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