The modern ACPI CPU hotplug interface was introduced in the following
series (aa1dd39ca307..679dd1a957df), released in v2.7.0:
1 abd49bc2ed docs: update ACPI CPU hotplug spec with new protocol
2 16bcab97eb pc: piix4/ich9: add 'cpu-hotplug-legacy' property
3 5e1b5d9388 acpi: cpuhp: add CPU devices AML with _STA method
4 ac35f13ba8 pc: acpi: introduce AcpiDeviceIfClass.madt_cpu hook
5 d2238cb678 acpi: cpuhp: implement hot-add parts of CPU hotplug
interface
6 8872c25a26 acpi: cpuhp: implement hot-remove parts of CPU hotplug
interface
7 76623d00ae acpi: cpuhp: add cpu._OST handling
8 679dd1a957 pc: use new CPU hotplug interface since 2.7 machine type
Before patch#1, "docs/specs/acpi_cpu_hotplug.txt" only specified 1-byte
accesses for the hotplug register block. Patch#1 preserved the same
restriction for the legacy register block, but:
- it specified DWORD accesses for some of the modern registers,
- in particular, the switch from the legacy block to the modern block
would require a DWORD write to the *legacy* block.
The latter functionality was then implemented in cpu_status_write()
[hw/acpi/cpu_hotplug.c], in patch#8.
Unfortunately, all DWORD accesses depended on a dormant bug: the one
introduced in earlier commit a014ed07bd ("memory: accept mismatching
sizes in memory_region_access_valid", 2013-05-29); first released in
v1.6.0. Due to commit a014ed07bd, the DWORD accesses to the *legacy*
CPU hotplug register block would work in spite of the above series *not*
relaxing "valid.max_access_size = 1" in "hw/acpi/cpu_hotplug.c":
> static const MemoryRegionOps AcpiCpuHotplug_ops = {
> .read = cpu_status_read,
> .write = cpu_status_write,
> .endianness = DEVICE_LITTLE_ENDIAN,
> .valid = {
> .min_access_size = 1,
> .max_access_size = 1,
> },
> };
Later, in commits e6d0c3ce68 ("acpi: cpuhp: introduce 'Command data 2'
field", 2020-01-22) and ae340aa3d2 ("acpi: cpuhp: spec: add typical
usecases", 2020-01-22), first released in v5.0.0, the modern CPU hotplug
interface (including the documentation) was extended with another DWORD
*read* access, namely to the "Command data 2" register, which would be
important for the guest to confirm whether it managed to switch the
register block from legacy to modern.
This functionality too silently depended on the bug from commit
a014ed07bd.
In commit 5d971f9e67 ('memory: Revert "memory: accept mismatching sizes
in memory_region_access_valid"', 2020-06-26), first released in v5.1.0,
the bug from commit a014ed07bd was fixed (the commit was reverted).
That swiftly exposed the bug in "AcpiCpuHotplug_ops", still present from
the v2.7.0 series quoted at the top -- namely the fact that
"valid.max_access_size = 1" didn't match what the guest was supposed to
do, according to the spec ("docs/specs/acpi_cpu_hotplug.txt").
The symptom is that the "modern interface negotiation protocol"
described in commit ae340aa3d256:
> + Use following steps to detect and enable modern CPU hotplug interface:
> + 1. Store 0x0 to the 'CPU selector' register,
> + attempting to switch to modern mode
> + 2. Store 0x0 to the 'CPU selector' register,
> + to ensure valid selector value
> + 3. Store 0x0 to the 'Command field' register,
> + 4. Read the 'Command data 2' register.
> + If read value is 0x0, the modern interface is enabled.
> + Otherwise legacy or no CPU hotplug interface available
falls apart for the guest: steps 1 and 2 are lost, because they are DWORD
writes; so no switching happens. Step 3 (a single-byte write) is not
lost, but it has no effect; see the condition in cpu_status_write() in
patch#8. And step 4 *misleads* the guest into thinking that the switch
worked: the DWORD read is lost again -- it returns zero to the guest
without ever reaching the device model, so the guest never learns the
switch didn't work.
This means that guest behavior centered on the "Command data 2" register
worked *only* in the v5.0.0 release; it got effectively regressed in
v5.1.0.
To make things *even more* complicated, the breakage was (and remains, as
of today) visible with TCG acceleration only. Commit 5d971f9e67 makes
no difference with KVM acceleration -- the DWORD accesses still work,
despite "valid.max_access_size = 1".
As commit 5d971f9e67 suggests, fix the problem by raising
"valid.max_access_size" to 4 -- the spec now clearly instructs the guest
to perform DWORD accesses to the legacy register block too, for enabling
(and verifying!) the modern block. In order to keep compatibility for the
device model implementation though, set "impl.max_access_size = 1", so
that wide accesses be split before they reach the legacy read/write
handlers, like they always have been on KVM, and like they were on TCG
before 5d971f9e67 (v5.1.0).
Tested with:
- OVMF IA32 + qemu-system-i386, CPU hotplug/hot-unplug with SMM,
intermixed with ACPI S3 suspend/resume, using KVM accel
(regression-test);
- OVMF IA32X64 + qemu-system-x86_64, CPU hotplug/hot-unplug with SMM,
intermixed with ACPI S3 suspend/resume, using KVM accel
(regression-test);
- OVMF IA32 + qemu-system-i386, SMM enabled, using TCG accel; verified the
register block switch and the present/possible CPU counting through the
modern hotplug interface, during OVMF boot (bugfix test);
- I do not have any testcase (guest payload) for regression-testing CPU
hotplug through the *legacy* CPU hotplug register block.
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Ani Sinha <ani@anisinha.ca>
Cc: Ard Biesheuvel <ardb@kernel.org>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: Philippe Mathieu-Daudé <philmd@linaro.org>
Cc: qemu-stable@nongnu.org
Ref: "IO port write width clamping differs between TCG and KVM"
Link: http://mid.mail-archive.com/aaedee84-d3ed-a4f9-21e7-d221a28d1683@redhat.com
Link: https://lists.gnu.org/archive/html/qemu-devel/2023-01/msg00199.html
Reported-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20230105161804.82486-1-lersek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
(cherry picked from commit dab30fbef3896bb652a09d46c37d3f55657cbcbb)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Failure to set pc_succ_insn may result in a TB covering zero bytes,
which triggers an assert within the code generator.
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1224
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20221203175744.151365-1-richard.henderson@linaro.org>
[ Changes by AF:
- Add missing run-plugin-test-noc-% line
]
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
(cherry picked from commit ec2918b467228e7634f1dd5f35033ad3021b6ef7)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
When installing shared libraries, as is the case for libvfio-user.so,
Meson will include relative symbolic links in the output of
"meson introspect --installed":
{
"libvfio-user.so": "/usr/local/lib64/libvfio-user.so",
...
}
In the case of scripts/symlink-install-tree.py, this will
be a symbolic link to a symbolic link but, in any case, there is
no issue in creating it.
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit f32eb0021a85efaca97f69b0e9201737562a8e4f)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
The check semihosting_enabled() wants to know if the guest is
currently in user mode. Unlike the other cases the test was inverted
causing us to block semihosting calls in non-EL0 modes.
Cc: qemu-stable@nongnu.org
Fixes: 19b26317e9 (target/arm: Honour -semihosting-config userspace=on)
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
(cherry picked from commit 9788d4c007cbde7cda1b7a577b8b836335eb2b73)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
It should be the variable rdl2 to revert the already-notified listeners.
Fixes: 2044969f0b ("virtio-mem: Implement RamDiscardManager interface")
Signed-off-by: Chenyi Qiang <chenyi.qiang@intel.com>
Message-Id: <20221228090312.17276-1-chenyi.qiang@intel.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: David Hildenbrand <david@redhat.com>
(cherry picked from commit 29f1b328e3b767cba2661920a8470738469b9e36)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
vmem->bitmap indexes the memory region of the virtio-mem backend at a
granularity of block_size. To calculate the index of target section offset,
the block_size should be divided instead of the bitmap_size.
Fixes: 2044969f0b ("virtio-mem: Implement RamDiscardManager interface")
Signed-off-by: Chenyi Qiang <chenyi.qiang@intel.com>
Message-Id: <20221216062231.11181-1-chenyi.qiang@intel.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: David Hildenbrand <david@redhat.com>
(cherry picked from commit b11cf32e07a2f7ff0d171b89497381a04c9d07e0)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
When vIOMMU is enabled, the vq->used_phys is actually the IOVA not
GPA. So we need to translate it to GPA before the syncing otherwise we
may hit the following crash since IOVA could be out of the scope of
the GPA log size. This could be noted when using virtio-IOMMU with
vhost using 1G memory.
Fixes: c471ad0e9b ("vhost_net: device IOTLB support")
Cc: qemu-stable@nongnu.org
Tested-by: Lei Yang <leiyang@redhat.com>
Reported-by: Yalan Zhang <yalzhang@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20221216033552.77087-1-jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
(cherry picked from commit 345cc1cbcbce2bab00abc2b88338d7d89c702d6b)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
The values in env->flags are a subset of tb->flags.
Restore only the bits that belong.
Cc: qemu-stable@nongnu.org
Fixes: ab419fd8a0 ("target/sh4: Fix TB_FLAG_UNALIGN")
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Message-ID: <20221212011345.GA2235238@roeck-us.net>
[rth: Reduce to only the the superh_cpu_synchronize_from_tb change]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
(cherry picked from commit bc2331635ce18ff068d2bb1e493bc546e1f786e1)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
