to address all of them as far as possible with macro magic. We don't have the technology to solve the bitfield and non-function context issues yet (__builtin_choose_expr() still throws a "no statement expression outside a function" error if it's only in the branch that's not chosen, unfortunately), so we'll have to provide alternative macros for use in those cases (and we'll avoid making __ALIGN_MASK() double-evaluation safe for now, since it would be annoying to do that there and having an alignment mask with side effects seems very unlikely). romcc can continue using unsafe versions since we're hopefully not writing a lot of new code for it. Sign warnings can be avoided in literal/variable comparisons by always using the type of the variable there. Shadowing is avoided by picking very explicit local variable names and using a special __COUNTER__ solution for MIN() and MAX() (the only ones of these you're likely to nest). Also add DIV_ROUND_UP() to libpayload since it's a generally quite useful thing to have. Change-Id: Iea35156c9aa9f6f2c7b8f00991418b746f44315d Signed-off-by: Julius Werner <jwerner@chromium.org> Reviewed-on: https://review.coreboot.org/c/coreboot/+/32027 Tested-by: build bot (Jenkins) <no-reply@coreboot.org> Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net> Reviewed-by: Patrick Georgi <pgeorgi@google.com>
2015-05-23 03:09:48 +02:00
__builtin_choose_expr(__builtin_constant_p(expr), fallback_expr, expr))
/* This creates a unique local variable name for use in macros. */
#define __TMPNAME_3(i) __tmpname_##i
#define __TMPNAME_2(i) __TMPNAME_3(i)
#define __TMPNAME __TMPNAME_2(__COUNTER__)
#endif