cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

design-patterns.rst (3199B)


      1=============================
      2Device Driver Design Patterns
      3=============================
      4
      5This document describes a few common design patterns found in device drivers.
      6It is likely that subsystem maintainers will ask driver developers to
      7conform to these design patterns.
      8
      91. State Container
     102. container_of()
     11
     12
     131. State Container
     14~~~~~~~~~~~~~~~~~~
     15
     16While the kernel contains a few device drivers that assume that they will
     17only be probed() once on a certain system (singletons), it is custom to assume
     18that the device the driver binds to will appear in several instances. This
     19means that the probe() function and all callbacks need to be reentrant.
     20
     21The most common way to achieve this is to use the state container design
     22pattern. It usually has this form::
     23
     24  struct foo {
     25      spinlock_t lock; /* Example member */
     26      (...)
     27  };
     28
     29  static int foo_probe(...)
     30  {
     31      struct foo *foo;
     32
     33      foo = devm_kzalloc(dev, sizeof(*foo), GFP_KERNEL);
     34      if (!foo)
     35          return -ENOMEM;
     36      spin_lock_init(&foo->lock);
     37      (...)
     38  }
     39
     40This will create an instance of struct foo in memory every time probe() is
     41called. This is our state container for this instance of the device driver.
     42Of course it is then necessary to always pass this instance of the
     43state around to all functions that need access to the state and its members.
     44
     45For example, if the driver is registering an interrupt handler, you would
     46pass around a pointer to struct foo like this::
     47
     48  static irqreturn_t foo_handler(int irq, void *arg)
     49  {
     50      struct foo *foo = arg;
     51      (...)
     52  }
     53
     54  static int foo_probe(...)
     55  {
     56      struct foo *foo;
     57
     58      (...)
     59      ret = request_irq(irq, foo_handler, 0, "foo", foo);
     60  }
     61
     62This way you always get a pointer back to the correct instance of foo in
     63your interrupt handler.
     64
     65
     662. container_of()
     67~~~~~~~~~~~~~~~~~
     68
     69Continuing on the above example we add an offloaded work::
     70
     71  struct foo {
     72      spinlock_t lock;
     73      struct workqueue_struct *wq;
     74      struct work_struct offload;
     75      (...)
     76  };
     77
     78  static void foo_work(struct work_struct *work)
     79  {
     80      struct foo *foo = container_of(work, struct foo, offload);
     81
     82      (...)
     83  }
     84
     85  static irqreturn_t foo_handler(int irq, void *arg)
     86  {
     87      struct foo *foo = arg;
     88
     89      queue_work(foo->wq, &foo->offload);
     90      (...)
     91  }
     92
     93  static int foo_probe(...)
     94  {
     95      struct foo *foo;
     96
     97      foo->wq = create_singlethread_workqueue("foo-wq");
     98      INIT_WORK(&foo->offload, foo_work);
     99      (...)
    100  }
    101
    102The design pattern is the same for an hrtimer or something similar that will
    103return a single argument which is a pointer to a struct member in the
    104callback.
    105
    106container_of() is a macro defined in <linux/kernel.h>
    107
    108What container_of() does is to obtain a pointer to the containing struct from
    109a pointer to a member by a simple subtraction using the offsetof() macro from
    110standard C, which allows something similar to object oriented behaviours.
    111Notice that the contained member must not be a pointer, but an actual member
    112for this to work.
    113
    114We can see here that we avoid having global pointers to our struct foo *
    115instance this way, while still keeping the number of parameters passed to the
    116work function to a single pointer.