Unmapping the virtual VMCB is performed near the end of the domain
destroy procedure but the mapped guest frame prevents domain destroy
from getting to that point. This means guests that call VMRUN cannot
be fully destroyed.
Move the unmap of the virtual VMCB earlier to fix the issue.
Fixes: bcf557675d ("x86: properly use map_domain_page() in nested HVM code")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
master commit: 4d069a84ee
master date: 2026-06-30 09:41:35 +0200
Fix the code which tries to pad the load segment to 2 MiB but only pads
it to a 1 MiB boundary.
This manifests as a page fault while scrubbing RAM during boot on certain
systems.
Xen fails to mark itself as reserved in the E820 (due to spanning multiple
regions), but does restrict the permissions in the directmap. The region of
RAM containing Xen is handed to physical memory manager as available for use,
and scrubbing hit the directmap protections.
Fixes: 4fb075201f ("x86/mkelf32: pad load segment to 2Mb boundary")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
master commit: 5416c455f6
master date: 2026-04-17 18:15:54 +0100
Commit 19ab8356ab ("tools: remove support for running a guest with
qemu-traditional") made passing --dm_version to mk_dsdt mandatory on
x86 but didn't fix all the invocations of it.
The previous default was qemu-xen-traditional so with that removed set
the dm_version to qemu-xen for this invocation.
Without this change, Xen fails to build on x86 when using
--enable-rombios.
Fixes: 19ab8356ab ("tools: remove support for running a guest with qemu-traditional")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
master commit: ea12f9d257
master date: 2026-03-27 18:56:42 +0000
If the config file is builtin, cfg.addr will be zero but Xen
unconditionally calls FreePages() on the address.
Xen may also call FreePages() with a zero address if blexit() is called
after this point since cfg.need_to_free is not set to false.
The UEFI specification does not say whether calling FreePages() with a
zero address is allowed so let's be cautious and use cfg.need_to_free
properly.
Fixes: 8a71d50ed4 ("efi: Enable booting unified hypervisor/kernel/initrd images")
Fixes: 04be2c3a06 ("efi/boot.c: add file.need_to_free")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by Daniel P. Smith <dpsmith@apertussolutions.com>
If the total returned by Xen is more than the number of elements
allocated, it means that the buffer was too small and so the data is
incomplete. Retry to get all the data.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
xc_pm_get_pxstat() requires the caller to allocate the pt and trans_pt
buffers but then calls xc_pm_get_max_px() to determine how big they are
(and hence how much Xen will copy into them). This is susceptible to
races if xc_pm_get_max_px() changes so avoid the problem by requiring
the caller to also pass in the size of the buffers.
Suggested-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
If set_px_pminfo is called a second time with a larger state_count than
the first call, calls to PMSTAT_get_pxstat will read beyond the end of
the pt and trans_pt buffers allocated in cpufreq_statistic_init() since
they would have been allocated with the original state_count.
Secondly, the states array leaks on each subsequent call of
set_px_pminfo.
Fix both these issues by ignoring subsequent calls to set_px_pminfo if
it completed successfully previously. Return success rather than an
error to avoid errors in the dom0 kernel log when reloading the
xen_acpi_processor module.
At the same time, fix a leak of the states array on error.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Check that the total number of states passed in and hence the size of
buffers is sufficient to avoid writing more than the caller has
allocated.
The interface is not explicit about whether getpx.total is expected to
be set by the caller in this case but since it is always set in
libxenctrl it seems reasonable to check it and make it explicit.
Fixes: c06a7db0c5 ("X86 and IA64: Update cpufreq statistic logic for supporting both x86 and ia64")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
In future, some code needs to generate a digest over several separate buffers
so export the necessary functions to do so. Adjust sha2_256_final() to have a
proper digest parameter.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
A lot of systems automatically add microcode to the initramfs so it can
be useful as a vendor policy to always scan for microcode. Add a Kconfig
option to allow setting the default behaviour.
The default behaviour is unchanged since the new option defaults to
"no".
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
The decompression code parses a huffman tree and counts the number of
symbols for a given bit length. In rare cases, there may be >= 256
symbols with a given bit length, causing the unsigned char to overflow.
This causes a decompression failure later when the code tries and fails to
find the bit length for a given symbol.
Since the maximum number of symbols is 258, use unsigned short instead.
Fixes: ab77e81f65 ("x86/dom0: support bzip2 and lzma compressed bzImage payloads")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
In a test, OVMF reported an error initializing the RTC without
indicating the precise nature of the error. The only plausible
explanation I can find is as follows:
As part of the initialization, OVMF reads register C and then reads
register A repatedly until the UIP flag is not set. If this takes longer
than 100 ms, OVMF fails and reports an error. This may happen with the
following sequence of events:
At guest time=0s, rtc_init() calls check_update_timer() which schedules
update_timer for t=(1 - 244us).
At t=1s, the update_timer function happens to have been called >= 244us
late. In the timer callback, it sets the UIP flag and schedules
update_timer2 for t=1s.
Before update_timer2 runs, the guest reads register C which calls
check_update_timer(). check_update_timer() stops the scheduled
update_timer2 and since the guest time is now outside of the update
cycle, it schedules update_timer for t=(2 - 244us).
The UIP flag will therefore be set for a whole second from t=1 to t=2
while the guest repeatedly reads register A waiting for the UIP flag to
clear. Fix it by clearing the UIP flag when scheduling update_timer.
I was able to reproduce this issue with a synthetic test and this
resolves the issue.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Remove Konrad from the livepatch maintainers list as he hasn't been
active for a few years.
At the same time, add Roger as a new maintainer since he has been
actively working on it for a while.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Not displaying INFO messages by default on release builds is not
helpful, as messages of that level are relevant for hypervisor
operation. For example messages related to livepatches applied and
reverted are of INFO level.
Custom builds that require less verbose output can adjust it using the
command line, but attempt to provide all relevant information by
default on release builds.
Adjust the loglevel of printks that don't have an associated level to
INFO instead of WARNING, since INFO will now be printed by default on
all builds.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
The FILE symbols in out-of-tree builds may be either a relative path to
the object dir or an absolute path depending on how the build is
invoked. Fix the paths for C files so that they are consistent with
in-tree builds - the path is relative to the "xen" directory (e.g.
common/irq.c).
This fixes livepatch builds when the original Xen build was out-of-tree
since livepatch-build always does in-tree builds. Note that this doesn't
fix the behaviour for Clang < 6 which always embeds full paths.
Fixes: 7115fa562f ("build: adding out-of-tree support to the xen build")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
This should only be called for the boot CPU to avoid calling _init code
after it has been unloaded.
Fixes: 062868a5a8 ("x86/amd: Work around CLFLUSH ordering on older parts")
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Part of XSA-295 (c/s 863e74eb2c) inadvertently re-positioned the brackets,
changing the logic. If the _set_status() call fails, the grant_map hypercall
would fail with a status of 1 (rc != GNTST_okay) instead of the expected
negative GNTST_* error.
This error path can be taken due to bad guest state, and causes net/blk-back
in Linux to crash.
This is XSA-316.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
The binary diffing algorithm used by xen-livepatch depends on having unique
symbols.
The livepatch loading algorithm used by Xen resolves relocations by symbol
name, and thus also depends on having unique symbols.
Introduce CONFIG_ENFORCE_UNIQUE_SYMBOLS to control failing the build if
duplicate symbols are found, and disable it in the RANDCONFIG build.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Release-acked-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Make handling in do_pm_op() more homogeneous: Before interpreting
op->cpuid as such, handle all operations not acting on a particular
CPU. Also expose the setting via xenpm.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Allow limiting the max C-state sub-state by appending to the max_cstate
command-line parameter. E.g. max_cstate=1,0
The limit only applies to the highest legal C-state. For example:
max_cstate = 1, max_csubstate = 0 ==> C0, C1 okay, but not C1E
max_cstate = 1, max_csubstate = 1 ==> C0, C1 and C1E okay, but not C2
max_cstate = 2, max_csubstate = 0 ==> C0, C1, C1E, C2 okay, but not C3
max_cstate = 2, max_csubstate = 1 ==> C0, C1, C1E, C2 okay, but not C3
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
When EFI booting the Dell PowerEdge R540 it consistently wanders into
the weeds and gets an invalid opcode in the EFI ResetSystem call. This
is the same bug which affects the PowerEdge R740 so fix it in the same
way: quirk this hardware to use the ACPI reboot method instead.
BIOS Information
Vendor: Dell Inc.
Version: 1.3.7
Release Date: 02/09/2018
System Information
Manufacturer: Dell Inc.
Product Name: PowerEdge R540
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
dm_op() fails with -EFAULT if the struct xen_dm_op given by the guest is
smaller than Xen's struct xen_dm_op. This is a problem because DMOP is
meant to be a stable ABI but it breaks whenever the size of struct
xen_dm_op changes.
To fix this, change how the copying to and from the guest is done. When
copying from the guest, first copy the header and inspect the op. Then,
only copy the correct amount needed for that op. When copying to the
guest, don't copy the header. Rather, copy only the correct amount
needed for that particular op.
So now the dm_op() will fail if the guest does not supply enough bytes
for the specific op. It will not fail if the guest supplies too many
bytes for the specific op, but Xen will not copy the extra bytes.
Remove some now unused macros and helper functions.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Remove the implementation of XEN_DOMCTL_pin_mem_cacheattr since it has
been replaced by a dmop. Change xc_domain_pin_memory_cacheattr() so
that it is only defined when XC_WANT_COMPAT_DEVICEMODEL_API is set and
have it call the new dmop. Leave the definitions of
XEN_DOMCTL_MEM_CACHEATTR_* since they are still used by QEMU.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Daniel De Graaf <dgdegra@tycho.nsa.gov>
Reviewed-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Provide XEN_DMOP_pin_memory_cacheattr to allow a deprivileged QEMU to
pin the caching type of RAM after moving the VRAM. It is equivalent to
XEN_DOMCTL_pin_memory_cacheattr.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Provide XEN_DMOP_relocate_memory, a limited version of
XENMEM_add_to_physmap to allow a deprivileged QEMU to move VRAM when a
guest programs its BAR. It is equivalent to XENMEM_add_to_physmap with
space == XENMAPSPACE_gmfn_range.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Paul Durrant <paul.durrant@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Implement support for restricting evtchn handles to a particular domain
on Linux by calling the IOCTL_EVTCHN_RESTRICT_DOMID ioctl (support added
in Linux v4.8).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Release-acked-by: Julien Grall <julien.grall@linaro.org>
Return 0 to indicate success rather than whatever rc was previously set
to (-EINVAL).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
After rombios transfers the data for a write, it checks the status and
fails if BSY is set. qemu-trad doesn't set BSY for PIO writes, but QEMU
upstream does, and this causes rombios to fail writes because they are
marked as BSY. Instead, wait for BSY to clear after a write.
INT 13 writes are probably rarely used these days, but they are used by
GRUB 2 to write to its environment file which happens by default on
Ubuntu.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Occasionally, on certain Broadwell CPUs MSR_IA32_LASTINTTOIP has been
observed to have the top three bits corrupted as though the MSR is using
the LBR_FORMAT_EIP_FLAGS_TSX format. This is incorrect and causes a
vmentry failure -- the MSR should contain an offset into the current
code segment. This is assumed to be erratum BDF14. Workaround the issue
by sign-extending into bits 48:63 for MSR_IA32_LASTINT{FROM,TO}IP.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Move microcode loading earlier for the boot CPU and secondary CPUs so
that it takes place before identify_cpu() is called for each CPU.
Without this, the detected features may be wrong if the new microcode
loading adjusts the feature bits. That could mean that some fixes (e.g.
d6e9f8d4f3 ("x86/vmx: fix vmentry failure with TSX bits in LBR"))
don't work as expected.
Previously during boot, the microcode loader was invoked for each
secondary CPU started and then again for each CPU as part of an
initcall. Simplify the code so that it is invoked exactly once for each
CPU during boot.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Tested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Release-acked-by: Julien Grall <julien.grall@arm.com>
Instead of using a locking order based on line numbers which interacts
poorly with trying to create a live patch, statically define the locking
order.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Dario Faggioli <dario.faggioli@citrix.com>
Acked-by: George Dunlap <george.dunlap@citrix.com>
Exit with 0 for success.
Exit with 1 for an error.
Exit with 2 if the operation should be retried for any reason (e.g. a
timeout or because another operation was in progress).
This allows a program or script driving xen-livepatch to determine if
the operation should be retried without parsing the output.
Acked-by: Wei Liu <wei.liu2@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
The default timeout in the hypervisor for a livepatch operation is 30 ms,
but xen-livepatch currently waits for up to 30 seconds for the operation
to complete. Instead, remove the retry loop and simply wait for 2 * 30 ms
for the operation to complete. The extra period is to account for the
time to actually start the operation.
Furthermore, have xen-livepatch set the hypervisor timeout rather than
relying on the hypervisor default since the tool doesn't know how long
it will be. Use nanosleep rather than usleep since usleep has been
removed from POSIX.1-2008.
Acked-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
The hypervisor expects the timeout from the hypercall to be in
nanoseconds, so document this correctly. Also correctly document
what happens when timeout is set to zero.
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Improving the output of xen-livepatch, which is currently hard to read,
especially when an error occurs.
Some examples of the changes:
Before:
$ xen-livepatch apply test
Performing apply:. completed
After:
$ xen-livepatch apply test
Applying test:. completed
Before:
$ xen-livepatch apply test2
test2 failed with 22(Invalid argument)
Performing apply: (no newline)
After:
$ xen-livepatch apply test2
Applying test2: failed
Error 22: Invalid argument
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Using both stdout and stderr interleaved without newlines can result in
strange output when using line buffered mode (e.g. a terminal) or when
fully buffered (e.g. redirected to a file). Set stdout to unbuffered mode
to fix this (stderr is always unbuffered by default).
Acked-by: Wei Liu <wei.liu2@citrix.com>
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Somewhat confusingly, before the action has been executed the patch is
expected to be in the "allow" state, not the "expected" state. The
check for this was correct but the subsequent error message was not.
Fix the error message to show this state correctly.
Before:
$ xen-livepatch unload test
test: in wrong state (APPLIED), expected (unknown)
After:
$ xen-livepatch unload test
test: in wrong state (APPLIED), expected (CHECKED)
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
When EFI booting the Dell OptiPlex 9020, it sometimes GP faults in the
EFI runtime instead of rebooting. Quirk this hardware to use the ACPI
reboot method instead.
dmidecode info:
BIOS Information
Vendor: Dell Inc.
Version: A15
Release Date: 11/08/2015
System Information
Manufacturer: Dell Inc.
Product Name: OptiPlex 9020
Version: 00
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Add hook functions which run during patch apply and patch revert.
Hook functions are used by livepatch payloads to manipulate data
structures during patching, etc.
One use case is the XSA91. As Martin mentions it:
"If we have shadow variables, we also need an unload hook to garbage
collect all the variables introduced by a hotpatch to prevent memory
leaks. Potentially, we also want to pre-reserve memory for static or
existing dynamic objects in the load-hook instead of on the fly.
For testing and debugging, various applications are possible.
In general, the hooks provide flexibility when having to deal with
unforeseen cases, but their application should be rarely required (<
10%)."
Furthermore include a test-case for it.
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
When loading patches, the code prevents loading a patch containing a new
symbol that duplicates a symbol from another loaded patch. However, the
check should also prevent loading a new symbol that duplicates a symbol
from the core hypervisor.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Add support for applying alternative sections within xsplice payload.
At payload load time, apply an alternative sections that are found.
Also we add an test-case exercising a rather useless alternative
(patching a NOP with a NOP) - but it does exercise the code-path.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Add support for exception tables contained within xSplice payloads. If an
exception occurs search either the main exception table or a particular
active payload's exception table depending on the instruction pointer.
Also we add an test-case to make sure we have an exception that
is handled.
To not grow the code-base if xSplice is not compiled in we add
certain #define to help in determining if code needs to be __init
or not.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Add support for handling bug frames contained with xsplice modules. If a
trap occurs search either the kernel bug table or an applied payload's
bug table depending on the instruction pointer.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Naturally the backtrace is presented when an instruction
hits an bug_frame or %p is used.
The payloads do not support bug_frames yet - however the functions
the payloads call could hit an BUG() or WARN().
The traps.c has logic to scan for it this - and eventually it will
find the correct bug_frame and the walk the stack using %p to print
the backtrace. For %p and symbols to print a string - the
'is_active_kernel_text' is consulted which uses an 'struct virtual_region'.
Therefore we register our start->end addresses so that
'is_active_kernel_text' will include our payload address.
We also register our symbol lookup table function so that it can
scan the list of payloads and retrieve the correct name.
Lastly we change vsprintf to take into account s and namebuf.
For core code they are the same, but for payloads they are different.
This gets us:
Xen call trace:
[<ffff82d080a00041>] revert_hook+0x31/0x35 [xen_hello_world]
[<ffff82d0801431bd>] xsplice.c#revert_payload+0x86/0xc6
[<ffff82d080143502>] check_for_xsplice_work+0x233/0x3cd
[<ffff82d08017a0b2>] domain.c#continue_idle_domain+0x9/0x1f
Which is great if payloads have similar or same symbol names.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
If in the payload we do not have the old_addr we can resolve
the virtual address based on the UNDEFined symbols.
We also use an boolean flag: new_symbol to track symbols. The usual
case this is used is by:
* A payload may introduce a new symbol
* A payload may override an existing symbol (introduced in Xen or another
payload)
* Overriding symbols must exist in the symtab for backtraces.
* A payload must always link against the object which defines the new symbol.
Considering that payloads may be loaded in any order it would be incorrect to
link against a payload which simply overrides a symbol because you could end
up with a chain of jumps which is inefficient and may result in the expected
function not being executed.
Since the payload we get is an relocatable image (partial linked ELF file)
we have to match up the symbols. We follow the ELF visibility rules for that
and for local symbols do what bintutils ld does.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Implement support for the apply, revert and replace actions.
To perform and action on a payload, the hypercall sets up a data
structure to schedule the work. A hook is added in the reset_stack_and_jump
to check for work and execute it if needed (specifically we check an
per-cpu flag to make this as quick as possible).
In this way, patches can be applied with all CPUs idle and without
stacks. The first CPU to run check_for_xsplice_work() becomes the
master and triggers a reschedule softirq to trigger all the other CPUs
to enter check_for_xsplice_work() with no stack. Once all CPUs
have rendezvoused, all CPUs disable their IRQs and NMIs are ignored.
The system is then quiscient and the master performs the action.
After this, all CPUs enable IRQs and NMIs are re-enabled.
Note that it is unsafe to patch do_nmi and the xSplice internal functions.
Patching functions on NMI/MCE path is liable to end in disaster on x86.
This is not addressed in this patch and is mentioned in the
design doc as a further TODO.
The action to perform is one of:
- APPLY: For each function in the module, store the first arch-specific
number bytes of the old function and replace it with a jump to the
new function. (on x86 it is 5 bytes, on ARM it will likey be 4 bytes).
- REVERT: Copy the previously stored bytes into the first arch-specific
number of bytes of the old function (again, 5 bytes on x86).
- REPLACE: Revert each applied module and then apply the new module.
To prevent a deadlock with any other barrier in the system, the master
will wait for up to 30ms before timing out.
Measurements found that the patch application to take about 100 μs on a
72 CPU system, whether idle or fully loaded.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Julien Grall <julien.grall@arm.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Add support for loading xsplice payloads. This is somewhat similar to
the Linux kernel module loader, implementing the following steps:
- Verify the elf file.
- Parse the elf file.
- Allocate a region of memory mapped within a free area of
[xen_virt_end, XEN_VIRT_END].
- Copy allocated sections into the new region. Split them in three
regions - .text, .data, and .rodata. MUST have at least .text.
- Resolve section symbols. All other symbols must be absolute addresses.
(Note that patch titled "xsplice,symbols: Implement symbol name resolution
on address" implements that)
- Perform relocations.
- Secure the the regions (.text,.data,.rodata) with proper permissions.
We capitalize on the vmalloc callback API (see patch titled:
"rm/x86/vmap: Add v[z|m]alloc_xen, and vm_init_type") to allocate
a region of memory within the [xen_virt_end, XEN_VIRT_END] for the code.
We also use the "x86/mm: Introduce modify_xen_mappings()"
to change the virtual address page-table permissions.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Julien Grall <julien.grall@arm.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
Add Elf routines and data structures in preparation for loading an
xSplice payload.
We make an assumption that the max number of sections an ELF payload
can have is 64. We can in future make this be dependent on the
names of the sections and verifying against a list, but for right now
this suffices.
Also we a whole lot of checks to make sure that the ELF payload
file is not corrupted nor that the offsets point past the file.
For most of the checks we print an message if the hypervisor is built
with debug enabled.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Release-acked-by: Wei Liu <wei.liu2@citrix.com>
GCC generates R_X86_64_64, R_X86_64_PC32, and R_X86_64_PLT32
relocations so those are the ones we need initially
to support xSplice.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Commit a929bee0e6 ("x86/vmx: Fix injection of #DB traps following
XSA-156") prevents an infinite loop in certain #DB traps. However, it
changed the behavior to not call hvm_hw_inject_trap() for #DB and #AC
traps which which means that the debug registers are not restored
correctly and nullified commit b56ae5b48c ("VMX: fix/adjust trap
injection").
To fix this, restore the original code path through hvm_inject_trap(),
but ensure that the struct hvm_trap is populated with all the required
data.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Kevin Tian <kevin.tian@intel.com>
Commit fc0c3fa2ad ("x86/IO-APIC: fix setup of Xen internally used IRQs
(take 2)") introduced a regression on some hardware where Xen would hang
during shutdown, repeating the following message:
APIC error on CPU0: 08(08), Receive accept error
This appears to be because an interrupt (in this case from the serial
console) destined for a CPU other than the boot CPU is left unhandled so
an APIC error on CPU 0 is generated instead.
To fix this, before taking down the non-boot CPUs, call fixup_irqs()
with a CPU mask of only the boot CPU to reset the IRQ affinities
correctly.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Previously, xenconsoled's daemonize function would do nothing if its
parent process is init (as it is under systemd but not sysv init).
This is confusing. Instead, always daemonize when asked to, but use the
"interactive" switch when running from the systemd service.
Because a pidfile is only written when daemonizing, drop the pidfile
parameters from the service file (systemd keeps track of the pids
anyway).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Since commit 191b3f3344 ("p2m/ept: enable PML in p2m-ept for
log-dirty"), the A and D bits of EPT paging entries are set
unconditionally, regardless of whether PML is enabled or not. This
causes a regression in Xen 4.6 on some processors due to Intel Errata
AVR41 -- HVM guests get severe memory corruption when the A bit is set
due to incorrect TLB flushing on mov to cr3. The erratum affects the
Atom C2000 family (Avoton).
To fix, do not set the A bit on this processor family.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Move feature suppression to feature detection code. Add command line
override.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Tested-by: Andrew Cooper <andrew.cooper3@citrix.com>
This contains the event machinery and state machines to write
non-checkpointed migration v2 stream (with the exception of the
xc_domain_save() handling which is spliced later in a bisectable way).
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
CC: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
---
v4: Don't make _init() mandatory, use FILLZERO(), adjust stream_success()
v3: Similar changes to the read side, following the read side review.
This contains the event machinery and state machines to read an act on
a non-checkpointed migration v2 stream (with the exception of the
xc_domain_restore() handling which is spliced later in a bisectable
way).
It also contains some boilerplate to help support checkpointed
streams, which shall be introduced in a later patch.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
CC: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
---
v4:
* Don't make _init() mandatory
* Don't unilaterally clobber rc from datacopier callback
* Comment adjustments
* More assertions
v3: More descriptions of the internal state, shuffle function order, add
an _init() call, condense error handling
The current code for allocating memory for PV guests batches the
hypercalls to allocate memory by allocating 1024*1024 extents of order 0
at a time. To make this faster, first try allocating extents of order 9
(2 MiB) before falling back to the order 0 allocating if the order 9
allocation fails.
On my test machine this reduced the time to start a 128 GiB PV guest by
about 60 seconds.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Constructing dom0 may take a few seconds, particularly if the slow VESA
graphics terminal is used. Process pending softirqs a few times to avoid
tripping a watchdog with a short timeout.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Move inclusion of xen/softirq.h (and at once clean up other includes).
Signed-off-by: Jan Beulich <jbeulich@suse.com>
After the first call to ExitBootServices(), avoid calling any boot
services (except GetMemoryMap() and ExitBootServices()) by setting
setting efi_bs to NULL and halting in blexit(). Only GetMemoryMap() and
ExitBootServices() are explicitly allowed to be called after the first
call to ExitBootServices() and so are are called via
SystemTable->BootServices.
Suggested-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
If calling ExitBootServices() fails, the required memory map size may
have increased. When initially allocating the memory map, allocate a
slightly larger buffer (by an arbitrary 8 entries) to fix this.
The ARM code path was already allocating a larger buffer than required,
so this moves the code to be common for all architectures.
This was seen on the following machine when using the iscsidxe UEFI
driver. The machine would consistently fail the first call to
ExitBootServices().
System Information
Manufacturer: Supermicro
Product Name: X10SLE-F/HF
BIOS Information
Vendor: American Megatrends Inc.
Version: 2.00
Release Date: 04/24/2014
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Roy Franz <roy.franz@linaro.org>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
When an interrupt is received during an EFI runtime service call, Xen
may call map_domain_page() while using the EFI runtime page tables.
This fails because, although the EFI runtime page tables are a
copy of the idle domain's page tables, current points at a different
domain's vCPU.
To fix this, return NULL from mapcache_current_vcpu() when using the EFI
runtime page tables which is treated equivalently to running in an idle
vCPU.
This issue can be reproduced by repeatedly calling GetVariable() from
dom0 while using VT-d, since VT-d frequently maps a page from interrupt
context.
Example call trace:
[<ffff82d0801615dc>] __find_next_zero_bit+0x28/0x60
[<ffff82d08016a10e>] map_domain_page+0x4c6/0x4eb
[<ffff82d080156ae6>] map_vtd_domain_page+0xd/0xf
[<ffff82d08015533a>] msi_msg_read_remap_rte+0xe3/0x1d8
[<ffff82d08014e516>] iommu_read_msi_from_ire+0x31/0x34
[<ffff82d08016ff6c>] set_msi_affinity+0x134/0x17a
[<ffff82d0801737b5>] move_masked_irq+0x5c/0x98
[<ffff82d080173816>] move_native_irq+0x25/0x36
[<ffff82d08016ffcb>] ack_nonmaskable_msi_irq+0x19/0x20
[<ffff82d08016ffdb>] ack_maskable_msi_irq+0x9/0x37
[<ffff82d080173e8b>] do_IRQ+0x251/0x635
[<ffff82d080234502>] common_interrupt+0x62/0x70
[<00000000df7ed2be>] 00000000df7ed2be
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Some EFI firmware implementations may place the SMBIOS table in RAM
marked as BootServicesData, which Xen does not consider as reserved.
When dom0 tries to access the SMBIOS, the region is not contained in the
initial P2M and it crashes with a page fault. To fix this, reserve the
SMBIOS region.
Also, fix the memcmp checks for existence of the SMBIOS.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
6d896e1357 ("tools/libxl: Extend datacopier to support reading into a
buffer") changed the semantics of the errnoval parameter for the datacopier
callback without updating all callers. Restore the original behavior for now.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Currently a datacopier may source its data from an fd or local buffer, but its
destination must be an fd. For migration v2, libxl needs to read from the
migration stream into a local buffer.
Implement a "read into local buffer" mode, invoked when readbuf is set and
writefd is -1. On success, the callback passes the number of bytes read.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
[Rewrite commit message]
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
CC: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Currently, a datacopier will unconditionally read until EOF on its read fd.
For migration v2, libxl needs to read records of a specific length out of the
migration stream, without reading any further data.
Introduce a parameter, bytes_to_read, which may be used to stop the datacopier
ahead of reaching EOF. If bytes_to_read is set to -1, then the datacopier will
read until EOF.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
[Rewrite commit message]
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
CC: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Wei Liu <wei.liu2@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
An individual datacopier_buf contains a static buffer of 1000 bytes.
Attempting to add prefixdata of more than 1000 bytes would overrun the buffer
and cause heap corruption.
Instead, split the prefixdata and chain together multiple datacopier buffers.
This allows for an arbitrary quantity of prefixdata to be added to a
datacopier.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
CC: Ian Campbell <Ian.Campbell@citrix.com>
CC: Ian Jackson <Ian.Jackson@eu.citrix.com>
CC: Wei Liu <wei.liu2@citrix.com>
If reboot= is specified on the command-line, don't apply reboot quirks
to allow the command-line option to take precedence.
This is a port of Linux commit 5955633e91bf ("x86/reboot: Skip DMI
checks if reboot set by user").
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Leverage (and make apply on top of) c643fb110a ("x86/EFI: allow
reboot= overrides when running under EFI").
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Copy the entire output buffer to the guest because some firmwares update
size on successful calls (contrary to the spec) and the buffer may
contain data beyond the output size that the firmware requires on a
subsequent GetNextVariableName() call (e.g. a NULL character).
Note that this shouldn't change the amount of data copied because on success, a
compliant firmware does not change size and so the entire buffer is copied
anyway. If size is changed, Xen does not copy the buffer.
Without this change, the following (simplified) sequence would occur:
GetNextVariableName: in \0, size 1024 || out AdminPw\0, size 7
GetNextVariableName: in AdminPw\0, size 1024 || out UserPw\0, size 6
GetNextVariableName: in UserPww\0, size 1024 || NOT FOUND
This was seen on an Intel S1200RP_SE with firmware
S1200RP.86B.02.02.0005.102320140911, version 4.6, date 2014-10-23.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Don't kill xenstored as part of the usual service shutdown process to
prevent hangs on shutdown where the kernel tries to unplug a VIF
after xenstored has exited.
In an ideal case with all guests cooperating, xendomains will have shut
down all guests before xenstored is killed.
However in the uncooperative case, malicious or crashed guests may still
be running after xendomains has exited and this should not block the
shutdown/reboot of dom0.
Xenstored has no state to sync to disk, and never used to be killed in
the sysvinit case; observe the warning in xencommons. Our testing has
shown regressions caused by the change in behaviour between sysvinit and
systemd when it comes to killing xenstored.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
[ ijc -- added systemd to title ]
Change NMI processing so that if watchdog=force is passed on the
command-line and the NMI is not caused by a perf counter overflow (i.e.
likely not a watchdog "tick"), the NMI is handled by the unknown NMI
handler.
This allows injection of NMIs from IPMI controllers that don't set the
IOCK/SERR bits to trigger the unknown NMI handler rather than be
ignored.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Fix command line parsing (don't enable the watchdog on e.g.
"watchdog=xyz").
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Use the C-state's type when tracing, not its index since the index is
not set by the mwait_idle driver.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
The existing mechanism of using interrupt frequency as a heuristic does
not work well for certain workloads. As an example, synchronous dd on a
small block size uses deep C-states because much of the time is spent
doing processing so the interrupt frequency is not too high, but when an
IOP is submitted, the interrupt occurs soon after going idle. This
causes exit latency to be a significant factor.
To fix this, add a new factor which limits the exit latency to be no
more than 10% of the decaying measured idle time. This improves
performance for workloads with a medium interrupt frequency but a short
idle duration.
In the workload given previously, throughput improves by 20% with this
patch.
This is not ported from the Linux menu governor since that uses load
average and number of IO wait processes to satisfy latency constraints.
If a process is in IO wait state, it compares the exit latency with the
predicted residency reduced by a factor of 10, which is somewhat similar
to what this patch does.
A side effect of this patch is to correctly limit the maximum idle time
used in the correction factor calculation. Previously data->measured_us
was used, and it was never set.
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Program the timer to the deadline of the closest timer if it is further
than 50us ahead, otherwise set it 50us ahead. This way a single event
fires on time rather than 50us late (as it would have previously) while
still preventing too many timer wakeups in the case of having many
timers scheduled close together.
(where 50us is the timer_slop)
Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Ross Lagerwall