| blob | ce0812b22bd9611efa7f2034e911dec9a2fe6a65 |
1 The purpose of the changes described here is to implement a more general
2 framework for multi-head on systems with more than one host-to-PCI bridge.
3 The changes also implement a basic port of XFree86 to SPARC Solaris.
5 These changes are derived from David S. Miller's submission #4653 to the
6 patch list. David Andrew of Sun Microsystems was also kind enough to
7 arrange for a hardware loan for development of these changes.
9 These changes are known to work on several SPARC SunOS and UltraSPARC
10 Linux configurations. Linux kernel work is in progress to port these
11 changes to Linux/PowerPC.
13 Several loose ends still need to be addressed before these changes can be
14 considered stable. The bulk of this note is devoted to enumerating what
15 remains to be done, along with other notes, broken down into various broad
16 categories.
18 SPARC SunOS (aka Solaris)
19 -------------------------
20 - An overview of this XFree86 port is available in README.Solaris.
21 - The keyboard map code in hw/xfree86/os-support/sunos/sun_kbdEv.c needs
22 to be extended to handle more than only the sun5 keyboard I targeted it
23 for. Even for the sun5, the map is incomplete as several keys are not
24 mapped. What is there is just barely usable.
25 - On exit, the server will zero out /dev/fb, but that might not be the
26 right thing to do for all primary adapters. This does however
27 appear to emulate the behaviour of Sun's commercial servers. It also
28 eliminates the need for output drivers to save and restore video memory
29 contents. (They still need to save/restore the mode timing however.)
30 This also chimes into a long-standing XFree86 policy to not save/restore
31 video memory contents if the mode on entry is found to be non-VGA, a
32 policy several existing drivers comply with.
33 - The SBUS drivers (sunbw2, suncg14, suncg3, suncg6, sunffb, sunleo and
34 suntcx), the common layer's SBUS code and the fbdev driver have all
35 only been compile tested. There are likely to be Linux'isms within
36 them that remain to be dealt with.
37 - It still needs to be verified whether or not this work adversely
38 affected support for ix86 Solaris.
40 UltraSPARC Linux
41 ----------------
42 - Although this code can be compiled using any Linux/SPARC64 kernel, it
43 can only run successfully using 2.4.12 or later.
44 - I haven't had time to sufficiently dig into XKB to properly configure it
45 for sun5 keyboards. Given XFree86 on Linux/SPARC has been around for a
46 while, it's likely someone has already done this, and I'd appreciate
47 receiving a copy of a working XF86Config input section.
49 PowerPC Linux
50 -------------
51 - As mentioned above, kernel work is in progress to port this PCI scheme
52 to Linux/PowerPC.
53 - Aside from kernel work, the inX() and outX() definitions in compiler.h
54 will need to be changed to do something akin to their SPARC definitions,
55 i.e. consider their port argument to be a virtual address.
57 Other Linux ports to multi-domain architectures
58 -----------------------------------------------
59 - Comments in os-support/bus/linuxPci.c document the kernel interface
60 required to port these changes. In short, Linux ports, such as Alpha
61 and mips, should follow SPARC and PowerPC's lead in providing support to
62 mmap() PCI devices through their /proc/bus/pci pseudo-files and to treat
63 such requests for host bridges as requests to mmap() space provided by
64 these bridges.
66 Other OS's
67 ----------
68 - In the right hands, either linuxPci.c or sparcPci.c can be used as a
69 guide for what would need to be done to port this scheme to other OS's.
70 Perhaps the largest difference between the two (in terms of interface to
71 the common layer) is that the SunOS port includes internally generated
72 domain numbers in PCITAG's, whereas the Linux port doesn't need to. The
73 remainder of the PCI code (which is OS-independent) can handle either
74 scheme.
75 - Required entry points are xf86GetPciDomain(), xf86MapDomainMemory(),
76 xf86MapDomainIO() and xf86ReadDomainMemory(). Replacements for
77 xf86BusAccWindowsFromOS(), xf86PciBusAccWindowsFromOS() and
78 xf86AccResFromOS() might also be required.
79 - Development of these changes has detected the fact that the XFree86 port
80 to the PowerMax OS is broken, and has been for some time, i.e. since
81 shortly after its introduction, back in the 3.9* days.
83 SPARC PCI (OS-independent)
84 --------------------------
85 - The "Simba" PCI-to-PCI bridge used in SPARC's does not implement VGA
86 routing, as defined in the PCI specs. Fortunately, OpenPROM seems to
87 always route VGA resources to the bus with PCI connectors, but this also
88 causes the common layer to not mark any PCI adapter as primary.
90 Multiple PCI domains (architecture- and OS-independent)
91 -------------------------------------------------------
92 - This implementation assumes every host-to-PCI bridge provides access to
93 a separate PCI domain. Each such domain provides three different
94 "address" spaces: PCI configuration, I/O and memory. The
95 implementation can also deal with situations where more than one PCI
96 domain share (different subsets of) the same PCI configuration space. I
97 have unconfirmed information that suggests it might be necessary to also
98 allow the sharing of PCI memory spaces.
99 - This implementation also assumes the CPU's physical address space
100 includes the entirety of each domain's I/O and memory spaces. I know
101 this'll need to be changed to deal with the so-called UniNorth bridge,
102 found on PowerPC's, which allows access to only a subset of the memory
103 space behind it.
104 - Ideally, the common layer should mark as primary up to one PCI adapter
105 per domain. This has yet to be done.
106 - Something needs to be done about PCI master aborts on primary buses.
107 For details on this, see my long-winded diatribe in sparcPci.c, and
108 related comments in linuxPci.c. Suffice it to say here that I see the
109 eventual implementation of host bridge drivers within XFree86 as
110 unavoidable at this point.
111 - DGA is broken on multi-domain platforms. The information passed to the
112 client to locate the framebuffer still needs to be revised. The best way
113 to deal with this is to change all drivers' OpenFramebuffer() function to
114 call a common layer routine to set the device name and displacements to be
115 returned to the DGA client.
117 Output drivers
118 --------------
119 Most drivers currently used on ix86 need(ed) source code changes.
120 - Calls to xf86ReadBIOS() and xf86MapVidMem() were replaced with calls to
121 xf86ReadDomainMemory() and xf86MapDomainMemory() respectively. Except
122 for the "ati" and "atimisc" modules, this has already been done.
123 - All ix86-style I/O port numbers need to be declared as an IOADDRESS, a
124 type defined in xf86Pci.h as "unsigned long". Such port numbers also
125 need to be offset by a displacement which is also defined as an
126 IOADDRESS. Before a driver's PreInit() is called, the common layer
127 makes this displacement available in ScrnInfoRec.domainIOBase. For
128 single-domain architectures, such as ix86, domainIOBase will always be
129 zero. Current use of vgaHWRec.PIOOffset has also been adjusted
130 accordingly. Some drivers have been changed to keep a copy of this
131 displacement in their private structure. Internally, an IOADDRESS is
132 actually a pointer that has been recasted to an unsigned long, but the
133 common layer "hides" this fact from the driver ABI, which means that I/O
134 port numbers, as seen by drivers, remain as integers rather than
135 addresses. Aside from the ati and atimisc modules, s3, sis and tseng
136 are the only modules left whose I/O still needs to be converted (I've
137 temporarily run out of steam).
138 - Note that these conversions are not necessarily sufficient to produce
139 drivers that will work on any given multi-domain architecture. A driver
140 that, for example, had endianness problems, still does. But, at least,
141 these conversions, along with the supporting common layer changes, make
142 PCI drivers more widely amenable to porting.
143 - rdinx(), wrinx(), modinx(), testrg(), testinx() and testinx2() are not
144 given enough information to allow for the relocation of their I/O. They
145 are consequently being deleted. The apm and ark drivers, the only
146 remaining callers of the first three, have been changed to use local
147 definitions instead. The last three (test*()) were already unused.
148 - As a temporary measure, these changes completely disable ISA-style
149 probing on SPARC's and PowerPC's. This means that driver calls to
150 xf86MatchIsaInstances(), while still valid, will always return detection
151 failure on SPARC's and PowerPC's. This will be dealt with when a more
152 general master abort handling scheme is implemented.
153 - I need to make a decision about the master abort issues mentionned above
154 before I can convert the "ati" and "atimisc" modules. Consequently,
155 these modules still need to be compiled with -DAVOID_CPIO on
156 multi-domain architectures, and support for Mach64 variants as
157 non-primary heads is not yet available.
159 $XFree86$