brintos

brintos / linux-shallow public Read only

0
0
Text · 8.0 KiB · bde3c0a Raw
192 lines · plain
1===========2VGA Arbiter3===========4 5Graphic devices are accessed through ranges in I/O or memory space. While most6modern devices allow relocation of such ranges, some "Legacy" VGA devices7implemented on PCI will typically have the same "hard-decoded" addresses as8they did on ISA. For more details see "PCI Bus Binding to IEEE Std 1275-19949Standard 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 for13the legacy VGA arbitration task (besides other bus management tasks) when more14than one legacy device co-exists on the same machine. But the problem happens15when these devices are trying to be accessed by different userspace clients16(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 to18control bus resources. Therefore an arbitration scheme outside of the X server19is needed to control the sharing of these resources. This document introduces20the operation of the VGA arbiter implemented for the Linux kernel.21 22vgaarb kernel/userspace ABI23---------------------------24 25The vgaarb is a module of the Linux Kernel. When it is initially loaded, it26scans all PCI devices and adds the VGA ones inside the arbitration. The27arbiter then enables/disables the decoding on different devices of the VGA28legacy instructions. Devices which do not want/need to use the arbiter may29explicitly tell it by calling vga_set_legacy_decoding().30 31The kernel exports a char device interface (/dev/vga_arbiter) to the clients,32which has the following semantics:33 34open35        Opens a user instance of the arbiter. By default, it's attached to the36        default VGA device of the system.37 38close39        Close a user instance. Release locks made by the user40 41read42        Return a string indicating the status of the target like:43 44        "<card_ID>,decodes=<io_state>,owns=<io_state>,locks=<io_state> (ic,mc)"45 46        An IO state string is of the form {io,mem,io+mem,none}, mc and47        ic are respectively mem and io lock counts (for debugging/48        diagnostic only). "decodes" indicate what the card currently49        decodes, "owns" indicates what is currently enabled on it, and50        "locks" indicates what is locked by this card. If the card is51        unplugged, we get "invalid" then for card_ID and an -ENODEV52        error is returned for any command until a new card is targeted.53 54 55write56        Write a command to the arbiter. List of commands:57 58        target <card_ID>59                switch target to card <card_ID> (see below)60        lock <io_state>61                acquires locks on target ("none" is an invalid io_state)62        trylock <io_state>63                non-blocking acquire locks on target (returns EBUSY if64                unsuccessful)65        unlock <io_state>66                release locks on target67        unlock all68                release all locks on target held by this user (not implemented69                yet)70        decodes <io_state>71                set the legacy decoding attributes for the card72 73        poll74                event if something changes on any card (not just the target)75 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 (TODO: not implemented yet). Currently,78        only PCI is supported as a prefix, but the userland API may support other bus79        types in the future, even if the current kernel implementation doesn't.80 81Note about locks:82 83The driver keeps track of which user has which locks on which card. It84supports stacking, like the kernel one. This complexifies the implementation85a bit, but makes the arbiter more tolerant to user space problems and able86to properly cleanup in all cases when a process dies.87Currently, a max of 16 cards can have locks simultaneously issued from88user space for a given user (file descriptor instance) of the arbiter.89 90In the case of devices hot-{un,}plugged, there is a hook - pci_notify() - to91notify them being added/removed in the system and automatically added/removed92in the arbiter.93 94There is also an in-kernel API of the arbiter in case DRM, vgacon, or other95drivers want to use it.96 97In-kernel interface98-------------------99 100.. kernel-doc:: include/linux/vgaarb.h101   :internal:102 103.. kernel-doc:: drivers/pci/vgaarb.c104   :export:105 106libpciaccess107------------108 109To use the vga arbiter char device it was implemented an API inside the110libpciaccess library. One field was added to struct pci_device (each device111on the system)::112 113    /* the type of resource decoded by the device */114    int vgaarb_rsrc;115 116Besides it, in pci_system were added::117 118    int vgaarb_fd;119    int vga_count;120    struct pci_device *vga_target;121    struct pci_device *vga_default_dev;122 123The vga_count is used to track how many cards are being arbitrated, so for124instance, if there is only one card, then it can completely escape arbitration.125 126These functions below acquire VGA resources for the given card and mark those127resources as locked. If the resources requested are "normal" (and not legacy)128resources, the arbiter will first check whether the card is doing legacy129decoding for that type of resource. If yes, the lock is "converted" into a130legacy resource lock. The arbiter will first look for all VGA cards that131might conflict and disable their IOs and/or Memory access, including VGA132forwarding on P2P bridges if necessary, so that the requested resources can133be used. Then, the card is marked as locking these resources and the IO and/or134Memory access is enabled on the card (including VGA forwarding on parent135P2P bridges if any). In the case of vga_arb_lock(), the function will block136if some conflicting card is already locking one of the required resources (or137any resource on a different bus segment, since P2P bridges don't differentiate138VGA memory and IO afaik). If the card already owns the resources, the function139succeeds.  vga_arb_trylock() will return (-EBUSY) instead of blocking. Nested140calls are supported (a per-resource counter is maintained).141 142Set the target device of this client. ::143 144    int  pci_device_vgaarb_set_target   (struct pci_device *dev);145 146For instance, in x86 if two devices on the same bus want to lock different147resources, both will succeed (lock). If devices are in different buses and148trying to lock different resources, only the first who tried succeeds. ::149 150    int  pci_device_vgaarb_lock         (void);151    int  pci_device_vgaarb_trylock      (void);152 153Unlock resources of device. ::154 155    int  pci_device_vgaarb_unlock       (void);156 157Indicates to the arbiter if the card decodes legacy VGA IOs, legacy VGA158Memory, both, or none. All cards default to both, the card driver (fbdev for159example) should tell the arbiter if it has disabled legacy decoding, so the160card can be left out of the arbitration process (and can be safe to take161interrupts at any time. ::162 163    int  pci_device_vgaarb_decodes      (int new_vgaarb_rsrc);164 165Connects to the arbiter device, allocates the struct ::166 167    int  pci_device_vgaarb_init         (void);168 169Close the connection ::170 171    void pci_device_vgaarb_fini         (void);172 173xf86VGAArbiter (X server implementation)174----------------------------------------175 176X server basically wraps all the functions that touch VGA registers somehow.177 178References179----------180 181Benjamin Herrenschmidt (IBM?) started this work when he discussed such design182with the Xorg community in 2005 [1, 2]. In the end of 2007, Paulo Zanoni and183Tiago Vignatti (both of C3SL/Federal University of Paraná) proceeded his work184enhancing the kernel code to adapt as a kernel module and also did the185implementation of the user space side [3]. Now (2009) Tiago Vignatti and Dave186Airlie finally put this work in shape and queued to Jesse Barnes' PCI tree.187 1880) https://cgit.freedesktop.org/xorg/xserver/commit/?id=4b42448a2388d40f257774fbffdccaea87bd03471891) https://lists.freedesktop.org/archives/xorg/2005-March/006663.html1902) https://lists.freedesktop.org/archives/xorg/2005-March/006745.html1913) https://lists.freedesktop.org/archives/xorg/2007-October/029507.html192