IO Page table growth is triggered by amd_iommu_map_page and grows to
upper level. I have tested it well for different devices (nic and gfx)
and different guests (linux and Win7) with different guest memory
sizes (512M, 1G, 4G and above).
Signed-off-by: Wei Wang <wei.wang2@amd.com>
Firstly, __cpupool_find_by_id() would dereference NULL, at the end of
an exact search if the search loop exited with *q==NULL. Fix this.
Secondly, provide suitable accessor functions so that no caller needs
to use the __-prefixed versions which take a boolean flag.
Signed-off-by: Keir Fraser <keir@xen.org>
Some cpupool actions didn't check the cpupool_id exactly. For some
actions this doesn't make any sense, so refuse those actions if the
specified cpupool doesn't exist.
Signed-off-by: Juergen Gross <juergen.gross@ts.fujitsu.com>
Added a spinlock across snb_vtd_ops_preamble() and
snb_vtd_ops_postamble() to make modifications to IGD registers atomic.
Continue keeping snb_igd_quirk default off.
Signed-off-by: Allen Kay <allen.m.kay@intel.com>
By default, vscsi expects to be passed the final device name (eg
/dev/st3) instead of one of the various udev symlinks (eg
/dev/tape/by-path/pci-0000:01:08.0-scsi-0:0:2:0-st). The following patch
resolves the path to the real path if the name starts with /dev/
Signed-off-by: James Harper <james.harper@bendigoit.com.au>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
GCC 4.2.4 cannot figure out that three variables aren't used before
initialisation:
xl_cmdimpl.c: In function `print_domain_vcpuinfo':
xl_cmdimpl.c:3351: warning: `firstset' may be used uninitialized in this function
[etc]
Signed-off-by: Kamala Narasimhan <kamala.narasimhan@citrix.com>
Acked-by: Andre Przywara <andre.przywara@amd.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Apply mtu size from bridge interface also in vif interface.
This depends on a kernel change which allows arbitrary mtu sizes until
the frontend driver has connected to the backend driver. Without this
kernel change, the vif mtu size will be limited to 1500 even with this
change to the vif-bridge script.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Acked-by: Ian Campbell <Ian.Campbell@eu.citrix.com>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Some of the cpupool_dprintk() calls are using undefined or
uninitialized variables. Correct the argument lists to be able to
define cpupool_printk as printk.
Signed-off-by: Juergen Gross <juergen.gross@ts.fujitsu.com>
Fixes ioemu stubdom build:
CC i386-stubdom/piix4acpi.o
[...]/stubdom/ioemu/hw/piix4acpi.c:272: error: expected ')' before '?' token
[...]/stubdom/ioemu/hw/piix4acpi.c:277: error: conflicting types for 'set_bit'
[...]/stubdom/../extras/mini-os/include/x86/mini-os/os.h:396: error: previous definition of 'set_bit' was here
[...]/stubdom/ioemu/hw/piix4acpi.c:282: error: conflicting types for 'clear_bit'
[...]/stubdom/../extras/mini-os/include/x86/mini-os/os.h:414: error: previous definition of 'clear_bit' was here
[...]/stubdom/ioemu/hw/piix4acpi.c: In function 'gpe_sts_write':
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Some of the hvmop memory structures have a shocking amount of
unnecesssary padding in them. Elements which can have only 3 values
are given 64 bits of memory, and then aligned (so that there is
padding behind them).
This patch resizes and reorganizes in the following way, (hopefully)
without introducing any differences between the layout for 32- and
64-bit.
xen_hvm_set_mem_type:
hvmmem_type -> 16 bits
nr -> 32 bits (limiting us to setting 16TB at a time)
xen_hvm_set_mem_access:
hvmmem_access -> 16 bits
nr -> 32 bits
xen_hvm_get_mem_access:
hvmmem_access -> 16 bits
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Fixes a size calculation bug when enabled bits in XFEATURE_MASK (xcr0)
aren't contiguous.
Current for_loop will stop when xcr0 feature bit is 0. But in reality,
the bits can be non-contiguous. One example is that LWP is bit 62 on
AMD platform. This patch iterates through all bits to calculate the
size for enabled features.
Signed-off-by: Wei Huang <wei.huang2@amd.com>
Firstly, they are only for use from privileged entities, so mark them
as usable only by code defining __XEN_TOOLS__.
Secondly, the check for must-not-operate-on-myself can be done more
cleanly.
Signed-off-by: Keir Fraser <keir@xen.org>
At some point, it seems that someone manually added Flask permission
definitions to one header file without updating the corresponding
policy configuration or the other related table. The end result is
that we can get uninterpretable AVC messages like this:
# xl dmesg | grep avc
(XEN) avc: denied { 0x4000000 } for domid=0
scontext=system_u:system_r:dom0_t tcontext=system_u:system_r:domU_t
tclass=domain
Fix this by updating the flask config and regenerating the headers
from it. In the future, this can be further improved by integrating
the automatic generation of the headers into the build process as is
presently done in SELinux.
Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
This patch turn-on SYSENTER MSRs interception for 32bit guest VMs on
AMD CPUs. With it, hvm_svm.guest_sysenter_xx fields always contain the
canonical version of SYSENTER MSRs and are used in guest save/restore.
The data fields in VMCB save area are updated as necessary.
Reported-by: James Harper <james.harper@bendigoit.com.au>
Signed-off-by: Wei Huang <wei.huang2@amd.com>
pci-detach triggers IO page table deallocation if the last passthru
device has been removed from pdev list, and this will result a BUG on
amd systems for next pci-attach. This patch fixes this issue.
Signed-off-by: Wei Wang <wei.wang2@amd.com>
When switching schedulers on a physical cpu due to a cpupool operation
check for a potential memory allocation failure and stop the operation
gracefully.
Signed-off-by: Juergen Gross <juergen.gross@ts.fujitsu.com>
The file we write the vnc password to must be rewound back to the
beginning, or the vnc viewer will simply get EOF.
When the syscalls for communicating the password to the vnc client
fail, bomb out with an error messsage rather than blundering on (and
probably producing a spurious password prompt).
Following this patch, xl vncviewer --autopass works, provided the qemu
patch for writing the password to xenstore has also been applied.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
The /local/domain/DOMID/device/vfb/0/backend path is irrelevant.
libxl does not create it, so the branch would never be taken.
Instead, simply read the target paths of interest.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
qemu only sets a password on its vnc display if the value for the -vnc
option has the ",password" modifier. The code for constructing
qemu-dm options was broken and only added this modifier for one of the
cases.
Unfortunately there does not appear to be any code for passing the vnc
password to upstream qemu (ie, in the case where
libxl_build_device_model_args_new is called). To avoid accidentally
running the domain without a password, check for this situation and
fail an assertion. This will have to be revisited after 4.1.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
When using pygrub we have to connect to the console twice: once at the
beginning to connect to pygrub and a second time after creating the pv
console to connect to the guest's console.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
The algorithm for printing the CPU affinity in a condensed way
looks for a set bit in a zero-byte:
for (i = 0; !(pcpumap & 1); ++i, pcpumap >>= 1)
Looking at the code I found that it is entirely broken if more than 8
CPUs are used. Beside that endless loop issue the output is totally
bogus except for the "any CPU" case, which is handled explicitly earlier.
I tried to fix it, but the whole approach does not work if the outer
loops actually iterates (executing more than once).
This fix reimplements the whole algorithm in a clean (though not much
optimized way). It survived some unit-testing.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
xend uses rtl8139, and we want xl to be compatible with xm. Some
older operating systems don't have e1000 drivers, and we want widest
compatibility rather than best performance (people who want good
performance are best advised to use PV-on-HVM drivers).
We'll probably switch to a new default when switching to upstream
qemu, in the Xen 4.2 release cycle.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Constantly m(un)locking memory can have significant overhead on
systems with large numbers of CPUs. This was previously fixed by
20841:fbe8f32fa257 but this was dropped during the transition to
hypercall buffers.
Introduce a small cache of single page hypercall buffer allocations
which can be resused to avoid this overhead.
Add some statistics tracking to the hypercall buffer allocations.
The cache size of 4 was chosen based on these statistics since they
indicated that 2 pages was sufficient to satisfy all concurrent single
page hypercall buffer allocations seen during "xl create", "xl
shutdown" and "xl destroy" of both a PV and HVM guest therefore 4
pages should cover the majority of important cases.
This fixes http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1719.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Reported-by: Zheng, Shaohui <shaohui.zheng@intel.com>
Tested-by: Haitao Shan <maillists.shan@gmail.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
The Xen build system is not currently compatible with the --as-needed
linker option. The proper fix for this is turning out to be rather
invasive to the build system so simply disable for now with the
intention of revisiting for the 4.2 release.
The --no-as-needed option is available at least since binutils 2.15
(released in May 2004) and hence I think can be unconditionally relied
on.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Reported-by: Nathan March <nathan@gt.net>
Tested-by: Nathan March <nathan@gt.net>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Shared libraries should be executable.
(rpm (4.9.0) doesn't automatically supply a "provides" entry for a
library unless it is executable. Non-executable libraries can cause
other trouble too.)
Signed-off-by: Michael Young <m.a.young@durham.ac.uk>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Update a comment about NetworkManager not supporting bridging in
Fedora 11 to refer instead to Fedora 14. Clarify the wording.
Signed-off-by: Michael Young <m.a.young@durham.ac.uk>
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Fix an anomaly in /etc/xen/scripts/network-route.
Currently this script contains
netdev=${netdev:-eth${vifnum}}
ie. netdev is set to eth${vifnum} by default. Unfortunately vifnum
is not set anywhere in the xen code so the default is actually the
broken "eth". And anyway the vif number (which is what vifnum ought
to be) is not relevant.
The patch changes the default to eth0 (which is what the comment at
the top of the file says it should be).
Signed-off-by: Michael Young <m.a.young@durham.ac.uk>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Attached patch disables pci_vtd_quirk for 32-bit Xen since 32-bit xen
does not support MMCFG access.
Signed-off-by: Allen Kay <allen.m.kay@intel.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Before the creation and population of a new CPU pool we have to clear
the poolid variable, which still contains the value from the previous
iteration.
This fixes the execution of xl cpupool-numa-split on machines with more
than two nodes.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Acked-by: juergen.gross@ts.fujitsu.com
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Though illegal characters on xl command lines are catched, the user
isn't currently informed which one was not right.
This patch fixes this by printing the faulting character.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
It's not clear why a userspace lzma decode would want to use that
particular value, what bearing it has on anything or why it would
assume it could use 1/3 of the total RAM in the system (potentially
quite a large amount of RAM) as opposed to any other limit number.
Instead, hardcode 32Mby.
This reverts 22830:c80960244942, removes the xc_get_physmem/physmem
function entirely, and replaces the expression at the call site with a
fixed constant.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <Ian.Campbell@eu.citrix.com>
Cc: Christoph Egger <Christoph.Egger@amd.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
libxl_domain_rename is where domain names are assigned. Therefore
this is where we check that no two domains have the same name. As a
special exception, domains whose names are "" are not considered to
clash.
We also take special care not to mind if we try to rename a domain to
the name it already has.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Previously calling libxl__devices_destroy on a half-constructed or
half-destroyed domain would sometimes complain along these lines:
libxl: error: libxl_device.c:327:libxl__devices_destroy /local/domain/29/device is empty
This is (a) not a reasonable thing to complain about and (b) not an
accurate description of all the things that that particular failure of
libxl__xs_directory might mean.
Change the code to check errno, so that if errno==ENOENT we silently
continue, not destroying any devices, and if errno!=ENOENT, properly
log the problem and fail.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>
Wei Gang
Wei Wang
Ian Jackson
Keir Fraser
James Harper
Kamala Narasimhan
Olaf Hering
Ian Campbell
Tim Deegan
Stefano Stabellini
Stefano Stabellini
Andre Przywara
Allen Kay
Michael Young