The credit2 scheduler tries to setup runqueues in such
a way that there is one of them per each socket. However,
that does not work. The issue is described in bug #36
"credit2 only uses one runqueue instead of one runq per
socket" (http://bugs.xenproject.org/xen/bug/36), and a
solution has been attempted by an old patch series:
http://lists.xen.org/archives/html/xen-devel/2014-08/msg02168.html
Here, we take advantage of the fact that now initialization
happens (for all schedulers) during CPU_STARTING, so we
have all the topology information available when necessary.
This is true for all the pCPUs _except_ the boot CPU. That
is not an issue, though. In fact, no runqueue exists yet
when the boot CPU is initialized, so we can just create
one and put the boot CPU in there.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>
As other schedulers are doing already: if the idle vcpu
is picked and scheduled, there is no need to reprogram the
scheduler timer to fire and invoke csched2_schedule()
again in future.
Tickling or external events will serve as pokes, when
necessary, but until we can, we should just stay idle.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reported-by: Tianyang Chen <tiche@seas.upenn.edu>
Suggested-by: George Dunlap <george.dunlap@citrix.com>
Acked-by: George Dunlap <george.dunlap@citrix.com>
In short, the point is making sure that the actual switch
of scheduler and the remapping of the scheduler's runqueue
lock occur in the same critical section, protected by the
"old" scheduler's lock (and not, e.g., in the free_pdata
hook, as it is now for Credit2 and RTDS).
Not doing so, is (at least) racy. In fact, for instance,
if we switch cpu X from, Credit2 to Credit, we do:
schedule_cpu_switch(x, csched2 --> csched):
//scheduler[x] is csched2
//schedule_lock[x] is csched2_lock
csched_alloc_pdata(x)
csched_init_pdata(x)
pcpu_schedule_lock(x) ----> takes csched2_lock
scheduler[X] = csched
pcpu_schedule_unlock(x) --> unlocks csched2_lock
[1]
csched2_free_pdata(x)
pcpu_schedule_lock(x) --> takes csched2_lock
schedule_lock[x] = csched_lock
spin_unlock(csched2_lock)
While, if we switch cpu X from, Credit to Credit2, we do:
schedule_cpu_switch(X, csched --> csched2):
//scheduler[x] is csched
//schedule_lock[x] is csched_lock
csched2_alloc_pdata(x)
csched2_init_pdata(x)
pcpu_schedule_lock(x) --> takes csched_lock
schedule_lock[x] = csched2_lock
spin_unlock(csched_lock)
[2]
pcpu_schedule_lock(x) ----> takes csched2_lock
scheduler[X] = csched2
pcpu_schedule_unlock(x) --> unlocks csched2_lock
csched_free_pdata(x)
And if we switch cpu X from RTDS to Credit2, we do:
schedule_cpu_switch(X, RTDS --> csched2):
//scheduler[x] is rtds
//schedule_lock[x] is rtds_lock
csched2_alloc_pdata(x)
csched2_init_pdata(x)
pcpu_schedule_lock(x) --> takes rtds_lock
schedule_lock[x] = csched2_lock
spin_unlock(rtds_lock)
pcpu_schedule_lock(x) ----> takes csched2_lock
scheduler[x] = csched2
pcpu_schedule_unlock(x) --> unlocks csched2_lock
rtds_free_pdata(x)
spin_lock(rtds_lock)
ASSERT(schedule_lock[x] == rtds_lock) [3]
schedule_lock[x] = DEFAULT_SCHEDULE_LOCK [4]
spin_unlock(rtds_lock)
So, the first problem is that, if anything related to
scheduling, and involving CPU, happens at [1] or [2], we:
- take csched2_lock,
- operate on Credit1 functions and data structures,
which is no good!
The second problem is that the ASSERT at [3] triggers, and
the third that at [4], we screw up the lock remapping we've
done for ourself in csched2_init_pdata()!
The first problem arises because there is a window during
which the lock is already the new one, but the scheduler is
still the old one. The other two, becase we let schedulers
mess with the lock (re)mapping done by others.
This patch, therefore, introduces a new hook in the scheduler
interface, called switch_sched, meant at being used when
switching scheduler on a CPU, and implements it for the
various schedulers, so that things are done in the proper
order and under the protection of the best suited (set of)
lock(s). It is necessary to add the hook (as compared to
keep doing things in generic code), because different
schedulers may have different locking schemes.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>
Acked-by: Robert VanVossen <robert.vanvossen@dornerworks.com>
create_domain() is of uint32_t return type, because on
success it returns the domid of the new domain, and
uint32_t is what we typically use for domid-s.
However, on failure, it returns ERROR_FAIL or ERROR_INVAL,
which are -3 and -6. Callers assign the return value to an
'int rc' variable and then check for '(rc < 0)'.
Although things work, and no tool (compiler, Coverity, ecc.)
is complaining, using 'int' as return type seems better.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
by making them more consistent with other examples in xl.
Affected functions are the ones related to console, vnc,
dump, destroy, shutdown, list, domid and domname.
Signed-off-by: Harmandeep Kaur <write.harmandeep@gmail.com>
Signed-off-by: Dario Fagggioli <dario.faggioli@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
by making them more consistent with other examples in xl.
While there, make freemem() of boolean return type, which
looks more natural, and add comment explaining why
parse_mem_size_kb() needs to diverge from the pattern.
Signed-off-by: Harmandeep Kaur <write.harmandeep@gmail.com>
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
The internals of the uuid_t struct don't match a big endian octet stream on
*BSD systems, which means that it cannot be directly casted to a
uint8_t[16].
In order to solve that change the type to be an unsigned char[16], which
doesn't imply any other change on Linux. On *BSDs change the helpers so that
the uuid is always stored as a big endian byte stream.
NB: tested on FreeBSD and Linux only.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Discussed-with: Ian Jackson <Ian.Jackson@eu.citrix.com>
Discussed-with: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Doing a 'xl migrate --debug domU host' on xen-4.5 adds a
XC_SAVE_ID_ENABLE_VERIFY_MODE marker, which is not handled.
Since using --debug is valid usage, handle it by logging the fact
instead of aborting the migration.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
Cc: Ian Jackson <ian.jackson@eu.citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Xen, being 64bit only, cannot run PV guests in vm86 mode. HVM guests however
can be running in vm86 mode, and common codepaths need to be able to cope.
The definition of vm86_mode() in x86_64/regs.h is incorrect, as the predicate
is used by non-PV codepaths.
One buggy use is in hvm/emulate.c. An HVM guest can be in vm86 mode, and
vm86_mode() sliently omits the check. Luckily, due to the VEX prefix decoding
logic in x86_emulate(), there is no path to the erronious use with EFLAGS_VM
set.
Another potentially problematic use is in show_guest_stack(). In principle,
show_guest_stack() is common code called for both PV and HVM vcpus. HVM vcpus
exit early (with no reasonable way of making the code generic), making this
part a PV-only codepath.
Open-code its use in emulate.c, matching the surrounding code. This causes
all other uses to be in PV-only codepaths, making the code to be logically
dead. Drop it completely, to avoid future misuse.
Part of resulting cleanup removes vm86attr from read_descriptor(), although
retaining one relevant piece of information; i.e. whether we are reading a
selector for an instruction fetch, or a data fetch.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Refer to SDM Volume 1 Extended Region of an XSAVE Area. The value returned
by ecx[1] with cpuid function 0xd and sub-function i (i>1) indicates
the alignment of the state component i when the compacted format of the
extended region of an xsave area is used.
So when hvm guest using CPUID eax=0xd, ecx=1 to get the size of area
used for compacted format, we need to take alignment into consideration.
tools side is fixed by
"tools/libxc: Calculate xstate cpuid leaf from guest information"
by Andrew Cooper
Signed-off-by: Shuai Ruan <shuai.ruan@intel.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
1. get_xsave_addr() will only be called when
xsave_area_compressed(xsave) is true. So drop the
conditional expression.
2. expand_xsave_states() will memset the area when
get NULL from get_xsave_addr().
Reported-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Shuai Ruan <shuai.ruan@intel.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
c/s 607044bf9 "build: avoid putting local absolute symbols in symbol tables"
breaks the build with Clang, as the command line argument isn't understood.
Clang does not appear to have any equivielent option, and already has
outstanding issues with duplicate symbols. Excluding this option makes the
problem no worse.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
That will turn out useful in following patches, where such
code will need to be called more than just once. Create an
helper now, and move the code there, to avoid mixing code
motion and functional changes later.
In Credit2, some style cleanup is also done.
No functional change intended.
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@eu.citrix.com>
In fact, if a scheduler needs per-pCPU information,
that needs to be initialized appropriately. So, we take
the code that is performing initializations from (right
now) .alloc_pdata, and use it for .init_pdata, leaving
only actualy allocations in the former, if any (which
is the case in RTDS and Credit1).
On the other hand, in Credit2, since we don't really
need any per-pCPU data allocation, everything that was
being done in .alloc_pdata, is now done in .init_pdata.
And the fact that now .alloc_pdata can be left undefined,
allows us to just get rid of it.
Still for Credit2, the fact that .init_pdata is called
during CPU_STARTING (rather than CPU_UP_PREPARE) kills
the need for the scheduler to setup a similar callback
itself, simplifying the code.
And thanks to such simplification, it is now also ok to
move some of the logic meant at double checking that a
cpu was (or was not) initialized, into ASSERTS (rather
than an if() and a BUG_ON).
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: Meng Xu <mengxu@cis.upenn.edu>
Reviewed-by: George Dunlap <george.dunlap@eu.citrix.com>
The .alloc_pdata scheduler hook must, before this change,
be implemented by all schedulers --even those ones that
don't need to allocate anything.
Make it possible to just use the SCHED_OP(), like for
the other hooks, by using ERR_PTR() and IS_ERR() for
error reporting. This:
- makes NULL a variant of success;
- allows for errors other than ENOMEM to be properly
communicated (if ever necessary).
This, in turn, means that schedulers not needing to
allocate any per-pCPU data, can avoid implementing the
hook. In fact, the artificial implementation of
.alloc_pdata in the ARINC653 is removed (and, while there,
nuke .free_pdata too, as it is equally useless).
Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: Meng Xu <mengxu@cis.upenn.edu>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: George Dunlap <george.dunlap@citrix.com>
Add the FreeBSD disk hotplug interface details to the block-scripts.txt
document.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>
The fields that are printed might not be set in the case of a failure, which
generates a segmentation fault.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
This code is extracted from the FreeBSD blkfront implementation.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
The current code in libxl assumed that vdev is equal to local device, but
this is only true for Linux systems. In other OSes the local device can use
a nomenclature completely different from the virtual device one.
Move the current libxl__devid_to_localdev Linux implementation out of the
OS-specific file and rename it to libxl__devid_to_vdev, and then make sure
local_device_attach_cb return the local device in the diskpath field.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
This factors out the nic hotplug specific code from the common code path in
order to make it easier to add support for disk hotplug scripts. It
shouldn't include any functional change.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
This is the default hotplug script for block devices. Its only job is to
copy the "params" blkback xenstore node to "physical-device".
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
FreeBSD blkback uses the physical-device-path xenstore node in order to
fetch the path to the underlying backing storage (either a block device or
raw image). This node is set by the hotplug scripts. Also clarify the usage
of the physical-device node.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
It seems the logic is meant to detect sector unaligned buffers for block
writes. The NOTing of the logic instead masks off any unaligned bits and
also would cause the function to always fail. It seems the function is not
used in any of the tools so that is probably why the problem is not seen.
In the vhd_read_block function it is correct.
Signed-off-by: Ross Philipson <ross.philipson@ainfosec.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Netlink is required when initialising COLO, so make sure only to compile
COLO only when netlink is available. Change the inclusion of
linux/netlink.h to netlink/netlink.h so that it doesn't use Linux kernel
header directly.
Provide necessary stub functions in case COLO is disabled. This should
fix build on FreeBSD because there is no netlink there. It would also
make libxl build properly when netlink is not present on a Linux
system.
Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
COLO and Remus net buffer support both depend on the availability of
libnl. Use a generic name.
No functional changes.
Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
They are only used there, no need to expose them to other parts of
libxl.
This is necessary to make libxl build on FreeBSD again because FreeBSD
doesn't have netlink.
Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
We need to separate COLO code from common code as clean as possible.
With this patch, all COLO structures are now in libxl_colo.h.
It does the following:
1. Move two typedefs for libxl__domain_create_state{,cb} back to
libxl_internal.h.
2. Move libxl__colo_save_state to libxl_colo.h.
3. Include libxl_internal.h in libxl_colo.h.
4. Move a bunch of colo typedefs to the top of libxl_internal.h.
5. Move the inclusion of libxl_colo.h to the right place in
libxl_internal.h.
Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Add a service routine checking whether a device model must be started
after adding a device to a domain.
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Add an internal service function to check for a running device model.
This can be used later when adding devices to a domain requiring a
device model for either printing an error message or starting the
device model in case it is not already running.
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Add a new pvusb backend type "qusb" which is provided by qemu. It can
be selected either by specifying the type directly in the configuration
or it is selected automatically by libxl in case there is no "usbback"
driver loaded.
The "qusb" backend is a full replacement for the kernel based backend
"vusb". The interface to the frontend is the very same including all
Xenstore paths, while the backend paths in Xenstore differ ("qusb"
instead ov "vusb").
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Add a function to query whether the device model is supporting a
specific backend type. The device model is writing the supported
backend types to Xenstore on startup. The new query function checks
for the appropriate entry to be present.
As not all versions of qemu are capable to indicate support of
specific backends the query function is to be called with an indicator
whether the default return value should be "supported" (in case qemu
doesn't know set any support information) or "not supported".
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
libxl__need_xenpv_qemu() is called with configuration data for console,
vfbs, disks and channels today in order to evaluate the need for
starting a device model for a pv domain.
The console data is local to the caller and setup in a way to never
require a device model. All other data is taken from the domain config
structure.
In order to support other device backends via qemu change the interface
of libxl__need_xenpv_qemu() to take the domain config structure as
input instead of the single device arrays.
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
In case libxl__need_xenpv_qemu() returns an error let domain creation
fail.
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
It takes a bit of thinking to parse the original statement. It looks
like it is missing an ',' - right after the 'braced' to make it
obvious that:
if (X)
return XYZ;
is OK, while multiple conditionals require braces.
Changing the 'apart from' with 'except' makes it more obvious
(I hope).
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Dario Faggioli
Uma Sharma
Harmandeep Kaur
Roger Pau Monne
Olaf Hering
Chunyan Liu
Ian Jackson
Andrew Cooper
Shuai Ruan
Chong Li
Ross Philipson
Wei Liu
Juergen Gross
Konrad Rzeszutek Wilk