mentos-team/MentOS

Motivation The kernel logging macros (pr_emerg, pr_crit, pr_err, pr_warning, pr_notice, pr_debug, …) all forward a format string and varargs to dbg_printf(), which carries no format attribute. As a result no format/argument mismatch anywhere in the kernel can be diagnosed by the compiler, no matter how strict the warning flags. This is not hypothetical: #211 is a live instance where %s is handed an integer file size, on an error path reached while handling malformed executables. That bug would almost certainly have been a compile-time warning had the attribute been present. Annotating one declaration would give the entire kernel this checking retroactively. Proposed behavior Annotate the variadic logging entry point so GCC and Clang verify format strings against their arguments: /* kernel/inc/io/debug.h */ void dbg_printf(const char *file, const char *fun, int line, char *header, short log_level, const char *format, ...) __attribute__((format(printf, 6, 7))); (ind

Documentation location kernel/src/mem/page_fault.c — the doc comment on the ERR_PRESENT macro (source documentation, not the Wiki). What is unclear or missing? The x86 page-fault error code's bit 0 means "the page was present": bit 0 = 0 -> the fault was caused by a non-present page bit 0 = 1 -> the fault was caused by a protection violation on a present page The macro's comment states the inverse: // Error code interpretation. #define ERR_PRESENT 0x01 ///< Page not present. #define ERR_RW 0x02 ///< Page is read only. #define ERR_USER 0x04 ///< Page is privileged. #define ERR_RESERVED 0x08 ///< Overwrote reserved bit. #define ERR_INST 0x10 ///< Instruction fetch. Read literally, ERR_PRESENT is documented as meaning "page not present", which is the opposite of what the bit encodes. A reader who trusts the comment will invert the meaning of every page-fault dump they look at. The neighbouring comments have the same loose style — ERR_RW is described as "Page is rea