vgaarbiter.rst 8.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191
  1. ===========
  2. VGA Arbiter
  3. ===========
  4. Graphic devices are accessed through ranges in I/O or memory space. While most
  5. modern devices allow relocation of such ranges, some "Legacy" VGA devices
  6. implemented on PCI will typically have the same "hard-decoded" addresses as
  7. they did on ISA. For more details see "PCI Bus Binding to IEEE Std 1275-1994
  8. Standard for Boot (Initialization Configuration) Firmware Revision 2.1"
  9. Section 7, Legacy Devices.
  10. The Resource Access Control (RAC) module inside the X server [0] existed for
  11. the legacy VGA arbitration task (besides other bus management tasks) when more
  12. than one legacy device co-exists on the same machine. But the problem happens
  13. when these devices are trying to be accessed by different userspace clients
  14. (e.g. two server in parallel). Their address assignments conflict. Moreover,
  15. ideally, being a userspace application, it is not the role of the X server to
  16. control bus resources. Therefore an arbitration scheme outside of the X server
  17. is needed to control the sharing of these resources. This document introduces
  18. the operation of the VGA arbiter implemented for the Linux kernel.
  19. vgaarb kernel/userspace ABI
  20. ---------------------------
  21. The vgaarb is a module of the Linux Kernel. When it is initially loaded, it
  22. scans all PCI devices and adds the VGA ones inside the arbitration. The
  23. arbiter then enables/disables the decoding on different devices of the VGA
  24. legacy instructions. Devices which do not want/need to use the arbiter may
  25. explicitly tell it by calling vga_set_legacy_decoding().
  26. The kernel exports a char device interface (/dev/vga_arbiter) to the clients,
  27. which has the following semantics:
  28. open
  29. Opens a user instance of the arbiter. By default, it's attached to the
  30. default VGA device of the system.
  31. close
  32. Close a user instance. Release locks made by the user
  33. read
  34. Return a string indicating the status of the target like:
  35. "<card_ID>,decodes=<io_state>,owns=<io_state>,locks=<io_state> (ic,mc)"
  36. An IO state string is of the form {io,mem,io+mem,none}, mc and
  37. ic are respectively mem and io lock counts (for debugging/
  38. diagnostic only). "decodes" indicate what the card currently
  39. decodes, "owns" indicates what is currently enabled on it, and
  40. "locks" indicates what is locked by this card. If the card is
  41. unplugged, we get "invalid" then for card_ID and an -ENODEV
  42. error is returned for any command until a new card is targeted.
  43. write
  44. Write a command to the arbiter. List of commands:
  45. target <card_ID>
  46. switch target to card <card_ID> (see below)
  47. lock <io_state>
  48. acquires locks on target ("none" is an invalid io_state)
  49. trylock <io_state>
  50. non-blocking acquire locks on target (returns EBUSY if
  51. unsuccessful)
  52. unlock <io_state>
  53. release locks on target
  54. unlock all
  55. release all locks on target held by this user (not implemented
  56. yet)
  57. decodes <io_state>
  58. set the legacy decoding attributes for the card
  59. poll
  60. event if something changes on any card (not just the target)
  61. card_ID is of the form "PCI:domain:bus:dev.fn". It can be set to "default"
  62. to go back to the system default card (TODO: not implemented yet). Currently,
  63. only PCI is supported as a prefix, but the userland API may support other bus
  64. types in the future, even if the current kernel implementation doesn't.
  65. Note about locks:
  66. The driver keeps track of which user has which locks on which card. It
  67. supports stacking, like the kernel one. This complexifies the implementation
  68. a bit, but makes the arbiter more tolerant to user space problems and able
  69. to properly cleanup in all cases when a process dies.
  70. Currently, a max of 16 cards can have locks simultaneously issued from
  71. user space for a given user (file descriptor instance) of the arbiter.
  72. In the case of devices hot-{un,}plugged, there is a hook - pci_notify() - to
  73. notify them being added/removed in the system and automatically added/removed
  74. in the arbiter.
  75. There is also an in-kernel API of the arbiter in case DRM, vgacon, or other
  76. drivers want to use it.
  77. In-kernel interface
  78. -------------------
  79. .. kernel-doc:: include/linux/vgaarb.h
  80. :internal:
  81. .. kernel-doc:: drivers/gpu/vga/vgaarb.c
  82. :export:
  83. libpciaccess
  84. ------------
  85. To use the vga arbiter char device it was implemented an API inside the
  86. libpciaccess library. One field was added to struct pci_device (each device
  87. on the system)::
  88. /* the type of resource decoded by the device */
  89. int vgaarb_rsrc;
  90. Besides it, in pci_system were added::
  91. int vgaarb_fd;
  92. int vga_count;
  93. struct pci_device *vga_target;
  94. struct pci_device *vga_default_dev;
  95. The vga_count is used to track how many cards are being arbitrated, so for
  96. instance, if there is only one card, then it can completely escape arbitration.
  97. These functions below acquire VGA resources for the given card and mark those
  98. resources as locked. If the resources requested are "normal" (and not legacy)
  99. resources, the arbiter will first check whether the card is doing legacy
  100. decoding for that type of resource. If yes, the lock is "converted" into a
  101. legacy resource lock. The arbiter will first look for all VGA cards that
  102. might conflict and disable their IOs and/or Memory access, including VGA
  103. forwarding on P2P bridges if necessary, so that the requested resources can
  104. be used. Then, the card is marked as locking these resources and the IO and/or
  105. Memory access is enabled on the card (including VGA forwarding on parent
  106. P2P bridges if any). In the case of vga_arb_lock(), the function will block
  107. if some conflicting card is already locking one of the required resources (or
  108. any resource on a different bus segment, since P2P bridges don't differentiate
  109. VGA memory and IO afaik). If the card already owns the resources, the function
  110. succeeds. vga_arb_trylock() will return (-EBUSY) instead of blocking. Nested
  111. calls are supported (a per-resource counter is maintained).
  112. Set the target device of this client. ::
  113. int pci_device_vgaarb_set_target (struct pci_device *dev);
  114. For instance, in x86 if two devices on the same bus want to lock different
  115. resources, both will succeed (lock). If devices are in different buses and
  116. trying to lock different resources, only the first who tried succeeds. ::
  117. int pci_device_vgaarb_lock (void);
  118. int pci_device_vgaarb_trylock (void);
  119. Unlock resources of device. ::
  120. int pci_device_vgaarb_unlock (void);
  121. Indicates to the arbiter if the card decodes legacy VGA IOs, legacy VGA
  122. Memory, both, or none. All cards default to both, the card driver (fbdev for
  123. example) should tell the arbiter if it has disabled legacy decoding, so the
  124. card can be left out of the arbitration process (and can be safe to take
  125. interrupts at any time. ::
  126. int pci_device_vgaarb_decodes (int new_vgaarb_rsrc);
  127. Connects to the arbiter device, allocates the struct ::
  128. int pci_device_vgaarb_init (void);
  129. Close the connection ::
  130. void pci_device_vgaarb_fini (void);
  131. xf86VGAArbiter (X server implementation)
  132. ----------------------------------------
  133. X server basically wraps all the functions that touch VGA registers somehow.
  134. References
  135. ----------
  136. Benjamin Herrenschmidt (IBM?) started this work when he discussed such design
  137. with the Xorg community in 2005 [1, 2]. In the end of 2007, Paulo Zanoni and
  138. Tiago Vignatti (both of C3SL/Federal University of Paraná) proceeded his work
  139. enhancing the kernel code to adapt as a kernel module and also did the
  140. implementation of the user space side [3]. Now (2009) Tiago Vignatti and Dave
  141. Airlie finally put this work in shape and queued to Jesse Barnes' PCI tree.
  142. 0) https://cgit.freedesktop.org/xorg/xserver/commit/?id=4b42448a2388d40f257774fbffdccaea87bd0347
  143. 1) https://lists.freedesktop.org/archives/xorg/2005-March/006663.html
  144. 2) https://lists.freedesktop.org/archives/xorg/2005-March/006745.html
  145. 3) https://lists.freedesktop.org/archives/xorg/2007-October/029507.html