view Documentation/sh/new-machine.txt @ 897:329ea0ccb344

balloon: try harder to balloon up under memory pressure.

Currently if the balloon driver is unable to increase the guest's
reservation it assumes the failure was due to reaching its full
allocation, gives up on the ballooning operation and records the limit
it reached as the "hard limit". The driver will not try again until
the target is set again (even to the same value).

However it is possible that ballooning has in fact failed due to
memory pressure in the host and therefore it is desirable to keep
attempting to reach the target in case memory becomes available. The
most likely scenario is that some guests are ballooning down while
others are ballooning up and therefore there is temporary memory
pressure while things stabilise. You would not expect a well behaved
toolstack to ask a domain to balloon to more than its allocation nor
would you expect it to deliberately over-commit memory by setting
balloon targets which exceed the total host memory.

This patch drops the concept of a hard limit and causes the balloon
driver to retry increasing the reservation on a timer in the same
manner as when decreasing the reservation.

Also if we partially succeed in increasing the reservation
(i.e. receive less pages than we asked for) then we may as well keep
those pages rather than returning them to Xen.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Fri Jun 05 14:01:20 2009 +0100 (2009-06-05)
parents 831230e53067
line source
2 Adding a new board to LinuxSH
3 ================================
5 Paul Mundt <lethal@linux-sh.org>
7 This document attempts to outline what steps are necessary to add support
8 for new boards to the LinuxSH port under the new 2.5 and 2.6 kernels. This
9 also attempts to outline some of the noticeable changes between the 2.4
10 and the 2.5/2.6 SH backend.
12 1. New Directory Structure
13 ==========================
15 The first thing to note is the new directory structure. Under 2.4, most
16 of the board-specific code (with the exception of stboards) ended up
17 in arch/sh/kernel/ directly, with board-specific headers ending up in
18 include/asm-sh/. For the new kernel, things are broken out by board type,
19 companion chip type, and CPU type. Looking at a tree view of this directory
20 heirarchy looks like the following:
22 Board-specific code:
24 .
25 |-- arch
26 | `-- sh
27 | `-- boards
28 | |-- adx
29 | | `-- board-specific files
30 | |-- bigsur
31 | | `-- board-specific files
32 | |
33 | ... more boards here ...
34 |
35 `-- include
36 `-- asm-sh
37 |-- adx
38 | `-- board-specific headers
39 |-- bigsur
40 | `-- board-specific headers
41 |
42 .. more boards here ...
44 It should also be noted that each board is required to have some certain
45 headers. At the time of this writing, io.h is the only thing that needs
46 to be provided for each board, and can generally just reference generic
47 functions (with the exception of isa_port2addr).
49 Next, for companion chips:
50 .
51 `-- arch
52 `-- sh
53 `-- cchips
54 `-- hd6446x
55 |-- hd64461
56 | `-- cchip-specific files
57 `-- hd64465
58 `-- cchip-specific files
60 ... and so on. Headers for the companion chips are treated the same way as
61 board-specific headers. Thus, include/asm-sh/hd64461 is home to all of the
62 hd64461-specific headers.
64 Finally, CPU family support is also abstracted:
65 .
66 |-- arch
67 | `-- sh
68 | |-- kernel
69 | | `-- cpu
70 | | |-- sh2
71 | | | `-- SH-2 generic files
72 | | |-- sh3
73 | | | `-- SH-3 generic files
74 | | `-- sh4
75 | | `-- SH-4 generic files
76 | `-- mm
77 | `-- This is also broken out per CPU family, so each family can
78 | have their own set of cache/tlb functions.
79 |
80 `-- include
81 `-- asm-sh
82 |-- cpu-sh2
83 | `-- SH-2 specific headers
84 |-- cpu-sh3
85 | `-- SH-3 specific headers
86 `-- cpu-sh4
87 `-- SH-4 specific headers
89 It should be noted that CPU subtypes are _not_ abstracted. Thus, these still
90 need to be dealt with by the CPU family specific code.
92 2. Adding a New Board
93 =====================
95 The first thing to determine is whether the board you are adding will be
96 isolated, or whether it will be part of a family of boards that can mostly
97 share the same board-specific code with minor differences.
99 In the first case, this is just a matter of making a directory for your
100 board in arch/sh/boards/ and adding rules to hook your board in with the
101 build system (more on this in the next section). However, for board families
102 it makes more sense to have a common top-level arch/sh/boards/ directory
103 and then populate that with sub-directories for each member of the family.
104 Both the Solution Engine and the hp6xx boards are an example of this.
106 After you have setup your new arch/sh/boards/ directory, remember that you
107 also must add a directory in include/asm-sh for headers localized to this
108 board. In order to interoperate seamlessly with the build system, it's best
109 to have this directory the same as the arch/sh/boards/ directory name,
110 though if your board is again part of a family, the build system has ways
111 of dealing with this, and you can feel free to name the directory after
112 the family member itself.
114 There are a few things that each board is required to have, both in the
115 arch/sh/boards and the include/asm-sh/ heirarchy. In order to better
116 explain this, we use some examples for adding an imaginary board. For
117 setup code, we're required at the very least to provide definitions for
118 get_system_type() and platform_setup(). For our imaginary board, this
119 might look something like:
121 /*
122 * arch/sh/boards/vapor/setup.c - Setup code for imaginary board
123 */
124 #include <linux/init.h>
126 const char *get_system_type(void)
127 {
128 return "FooTech Vaporboard";
129 }
131 int __init platform_setup(void)
132 {
133 /*
134 * If our hardware actually existed, we would do real
135 * setup here. Though it's also sane to leave this empty
136 * if there's no real init work that has to be done for
137 * this board.
138 */
140 /*
141 * Presume all FooTech boards have the same broken timer,
142 * and also presume that we've defined foo_timer_init to
143 * do something useful.
144 */
145 board_time_init = foo_timer_init;
147 /* Start-up imaginary PCI ... */
149 /* And whatever else ... */
151 return 0;
152 }
154 Our new imaginary board will also have to tie into the machvec in order for it
155 to be of any use. Currently the machvec is slowly on its way out, but is still
156 required for the time being. As such, let us take a look at what needs to be
157 done for the machvec assignment.
159 machvec functions fall into a number of categories:
161 - I/O functions to IO memory (inb etc) and PCI/main memory (readb etc).
162 - I/O remapping functions (ioremap etc)
163 - some initialisation functions
164 - a 'heartbeat' function
165 - some miscellaneous flags
167 The tree can be built in two ways:
168 - as a fully generic build. All drivers are linked in, and all functions
169 go through the machvec
170 - as a machine specific build. In this case only the required drivers
171 will be linked in, and some macros may be redefined to not go through
172 the machvec where performance is important (in particular IO functions).
174 There are three ways in which IO can be performed:
175 - none at all. This is really only useful for the 'unknown' machine type,
176 which us designed to run on a machine about which we know nothing, and
177 so all all IO instructions do nothing.
178 - fully custom. In this case all IO functions go to a machine specific
179 set of functions which can do what they like
180 - a generic set of functions. These will cope with most situations,
181 and rely on a single function, mv_port2addr, which is called through the
182 machine vector, and converts an IO address into a memory address, which
183 can be read from/written to directly.
185 Thus adding a new machine involves the following steps (I will assume I am
186 adding a machine called vapor):
188 - add a new file include/asm-sh/vapor/io.h which contains prototypes for
189 any machine specific IO functions prefixed with the machine name, for
190 example vapor_inb. These will be needed when filling out the machine
191 vector.
193 This is the minimum that is required, however there are ample
194 opportunities to optimise this. In particular, by making the prototypes
195 inline function definitions, it is possible to inline the function when
196 building machine specific versions. Note that the machine vector
197 functions will still be needed, so that a module built for a generic
198 setup can be loaded.
200 - add a new file arch/sh/boards/vapor/mach.c. This contains the definition
201 of the machine vector. When building the machine specific version, this
202 will be the real machine vector (via an alias), while in the generic
203 version is used to initialise the machine vector, and then freed, by
204 making it initdata. This should be defined as:
206 struct sh_machine_vector mv_vapor __initmv = {
207 .mv_name = "vapor",
208 }
209 ALIAS_MV(vapor)
211 - finally add a file arch/sh/boards/vapor/io.c, which contains
212 definitions of the machine specific io functions.
214 A note about initialisation functions. Three initialisation functions are
215 provided in the machine vector:
216 - mv_arch_init - called very early on from setup_arch
217 - mv_init_irq - called from init_IRQ, after the generic SH interrupt
218 initialisation
219 - mv_init_pci - currently not used
221 Any other remaining functions which need to be called at start up can be
222 added to the list using the __initcalls macro (or module_init if the code
223 can be built as a module). Many generic drivers probe to see if the device
224 they are targeting is present, however this may not always be appropriate,
225 so a flag can be added to the machine vector which will be set on those
226 machines which have the hardware in question, reducing the probe to a
227 single conditional.
229 3. Hooking into the Build System
230 ================================
232 Now that we have the corresponding directories setup, and all of the
233 board-specific code is in place, it's time to look at how to get the
234 whole mess to fit into the build system.
236 Large portions of the build system are now entirely dynamic, and merely
237 require the proper entry here and there in order to get things done.
239 The first thing to do is to add an entry to arch/sh/Kconfig, under the
240 "System type" menu:
242 config SH_VAPOR
243 bool "Vapor"
244 help
245 select Vapor if configuring for a FooTech Vaporboard.
247 next, this has to be added into arch/sh/Makefile. All boards require a
248 machdir-y entry in order to be built. This entry needs to be the name of
249 the board directory as it appears in arch/sh/boards, even if it is in a
250 sub-directory (in which case, all parent directories below arch/sh/boards/
251 need to be listed). For our new board, this entry can look like:
253 machdir-$(CONFIG_SH_VAPOR) += vapor
255 provided that we've placed everything in the arch/sh/boards/vapor/ directory.
257 Next, the build system assumes that your include/asm-sh directory will also
258 be named the same. If this is not the case (as is the case with multiple
259 boards belonging to a common family), then the directory name needs to be
260 implicitly appended to incdir-y. The existing code manages this for the
261 Solution Engine and hp6xx boards, so see these for an example.
263 Once that is taken care of, it's time to add an entry for the mach type.
264 This is done by adding an entry to the end of the arch/sh/tools/mach-types
265 list. The method for doing this is self explanatory, and so we won't waste
266 space restating it here. After this is done, you will be able to use
267 implicit checks for your board if you need this somewhere throughout the
268 common code, such as:
270 /* Make sure we're on the FooTech Vaporboard */
271 if (!mach_is_vapor())
272 return -ENODEV;
274 also note that the mach_is_boardname() check will be implicitly forced to
275 lowercase, regardless of the fact that the mach-types entries are all
276 uppercase. You can read the script if you really care, but it's pretty ugly,
277 so you probably don't want to do that.
279 Now all that's left to do is providing a defconfig for your new board. This
280 way, other people who end up with this board can simply use this config
281 for reference instead of trying to guess what settings are supposed to be
282 used on it.
284 Also, as soon as you have copied over a sample .config for your new board
285 (assume arch/sh/configs/vapor_defconfig), you can also use this directly as a
286 build target, and it will be implicitly listed as such in the help text.
288 Looking at the 'make help' output, you should now see something like:
290 Architecture specific targets (sh):
291 zImage - Compressed kernel image (arch/sh/boot/zImage)
292 adx_defconfig - Build for adx
293 cqreek_defconfig - Build for cqreek
294 dreamcast_defconfig - Build for dreamcast
295 ...
296 vapor_defconfig - Build for vapor
298 which then allows you to do:
300 $ make ARCH=sh CROSS_COMPILE=sh4-linux- vapor_defconfig vmlinux
302 which will in turn copy the defconfig for this board, run it through
303 oldconfig (prompting you for any new options since the time of creation),
304 and start you on your way to having a functional kernel for your new
305 board.