1Modelling a clock tree in QEMU 2============================== 3 4What are clocks? 5---------------- 6 7Clocks are QOM objects developed for the purpose of modelling the 8distribution of clocks in QEMU. 9 10They allow us to model the clock distribution of a platform and detect 11configuration errors in the clock tree such as badly configured PLL, clock 12source selection or disabled clock. 13 14The object is *Clock* and its QOM name is ``clock`` (in C code, the macro 15``TYPE_CLOCK``). 16 17Clocks are typically used with devices where they are used to model inputs 18and outputs. They are created in a similar way to GPIOs. Inputs and outputs 19of different devices can be connected together. 20 21In these cases a Clock object is a child of a Device object, but this 22is not a requirement. Clocks can be independent of devices. For 23example it is possible to create a clock outside of any device to 24model the main clock source of a machine. 25 26Here is an example of clocks:: 27 28 +---------+ +----------------------+ +--------------+ 29 | Clock 1 | | Device B | | Device C | 30 | | | +-------+ +-------+ | | +-------+ | 31 | |>>-+-->>|Clock 2| |Clock 3|>>--->>|Clock 6| | 32 +---------+ | | | (in) | | (out) | | | | (in) | | 33 | | +-------+ +-------+ | | +-------+ | 34 | | +-------+ | +--------------+ 35 | | |Clock 4|>> 36 | | | (out) | | +--------------+ 37 | | +-------+ | | Device D | 38 | | +-------+ | | +-------+ | 39 | | |Clock 5|>>--->>|Clock 7| | 40 | | | (out) | | | | (in) | | 41 | | +-------+ | | +-------+ | 42 | +----------------------+ | | 43 | | +-------+ | 44 +----------------------------->>|Clock 8| | 45 | | (in) | | 46 | +-------+ | 47 +--------------+ 48 49Clocks are defined in the ``include/hw/clock.h`` header and device 50related functions are defined in the ``include/hw/qdev-clock.h`` 51header. 52 53The clock state 54--------------- 55 56The state of a clock is its period; it is stored as an integer 57representing it in units of 2 :sup:`-32` ns. The special value of 0 is used to 58represent the clock being inactive or gated. The clocks do not model 59the signal itself (pin toggling) or other properties such as the duty 60cycle. 61 62All clocks contain this state: outputs as well as inputs. This allows 63the current period of a clock to be fetched at any time. When a clock 64is updated, the value is immediately propagated to all connected 65clocks in the tree. 66 67To ease interaction with clocks, helpers with a unit suffix are defined for 68every clock state setter or getter. The suffixes are: 69 70- ``_ns`` for handling periods in nanoseconds 71- ``_hz`` for handling frequencies in hertz 72 73The 0 period value is converted to 0 in hertz and vice versa. 0 always means 74that the clock is disabled. 75 76Adding a new clock 77------------------ 78 79Adding clocks to a device must be done during the init method of the Device 80instance. 81 82To add an input clock to a device, the function ``qdev_init_clock_in()`` 83must be used. It takes the name, a callback, an opaque parameter 84for the callback and a mask of events when the callback should be 85called (this will be explained in a following section). 86Output is simpler; only the name is required. Typically:: 87 88 qdev_init_clock_in(DEVICE(dev), "clk_in", clk_in_callback, dev, ClockUpdate); 89 qdev_init_clock_out(DEVICE(dev), "clk_out"); 90 91Both functions return the created Clock pointer, which should be saved in the 92device's state structure for further use. 93 94These objects will be automatically deleted by the QOM reference mechanism. 95 96Note that it is possible to create a static array describing clock inputs and 97outputs. The function ``qdev_init_clocks()`` must be called with the array as 98parameter to initialize the clocks: it has the same behaviour as calling the 99``qdev_init_clock_in/out()`` for each clock in the array. To ease the array 100construction, some macros are defined in ``include/hw/qdev-clock.h``. 101As an example, the following creates 2 clocks to a device: one input and one 102output. 103 104.. code-block:: c 105 106 /* device structure containing pointers to the clock objects */ 107 typedef struct MyDeviceState { 108 DeviceState parent_obj; 109 Clock *clk_in; 110 Clock *clk_out; 111 } MyDeviceState; 112 113 /* 114 * callback for the input clock (see "Callback on input clock 115 * change" section below for more information). 116 */ 117 static void clk_in_callback(void *opaque, ClockEvent event); 118 119 /* 120 * static array describing clocks: 121 * + a clock input named "clk_in", whose pointer is stored in 122 * the clk_in field of a MyDeviceState structure with callback 123 * clk_in_callback. 124 * + a clock output named "clk_out" whose pointer is stored in 125 * the clk_out field of a MyDeviceState structure. 126 */ 127 static const ClockPortInitArray mydev_clocks = { 128 QDEV_CLOCK_IN(MyDeviceState, clk_in, clk_in_callback, ClockUpdate), 129 QDEV_CLOCK_OUT(MyDeviceState, clk_out), 130 QDEV_CLOCK_END 131 }; 132 133 /* device initialization function */ 134 static void mydev_init(Object *obj) 135 { 136 /* cast to MyDeviceState */ 137 MyDeviceState *mydev = MYDEVICE(obj); 138 /* create and fill the pointer fields in the MyDeviceState */ 139 qdev_init_clocks(mydev, mydev_clocks); 140 [...] 141 } 142 143An alternative way to create a clock is to simply call 144``object_new(TYPE_CLOCK)``. In that case the clock will neither be an 145input nor an output of a device. After the whole QOM hierarchy of the 146clock has been set ``clock_setup_canonical_path()`` should be called. 147 148At creation, the period of the clock is 0: the clock is disabled. You can 149change it using ``clock_set_ns()`` or ``clock_set_hz()``. 150 151Note that if you are creating a clock with a fixed period which will never 152change (for example the main clock source of a board), then you'll have 153nothing else to do. This value will be propagated to other clocks when 154connecting the clocks together and devices will fetch the right value during 155the first reset. 156 157Clock callbacks 158--------------- 159 160You can give a clock a callback function in several ways: 161 162 * by passing it as an argument to ``qdev_init_clock_in()`` 163 * as an argument to the ``QDEV_CLOCK_IN()`` macro initializing an 164 array to be passed to ``qdev_init_clocks()`` 165 * by directly calling the ``clock_set_callback()`` function 166 167The callback function must be of this type: 168 169.. code-block:: c 170 171 typedef void ClockCallback(void *opaque, ClockEvent event); 172 173The ``opaque`` argument is the pointer passed to ``qdev_init_clock_in()`` 174or ``clock_set_callback()``; for ``qdev_init_clocks()`` it is the 175``dev`` device pointer. 176 177The ``event`` argument specifies why the callback has been called. 178When you register the callback you specify a mask of ClockEvent values 179that you are interested in. The callback will only be called for those 180events. 181 182The events currently supported are: 183 184 * ``ClockPreUpdate`` : called when the input clock's period is about to 185 update. This is useful if the device needs to do some action for 186 which it needs to know the old value of the clock period. During 187 this callback, Clock API functions like ``clock_get()`` or 188 ``clock_ticks_to_ns()`` will use the old period. 189 * ``ClockUpdate`` : called after the input clock's period has changed. 190 During this callback, Clock API functions like ``clock_ticks_to_ns()`` 191 will use the new period. 192 193Note that a clock only has one callback: it is not possible to register 194different functions for different events. You must register a single 195callback which listens for all of the events you are interested in, 196and use the ``event`` argument to identify which event has happened. 197 198Retrieving clocks from a device 199------------------------------- 200 201``qdev_get_clock_in()`` and ``dev_get_clock_out()`` are available to 202get the clock inputs or outputs of a device. For example: 203 204.. code-block:: c 205 206 Clock *clk = qdev_get_clock_in(DEVICE(mydev), "clk_in"); 207 208or: 209 210.. code-block:: c 211 212 Clock *clk = qdev_get_clock_out(DEVICE(mydev), "clk_out"); 213 214Connecting two clocks together 215------------------------------ 216 217To connect two clocks together, use the ``clock_set_source()`` function. 218Given two clocks ``clk1``, and ``clk2``, ``clock_set_source(clk2, clk1);`` 219configures ``clk2`` to follow the ``clk1`` period changes. Every time ``clk1`` 220is updated, ``clk2`` will be updated too. 221 222When connecting clock between devices, prefer using the 223``qdev_connect_clock_in()`` function to set the source of an input 224device clock. For example, to connect the input clock ``clk2`` of 225``devB`` to the output clock ``clk1`` of ``devA``, do: 226 227.. code-block:: c 228 229 qdev_connect_clock_in(devB, "clk2", qdev_get_clock_out(devA, "clk1")) 230 231We used ``qdev_get_clock_out()`` above, but any clock can drive an 232input clock, even another input clock. The following diagram shows 233some examples of connections. Note also that a clock can drive several 234other clocks. 235 236:: 237 238 +------------+ +--------------------------------------------------+ 239 | Device A | | Device B | 240 | | | +---------------------+ | 241 | | | | Device C | | 242 | +-------+ | | +-------+ | +-------+ +-------+ | +-------+ | 243 | |Clock 1|>>-->>|Clock 2|>>+-->>|Clock 3| |Clock 5|>>>>|Clock 6|>> 244 | | (out) | | | | (in) | | | | (in) | | (out) | | | (out) | | 245 | +-------+ | | +-------+ | | +-------+ +-------+ | +-------+ | 246 +------------+ | | +---------------------+ | 247 | | | 248 | | +--------------+ | 249 | | | Device D | | 250 | | | +-------+ | | 251 | +-->>|Clock 4| | | 252 | | | (in) | | | 253 | | +-------+ | | 254 | +--------------+ | 255 +--------------------------------------------------+ 256 257In the above example, when *Clock 1* is updated by *Device A*, three 258clocks get the new clock period value: *Clock 2*, *Clock 3* and *Clock 4*. 259 260It is not possible to disconnect a clock or to change the clock connection 261after it is connected. 262 263Unconnected input clocks 264------------------------ 265 266A newly created input clock is disabled (period of 0). This means the 267clock will be considered as disabled until the period is updated. If 268the clock remains unconnected it will always keep its initial value 269of 0. If this is not the desired behaviour, ``clock_set()``, 270``clock_set_ns()`` or ``clock_set_hz()`` should be called on the Clock 271object during device instance init. For example: 272 273.. code-block:: c 274 275 clk = qdev_init_clock_in(DEVICE(dev), "clk-in", clk_in_callback, 276 dev, ClockUpdate); 277 /* set initial value to 10ns / 100MHz */ 278 clock_set_ns(clk, 10); 279 280To enforce that the clock is wired up by the board code, you can 281call ``clock_has_source()`` in your device's realize method: 282 283.. code-block:: c 284 285 if (!clock_has_source(s->clk)) { 286 error_setg(errp, "MyDevice: clk input must be connected"); 287 return; 288 } 289 290Note that this only checks that the clock has been wired up; it is 291still possible that the output clock connected to it is disabled 292or has not yet been configured, in which case the period will be 293zero. You should use the clock callback to find out when the clock 294period changes. 295 296Fetching clock frequency/period 297------------------------------- 298 299To get the current state of a clock, use the functions ``clock_get()`` 300or ``clock_get_hz()``. 301 302``clock_get()`` returns the period of the clock in its fully precise 303internal representation, as an unsigned 64-bit integer in units of 3042^-32 nanoseconds. (For many purposes ``clock_ticks_to_ns()`` will 305be more convenient; see the section below on expiry deadlines.) 306 307``clock_get_hz()`` returns the frequency of the clock, rounded to the 308next lowest integer. This implies some inaccuracy due to the rounding, 309so be cautious about using it in calculations. 310 311It is also possible to register a callback on clock frequency changes. 312Here is an example, which assumes that ``clock_callback`` has been 313specified as the callback for the ``ClockUpdate`` event: 314 315.. code-block:: c 316 317 void clock_callback(void *opaque, ClockEvent event) { 318 MyDeviceState *s = (MyDeviceState *) opaque; 319 /* 320 * 'opaque' is the argument passed to qdev_init_clock_in(); 321 * usually this will be the device state pointer. 322 */ 323 324 /* do something with the new period */ 325 fprintf(stdout, "device new period is %" PRIu64 "* 2^-32 ns\n", 326 clock_get(dev->my_clk_input)); 327 } 328 329If you are only interested in the frequency for displaying it to 330humans (for instance in debugging), use ``clock_display_freq()``, 331which returns a prettified string-representation, e.g. "33.3 MHz". 332The caller must free the string with g_free() after use. 333 334Calculating expiry deadlines 335---------------------------- 336 337A commonly required operation for a clock is to calculate how long 338it will take for the clock to tick N times; this can then be used 339to set a timer expiry deadline. Use the function ``clock_ticks_to_ns()``, 340which takes an unsigned 64-bit count of ticks and returns the length 341of time in nanoseconds required for the clock to tick that many times. 342 343It is important not to try to calculate expiry deadlines using a 344shortcut like multiplying a "period of clock in nanoseconds" value 345by the tick count, because clocks can have periods which are not a 346whole number of nanoseconds, and the accumulated error in the 347multiplication can be significant. 348 349For a clock with a very long period and a large number of ticks, 350the result of this function could in theory be too large to fit in 351a 64-bit value. To avoid overflow in this case, ``clock_ticks_to_ns()`` 352saturates the result to INT64_MAX (because this is the largest valid 353input to the QEMUTimer APIs). Since INT64_MAX nanoseconds is almost 354300 years, anything with an expiry later than that is in the "will 355never happen" category. Callers of ``clock_ticks_to_ns()`` should 356therefore generally not special-case the possibility of a saturated 357result but just allow the timer to be set to that far-future value. 358(If you are performing further calculations on the returned value 359rather than simply passing it to a QEMUTimer function like 360``timer_mod_ns()`` then you should be careful to avoid overflow 361in those calculations, of course.) 362 363Obtaining tick counts 364--------------------- 365 366For calculations where you need to know the number of ticks in 367a given duration, use ``clock_ns_to_ticks()``. This function handles 368possible non-whole-number-of-nanoseconds periods and avoids 369potential rounding errors. It will return '0' if the clock is stopped 370(i.e. it has period zero). If the inputs imply a tick count that 371overflows a 64-bit value (a very long duration for a clock with a 372very short period) the output value is truncated, so effectively 373the 64-bit output wraps around. 374 375Changing a clock period 376----------------------- 377 378A device can change its outputs using the ``clock_update()``, 379``clock_update_ns()`` or ``clock_update_hz()`` function. It will trigger 380updates on every connected input. 381 382For example, let's say that we have an output clock *clkout* and we 383have a pointer to it in the device state because we did the following 384in init phase: 385 386.. code-block:: c 387 388 dev->clkout = qdev_init_clock_out(DEVICE(dev), "clkout"); 389 390Then at any time (apart from the cases listed below), it is possible to 391change the clock value by doing: 392 393.. code-block:: c 394 395 clock_update_hz(dev->clkout, 1000 * 1000 * 1000); /* 1GHz */ 396 397Because updating a clock may trigger any side effects through 398connected clocks and their callbacks, this operation must be done 399while holding the qemu io lock. 400 401For the same reason, one can update clocks only when it is allowed to have 402side effects on other objects. In consequence, it is forbidden: 403 404* during migration, 405* and in the enter phase of reset. 406 407Note that calling ``clock_update[_ns|_hz]()`` is equivalent to calling 408``clock_set[_ns|_hz]()`` (with the same arguments) then 409``clock_propagate()`` on the clock. Thus, setting the clock value can 410be separated from triggering the side-effects. This is often required 411to factorize code to handle reset and migration in devices. 412 413Aliasing clocks 414--------------- 415 416Sometimes, one needs to forward, or inherit, a clock from another 417device. Typically, when doing device composition, a device might 418expose a sub-device's clock without interfering with it. The function 419``qdev_alias_clock()`` can be used to achieve this behaviour. Note 420that it is possible to expose the clock under a different name. 421``qdev_alias_clock()`` works for both input and output clocks. 422 423For example, if device B is a child of device A, 424``device_a_instance_init()`` may do something like this: 425 426.. code-block:: c 427 428 void device_a_instance_init(Object *obj) 429 { 430 AState *A = DEVICE_A(obj); 431 BState *B; 432 /* create object B as child of A */ 433 [...] 434 qdev_alias_clock(B, "clk", A, "b_clk"); 435 /* 436 * Now A has a clock "b_clk" which is an alias to 437 * the clock "clk" of its child B. 438 */ 439 } 440 441This function does not return any clock object. The new clock has the 442same direction (input or output) as the original one. This function 443only adds a link to the existing clock. In the above example, object B 444remains the only object allowed to use the clock and device A must not 445try to change the clock period or set a callback to the clock. This 446diagram describes the example with an input clock:: 447 448 +--------------------------+ 449 | Device A | 450 | +--------------+ | 451 | | Device B | | 452 | | +-------+ | | 453 >>"b_clk">>>| "clk" | | | 454 | (in) | | (in) | | | 455 | | +-------+ | | 456 | +--------------+ | 457 +--------------------------+ 458 459Migration 460--------- 461 462Clock state is not migrated automatically. Every device must handle its 463clock migration. Alias clocks must not be migrated. 464 465To ensure clock states are restored correctly during migration, there 466are two solutions. 467 468Clock states can be migrated by adding an entry into the device 469vmstate description. You should use the ``VMSTATE_CLOCK`` macro for this. 470This is typically used to migrate an input clock state. For example: 471 472.. code-block:: c 473 474 MyDeviceState { 475 DeviceState parent_obj; 476 [...] /* some fields */ 477 Clock *clk; 478 }; 479 480 VMStateDescription my_device_vmstate = { 481 .name = "my_device", 482 .fields = (VMStateField[]) { 483 [...], /* other migrated fields */ 484 VMSTATE_CLOCK(clk, MyDeviceState), 485 VMSTATE_END_OF_LIST() 486 } 487 }; 488 489The second solution is to restore the clock state using information already 490at our disposal. This can be used to restore output clock states using the 491device state. The functions ``clock_set[_ns|_hz]()`` can be used during the 492``post_load()`` migration callback. 493 494When adding clock support to an existing device, if you care about 495migration compatibility you will need to be careful, as simply adding 496a ``VMSTATE_CLOCK()`` line will break compatibility. Instead, you can 497put the ``VMSTATE_CLOCK()`` line into a vmstate subsection with a 498suitable ``needed`` function, and use ``clock_set()`` in a 499``pre_load()`` function to set the default value that will be used if 500the source virtual machine in the migration does not send the clock 501state. 502 503Care should be taken not to use ``clock_update[_ns|_hz]()`` or 504``clock_propagate()`` during the whole migration procedure because it 505will trigger side effects to other devices in an unknown state. 506