ia64/linux-2.6.18-xen.hg

view Documentation/sx.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
children
line source
2 sx.txt -- specialix SX/SI multiport serial driver readme.
6 Copyright (C) 1997 Roger Wolff (R.E.Wolff@BitWizard.nl)
8 Specialix pays for the development and support of this driver.
9 Please DO contact support@specialix.co.uk if you require
10 support.
12 This driver was developed in the BitWizard linux device
13 driver service. If you require a linux device driver for your
14 product, please contact devices@BitWizard.nl for a quote.
16 (History)
17 There used to be an SI driver by Simon Allan. This is a complete
18 rewrite from scratch. Just a few lines-of-code have been snatched.
20 (Sources)
21 Specialix document number 6210028: SX Host Card and Download Code
22 Software Functional Specification.
24 (Copying)
25 This program is free software; you can redistribute it and/or
26 modify it under the terms of the GNU General Public License as
27 published by the Free Software Foundation; either version 2 of
28 the License, or (at your option) any later version.
30 This program is distributed in the hope that it will be
31 useful, but WITHOUT ANY WARRANTY; without even the implied
32 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
33 PURPOSE. See the GNU General Public License for more details.
35 You should have received a copy of the GNU General Public
36 License along with this program; if not, write to the Free
37 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139,
38 USA.
40 (Addendum)
41 I'd appreciate it that if you have fixes, that you send them
42 to me first.
45 Introduction
46 ============
48 This file contains some random information, that I like to have online
49 instead of in a manual that can get lost. Ever misplace your Linux
50 kernel sources? And the manual of one of the boards in your computer?
53 Theory of operation
54 ===================
56 An important thing to know is that the driver itself doesn't have the
57 firmware for the card. This means that you need the separate package
58 "sx_firmware". For now you can get the source at
60 ftp://ftp.bitwizard.nl/specialix/sx_firmware_<version>.tgz
62 The firmware load needs a "misc" device, so you'll need to enable the
63 "Support for user misc device modules" in your kernel configuration.
64 The misc device needs to be called "/dev/specialix_sxctl". It needs
65 misc major 10, and minor number 167 (assigned by HPA). The section
66 on creating device files below also creates this device.
68 After loading the sx.o module into your kernel, the driver will report
69 the number of cards detected, but because it doesn't have any
70 firmware, it will not be able to determine the number of ports. Only
71 when you then run "sx_firmware" will the firmware be downloaded and
72 the rest of the driver initialized. At that time the sx_firmware
73 program will report the number of ports installed.
75 In contrast with many other multi port serial cards, some of the data
76 structures are only allocated when the card knows the number of ports
77 that are connected. This means we won't waste memory for 120 port
78 descriptor structures when you only have 8 ports. If you experience
79 problems due to this, please report them: I haven't seen any.
82 Interrupts
83 ==========
85 A multi port serial card, would generate a horrendous amount of
86 interrupts if it would interrupt the CPU for every received
87 character. Even more than 10 years ago, the trick not to use
88 interrupts but to poll the serial cards was invented.
90 The SX card allow us to do this two ways. First the card limits its
91 own interrupt rate to a rate that won't overwhelm the CPU. Secondly,
92 we could forget about the cards interrupt completely and use the
93 internal timer for this purpose.
95 Polling the card can take up to a few percent of your CPU. Using the
96 interrupts would be better if you have most of the ports idle. Using
97 timer-based polling is better if your card almost always has work to
98 do. You save the separate interrupt in that case.
100 In any case, it doesn't really matter all that much.
102 The most common problem with interrupts is that for ISA cards in a PCI
103 system the BIOS has to be told to configure that interrupt as "legacy
104 ISA". Otherwise the card can pull on the interrupt line all it wants
105 but the CPU won't see this.
107 If you can't get the interrupt to work, remember that polling mode is
108 more efficient (provided you actually use the card intensively).
111 Allowed Configurations
112 ======================
114 Some configurations are disallowed. Even though at a glance they might
115 seem to work, they are known to lockup the bus between the host card
116 and the device concentrators. You should respect the drivers decision
117 not to support certain configurations. It's there for a reason.
119 Warning: Seriously technical stuff ahead. Executive summary: Don't use
120 SX cards except configured at a 64k boundary. Skip the next paragraph.
122 The SX cards can theoretically be placed at a 32k boundary. So for
123 instance you can put an SX card at 0xc8000-0xd7fff. This is not a
124 "recommended configuration". ISA cards have to tell the bus controller
125 how they like their timing. Due to timing issues they have to do this
126 based on which 64k window the address falls into. This means that the
127 32k window below and above the SX card have to use exactly the same
128 timing as the SX card. That reportedly works for other SX cards. But
129 you're still left with two useless 32k windows that should not be used
130 by anybody else.
133 Configuring the driver
134 ======================
136 PCI cards are always detected. The driver auto-probes for ISA cards at
137 some sensible addresses. Please report if the auto-probe causes trouble
138 in your system, or when a card isn't detected.
140 I'm afraid I haven't implemented "kernel command line parameters" yet.
141 This means that if the default doesn't work for you, you shouldn't use
142 the compiled-into-the-kernel version of the driver. Use a module
143 instead. If you convince me that you need this, I'll make it for
144 you. Deal?
146 I'm afraid that the module parameters are a bit clumsy. If you have a
147 better idea, please tell me.
149 You can specify several parameters:
151 sx_poll: number of jiffies between timer-based polls.
153 Set this to "0" to disable timer based polls.
154 Initialization of cards without a working interrupt
155 will fail.
157 Set this to "1" if you want a polling driver.
158 (on Intel: 100 polls per second). If you don't use
159 fast baud rates, you might consider a value like "5".
160 (If you don't know how to do the math, use 1).
162 sx_slowpoll: Number of jiffies between timer-based polls.
163 Set this to "100" to poll once a second.
164 This should get the card out of a stall if the driver
165 ever misses an interrupt. I've never seen this happen,
166 and if it does, that's a bug. Tell me.
168 sx_maxints: Number of interrupts to request from the card.
169 The card normally limits interrupts to about 100 per
170 second to offload the host CPU. You can increase this
171 number to reduce latency on the card a little.
172 Note that if you give a very high number you can overload
173 your CPU as well as the CPU on the host card. This setting
174 is inaccurate and not recommended for SI cards (But it
175 works).
177 sx_irqmask: The mask of allowable IRQs to use. I suggest you set
178 this to 0 (disable IRQs all together) and use polling if
179 the assignment of IRQs becomes problematic. This is defined
180 as the sum of (1 << irq) 's that you want to allow. So
181 sx_irqmask of 8 (1 << 3) specifies that only irq 3 may
182 be used by the SX driver. If you want to specify to the
183 driver: "Either irq 11 or 12 is ok for you to use", then
184 specify (1 << 11) | (1 << 12) = 0x1800 .
186 sx_debug: You can enable different sorts of debug traces with this.
187 At "-1" all debugging traces are active. You'll get several
188 times more debugging output than you'll get characters
189 transmitted.
192 Baud rates
193 ==========
195 Theoretically new SXDCs should be capable of more than 460k
196 baud. However the line drivers usually give up before that. Also the
197 CPU on the card may not be able to handle 8 channels going at full
198 blast at that speed. Moreover, the buffers are not large enough to
199 allow operation with 100 interrupts per second. You'll have to realize
200 that the card has a 256 byte buffer, so you'll have to increase the
201 number of interrupts per second if you have more than 256*100 bytes
202 per second to transmit. If you do any performance testing in this
203 area, I'd be glad to hear from you...
205 (Psst Linux users..... I think the Linux driver is more efficient than
206 the driver for other OSes. If you can and want to benchmark them
207 against each other, be my guest, and report your findings...... :-)
210 Ports and devices
211 =================
213 Port 0 is the top connector on the module closest to the host
214 card. Oh, the ports on the SXDCs and TAs are labelled from 1 to 8
215 instead of from 0 to 7, as they are numbered by linux. I'm stubborn in
216 this: I know for sure that I wouldn't be able to calculate which port
217 is which anymore if I would change that....
220 Devices:
222 You should make the device files as follows:
224 #!/bin/sh
225 # (I recommend that you cut-and-paste this into a file and run that)
226 cd /dev
227 t=0
228 mknod specialix_sxctl c 10 167
229 while [ $t -lt 64 ]
230 do
231 echo -n "$t "
232 mknod ttyX$t c 32 $t
233 mknod cux$t c 33 $t
234 t=`expr $t + 1`
235 done
236 echo ""
237 rm /etc/psdevtab
238 ps > /dev/null
241 This creates 64 devices. If you have more, increase the constant on
242 the line with "while". The devices start at 0, as is customary on
243 Linux. Specialix seems to like starting the numbering at 1.
245 If your system doesn't come with these devices pre-installed, bug your
246 linux-vendor about this. They should have these devices
247 "pre-installed" before the new millennium. The "ps" stuff at the end
248 is to "tell" ps that the new devices exist.
250 Officially the maximum number of cards per computer is 4. This driver
251 however supports as many cards in one machine as you want. You'll run
252 out of interrupts after a few, but you can switch to polled operation
253 then. At about 256 ports (More than 8 cards), we run out of minor
254 device numbers. Sorry. I suggest you buy a second computer.... (Or
255 switch to RIO).
257 ------------------------------------------------------------------------
260 Fixed bugs and restrictions:
261 - Hangup processing.
262 -- Done.
264 - the write path in generic_serial (lockup / oops).
265 -- Done (Ugly: not the way I want it. Copied from serial.c).
267 - write buffer isn't flushed at close.
268 -- Done. I still seem to lose a few chars at close.
269 Sorry. I think that this is a firmware issue. (-> Specialix)
271 - drain hardware before changing termios
272 - Change debug on the fly.
273 - ISA free irq -1. (no firmware loaded).
274 - adding c8000 as a probe address. Added warning.
275 - Add a RAMtest for the RAM on the card.c
276 - Crash when opening a port "way" of the number of allowed ports.
277 (for example opening port 60 when there are only 24 ports attached)
278 - Sometimes the use-count strays a bit. After a few hours of
279 testing the use count is sometimes "3". If you are not like
280 me and can remember what you did to get it that way, I'd
281 appreciate an Email. Possibly fixed. Tell me if anyone still
282 sees this.
283 - TAs don't work right if you don't connect all the modem control
284 signals. SXDCs do. T225 firmware problem -> Specialix.
285 (Mostly fixed now, I think. Tell me if you encounter this!)
287 Bugs & restrictions:
289 - Arbitrary baud rates. Requires firmware update. (-> Specialix)
291 - Low latency (mostly firmware, -> Specialix)