This reverts commit a0035ecc0d.
Even with this fix there is a period between the flush and the unmap where
processor may speculate data into the cache. The solution is to map this
region uncached or to use the HCR.DC bit to mark all guest accesses cached.
89eb02c220 "xen: arm: force guest memory accesses to cacheable when MMU is
disabled" has arranged to do the latter.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
On ARM guest OSes are started with MMU and Caches disables (as they are on
native) however caching is enabled in the domain running the builder and
therefore we must ensure cache consistency.
The existing solution to this problem (a0035ecc0d "tools: libxc: flush data
cache after loading images into guest memory") is to flush the caches after
loading the various blobs into guest RAM. However this approach has two short
comings:
- The cache flush primitives available to userspace on arm32 are not
sufficient for our needs.
- There is a race between the cache flush and the unmap of the guest page
where the processor might speculatively dirty the cache line again.
(of these the second is the more fundamental)
This patch makes use of the the hardware functionality to force all accesses
made from guest mode to be cached (the HCR.DC == default cached bit). This
means that we don't need to worry about the domain builder's writes being
cached because the guests "uncached" accesses will actually be cached.
Unfortunately the use of HCR.DC is incompatible with the guest enabling its
MMU (SCTLR.M bit). Therefore we must trap accesses to the SCTLR so that we can
detect when this happens and disable HCR.DC. This is done with the HCR.TVM
(trap virtual memory controls) bit which also causes various other registers
to be trapped, all of which can be passed straight through to the underlying
register. Once the guest has enabled its MMU we no longer need to trap so
there is no ongoing overhead. In my tests Linux makes about half a dozen
accesses to these registers before the MMU is enabled, I would expect other
OSes to behave similarly (the sequence of writes needed to setup the MMU is
pretty obvious).
Apart from this unfortunate need to trap these accesses this approach is
incompatible with guests which attempt to do DMA operations with their MMU
disabled. In practice this means guests with passthrough which we do not yet
support. Since a typical guest (including dom0) does not access devices which
require DMA until after it is fully up and running with paging enabled the
main risk is to in-guest firmware which does DMA i.e. running EFI in a guest,
with a disk passed through and booting from that disk. Since we know that dom0
is not using any such firmware and we do not support device passthrough to
guests yet we can live with this restriction. Once passthrough is implemented
this will need to be revisited.
The patch includes a couple of seemingly unrelated but necessary changes:
- HSR_SYSREG_CRN_MASK was incorrectly defined, which happened to be benign
with the existing set of system register we handled, but broke with the new
ones introduced here.
- The defines used to decode the HSR system register fields were named the
same as the register. This breaks the accessor macros. This had gone
unnoticed because the handling of the existing trapped registers did not
require accessing the underlying hardware register. Rename those constants
with an HSR_SYSREG prefix (in line with HSR_CP32/64 for 32-bit registers).
This patch has survived thousands of boot loops on a Midway system.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Julien Grall <julien.grall@linaro.org>
Scrub heap pages was disabled because it was slow on the models. Now that Xen
supports real hardware, it's possible to enable by default scrubbing.
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
This allows the application to pass a token to libxl in the fd/timeout
registration callbacks, which it receives back in modification or
deregistration callbacks.
It turns out that this is essential for timeout handling, in order to
identify which timeout to change on a modify event.
Signed-off-by: Rob Hoes <rob.hoes@citrix.com>
Acked-by: David Scott <dave.scott@eu.citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Commit 7113a45451 (kexec/x86: do not map
crash kernel area) causes fatal page faults when loading a crash
image. The attempt to zero the first control page allocated from the
crash region will fault as the VA return by map_domain_page() has no
mapping.
The fault will occur on non-debug builds of Xen when the crash area is
below 5 TiB (which will be most systems).
The assumption that the crash area mapping was not used is incorrect.
map_domain_page() is used when loading an image and building the
image's page tables to temporarily map the crash area, thus the
mapping is required if the crash area is in the direct map area.
Reintroduce the mapping, but only the portions of the crash area that
are within the direct map area.
Reported-by: Don Slutz <dslutz@verizon.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Tested-by: Don Slutz <dslutz@verizon.com>
Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>
Tested-by: Daniel Kiper <daniel.kiper@oracle.com>
This is really just a band aid - kexec shouldn't rely on the crash area
being always mapped when in the direct mapping range (and it didn't use
to in its previous form). That's primarily because map_domain_page()
(needed when the area is outside the direct mapping range) may be
unusable when wanting to kexec due to a crash, but also because in the
case of PFN compression the kexec range (if specified on the command
line) could fall into a hole between used memory ranges (while we're
currently only ignoring memory at the top of the physical address
space, it's pretty clear that sooner or later we will want that
selection to become more sophisticated in order to maximize the memory
made use of).
Acked-by: Jan Beulich <jbeulich@suse.com>
Using a 1G hvm domU (in grub) and gdbsx:
(gdb) set arch i8086
warning: A handler for the OS ABI "GNU/Linux" is not built into this configuration
of GDB. Attempting to continue with the default i8086 settings.
The target architecture is assumed to be i8086
(gdb) target remote localhost:9999
Remote debugging using localhost:9999
Remote debugging from host 127.0.0.1
0x0000d475 in ?? ()
(gdb) x/1xh 0x6ae9168b
Will reproduce this bug.
With a debug=y build you will get:
Assertion '!preempt_count()' failed at preempt.c:37
For a debug=n build you will get a dom0 VCPU hung (at some point) in:
[ffff82c4c0126eec] _write_lock+0x3c/0x50
ffff82c4c01e43a0 __get_gfn_type_access+0x150/0x230
ffff82c4c0158885 dbg_rw_mem+0x115/0x360
ffff82c4c0158fc8 arch_do_domctl+0x4b8/0x22f0
ffff82c4c01709ed get_page+0x2d/0x100
ffff82c4c01031aa do_domctl+0x2ba/0x11e0
ffff82c4c0179662 do_mmuext_op+0x8d2/0x1b20
ffff82c4c0183598 __update_vcpu_system_time+0x288/0x340
ffff82c4c015c719 continue_nonidle_domain+0x9/0x30
ffff82c4c012938b add_entry+0x4b/0xb0
ffff82c4c02223f9 syscall_enter+0xa9/0xae
And gdb output:
(gdb) x/1xh 0x6ae9168b
0x6ae9168b: 0x3024
(gdb) x/1xh 0x6ae9168b
0x6ae9168b: Ignoring packet error, continuing...
Reply contains invalid hex digit 116
The 1st one worked because the p2m.lock is recursive and the PCPU
had not yet changed.
crash reports (for example):
crash> mm_rwlock_t 0xffff83083f913010
struct mm_rwlock_t {
lock = {
raw = {
lock = 2147483647
},
debug = {<No data fields>}
},
unlock_level = 0,
recurse_count = 1,
locker = 1,
locker_function = 0xffff82c4c022c640 <__func__.13514> "__get_gfn_type_access"
}
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Don Slutz <dslutz@verizon.com>
Acked-by: Mukesh Rathor <mukesh.rathor@oracle.com>
This will change initiate_domain_create to walk through NIC definitions
and automatically assign devids to those which have not assigned one.
The devids are needed later in domcreate_launch_dm (for HVM domains
using emulated NICs). The command string for starting the device-model
has those ids as part of its arguments.
Assignment of devids in the hotplug case is handled by libxl_device_nic_add
but that would be called too late in the startup case.
I also moved the call to libxl__device_nic_setdefault here as this seems
to be the only path leading there and avoids doing the loop a third time.
The two loops are trying to handle a case where the caller sets some devids
(not sure that should be valid) but leaves some unset.
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
The errors have been incorrectly identifying their function since c/s
861aef6e15 which did a lot of error handling
cleanup.
Use __func__ to ensure the name remains correct in the future.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
ePAR specifies that if the property "ranges" doesn't exist in a bus node:
"it is assumed that no mapping exists between children of node and the parent
address space".
Modify dt_number_of_address to check if the list of ranges are valid. Return
0 (ie there is zero range) if the list is not valid.
This patch has been tested on the Arndale where the bug can occur with the
'/hdmi' node.
Reported-by: <tsahee@gmx.com>
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
The original code works fine on 64-bit, but on 32-bit, the OCaml int (which is
1 bit smaller than the C int) is likely to overflow.
Signed-off-by: Rob Hoes <rob.hoes@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
This should help to reduce a CPU hotplug race window where a cpu hotplug
event while not be seen by the OS.
When hotplugging more than one vcpu, some of those vcpus might not be
seen as plugged by the guest.
This is what is currently happenning:
1. hw adds cpu, sets GPE.2 bit and sends SCI
2. OSPM gets SCI, reads GPE00.sts and masks GPE.2 bit in GPE00.en
3. OSPM executes _L02 (level-triggered event associate to cpu hotplug)
4. hw adds second cpu and sets GPE.2 bit but SCI is not asserted
since GPE00.en masks event
5. OSPM resets GPE.2 bit in GPE00.sts and umasks it in GPE00.en
as result event for step 4 is lost because step 5 clears it and OS
will not see added second cpu.
ACPI 50 spec: 5.6.4 General-Purpose Event Handling
defines GPE event handling as following:
1. Disables the interrupt source (GPEx_BLK EN bit).
2. If an edge event, clears the status bit.
3. Performs one of the following:
* Dispatches to an ACPI-aware device driver.
* Queues the matching control method for execution.
* Manages a wake event using device _PRW objects.
4. If a level event, clears the status bit.
5. Enables the interrupt source.
So, by using edge-triggered General-Purpose Event instead of a
level-triggered GPE, OSPM is less likely to clear the status bit of the
addition of the second CPU. On step 5, QEMU will resend an interrupt if
the status bit is set.
This description apply also for PCI hotplug since the same step are
followed by QEMU, so we also change the GPE event type for PCI hotplug.
This does not apply to qemu-xen-traditional because it does not resend
an interrupt if necessary as a result of step 5.
Patch and description inspired by SeaBIOS's commit:
Replace level gpe event with edge gpe event for hot-plug handlers
9c6635bd48d39a1d17d0a73df6e577ef6bd0037c
from Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
It is possible that hvm_sr_handlers[typecode].save does not use all
the provided room. Also it can use variable sized records. In both
cases, using:
instance * hvm_sr_handlers[typecode].size
does not select the correct instance. Add code to search for the
correct instance.
Signed-off-by: Don Slutz <dslutz@verizon.com>
Release-acked-by: George Dunlap <george.dunlap@eu.citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
The use of "range" here wasn't really correct - there are no ranges
involved. As the comment in the public header already correctly said,
all this is about is batching of XENMEM_add_to_physmap calls (with
the addition of having a way to specify a foreign domain for
XENMAPSPACE_gmfn_foreign).
Suggested-by: Ian Campbell <Ian.Campbell@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Keir Fraser <keir@xen.org>
Use function copy_to_guest_offset/copy_to_guest directly and check their return
value.
This also fixes CID 1132754, and 1132755:
"Unchecked return value
If the function returns an error value, the error value may be mistaken for a
normal value. In tmem_copy_to_client_buf_offset: Value returned from a function
is not checked for errors before being used (CWE-252)"
And CID 1055125, 1055126, 1055127, 1055128, 1055129, 1055130
"Unchecked return value
If the function returns an error value, the error value may be mistaken for a
normal value. In <functions changed>: Value returned from a function is not
checked for errors before being used (CWE-252)"
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Rename tmem_relinquish_npages() to tmem_flush_npages() to
distinguish it from tmem_relinquish_pages().
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
tmem_ensure_avail_pages() doesn't return a value which is incorrect because
the caller need to confirm whether there is enough memory.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
They are several one line functions in tmem_xen.h which are useless, this patch
embeded them into tmem.c directly.
Also modify void *tmem in struct domain to struct client *tmem_client in order
to make things more straightforward.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Keir Fraser <keir@xen.org>
The two parameters of __tmem_alloc_page() can be reduced.
tmem_called_from_tmem() was also dropped by this patch.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
There are two allocate path for each persistant and ephemeral pool.
This path try to refactor those allocate/free functions with better name and
more readable call layer. Also added more comment.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
tmem_lock_all is used for debug only, remove it from upstream to make
tmem source code more readable and easier to maintain.
And no_evict is meaningless without tmem_lock_all, this patch removes it
also.
This also fixes CID 1055654 Thread deadlock
[ Two threads will be stuck waiting forever if each holds a lock the other needs to acquire.
In alloc_heap_pages: Threads may try to acquire two locks in different orders, potentially
causing deadlock (CWE-833)]
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
It's meaningless to define EXPORT/FORWARD and nobody uses them.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Can use !is_persistent() and !is_shared() to replace them directly.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reorg code logic of do_tmem_put() to make it more readable and clean.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Tmem only takes page as a unit, so parameters tmem_offset, pfn_offset and len in
do_tmem_put/get() are meaningless. All of the callers are using the same
value(tmem_offset=0, pfn_offset=0, len=PAGE_SIZE).
This patch simplifies tmem ignoring those useless parameters and use the default
value directly.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
"SENTINELS" and "DECL_CYC_COUNTER" are hacky code for debugging, there are not
suitable exist in upstream code.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
TMEM_READ/TMEM_WRITE/TMEM_XCHG/TMEM_NEW_PAGE are never used, drop them to make
things simple and clean.
To be clear - we are bit lucky here - as none of the other implementors
of the tmem API are using it (Windows GPLPV code, SLES11, Linux upstream).
The spec says that the operations can return an error code (-ENOSYS for
example) so we are OK doing that.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
An event may still be the tail of a queue even if the queue is now
empty (an 'old tail' event). There is logic to handle the case when
this old tail event needs to be added to the now empty queue (by
checking for q->tail == port).
However, this does not cover all cases.
1. An old tail may be re-added simultaneously with another event.
LINKED is set on the old tail, and the other CPU may misinterpret
this as the old tail still being valid and set LINK instead of
HEAD. All events on this queue will then be lost.
2. If the old tail event on queue A is moved to a different queue B
(by changing its VCPU or priority), the event may then be linked
onto queue B. When another event is linked onto queue A it will
check the old tail, see that it is linked (but on queue B) and
overwrite the LINK field, corrupting both queues.
When an event is linked, save the vcpu id and priority of the queue it
is being linked onto. Use this when linking an event to check if it
is an unlinked old tail event. If it is an old tail event, the old
queue is empty and old_q->tail is invalidated to ensure adding another
event to old_q will update HEAD. The tail is invalidated by setting
it to 0 since the event 0 is never linked.
The old_q->lock is held while setting LINKED to avoid the race with
the test of LINKED in evtchn_fifo_set_link().
Since a event channel may move queues after old_q->lock is acquired,
we must check that we have the correct lock and retry if not. Since
changing VCPUs or priority is expected to be rare events that are
serialized in the guest, we try at most 3 times before dropping the
event. This prevents a malicious guest from repeatedly adjusting
priority to prevent another domain from acquiring old_q->lock.
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
Event channel ports that are reused or that were not in the initial
bucket would have a non-default priority.
Add an init evtchn_port_op hook and use this to set the priority when
an event channel is bound.
Within this new evtchn_fifo_init() call, also check if the event is
already on a queue and print a warning, as this event may have its
first event delivered on a queue with the wrong VCPU or priority.
This guest is expected to prevent this (if it cares) by not unbinding
events that are still linked.
Reported-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>
This too can take an arbitrary amount of time.
In fact, the bulk of the work is being moved to a tasklet, as handling
the necessary preemption logic in line seems close to impossible given
that the teardown may also be invoked on error paths.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
We appear to have somehow missed these. Linux doesn't actually use them and
none of the processors I've looked at actually define any bits in them (so
they are UNK/SBZP) but it is good form to context switch them anyway.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Julien Grall <julien.grall@linaro.org>
Certain AMD systems could have upto 0x10000 ivrs_bdf_entries.
However, the loop variable (bdf) is declared as u16 which causes
inifinite loop when parsing IOMMU event log with IO_PAGE_FAULT event.
This patch changes the variable to u32 instead.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Free the allocated structure rather than the ACPI table ATS entry.
On further analysis, there is another memory leak. acpi_parse_dev_scope()
could allocate scope->devices, and return with -ENOMEM. All callers of
acpi_parse_dev_scope() would then free the underlying structure, loosing the
pointer.
These errors can only actually be reached through acpi_parse_dev_scope()
(which passes type = DMAR_TYPE), but I am quite surprised Coverity didn't spot
it.
Coverity-ID: 1146949
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
It is possible to free the mc_old buffer and then store it for use in the case
of resume.
This keeps the old semantics of being able to return an error even after a
successful microcode application.
Coverity-ID 1146953
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Tweak the logic slightly to return the real errors from
get_iommu_{,msi_}capabilities(), which at the moment is no functional change.
Coverity-ID: 1146950
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
The data type for DT compatible list should be:
const char * const[] __initconst
Fix every serial drivers which support device tree.
Spotted-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Ns16550a devices are Ns16550 devices with additional capabilities.
Decare XEN is compatible with this device, to be able to use unmodified
devicetrees.
Signed-off-by: Tsahee Zidenberg <tsahee@gmx.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Julien Grall <julien.grall@linaro.org>
Provide semantic documentation for how the libxc calls relate to the
hypervisor interface, and how they are to be used.
Also document the bug (present at least in Linux 3.12) that setting
the evtchn fd to nonblocking doesn't in fact make xc_evtchn_pending
nonblocking, and describe the appropriate workaround.
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
CC: Jan Beulich <JBeulich@suse.com>
CC: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Document the event channel protocol in xenstore-paths.markdown, in the
section for ~/device/suspend/event-channel.
Protocol reverse-engineered from commentary and commit messages of
4539594d46 Add facility to get notification of domain suspend ...
17636f47a4 Teach xc_save to use event-channel-based ...
and implementations in
xc_save (current version)
libxl (current version)
linux-2.6.18-xen (mercurial 1241:2993033a77ca)
Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
CC: Keir Fraser <keir@xen.org>
CC: Shriram Rajagopalan <rshriram@cs.ubc.ca>
Jan Beulich
Ian Campbell
Julien Grall
Rob Hoes
David Vrabel
Andrew Cooper
Stefan Bader
Wei Liu
Don Slutz
Anthony PERARD
Bob Liu
Suravee Suthikulpanit
Tsahee Zidenberg
Ian Jackson