HACKING

   1 1. Preprocessor
   2
   3 1.1. Variadic macros
   4
   5 For variadic macros, stick with this C99-like syntax:
   6
   7 #define DPRINTF(fmt, ...)                                       \
   8     do { printf("IRQ: " fmt, ## __VA_ARGS__); } while (0)
   9
  10 1.2. Include directives
  11
  12 Order include directives as follows:
  13
  14 #include "qemu/osdep.h"  /* Always first... */
  15 #include <...>           /* then system headers... */
  16 #include "..."           /* and finally QEMU headers. */
  17
  18 The "qemu/osdep.h" header contains preprocessor macros that affect the behavior
  19 of core system headers like <stdint.h>.  It must be the first include so that
  20 core system headers included by external libraries get the preprocessor macros
  21 that QEMU depends on.
  22
  23 Do not include "qemu/osdep.h" from header files since the .c file will have
  24 already included it.
  25
  26 2. C types
  27
  28 It should be common sense to use the right type, but we have collected
  29 a few useful guidelines here.
  30
  31 2.1. Scalars
  32
  33 If you're using "int" or "long", odds are good that there's a better type.
  34 If a variable is counting something, it should be declared with an
  35 unsigned type.
  36
  37 If it's host memory-size related, size_t should be a good choice (use
  38 ssize_t only if required). Guest RAM memory offsets must use ram_addr_t,
  39 but only for RAM, it may not cover whole guest address space.
  40
  41 If it's file-size related, use off_t.
  42 If it's file-offset related (i.e., signed), use off_t.
  43 If it's just counting small numbers use "unsigned int";
  44 (on all but oddball embedded systems, you can assume that that
  45 type is at least four bytes wide).
  46
  47 In the event that you require a specific width, use a standard type
  48 like int32_t, uint32_t, uint64_t, etc.  The specific types are
  49 mandatory for VMState fields.
  50
  51 Don't use Linux kernel internal types like u32, __u32 or __le32.
  52
  53 Use hwaddr for guest physical addresses except pcibus_t
  54 for PCI addresses.  In addition, ram_addr_t is a QEMU internal address
  55 space that maps guest RAM physical addresses into an intermediate
  56 address space that can map to host virtual address spaces.  Generally
  57 speaking, the size of guest memory can always fit into ram_addr_t but
  58 it would not be correct to store an actual guest physical address in a
  59 ram_addr_t.
  60
  61 For CPU virtual addresses there are several possible types.
  62 vaddr is the best type to use to hold a CPU virtual address in
  63 target-independent code. It is guaranteed to be large enough to hold a
  64 virtual address for any target, and it does not change size from target
  65 to target. It is always unsigned.
  66 target_ulong is a type the size of a virtual address on the CPU; this means
  67 it may be 32 or 64 bits depending on which target is being built. It should
  68 therefore be used only in target-specific code, and in some
  69 performance-critical built-per-target core code such as the TLB code.
  70 There is also a signed version, target_long.
  71 abi_ulong is for the *-user targets, and represents a type the size of
  72 'void *' in that target's ABI. (This may not be the same as the size of a
  73 full CPU virtual address in the case of target ABIs which use 32 bit pointers
  74 on 64 bit CPUs, like sparc32plus.) Definitions of structures that must match
  75 the target's ABI must use this type for anything that on the target is defined
  76 to be an 'unsigned long' or a pointer type.
  77 There is also a signed version, abi_long.
  78
  79 Of course, take all of the above with a grain of salt.  If you're about
  80 to use some system interface that requires a type like size_t, pid_t or
  81 off_t, use matching types for any corresponding variables.
  82
  83 Also, if you try to use e.g., "unsigned int" as a type, and that
  84 conflicts with the signedness of a related variable, sometimes
  85 it's best just to use the *wrong* type, if "pulling the thread"
  86 and fixing all related variables would be too invasive.
  87
  88 Finally, while using descriptive types is important, be careful not to
  89 go overboard.  If whatever you're doing causes warnings, or requires
  90 casts, then reconsider or ask for help.
  91
  92 2.2. Pointers
  93
  94 Ensure that all of your pointers are "const-correct".
  95 Unless a pointer is used to modify the pointed-to storage,
  96 give it the "const" attribute.  That way, the reader knows
  97 up-front that this is a read-only pointer.  Perhaps more
  98 importantly, if we're diligent about this, when you see a non-const
  99 pointer, you're guaranteed that it is used to modify the storage
 100 it points to, or it is aliased to another pointer that is.
 101
 102 2.3. Typedefs
 103 Typedefs are used to eliminate the redundant 'struct' keyword.
 104
 105 2.4. Reserved namespaces in C and POSIX
 106 Underscore capital, double underscore, and underscore 't' suffixes should be
 107 avoided.
 108
 109 3. Low level memory management
 110
 111 Use of the malloc/free/realloc/calloc/valloc/memalign/posix_memalign
 112 APIs is not allowed in the QEMU codebase. Instead of these routines,
 113 use the GLib memory allocation routines g_malloc/g_malloc0/g_new/
 114 g_new0/g_realloc/g_free or QEMU's qemu_memalign/qemu_blockalign/qemu_vfree
 115 APIs.
 116
 117 Please note that g_malloc will exit on allocation failure, so there
 118 is no need to test for failure (as you would have to with malloc).
 119 Calling g_malloc with a zero size is valid and will return NULL.
 120
 121 Prefer g_new(T, n) instead of g_malloc(sizeof(T) * n) for the following
 122 reasons:
 123
 124   a. It catches multiplication overflowing size_t;
 125   b. It returns T * instead of void *, letting compiler catch more type
 126      errors.
 127
 128 Declarations like T *v = g_malloc(sizeof(*v)) are acceptable, though.
 129
 130 Memory allocated by qemu_memalign or qemu_blockalign must be freed with
 131 qemu_vfree, since breaking this will cause problems on Win32.
 132
 133 4. String manipulation
 134
 135 Do not use the strncpy function.  As mentioned in the man page, it does *not*
 136 guarantee a NULL-terminated buffer, which makes it extremely dangerous to use.
 137 It also zeros trailing destination bytes out to the specified length.  Instead,
 138 use this similar function when possible, but note its different signature:
 139 void pstrcpy(char *dest, int dest_buf_size, const char *src)
 140
 141 Don't use strcat because it can't check for buffer overflows, but:
 142 char *pstrcat(char *buf, int buf_size, const char *s)
 143
 144 The same limitation exists with sprintf and vsprintf, so use snprintf and
 145 vsnprintf.
 146
 147 QEMU provides other useful string functions:
 148 int strstart(const char *str, const char *val, const char **ptr)
 149 int stristart(const char *str, const char *val, const char **ptr)
 150 int qemu_strnlen(const char *s, int max_len)
 151
 152 There are also replacement character processing macros for isxyz and toxyz,
 153 so instead of e.g. isalnum you should use qemu_isalnum.
 154
 155 Because of the memory management rules, you must use g_strdup/g_strndup
 156 instead of plain strdup/strndup.
 157
 158 5. Printf-style functions
 159
 160 Whenever you add a new printf-style function, i.e., one with a format
 161 string argument and following "..." in its prototype, be sure to use
 162 gcc's printf attribute directive in the prototype.
 163
 164 This makes it so gcc's -Wformat and -Wformat-security options can do
 165 their jobs and cross-check format strings with the number and types
 166 of arguments.
 167
 168 6. C standard, implementation defined and undefined behaviors
 169
 170 C code in QEMU should be written to the C99 language specification. A copy
 171 of the final version of the C99 standard with corrigenda TC1, TC2, and TC3
 172 included, formatted as a draft, can be downloaded from:
 173  http://www.open-std.org/jtc1/sc22/WG14/www/docs/n1256.pdf
 174
 175 The C language specification defines regions of undefined behavior and
 176 implementation defined behavior (to give compiler authors enough leeway to
 177 produce better code).  In general, code in QEMU should follow the language
 178 specification and avoid both undefined and implementation defined
 179 constructs. ("It works fine on the gcc I tested it with" is not a valid
 180 argument...) However there are a few areas where we allow ourselves to
 181 assume certain behaviors because in practice all the platforms we care about
 182 behave in the same way and writing strictly conformant code would be
 183 painful. These are:
 184  * you may assume that integers are 2s complement representation
 185  * you may assume that right shift of a signed integer duplicates
 186    the sign bit (ie it is an arithmetic shift, not a logical shift)
 187
 188 In addition, QEMU assumes that the compiler does not use the latitude
 189 given in C99 and C11 to treat aspects of signed '<<' as undefined, as
 190 documented in the GNU Compiler Collection manual starting at version 4.0.
 191
 192 7. Error handling and reporting
 193
 194 7.1 Reporting errors to the human user
 195
 196 Do not use printf(), fprintf() or monitor_printf().  Instead, use
 197 error_report() or error_vreport() from error-report.h.  This ensures the
 198 error is reported in the right place (current monitor or stderr), and in
 199 a uniform format.
 200
 201 Use error_printf() & friends to print additional information.
 202
 203 error_report() prints the current location.  In certain common cases
 204 like command line parsing, the current location is tracked
 205 automatically.  To manipulate it manually, use the loc_*() from
 206 error-report.h.
 207
 208 7.2 Propagating errors
 209
 210 An error can't always be reported to the user right where it's detected,
 211 but often needs to be propagated up the call chain to a place that can
 212 handle it.  This can be done in various ways.
 213
 214 The most flexible one is Error objects.  See error.h for usage
 215 information.
 216
 217 Use the simplest suitable method to communicate success / failure to
 218 callers.  Stick to common methods: non-negative on success / -1 on
 219 error, non-negative / -errno, non-null / null, or Error objects.
 220
 221 Example: when a function returns a non-null pointer on success, and it
 222 can fail only in one way (as far as the caller is concerned), returning
 223 null on failure is just fine, and certainly simpler and a lot easier on
 224 the eyes than propagating an Error object through an Error ** parameter.
 225
 226 Example: when a function's callers need to report details on failure
 227 only the function really knows, use Error **, and set suitable errors.
 228
 229 Do not report an error to the user when you're also returning an error
 230 for somebody else to handle.  Leave the reporting to the place that
 231 consumes the error returned.
 232
 233 7.3 Handling errors
 234
 235 Calling exit() is fine when handling configuration errors during
 236 startup.  It's problematic during normal operation.  In particular,
 237 monitor commands should never exit().
 238
 239 Do not call exit() or abort() to handle an error that can be triggered
 240 by the guest (e.g., some unimplemented corner case in guest code
 241 translation or device emulation).  Guests should not be able to
 242 terminate QEMU.
 243
 244 Note that &error_fatal is just another way to exit(1), and &error_abort
 245 is just another way to abort().