I found that xend's memory usage grows considerably when running a
script such as
while true; do xm new foo ; sleep 2s; xm delete foo ; sleep 2s; done
XendAPIStore maintains a list of class instances and in the case of
new/create operation, an associated VMMetrics object is stashed in the
list but never removed on delete/shutdown. This patch removes the
instance by invoking XendBase.destroy() method.
This patch adds 'type vnc' to vfb device sexpr for HVM guests. PV
guests already contain this entry in sexpr, e.g.
(device
(vfb
(vncunused 1)
...
(type vnc)
...
)
)
Some tools, such as libvirt, look for device/vfb/type in sexpr before
publishing VNC port. More importantly, this patch provides
consistency in vfb device sexpr between HVM and PV guests.
x86: Clean up asm keyword usage (asm volatile rather than __asm__
__volatile__ in most places) and ensure we use volatile keyword
wherever we have an asm stmt that produces outputs but has other
unspecified side effects or dependencies other than the
explicitly-stated inputs.
Also added volatile in a few places where its not strictly necessary
but where it's unlikely to produce worse code and it makes our
intentions perfectly clear.
The original problem this patch fixes was tracked down by Joseph
Cihula <joseph.cihula@intel.com>.
x86: Cleanup system restart code, and wait 10ms for APs to offline. Signed-off-by: Joseph Cihula <joseph.cihula@intel.com> Signed-off-by: Keir Fraser <keir@xensource.com>
svm: Greatly reduce total number of CR8 intercepts
This patch reduces the number of CR8 intercept to a fraction of the
number of CR8 intercepts without. First, CR8 read intercepts are
completely disabled since the SVM vTPR is kept kept in sync with the
HVM vLAPIC TPR. Second, CR8 write intercepts are enabled and disabled
based upon certain conditions. Most of the time, CR8 write intercepts
are disabled. They are enabled only when there is a pending interrupt
that can't be delivered because of either the current ISR or TPR (aka
PPR) because this is the only time the TPR matters.
With this patch, the number of CR8 intercepts dropped from around
10,000,000 to around 6,000 during boot of Windows 2003 Server 64-bit
(this is a rough estimate).
Tim Deegan [Mon, 10 Sep 2007 13:42:30 +0000 (14:42 +0100)]
[HVM] Add type information to the p2m map.
This is a base for memory tricks like page sharing, copy-on-write, lazy
allocation etc. It should also make pass-through MMIO easier to
implement in the p2m. Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>
xc_map_foreign_pages(), a convenient alternative to xc_map_foreign_batch()
xc_map_foreign_batch() can succeed partially. It is awkward to use
when you're only interested in complete success. Provide new
xc_map_foreign_pages() convenience function for that kind of use.
Also convert two obvious calls to use it.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Fix error message and wait time for xm block-detach command.
- Wait time
When xm requests a block device detach to xend, xm makes two
requests. At first, xm requests the block device detach by device
class 'vbd'. Next, xm requests the block device detaching by
device class 'tap'.
As a result, the wait time is 200 seconds because each of
the block device detaching requests causes time-out.
- Misleading error message
Because the last request is by device class 'tap' to xend,
the keyword "(tap)" is included in the error message.
This patch fixes the number of times of the block device detaching
request to one time. At first, xm makes inquiries about device
class of a detaching target device to xend. Then xm requires the
block device detaching by xend returned device class. The wait
time becomes 100 seconds because the block device detaching request
is one time. And the error message is also fixed.
The type of "TRIGGER_TYPE" is dictionary. domain_send_trigger()
refers to the keys of "TRIGGER_TYPE" without using keys() currently.
This patch adds keys() there.
PV-on-HVM: Fix non-SMP build warning for PV-on-HVM drivers.
smp_call_function() compiles to nothing on non-SMP, so we had a
defined-but-not-used static function.
Based on an original patch by: Signed-off-by: Ben Guthro <bguthro@virtualiron.com> Signed-off-by: Robert Phillips <rphillips@virtualiron.com> Signed-off-by: Keir Fraser <keir@xensource.com>
[XSM] correctly rename, relocate and patch the security.py file
This patch corrects an error in the original XSM tools patch. The
original patch did not rename, relocate and patch the security.py file
from changeset 15730:256160ff19b7. This patch addresses this issue as
well as any updates made to security.py in xen-staging during the
merge of XSM.
Signed-off-by: George Coker <gscoker@alpha.ncsc.mil>
[XSM] correctly located update_va_mapping hook within x86 ifdefs
The following patch addresses the issue where the XSM update_va_mapping
hook was not correctly located inside the x86 ifdefs. Included are
updates for the dummy and flask modules which are also effected by
this issue.
- update_va_mapping ifdef fix for xsm
- update_va_mapping ifdef fix for dummy module
- update_va_mapping ifdef fix for flask module
Signed-off-by: George Coker <gscoker@alpha.ncsc.mil>
Alex Williamson [Thu, 6 Sep 2007 15:05:26 +0000 (09:05 -0600)]
[IA64] Supress warning of __assign_domain_page().
On Tiger, the following memory region triggers a warning.
It seems a false-positive warning caused by c/s 13123:90db0f68b121.
so suppress the warning in such a case.
Implement x86 continuable domain destroy.
This patch addresses the following bug report.
http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1037 Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
xend: Always build lowlevel security modules, as there are some
dependencies on these from with xend proper. Signed-off-by: Keir Fraser <keir@xensource.com>
Print the following error when you give a wrong disk type to xm commands:
# xm create /xen/vm1.conf disk='tap:xxx:/xen/root-vm1.img,hda1,w'
Using config file "/xen/vm1.conf".
Error: tap:xxx not a valid disk type
# xm block-attach vm2 tap:yyy:/xen/second.img hdb1 w
Error: tap:yyy not a valid disk type
Usage: xm block-attach <Domain> <BackDev> <FrontDev> <Mode>
[BackDomain]
DEFINE_XEN_GUEST_HANDLE is evaluated by the pre-processor twice. Do this also for XEN_GUEST_HANDLE.
This fixes a build error for PV guests (OpenBSD, NetBSD) where the use
of XEN_GUEST_HANDLE(uint8_t) leads to a build error because uint8_t is
a #define instead a typedef.
Signed-off-by: Christoph Egger <Christoph.Egger@amd.com>
Fix blktap script for versions of readlink command that do not handle
regular files. Signed-off-by: Ben Guthro <bguthro@virtualiron.com> Signed-off-by: Josh Nicholas <jnicholas@virtualiron.com>
Tim Deegan [Fri, 31 Aug 2007 10:06:22 +0000 (11:06 +0100)]
[HVM] Shadow: don't shadow the p2m table.
For HVM vcpus with paging disabled, we used to shadow the p2m table,
and skip the p2m lookup to go from gfn to mfn. Instead, we now
provide a simple pagetable that gives a one-to-one mapping of 4GB, and
shadow that, making the translations from gfn to mfn via the p2m.
This removes the paging-disabled special-case code from the shadow
fault handler, and allows us to expand the p2m interface, since all HVM
translations now go through the same p2m lookups. Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>
Support extensions to Intel architecture for TXT/SMX. Signed-off-by: Joseph Cihula <joseph.cihula@intel.com> Signed-off-by: Keir Fraser <keir@xensource.com>
Made the QEMU IDE disk tolerate more power-mgmt commands, specifically:
WIN_SETFEATURE(EN_AAM) - enable automatic acoustic mgmt
WIN_SETFEATURE(DIS_AAM) - disable automatic acoustic mgmt
WIN_SETFEATURE(DIS_APM) - disable automatic power mgmt
WIN_STANDBY
WIN_SETIDLE1
WIN_SLEEPNOW1
WIN_STANDBY2
WIN_SETIDLE2
WIN_SLEEPNOW2
WIN_STANDBYNOW2
These are all essentially no-ops, like existing support for EN_RLA
(enable read lookahead) and WIN_STANDBYNOW1.
This fixes a crash in the SLES9-SP3 64bit kernel when the powersaved
was started (with ACPI or ACM). This guest really only needs EN_AAM,
DIS_APM, and WIN_SETIDLE1 support, but the others seemed sensible to
include.
I've excluded EN_APM since I'm unsure of what that's agreeing to do.
It's probably ok to include.
Signed-off-by: Ben Guthro <bguthro@virtualiron.com> Signed-off-by: David Lively <dlively@virtualiron.com>
Introduce xc_core_arch_gpfn_may_present() hook.
On ia64 trying to map a foreign domain page which isn't allocated
cause annoying warning message. xm dump-core results in too many
warnings. With this hook, ia64 dump-core can suprress warning. Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Skipping image name at start of command line is an
architecture-specific action. Definitely not required on IA64, for now
make it just x86 specific. Signed-off-by: Keir Fraser <keir@xensource.com>
hvm: Intercept reads of MSR_IA32_MCG_CAP and return 0.
This indicates no machine check "units", which agrees more closely
with Xen's super-minimal machine check architecture (just enough to
allow Windows to run). This fixes a bug that occurs when migrating a
RHEL4-64bit guest to a host with fewer machine check units than the
original host. These host physical details shouldn't be leaking
through to guests.
Signed-off-by: David Lively <dlively@virtualiron.com>
x86: Remove (most) Centaur CPU support. Only VIA C7 can work, as it
has CMOV support. Leave a small amount of centaur.c around to support
that. MTRR code goes entirely, as 686-class Centaur CPUs have generic
MTRR support.
Don't carry dead code needlessly: this is a family 5 CPU, which Xen
doesn't support. Perhaps, other CPUs' files could use some cleanup in
that respect, too, but there it would increase the delta to the Linux
origin of these files, while here the entire file can go away.
Alex Williamson [Tue, 28 Aug 2007 18:25:37 +0000 (12:25 -0600)]
[IA64] Move nvram from /usr to /var
Presently nvram is stored in /usr/lib/xen/boot/nvram_<domain> next to the guest
firmware. This violates the FHS because /usr might be mounted read-only. This
patch moves the nvram storage to /var/lib/xen/nvram/nvram_<domain>
Also clean up:
- references to stat_buf assumed that stat() had succeeded; use access()
instead since it's easier and doesn't require stat_buf at all
- nvram_path[PATH_MAX] instead of nvram_path[100]
- strncpy(..., strlen(src)) is meaningless, re-order length tests to work
correctly
ioemu: Avoid struct members clashing with POSIX apis
The TPM code in tools/ioemu/hw/tpm_tis.c has a struct containing a
number of function pointers with names open, close, read, write which
are the same as various POSIX apis already #included in the
file. POSIX allows these functions to be defined as macros and latest
GCC/glibc does indeed define them as macros depending on compiler
flags. This causes compile errors when deferencing the struct
members. The solution is either to change calls like ctx->open () to
be (* ctx->open) (), or simply to rename the struct members. Since
this struct was only used inside that one file I simply renamed them.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When I tested xm create command, I saw the following error message.
I expected an error message "Error: (12, 'Cannot allocate memory')"
because I intentionally caused a memory shortage on the test.
But the error message was different from my expectation.
# xm create /xen/HVMdomain.1
Using config file "/xen/HVMdomain.1".
Error: an integer is required
I looked at xend.log to examine the cause why the error message was
shown. (Could you see the attached xend.log?)
xend had the error message "Error: (12, 'Cannot allocate memory')"
first. But xend changed the error message to "Error: an integer is
required" halfway. I'm not sure about the cause why an exception
occurred in logging processing. But when I applied an attached patch,
I confirmed that the error message that I expected was shown. The
patch does not call xc.domain_destroy_hook() if self.domid is None.
ioemu: error checkin when setting up the Cirrus Logic video device.
set_mm_mapping() may fail because of xc_domain_populate_physmap(). In
this case, we should not blindly go on; the xc_map_foreign_batch()
that follows will cause a page fault and, at best, get mapped in a
zeroed page from the dom0 (which is not what we want). While I'm in
here, fix a memory leak on an error path.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
Fix VNC server after HVM save/restore/migrate, when no vncpasswd specified.
The basic problem was that the "image" section of the sxp had a "None"
in it, which means that on the restore (either on the local machine or
remote machine), the vnc server in the qemu device model literally
needed the string "None" to properly connect. This simple patch only
puts a vncpasswd entry in the image if it is *not* None in the python
code, thus avoiding the whole issue.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
projects / xen.git / log