11. Preprocessor
   3For variadic macros, stick with this C99-like syntax:
   5#define DPRINTF(fmt, ...)                                       \
   6    do { printf("IRQ: " fmt, ## __VA_ARGS__); } while (0)
   82. C types
  10It should be common sense to use the right type, but we have collected
  11a few useful guidelines here.
  132.1. Scalars
  15If you're using "int" or "long", odds are good that there's a better type.
  16If a variable is counting something, it should be declared with an
  17unsigned type.
  19If it's host memory-size related, size_t should be a good choice (use
  20ssize_t only if required). Guest RAM memory offsets must use ram_addr_t,
  21but only for RAM, it may not cover whole guest address space.
  23If it's file-size related, use off_t.
  24If it's file-offset related (i.e., signed), use off_t.
  25If it's just counting small numbers use "unsigned int";
  26(on all but oddball embedded systems, you can assume that that
  27type is at least four bytes wide).
  29In the event that you require a specific width, use a standard type
  30like int32_t, uint32_t, uint64_t, etc.  The specific types are
  31mandatory for VMState fields.
  33Don't use Linux kernel internal types like u32, __u32 or __le32.
  35Use hwaddr for guest physical addresses except pcibus_t
  36for PCI addresses.  In addition, ram_addr_t is a QEMU internal address
  37space that maps guest RAM physical addresses into an intermediate
  38address space that can map to host virtual address spaces.  Generally
  39speaking, the size of guest memory can always fit into ram_addr_t but
  40it would not be correct to store an actual guest physical address in a
  43For CPU virtual addresses there are several possible types.
  44vaddr is the best type to use to hold a CPU virtual address in
  45target-independent code. It is guaranteed to be large enough to hold a
  46virtual address for any target, and it does not change size from target
  47to target. It is always unsigned.
  48target_ulong is a type the size of a virtual address on the CPU; this means
  49it may be 32 or 64 bits depending on which target is being built. It should
  50therefore be used only in target-specific code, and in some
  51performance-critical built-per-target core code such as the TLB code.
  52There is also a signed version, target_long.
  53abi_ulong is for the *-user targets, and represents a type the size of
  54'void *' in that target's ABI. (This may not be the same as the size of a
  55full CPU virtual address in the case of target ABIs which use 32 bit pointers
  56on 64 bit CPUs, like sparc32plus.) Definitions of structures that must match
  57the target's ABI must use this type for anything that on the target is defined
  58to be an 'unsigned long' or a pointer type.
  59There is also a signed version, abi_long.
  61Of course, take all of the above with a grain of salt.  If you're about
  62to use some system interface that requires a type like size_t, pid_t or
  63off_t, use matching types for any corresponding variables.
  65Also, if you try to use e.g., "unsigned int" as a type, and that
  66conflicts with the signedness of a related variable, sometimes
  67it's best just to use the *wrong* type, if "pulling the thread"
  68and fixing all related variables would be too invasive.
  70Finally, while using descriptive types is important, be careful not to
  71go overboard.  If whatever you're doing causes warnings, or requires
  72casts, then reconsider or ask for help.
  742.2. Pointers
  76Ensure that all of your pointers are "const-correct".
  77Unless a pointer is used to modify the pointed-to storage,
  78give it the "const" attribute.  That way, the reader knows
  79up-front that this is a read-only pointer.  Perhaps more
  80importantly, if we're diligent about this, when you see a non-const
  81pointer, you're guaranteed that it is used to modify the storage
  82it points to, or it is aliased to another pointer that is.
  842.3. Typedefs
  85Typedefs are used to eliminate the redundant 'struct' keyword.
  872.4. Reserved namespaces in C and POSIX
  88Underscore capital, double underscore, and underscore 't' suffixes should be
  913. Low level memory management
  93Use of the malloc/free/realloc/calloc/valloc/memalign/posix_memalign
  94APIs is not allowed in the QEMU codebase. Instead of these routines,
  95use the GLib memory allocation routines g_malloc/g_malloc0/g_new/
  96g_new0/g_realloc/g_free or QEMU's qemu_memalign/qemu_blockalign/qemu_vfree
  99Please note that g_malloc will exit on allocation failure, so there
 100is no need to test for failure (as you would have to with malloc).
 101Calling g_malloc with a zero size is valid and will return NULL.
 103Memory allocated by qemu_memalign or qemu_blockalign must be freed with
 104qemu_vfree, since breaking this will cause problems on Win32.
 1064. String manipulation
 108Do not use the strncpy function.  As mentioned in the man page, it does *not*
 109guarantee a NULL-terminated buffer, which makes it extremely dangerous to use.
 110It also zeros trailing destination bytes out to the specified length.  Instead,
 111use this similar function when possible, but note its different signature:
 112void pstrcpy(char *dest, int dest_buf_size, const char *src)
 114Don't use strcat because it can't check for buffer overflows, but:
 115char *pstrcat(char *buf, int buf_size, const char *s)
 117The same limitation exists with sprintf and vsprintf, so use snprintf and
 120QEMU provides other useful string functions:
 121int strstart(const char *str, const char *val, const char **ptr)
 122int stristart(const char *str, const char *val, const char **ptr)
 123int qemu_strnlen(const char *s, int max_len)
 125There are also replacement character processing macros for isxyz and toxyz,
 126so instead of e.g. isalnum you should use qemu_isalnum.
 128Because of the memory management rules, you must use g_strdup/g_strndup
 129instead of plain strdup/strndup.
 1315. Printf-style functions
 133Whenever you add a new printf-style function, i.e., one with a format
 134string argument and following "..." in its prototype, be sure to use
 135gcc's printf attribute directive in the prototype.
 137This makes it so gcc's -Wformat and -Wformat-security options can do
 138their jobs and cross-check format strings with the number and types
 139of arguments.
 1416. C standard, implementation defined and undefined behaviors
 143C code in QEMU should be written to the C99 language specification. A copy
 144of the final version of the C99 standard with corrigenda TC1, TC2, and TC3
 145included, formatted as a draft, can be downloaded from:
 146 http://www.open-std.org/jtc1/sc22/WG14/www/docs/n1256.pdf
 148The C language specification defines regions of undefined behavior and
 149implementation defined behavior (to give compiler authors enough leeway to
 150produce better code).  In general, code in QEMU should follow the language
 151specification and avoid both undefined and implementation defined
 152constructs. ("It works fine on the gcc I tested it with" is not a valid
 153argument...) However there are a few areas where we allow ourselves to
 154assume certain behaviors because in practice all the platforms we care about
 155behave in the same way and writing strictly conformant code would be
 156painful. These are:
 157 * you may assume that integers are 2s complement representation
 158 * you may assume that right shift of a signed integer duplicates
 159   the sign bit (ie it is an arithmetic shift, not a logical shift)
 1617. Error handling and reporting
 1637.1 Reporting errors to the human user
 165Do not use printf(), fprintf() or monitor_printf().  Instead, use
 166error_report() or error_vreport() from error-report.h.  This ensures the
 167error is reported in the right place (current monitor or stderr), and in
 168a uniform format.
 170Use error_printf() & friends to print additional information.
 172error_report() prints the current location.  In certain common cases
 173like command line parsing, the current location is tracked
 174automatically.  To manipulate it manually, use the loc_*() from
 1777.2 Propagating errors
 179An error can't always be reported to the user right where it's detected,
 180but often needs to be propagated up the call chain to a place that can
 181handle it.  This can be done in various ways.
 183The most flexible one is Error objects.  See error.h for usage
 186Use the simplest suitable method to communicate success / failure to
 187callers.  Stick to common methods: non-negative on success / -1 on
 188error, non-negative / -errno, non-null / null, or Error objects.
 190Example: when a function returns a non-null pointer on success, and it
 191can fail only in one way (as far as the caller is concerned), returning
 192null on failure is just fine, and certainly simpler and a lot easier on
 193the eyes than propagating an Error object through an Error ** parameter.
 195Example: when a function's callers need to report details on failure
 196only the function really knows, use Error **, and set suitable errors.
 198Do not report an error to the user when you're also returning an error
 199for somebody else to handle.  Leave the reporting to the place that
 200consumes the error returned.
 2027.3 Handling errors
 204Calling exit() is fine when handling configuration errors during
 205startup.  It's problematic during normal operation.  In particular,
 206monitor commands should never exit().
 208Do not call exit() or abort() to handle an error that can be triggered
 209by the guest (e.g., some unimplemented corner case in guest code
 210translation or device emulation).  Guests should not be able to
 211terminate QEMU.
 213Note that &error_fatal is just another way to exit(1), and &error_abort
 214is just another way to abort().