TEXT   59

devres.txt

Guest on 1st August 2021 06:08:05 AM

  1. Devres - Managed Device Resource
  2. ================================
  3.  
  4. Tejun Heo       <teheo@suse.de>
  5.  
  6. First draft     10 January
  7.  
  8.  
  9. 1. Intro                        : Huh? Devres?
  10. 2. Devres                       : Devres in a nutshell
  11. 3. Devres Group                 : Group devres'es and release them together
  12. 4. Details                      : Life time rules, calling context, ...
  13. 5. Overhead                     : How much do we have to pay for this?
  14. 6. List of managed interfaces   : Currently implemented managed interfaces
  15.  
  16.  
  17.   1. Intro
  18.   --------
  19.  
  20. devres came up while trying to convert libata to use iomap.  Each
  21. iomapped address should be kept and unmapped on driver detach.  For
  22. example, a plain SFF ATA controller (that is, good old PCI IDE) in
  23. native mode makes use of 5 PCI BARs and all of them should be
  24. maintained.
  25.  
  26. As with many other device drivers, libata low level drivers have
  27. sufficient bugs in ->remove and ->probe failure path.  Well, yes,
  28. that's probably because libata low level driver developers are lazy
  29. bunch, but aren't all low level driver developers?  After spending a
  30. day fiddling with braindamaged hardware with no document or
  31. braindamaged document, if it's finally working, well, it's working.
  32.  
  33. For one reason or another, low level drivers don't receive as much
  34. attention or testing as core code, and bugs on driver detach or
  35. initialization failure don't happen often enough to be noticeable.
  36. Init failure path is worse because it's much less travelled while
  37. needs to handle multiple entry points.
  38.  
  39. So, many low level drivers end up leaking resources on driver detach
  40. and having half broken failure path implementation in ->probe() which
  41. would leak resources or even cause oops when failure occurs.  iomap
  42. adds more to this mix.  So do msi and msix.
  43.  
  44.  
  45.   2. Devres
  46.   ---------
  47.  
  48. devres is basically linked list of arbitrarily sized memory areas
  49. associated with a struct device.  Each devres entry is associated with
  50. a release function.  A devres can be released in several ways.  No
  51. matter what, all devres entries are released on driver detach.  On
  52. release, the associated release function is invoked and then the
  53. devres entry is freed.
  54.  
  55. Managed interface is created for resources commonly used by device
  56. drivers using devres.  For example, coherent DMA memory is acquired
  57. using dma_alloc_coherent().  The managed version is called
  58. dmam_alloc_coherent().  It is identical to dma_alloc_coherent() except
  59. for the DMA memory allocated using it is managed and will be
  60. automatically released on driver detach.  Implementation looks like
  61. the following.
  62.  
  63.   struct dma_devres {
  64.         size_t          size;
  65.         void            *vaddr;
  66.         dma_addr_t      dma_handle;
  67.   };
  68.  
  69.   static void dmam_coherent_release(struct device *dev, void *res)
  70.   {
  71.         struct dma_devres *this = res;
  72.  
  73.         dma_free_coherent(dev, this->size, this->vaddr, this->dma_handle);
  74.   }
  75.  
  76.   dmam_alloc_coherent(dev, size, dma_handle, gfp)
  77.   {
  78.         struct dma_devres *dr;
  79.         void *vaddr;
  80.  
  81.         dr = devres_alloc(dmam_coherent_release, sizeof(*dr), gfp);
  82.         ...
  83.  
  84.         /* alloc DMA memory as usual */
  85.         vaddr = dma_alloc_coherent(...);
  86.         ...
  87.  
  88.         /* record size, vaddr, dma_handle in dr */
  89.         dr->vaddr = vaddr;
  90.         ...
  91.  
  92.         devres_add(dev, dr);
  93.  
  94.         return vaddr;
  95.   }
  96.  
  97. If a driver uses dmam_alloc_coherent(), the area is guaranteed to be
  98. freed whether initialization fails half-way or the device gets
  99. detached.  If most resources are acquired using managed interface, a
  100. driver can have much simpler init and exit code.  Init path basically
  101. looks like the following.
  102.  
  103.   my_init_one()
  104.   {
  105.         struct mydev *d;
  106.  
  107.         d = devm_kzalloc(dev, sizeof(*d), GFP_KERNEL);
  108.         if (!d)
  109.                 return -ENOMEM;
  110.  
  111.         d->ring = dmam_alloc_coherent(...);
  112.         if (!d->ring)
  113.                 return -ENOMEM;
  114.  
  115.         if (check something)
  116.                 return -EINVAL;
  117.         ...
  118.  
  119.         return register_to_upper_layer(d);
  120.   }
  121.  
  122. And exit path,
  123.  
  124.   my_remove_one()
  125.   {
  126.         unregister_from_upper_layer(d);
  127.         shutdown_my_hardware();
  128.   }
  129.  
  130. As shown above, low level drivers can be simplified a lot by using
  131. devres.  Complexity is shifted from less maintained low level drivers
  132. to better maintained higher layer.  Also, as init failure path is
  133. shared with exit path, both can get more testing.
  134.  
  135. Note though that when converting current calls or assignments to
  136. managed devm_* versions it is up to you to check if internal operations
  137. like allocating memory, have failed. Managed resources pertains to the
  138. freeing of these resources *only* - all other checks needed are still
  139. on you. In some cases this may mean introducing checks that were not
  140. necessary before moving to the managed devm_* calls.
  141.  
  142.  
  143.   3. Devres group
  144.   ---------------
  145.  
  146. Devres entries can be grouped using devres group.  When a group is
  147. released, all contained normal devres entries and properly nested
  148. groups are released.  One usage is to rollback series of acquired
  149. resources on failure.  For example,
  150.  
  151.   if (!devres_open_group(dev, NULL, GFP_KERNEL))
  152.         return -ENOMEM;
  153.  
  154.   acquire A;
  155.   if (failed)
  156.         goto err;
  157.  
  158.   acquire B;
  159.   if (failed)
  160.         goto err;
  161.   ...
  162.  
  163.   devres_remove_group(dev, NULL);
  164.   return 0;
  165.  
  166.  err:
  167.   devres_release_group(dev, NULL);
  168.   return err_code;
  169.  
  170. As resource acquisition failure usually means probe failure, constructs
  171. like above are usually useful in midlayer driver (e.g. libata core
  172. layer) where interface function shouldn't have side effect on failure.
  173. For LLDs, just returning error code suffices in most cases.
  174.  
  175. Each group is identified by void *id.  It can either be explicitly
  176. specified by @id argument to devres_open_group() or automatically
  177. created by passing NULL as @id as in the above example.  In both
  178. cases, devres_open_group() returns the group's id.  The returned id
  179. can be passed to other devres functions to select the target group.
  180. If NULL is given to those functions, the latest open group is
  181. selected.
  182.  
  183. For example, you can do something like the following.
  184.  
  185.   int my_midlayer_create_something()
  186.   {
  187.         if (!devres_open_group(dev, my_midlayer_create_something, GFP_KERNEL))
  188.                 return -ENOMEM;
  189.  
  190.         ...
  191.  
  192.         devres_close_group(dev, my_midlayer_create_something);
  193.         return 0;
  194.   }
  195.  
  196.   void my_midlayer_destroy_something()
  197.   {
  198.         devres_release_group(dev, my_midlayer_create_something);
  199.   }
  200.  
  201.  
  202.   4. Details
  203.   ----------
  204.  
  205. Lifetime of a devres entry begins on devres allocation and finishes
  206. when it is released or destroyed (removed and freed) - no reference
  207. counting.
  208.  
  209. devres core guarantees atomicity to all basic devres operations and
  210. has support for single-instance devres types (atomic
  211. lookup-and-add-if-not-found).  Other than that, synchronizing
  212. concurrent accesses to allocated devres data is caller's
  213. responsibility.  This is usually non-issue because bus ops and
  214. resource allocations already do the job.
  215.  
  216. For an example of single-instance devres type, read pcim_iomap_table()
  217. in lib/devres.c.
  218.  
  219. All devres interface functions can be called without context if the
  220. right gfp mask is given.
  221.  
  222.  
  223.   5. Overhead
  224.   -----------
  225.  
  226. Each devres bookkeeping info is allocated together with requested data
  227. area.  With debug option turned off, bookkeeping info occupies 16
  228. bytes on 32bit machines and 24 bytes on 64bit (three pointers rounded
  229. up to ull alignment).  If singly linked list is used, it can be
  230. reduced to two pointers (8 bytes on 32bit, 16 bytes on 64bit).
  231.  
  232. Each devres group occupies 8 pointers.  It can be reduced to 6 if
  233. singly linked list is used.
  234.  
  235. Memory space overhead on ahci controller with two ports is between 300
  236. and 400 bytes on 32bit machine after naive conversion (we can
  237. certainly invest a bit more effort into libata core layer).
  238.  
  239.  
  240.   6. List of managed interfaces
  241.   -----------------------------
  242.  
  243. CLOCK
  244.   devm_clk_get()
  245.   devm_clk_get_optional()
  246.   devm_clk_put()
  247.   devm_clk_hw_register()
  248.   devm_of_clk_add_hw_provider()
  249.   devm_clk_hw_register_clkdev()
  250.  
  251. DMA
  252.   dmaenginem_async_device_register()
  253.   dmam_alloc_coherent()
  254.   dmam_alloc_attrs()
  255.   dmam_free_coherent()
  256.   dmam_pool_create()
  257.   dmam_pool_destroy()
  258.  
  259. DRM
  260.   devm_drm_dev_init()
  261.  
  262. GPIO
  263.   devm_gpiod_get()
  264.   devm_gpiod_get_index()
  265.   devm_gpiod_get_index_optional()
  266.   devm_gpiod_get_optional()
  267.   devm_gpiod_put()
  268.   devm_gpiod_unhinge()
  269.   devm_gpiochip_add_data()
  270.   devm_gpio_request()
  271.   devm_gpio_request_one()
  272.   devm_gpio_free()
  273.  
  274. I2C
  275.   devm_i2c_new_dummy_device()
  276.  
  277. IIO
  278.   devm_iio_device_alloc()
  279.   devm_iio_device_free()
  280.   devm_iio_device_register()
  281.   devm_iio_device_unregister()
  282.   devm_iio_kfifo_allocate()
  283.   devm_iio_kfifo_free()
  284.   devm_iio_triggered_buffer_setup()
  285.   devm_iio_triggered_buffer_cleanup()
  286.   devm_iio_trigger_alloc()
  287.   devm_iio_trigger_free()
  288.   devm_iio_trigger_register()
  289.   devm_iio_trigger_unregister()
  290.   devm_iio_channel_get()
  291.   devm_iio_channel_release()
  292.   devm_iio_channel_get_all()
  293.   devm_iio_channel_release_all()
  294.  
  295. INPUT
  296.   devm_input_allocate_device()
  297.  
  298. IO region
  299.   devm_release_mem_region()
  300.   devm_release_region()
  301.   devm_release_resource()
  302.   devm_request_mem_region()
  303.   devm_request_region()
  304.   devm_request_resource()
  305.  
  306. IOMAP
  307.   devm_ioport_map()
  308.   devm_ioport_unmap()
  309.   devm_ioremap()
  310.   devm_ioremap_nocache()
  311.   devm_ioremap_wc()
  312.   devm_ioremap_resource() : checks resource, requests memory region, ioremaps
  313.   devm_iounmap()
  314.   pcim_iomap()
  315.   pcim_iomap_regions()  : do request_region() and iomap() on multiple BARs
  316.   pcim_iomap_table()    : array of mapped addresses indexed by BAR
  317.   pcim_iounmap()
  318.  
  319. IRQ
  320.   devm_free_irq()
  321.   devm_request_any_context_irq()
  322.   devm_request_irq()
  323.   devm_request_threaded_irq()
  324.   devm_irq_alloc_descs()
  325.   devm_irq_alloc_desc()
  326.   devm_irq_alloc_desc_at()
  327.   devm_irq_alloc_desc_from()
  328.   devm_irq_alloc_descs_from()
  329.   devm_irq_alloc_generic_chip()
  330.   devm_irq_setup_generic_chip()
  331.   devm_irq_sim_init()
  332.  
  333. LED
  334.   devm_led_classdev_register()
  335.   devm_led_classdev_unregister()
  336.  
  337. MDIO
  338.   devm_mdiobus_alloc()
  339.   devm_mdiobus_alloc_size()
  340.   devm_mdiobus_free()
  341.  
  342. MEM
  343.   devm_free_pages()
  344.   devm_get_free_pages()
  345.   devm_kasprintf()
  346.   devm_kcalloc()
  347.   devm_kfree()
  348.   devm_kmalloc()
  349.   devm_kmalloc_array()
  350.   devm_kmemdup()
  351.   devm_kstrdup()
  352.   devm_kvasprintf()
  353.   devm_kzalloc()
  354.  
  355. MFD
  356.   devm_mfd_add_devices()
  357.  
  358. MUX
  359.   devm_mux_chip_alloc()
  360.   devm_mux_chip_register()
  361.   devm_mux_control_get()
  362.  
  363. PER-CPU MEM
  364.   devm_alloc_percpu()
  365.   devm_free_percpu()
  366.  
  367. PCI
  368.   devm_pci_alloc_host_bridge()  : managed PCI host bridge allocation
  369.   devm_pci_remap_cfgspace()     : ioremap PCI configuration space
  370.   devm_pci_remap_cfg_resource() : ioremap PCI configuration space resource
  371.   pcim_enable_device()          : after success, all PCI ops become managed
  372.   pcim_pin_device()             : keep PCI device enabled after release
  373.  
  374. PHY
  375.   devm_usb_get_phy()
  376.   devm_usb_put_phy()
  377.  
  378. PINCTRL
  379.   devm_pinctrl_get()
  380.   devm_pinctrl_put()
  381.   devm_pinctrl_register()
  382.   devm_pinctrl_unregister()
  383.  
  384. POWER
  385.   devm_reboot_mode_register()
  386.   devm_reboot_mode_unregister()
  387.  
  388. PWM
  389.   devm_pwm_get()
  390.   devm_pwm_put()
  391.  
  392. REGULATOR
  393.   devm_regulator_bulk_get()
  394.   devm_regulator_get()
  395.   devm_regulator_put()
  396.   devm_regulator_register()
  397.  
  398. RESET
  399.   devm_reset_control_get()
  400.   devm_reset_controller_register()
  401.  
  402. SERDEV
  403.   devm_serdev_device_open()
  404.  
  405. SLAVE DMA ENGINE
  406.   devm_acpi_dma_controller_register()
  407.  
  408. SPI
  409.   devm_spi_register_master()
  410.  
  411. WATCHDOG
  412.   devm_watchdog_register_device()

Raw Paste


Login or Register to edit or fork this paste. It's free.