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 an userspace application, it is not the role of the the X
18server to control bus resources. Therefore an arbitration scheme outside of
19the X server is needed to control the sharing of these resources. This
20document introduces the operation of the VGA arbiter implemented for Linux
21kernel.
22
23----------------------------------------------------------------------------
24
25I.  Details and Theory of Operation
26        I.1 vgaarb
27        I.2 libpciaccess
28        I.3 xf86VGAArbiter (X server implementation)
29II. Credits
30III.References
31
32
33I. Details and Theory of Operation
34==================================
35
36I.1 vgaarb
37----------
38
39The vgaarb is a module of the Linux Kernel. When it is initially loaded, it
40scans all PCI devices and adds the VGA ones inside the arbitration. The
41arbiter then enables/disables the decoding on different devices of the VGA
42legacy instructions. Device which do not want/need to use the arbiter may
43explicitly tell it by calling vga_set_legacy_decoding().
44
45The kernel exports a char device interface (/dev/vga_arbiter) to the clients,
46which has the following semantics:
47
48 open       : open user instance of the arbiter. By default, it's attached to
49              the default VGA device of the system.
50
51 close      : close user instance. Release locks made by the user
52
53 read       : return a string indicating the status of the target like:
54
55              "<card_ID>,decodes=<io_state>,owns=<io_state>,locks=<io_state> (ic,mc)"
56
57              An IO state string is of the form {io,mem,io+mem,none}, mc and
58              ic are respectively mem and io lock counts (for debugging/
59              diagnostic only). "decodes" indicate what the card currently
60              decodes, "owns" indicates what is currently enabled on it, and
61              "locks" indicates what is locked by this card. If the card is
62              unplugged, we get "invalid" then for card_ID and an -ENODEV
63              error is returned for any command until a new card is targeted.
64
65
66 write       : write a command to the arbiter. List of commands:
67
68  target <card_ID>   : switch target to card <card_ID> (see below)
69  lock <io_state>    : acquires locks on target ("none" is an invalid io_state)
70  trylock <io_state> : non-blocking acquire locks on target (returns EBUSY if
71                       unsuccessful)
72  unlock <io_state>  : release locks on target
73  unlock all         : release all locks on target held by this user (not
74                       implemented yet)
75  decodes <io_state> : set the legacy decoding attributes for the card
76
77  poll               : event if something changes on any card (not just the
78                       target)
79
80  card_ID is of the form "PCI:domain:bus:dev.fn". It can be set to "default"
81  to go back to the system default card (TODO: not implemented yet). Currently,
82  only PCI is supported as a prefix, but the userland API may support other bus
83  types in the future, even if the current kernel implementation doesn't.
84
85Note about locks:
86
87The driver keeps track of which user has which locks on which card. It
88supports stacking, like the kernel one. This complexifies the implementation
89a bit, but makes the arbiter more tolerant to user space problems and able
90to properly cleanup in all cases when a process dies.
91Currently, a max of 16 cards can have locks simultaneously issued from
92user space for a given user (file descriptor instance) of the arbiter.
93
94In the case of devices hot-{un,}plugged, there is a hook - pci_notify() - to
95notify them being added/removed in the system and automatically added/removed
96in the arbiter.
97
98There's also a in-kernel API of the arbiter in the case of DRM, vgacon and
99others which may use the arbiter.
100
101
102I.2 libpciaccess
103----------------
104
105To use the vga arbiter char device it was implemented an API inside the
106libpciaccess library. One field was added to struct pci_device (each device
107on the system):
108
109    /* the type of resource decoded by the device */
110    int vgaarb_rsrc;
111
112Besides it, in pci_system were added:
113
114    int vgaarb_fd;
115    int vga_count;
116    struct pci_device *vga_target;
117    struct pci_device *vga_default_dev;
118
119
120The vga_count is usually need to keep informed how many cards are being
121arbitrated, so for instance if there's only one then it can totally escape the
122scheme.
123
124
125These functions below acquire VGA resources for the given card and mark those
126resources as locked. If the resources requested are "normal" (and not legacy)
127resources, the arbiter will first check whether the card is doing legacy
128decoding for that type of resource. If yes, the lock is "converted" into a
129legacy resource lock. The arbiter will first look for all VGA cards that
130might conflict and disable their IOs and/or Memory access, including VGA
131forwarding on P2P bridges if necessary, so that the requested resources can
132be used. Then, the card is marked as locking these resources and the IO and/or
133Memory access is enabled on the card (including VGA forwarding on parent
134P2P bridges if any). In the case of vga_arb_lock(), the function will block
135if some conflicting card is already locking one of the required resources (or
136any resource on a different bus segment, since P2P bridges don't differentiate
137VGA memory and IO afaik). If the card already owns the resources, the function
138succeeds.  vga_arb_trylock() will return (-EBUSY) instead of blocking. Nested
139calls are supported (a per-resource counter is maintained).
140
141
142Set the target device of this client.
143    int  pci_device_vgaarb_set_target   (struct pci_device *dev);
144
145
146For instance, in x86 if two devices on the same bus want to lock different
147resources, both will succeed (lock). If devices are in different buses and
148trying to lock different resources, only the first who tried succeeds.
149    int  pci_device_vgaarb_lock         (void);
150    int  pci_device_vgaarb_trylock      (void);
151
152Unlock resources of device.
153    int  pci_device_vgaarb_unlock       (void);
154
155Indicates to the arbiter if the card decodes legacy VGA IOs, legacy VGA
156Memory, both, or none. All cards default to both, the card driver (fbdev for
157example) should tell the arbiter if it has disabled legacy decoding, so the
158card can be left out of the arbitration process (and can be safe to take
159interrupts at any time.
160    int  pci_device_vgaarb_decodes      (int new_vgaarb_rsrc);
161
162Connects to the arbiter device, allocates the struct
163    int  pci_device_vgaarb_init         (void);
164
165Close the connection
166    void pci_device_vgaarb_fini         (void);
167
168
169I.3 xf86VGAArbiter (X server implementation)
170--------------------------------------------
171
172(TODO)
173
174X server basically wraps all the functions that touch VGA registers somehow.
175
176
177II. Credits
178===========
179
180Benjamin Herrenschmidt (IBM?) started this work when he discussed such design
181with the Xorg community in 2005 [1, 2]. In the end of 2007, Paulo Zanoni and
182Tiago Vignatti (both of C3SL/Federal University of Paran�) proceeded his work
183enhancing the kernel code to adapt as a kernel module and also did the
184implementation of the user space side [3]. Now (2009) Tiago Vignatti and Dave
185Airlie finally put this work in shape and queued to Jesse Barnes' PCI tree.
186
187
188III. References
189==============
190
191[0] http://cgit.freedesktop.org/xorg/xserver/commit/?id=4b42448a2388d40f257774fbffdccaea87bd0347
192[1] http://lists.freedesktop.org/archives/xorg/2005-March/006663.html
193[2] http://lists.freedesktop.org/archives/xorg/2005-March/006745.html
194[3] http://lists.freedesktop.org/archives/xorg/2007-October/029507.html
195