ia64/linux-2.6.18-xen.hg

view Documentation/cciss.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
1 This driver is for Compaq's SMART Array Controllers.
3 Supported Cards:
4 ----------------
6 This driver is known to work with the following cards:
8 * SA 5300
9 * SA 5i
10 * SA 532
11 * SA 5312
12 * SA 641
13 * SA 642
14 * SA 6400
15 * SA 6400 U320 Expansion Module
16 * SA 6i
17 * SA P600
18 * SA P800
19 * SA E400
20 * SA P400i
21 * SA E200
22 * SA E200i
23 * SA E500
25 If nodes are not already created in the /dev/cciss directory, run as root:
27 # cd /dev
28 # ./MAKEDEV cciss
30 Device Naming:
31 --------------
33 You need some entries in /dev for the cciss device. The MAKEDEV script
34 can make device nodes for you automatically. Currently the device setup
35 is as follows:
37 Major numbers:
38 104 cciss0
39 105 cciss1
40 106 cciss2
41 105 cciss3
42 108 cciss4
43 109 cciss5
44 110 cciss6
45 111 cciss7
47 Minor numbers:
48 b7 b6 b5 b4 b3 b2 b1 b0
49 |----+----| |----+----|
50 | |
51 | +-------- Partition ID (0=wholedev, 1-15 partition)
52 |
53 +-------------------- Logical Volume number
55 The device naming scheme is:
56 /dev/cciss/c0d0 Controller 0, disk 0, whole device
57 /dev/cciss/c0d0p1 Controller 0, disk 0, partition 1
58 /dev/cciss/c0d0p2 Controller 0, disk 0, partition 2
59 /dev/cciss/c0d0p3 Controller 0, disk 0, partition 3
61 /dev/cciss/c1d1 Controller 1, disk 1, whole device
62 /dev/cciss/c1d1p1 Controller 1, disk 1, partition 1
63 /dev/cciss/c1d1p2 Controller 1, disk 1, partition 2
64 /dev/cciss/c1d1p3 Controller 1, disk 1, partition 3
66 SCSI tape drive and medium changer support
67 ------------------------------------------
69 SCSI sequential access devices and medium changer devices are supported and
70 appropriate device nodes are automatically created. (e.g.
71 /dev/st0, /dev/st1, etc. See the "st" man page for more details.)
72 You must enable "SCSI tape drive support for Smart Array 5xxx" and
73 "SCSI support" in your kernel configuration to be able to use SCSI
74 tape drives with your Smart Array 5xxx controller.
76 Additionally, note that the driver will not engage the SCSI core at init
77 time. The driver must be directed to dynamically engage the SCSI core via
78 the /proc filesystem entry which the "block" side of the driver creates as
79 /proc/driver/cciss/cciss* at runtime. This is because at driver init time,
80 the SCSI core may not yet be initialized (because the driver is a block
81 driver) and attempting to register it with the SCSI core in such a case
82 would cause a hang. This is best done via an initialization script
83 (typically in /etc/init.d, but could vary depending on distibution).
84 For example:
86 for x in /proc/driver/cciss/cciss[0-9]*
87 do
88 echo "engage scsi" > $x
89 done
91 Once the SCSI core is engaged by the driver, it cannot be disengaged
92 (except by unloading the driver, if it happens to be linked as a module.)
94 Note also that if no sequential access devices or medium changers are
95 detected, the SCSI core will not be engaged by the action of the above
96 script.
98 Hot plug support for SCSI tape drives
99 -------------------------------------
101 Hot plugging of SCSI tape drives is supported, with some caveats.
102 The cciss driver must be informed that changes to the SCSI bus
103 have been made, in addition to and prior to informing the SCSI
104 mid layer. This may be done via the /proc filesystem. For example:
106 echo "rescan" > /proc/scsi/cciss0/1
108 This causes the adapter to query the adapter about changes to the
109 physical SCSI buses and/or fibre channel arbitrated loop and the
110 driver to make note of any new or removed sequential access devices
111 or medium changers. The driver will output messages indicating what
112 devices have been added or removed and the controller, bus, target and
113 lun used to address the device. Once this is done, the SCSI mid layer
114 can be informed of changes to the virtual SCSI bus which the driver
115 presents to it in the usual way. For example:
117 echo scsi add-single-device 3 2 1 0 > /proc/scsi/scsi
119 to add a device on controller 3, bus 2, target 1, lun 0. Note that
120 the driver makes an effort to preserve the devices positions
121 in the virtual SCSI bus, so if you are only moving tape drives
122 around on the same adapter and not adding or removing tape drives
123 from the adapter, informing the SCSI mid layer may not be necessary.
125 Note that the naming convention of the /proc filesystem entries
126 contains a number in addition to the driver name. (E.g. "cciss0"
127 instead of just "cciss" which you might expect.)
129 Note: ONLY sequential access devices and medium changers are presented
130 as SCSI devices to the SCSI mid layer by the cciss driver. Specifically,
131 physical SCSI disk drives are NOT presented to the SCSI mid layer. The
132 physical SCSI disk drives are controlled directly by the array controller
133 hardware and it is important to prevent the kernel from attempting to directly
134 access these devices too, as if the array controller were merely a SCSI
135 controller in the same way that we are allowing it to access SCSI tape drives.
137 SCSI error handling for tape drives and medium changers
138 -------------------------------------------------------
140 The linux SCSI mid layer provides an error handling protocol which
141 kicks into gear whenever a SCSI command fails to complete within a
142 certain amount of time (which can vary depending on the command).
143 The cciss driver participates in this protocol to some extent. The
144 normal protocol is a four step process. First the device is told
145 to abort the command. If that doesn't work, the device is reset.
146 If that doesn't work, the SCSI bus is reset. If that doesn't work
147 the host bus adapter is reset. Because the cciss driver is a block
148 driver as well as a SCSI driver and only the tape drives and medium
149 changers are presented to the SCSI mid layer, and unlike more
150 straightforward SCSI drivers, disk i/o continues through the block
151 side during the SCSI error recovery process, the cciss driver only
152 implements the first two of these actions, aborting the command, and
153 resetting the device. Additionally, most tape drives will not oblige
154 in aborting commands, and sometimes it appears they will not even
155 obey a reset coommand, though in most circumstances they will. In
156 the case that the command cannot be aborted and the device cannot be
157 reset, the device will be set offline.
159 In the event the error handling code is triggered and a tape drive is
160 successfully reset or the tardy command is successfully aborted, the
161 tape drive may still not allow i/o to continue until some command
162 is issued which positions the tape to a known position. Typically you
163 must rewind the tape (by issuing "mt -f /dev/st0 rewind" for example)
164 before i/o can proceed again to a tape drive which was reset.