HACKING

   1 1. Preprocessor
   2
   3 For variadic macros, stick with this C99-like syntax:
   4
   5 #define DPRINTF(fmt, ...)                                       \
   6     do { printf("IRQ: " fmt, ## __VA_ARGS__); } while (0)
   7
   8 2. C types
   9
  10 It should be common sense to use the right type, but we have collected
  11 a few useful guidelines here.
  12
  13 2.1. Scalars
  14
  15 If you're using "int" or "long", odds are good that there's a better type.
  16 If a variable is counting something, it should be declared with an
  17 unsigned type.
  18
  19 If it's host memory-size related, size_t should be a good choice (use
  20 ssize_t only if required). Guest RAM memory offsets must use ram_addr_t,
  21 but only for RAM, it may not cover whole guest address space.
  22
  23 If it's file-size related, use off_t.
  24 If it's file-offset related (i.e., signed), use off_t.
  25 If it's just counting small numbers use "unsigned int";
  26 (on all but oddball embedded systems, you can assume that that
  27 type is at least four bytes wide).
  28
  29 In the event that you require a specific width, use a standard type
  30 like int32_t, uint32_t, uint64_t, etc.  The specific types are
  31 mandatory for VMState fields.
  32
  33 Don't use Linux kernel internal types like u32, __u32 or __le32.
  34
  35 Use hwaddr for guest physical addresses except pcibus_t
  36 for PCI addresses.  In addition, ram_addr_t is a QEMU internal address
  37 space that maps guest RAM physical addresses into an intermediate
  38 address space that can map to host virtual address spaces.  Generally
  39 speaking, the size of guest memory can always fit into ram_addr_t but
  40 it would not be correct to store an actual guest physical address in a
  41 ram_addr_t.
  42
  43 For CPU virtual addresses there are several possible types.
  44 vaddr is the best type to use to hold a CPU virtual address in
  45 target-independent code. It is guaranteed to be large enough to hold a
  46 virtual address for any target, and it does not change size from target
  47 to target. It is always unsigned.
  48 target_ulong is a type the size of a virtual address on the CPU; this means
  49 it may be 32 or 64 bits depending on which target is being built. It should
  50 therefore be used only in target-specific code, and in some
  51 performance-critical built-per-target core code such as the TLB code.
  52 There is also a signed version, target_long.
  53 abi_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
  55 full CPU virtual address in the case of target ABIs which use 32 bit pointers
  56 on 64 bit CPUs, like sparc32plus.) Definitions of structures that must match
  57 the target's ABI must use this type for anything that on the target is defined
  58 to be an 'unsigned long' or a pointer type.
  59 There is also a signed version, abi_long.
  60
  61 Of course, take all of the above with a grain of salt.  If you're about
  62 to use some system interface that requires a type like size_t, pid_t or
  63 off_t, use matching types for any corresponding variables.
  64
  65 Also, if you try to use e.g., "unsigned int" as a type, and that
  66 conflicts with the signedness of a related variable, sometimes
  67 it's best just to use the *wrong* type, if "pulling the thread"
  68 and fixing all related variables would be too invasive.
  69
  70 Finally, while using descriptive types is important, be careful not to
  71 go overboard.  If whatever you're doing causes warnings, or requires
  72 casts, then reconsider or ask for help.
  73
  74 2.2. Pointers
  75
  76 Ensure that all of your pointers are "const-correct".
  77 Unless a pointer is used to modify the pointed-to storage,
  78 give it the "const" attribute.  That way, the reader knows
  79 up-front that this is a read-only pointer.  Perhaps more
  80 importantly, if we're diligent about this, when you see a non-const
  81 pointer, you're guaranteed that it is used to modify the storage
  82 it points to, or it is aliased to another pointer that is.
  83
  84 2.3. Typedefs
  85 Typedefs are used to eliminate the redundant 'struct' keyword.
  86
  87 2.4. Reserved namespaces in C and POSIX
  88 Underscore capital, double underscore, and underscore 't' suffixes should be
  89 avoided.
  90
  91 3. Low level memory management
  92
  93 Use of the malloc/free/realloc/calloc/valloc/memalign/posix_memalign
  94 APIs is not allowed in the QEMU codebase. Instead of these routines,
  95 use the GLib memory allocation routines g_malloc/g_malloc0/g_new/
  96 g_new0/g_realloc/g_free or QEMU's qemu_memalign/qemu_blockalign/qemu_vfree
  97 APIs.
  98
  99 Please note that g_malloc will exit on allocation failure, so there
 100 is no need to test for failure (as you would have to with malloc).
 101 Calling g_malloc with a zero size is valid and will return NULL.
 102
 103 Memory allocated by qemu_memalign or qemu_blockalign must be freed with
 104 qemu_vfree, since breaking this will cause problems on Win32.
 105
 106 4. String manipulation
 107
 108 Do not use the strncpy function.  As mentioned in the man page, it does *not*
 109 guarantee a NULL-terminated buffer, which makes it extremely dangerous to use.
 110 It also zeros trailing destination bytes out to the specified length.  Instead,
 111 use this similar function when possible, but note its different signature:
 112 void pstrcpy(char *dest, int dest_buf_size, const char *src)
 113
 114 Don't use strcat because it can't check for buffer overflows, but:
 115 char *pstrcat(char *buf, int buf_size, const char *s)
 116
 117 The same limitation exists with sprintf and vsprintf, so use snprintf and
 118 vsnprintf.
 119
 120 QEMU provides other useful string functions:
 121 int strstart(const char *str, const char *val, const char **ptr)
 122 int stristart(const char *str, const char *val, const char **ptr)
 123 int qemu_strnlen(const char *s, int max_len)
 124
 125 There are also replacement character processing macros for isxyz and toxyz,
 126 so instead of e.g. isalnum you should use qemu_isalnum.
 127
 128 Because of the memory management rules, you must use g_strdup/g_strndup
 129 instead of plain strdup/strndup.
 130
 131 5. Printf-style functions
 132
 133 Whenever you add a new printf-style function, i.e., one with a format
 134 string argument and following "..." in its prototype, be sure to use
 135 gcc's printf attribute directive in the prototype.
 136
 137 This makes it so gcc's -Wformat and -Wformat-security options can do
 138 their jobs and cross-check format strings with the number and types
 139 of arguments.
 140
 141 6. C standard, implementation defined and undefined behaviors
 142
 143 C code in QEMU should be written to the C99 language specification. A copy
 144 of the final version of the C99 standard with corrigenda TC1, TC2, and TC3
 145 included, formatted as a draft, can be downloaded from:
 146  http://www.open-std.org/jtc1/sc22/WG14/www/docs/n1256.pdf
 147
 148 The C language specification defines regions of undefined behavior and
 149 implementation defined behavior (to give compiler authors enough leeway to
 150 produce better code).  In general, code in QEMU should follow the language
 151 specification and avoid both undefined and implementation defined
 152 constructs. ("It works fine on the gcc I tested it with" is not a valid
 153 argument...) However there are a few areas where we allow ourselves to
 154 assume certain behaviors because in practice all the platforms we care about
 155 behave in the same way and writing strictly conformant code would be
 156 painful. 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)
 160
 161 In addition, QEMU assumes that the compiler does not use the latitude
 162 given in C99 and C11 to treat aspects of signed '<<' as undefined, as
 163 documented in the GNU Compiler Collection manual starting at version 4.0.
 164
 165 7. Error handling and reporting
 166
 167 7.1 Reporting errors to the human user
 168
 169 Do not use printf(), fprintf() or monitor_printf().  Instead, use
 170 error_report() or error_vreport() from error-report.h.  This ensures the
 171 error is reported in the right place (current monitor or stderr), and in
 172 a uniform format.
 173
 174 Use error_printf() & friends to print additional information.
 175
 176 error_report() prints the current location.  In certain common cases
 177 like command line parsing, the current location is tracked
 178 automatically.  To manipulate it manually, use the loc_*() from
 179 error-report.h.
 180
 181 7.2 Propagating errors
 182
 183 An error can't always be reported to the user right where it's detected,
 184 but often needs to be propagated up the call chain to a place that can
 185 handle it.  This can be done in various ways.
 186
 187 The most flexible one is Error objects.  See error.h for usage
 188 information.
 189
 190 Use the simplest suitable method to communicate success / failure to
 191 callers.  Stick to common methods: non-negative on success / -1 on
 192 error, non-negative / -errno, non-null / null, or Error objects.
 193
 194 Example: when a function returns a non-null pointer on success, and it
 195 can fail only in one way (as far as the caller is concerned), returning
 196 null on failure is just fine, and certainly simpler and a lot easier on
 197 the eyes than propagating an Error object through an Error ** parameter.
 198
 199 Example: when a function's callers need to report details on failure
 200 only the function really knows, use Error **, and set suitable errors.
 201
 202 Do not report an error to the user when you're also returning an error
 203 for somebody else to handle.  Leave the reporting to the place that
 204 consumes the error returned.
 205
 206 7.3 Handling errors
 207
 208 Calling exit() is fine when handling configuration errors during
 209 startup.  It's problematic during normal operation.  In particular,
 210 monitor commands should never exit().
 211
 212 Do not call exit() or abort() to handle an error that can be triggered
 213 by the guest (e.g., some unimplemented corner case in guest code
 214 translation or device emulation).  Guests should not be able to
 215 terminate QEMU.
 216
 217 Note that &error_fatal is just another way to exit(1), and &error_abort
 218 is just another way to abort().