1
2VGA Arbiter
3===========
4
5Graphic devices are accessed through ranges in I/O or memory space. While most
6modern devices allow relocation of such ranges, some "Legacy" VGA devices
7implemented on PCI will typically have the same "hard-decoded" addresses as
8they did on ISA. For more details see "PCI Bus Binding to IEEE Std 1275-1994
9Standard for Boot (Initialization Configuration) Firmware Revision 2.1"
10Section 7, Legacy Devices.
11
12The Resource Access Control (RAC) module inside the X server [0] existed for
13the legacy VGA arbitration task (besides other bus management tasks) when more
14than one legacy device co-exists on the same machine. But the problem happens
15when these devices are trying to be accessed by different userspace clients
16(e.g. two server in parallel). Their address assignments conflict. Moreover,
17ideally, being a userspace application, it is not the role of the X server to
18control bus resources. Therefore an arbitration scheme outside of the X server
19is needed to control the sharing of these resources. This document introduces
20the operation of the VGA arbiter implemented for the Linux kernel.
21
22----------------------------------------------------------------------------
23
24I.  Details and Theory of Operation
25        I.1 vgaarb
26        I.2 libpciaccess
27        I.3 xf86VGAArbiter (X server implementation)
28II. Credits
29III.References
30
31
32I. Details and Theory of Operation
33==================================
34
35I.1 vgaarb
36----------
37
38The vgaarb is a module of the Linux Kernel. When it is initially loaded, it
39scans all PCI devices and adds the VGA ones inside the arbitration. The
40arbiter then enables/disables the decoding on different devices of the VGA
41legacy instructions. Devices which do not want/need to use the arbiter may
42explicitly tell it by calling vga_set_legacy_decoding().
43
44The kernel exports a char device interface (/dev/vga_arbiter) to the clients,
45which has the following semantics:
46
47 open       : open user instance of the arbiter. By default, it's attached to
48              the default VGA device of the system.
49
50 close      : close user instance. Release locks made by the user
51
52 read       : return a string indicating the status of the target like:
53
54              "<card_ID>,decodes=<io_state>,owns=<io_state>,locks=<io_state> (ic,mc)"
55
56              An IO state string is of the form {io,mem,io+mem,none}, mc and
57              ic are respectively mem and io lock counts (for debugging/
58              diagnostic only). "decodes" indicate what the card currently
59              decodes, "owns" indicates what is currently enabled on it, and
60              "locks" indicates what is locked by this card. If the card is
61              unplugged, we get "invalid" then for card_ID and an -ENODEV
62              error is returned for any command until a new card is targeted.
63
64
65 write       : write a command to the arbiter. List of commands:
66
67  target <card_ID>   : switch target to card <card_ID> (see below)
68  lock <io_state>    : acquires locks on target ("none" is an invalid io_state)
69  trylock <io_state> : non-blocking acquire locks on target (returns EBUSY if
70                       unsuccessful)
71  unlock <io_state>  : release locks on target
72  unlock all         : release all locks on target held by this user (not
73                       implemented yet)
74  decodes <io_state> : set the legacy decoding attributes for the card
75
76  poll               : event if something changes on any card (not just the
77                       target)
78
79  card_ID is of the form "PCI:domain:bus:dev.fn". It can be set to "default"
80  to go back to the system default card (TODO: not implemented yet). Currently,
81  only PCI is supported as a prefix, but the userland API may support other bus
82  types in the future, even if the current kernel implementation doesn't.
83
84Note about locks:
85
86The driver keeps track of which user has which locks on which card. It
87supports stacking, like the kernel one. This complexifies the implementation
88a bit, but makes the arbiter more tolerant to user space problems and able
89to properly cleanup in all cases when a process dies.
90Currently, a max of 16 cards can have locks simultaneously issued from
91user space for a given user (file descriptor instance) of the arbiter.
92
93In the case of devices hot-{un,}plugged, there is a hook - pci_notify() - to
94notify them being added/removed in the system and automatically added/removed
95in the arbiter.
96
97There is also an in-kernel API of the arbiter in case DRM, vgacon, or other
98drivers want to use it.
99
100
101I.2 libpciaccess
102----------------
103
104To use the vga arbiter char device it was implemented an API inside the
105libpciaccess library. One field was added to struct pci_device (each device
106on the system):
107
108    /* the type of resource decoded by the device */
109    int vgaarb_rsrc;
110
111Besides it, in pci_system were added:
112
113    int vgaarb_fd;
114    int vga_count;
115    struct pci_device *vga_target;
116    struct pci_device *vga_default_dev;
117
118
119The vga_count is used to track how many cards are being arbitrated, so for
120instance, if there is only one card, then it can completely escape arbitration.
121
122
123These functions below acquire VGA resources for the given card and mark those
124resources as locked. If the resources requested are "normal" (and not legacy)
125resources, the arbiter will first check whether the card is doing legacy
126decoding for that type of resource. If yes, the lock is "converted" into a
127legacy resource lock. The arbiter will first look for all VGA cards that
128might conflict and disable their IOs and/or Memory access, including VGA
129forwarding on P2P bridges if necessary, so that the requested resources can
130be used. Then, the card is marked as locking these resources and the IO and/or
131Memory access is enabled on the card (including VGA forwarding on parent
132P2P bridges if any). In the case of vga_arb_lock(), the function will block
133if some conflicting card is already locking one of the required resources (or
134any resource on a different bus segment, since P2P bridges don't differentiate
135VGA memory and IO afaik). If the card already owns the resources, the function
136succeeds.  vga_arb_trylock() will return (-EBUSY) instead of blocking. Nested
137calls are supported (a per-resource counter is maintained).
138
139
140Set the target device of this client.
141    int  pci_device_vgaarb_set_target   (struct pci_device *dev);
142
143
144For instance, in x86 if two devices on the same bus want to lock different
145resources, both will succeed (lock). If devices are in different buses and
146trying to lock different resources, only the first who tried succeeds.
147    int  pci_device_vgaarb_lock         (void);
148    int  pci_device_vgaarb_trylock      (void);
149
150Unlock resources of device.
151    int  pci_device_vgaarb_unlock       (void);
152
153Indicates to the arbiter if the card decodes legacy VGA IOs, legacy VGA
154Memory, both, or none. All cards default to both, the card driver (fbdev for
155example) should tell the arbiter if it has disabled legacy decoding, so the
156card can be left out of the arbitration process (and can be safe to take
157interrupts at any time.
158    int  pci_device_vgaarb_decodes      (int new_vgaarb_rsrc);
159
160Connects to the arbiter device, allocates the struct
161    int  pci_device_vgaarb_init         (void);
162
163Close the connection
164    void pci_device_vgaarb_fini         (void);
165
166
167I.3 xf86VGAArbiter (X server implementation)
168--------------------------------------------
169
170(TODO)
171
172X server basically wraps all the functions that touch VGA registers somehow.
173
174
175II. Credits
176===========
177
178Benjamin Herrenschmidt (IBM?) started this work when he discussed such design
179with the Xorg community in 2005 [1, 2]. In the end of 2007, Paulo Zanoni and
180Tiago Vignatti (both of C3SL/Federal University of Paraná) proceeded his work
181enhancing the kernel code to adapt as a kernel module and also did the
182implementation of the user space side [3]. Now (2009) Tiago Vignatti and Dave
183Airlie finally put this work in shape and queued to Jesse Barnes' PCI tree.
184
185
186III. References
187==============
188
189[0] http://cgit.freedesktop.org/xorg/xserver/commit/?id=4b42448a2388d40f257774fbffdccaea87bd0347
190[1] http://lists.freedesktop.org/archives/xorg/2005-March/006663.html
191[2] http://lists.freedesktop.org/archives/xorg/2005-March/006745.html
192[3] http://lists.freedesktop.org/archives/xorg/2007-October/029507.html
193