Description:
For coroutine live patch, we need find all coroutines stack and check them
before patching. There is no structure to manage all coroutines in qemu. So we
add a list which contain all running coroutines to accelerate libcare live
patch.
Signed-off-by: jiang-dawei15 <jiangdawei15@huawei.com>
Signed-off-by: yezengruan <yezengruan@huawei.com>
This option changes the thread local storage (TLS) model. Thread-local storage
is a mechanism by which variables are allocated in a way that causes one instance
of the variable per extant thread.
i.global-dynamic
Generates a generic TLS code. The code can be used everywhere and the code can access
variables defined anywhere else. This setting causes the largest size code to be generated
and uses the most run time to produce.
ii.local-dynamic
Generates an optimized TLS code. To use this setting, the thread-local variables must be
defined in the same object in which they are referenced.
iii.initial-exec
Generates a restrictive, optimized TLS code. To use this setting, the thread-local variables
accessed must be defined in one of the modules available to the program.
iv.local-exec
Generates the most restrictive TLS code. To use this setting, the thread-local variables
must be defined in the executable.
Optimize qemu cflags with '-ftls-model=initial-exec' which means we use initial-exec
mode.
The virtiofsd currently crashes when used with glibc 2.35.
That is due to the rseq system call being added to every thread
creation [1][2].
[1]: https://www.efficios.com/blog/2019/02/08/linux-restartable-sequences/
[2]: https://sourceware.org/pipermail/libc-alpha/2022-February/136040.html
This happens not at daemon start, but when a guest connects
/usr/lib/qemu/virtiofsd -f --socket-path=/tmp/testvfsd -o sandbox=chroot \
-o source=/var/guests/j-virtiofs --socket-group=kvm
virtio_session_mount: Waiting for vhost-user socket connection...
# start ok, now guest will connect
virtio_session_mount: Received vhost-user socket connection
virtio_loop: Entry
fv_queue_set_started: qidx=0 started=1
fv_queue_set_started: qidx=1 started=1
Bad system call (core dumped)
We have to put rseq on the seccomp allowlist to avoid that the daemon
is crashing in this case.
Reported-by: Michael Hudson-Doyle <michael.hudson@canonical.com>
Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-id: 20220209111456.3328420-1-christian.ehrhardt@canonical.com
[Moved rseq to its alphabetically ordered position in the seccomp
allowlist.
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: qinyu <qinyu16@huawei.com>
Fix commit 391dd8f1("scsi-bus: Refactor the code that retries requests"),
which split scsi_dma_restart_bh(), but the object_unref() belongs to
scsi_dma_restart_bh().
So, we should mv object_unref() from scsi_retry_requests() to
scsi_dma_restart_bh().
Signed-off-by: Yan Wang <wangyan122@huawei.com>
pl011-reset-read-FIFO-when-UARTTIMSC-0-UARTICR-0xfff.patch
qcow2-fix-memory-leak-in-qcow2_read_extensions.patch
scsi-disk-define-props-in-scsi_block_disk-to-avoid-m.patch
pcie-Add-pcie-root-port-fast-plug-unplug-feature.patch
pcie-Compat-with-devices-which-do-not-support-Link-W.patch
Signed-off-by: Yan Wang <wangyan122@huawei.com>
We hack into PCI_EXP_LNKCAP to support device fast plug/unplug
for pcie-root-port. However some devices like ioh3420 does not
suport it, so PCI_EXP_LNKCAP is not set for such devices.
Signed-off-by: Ying Fang <fangying1@huawei.com>
Signed-off-by: Yan Wang <wangyan122@huawei.com>
If a device is plugged in the pcie-root-port when VM kernel is
booting, the kernel may wrongly disable the device.
This bug was brought in by two patches of the linux kernel:
https://patchwork.kernel.org/patch/10575355/https://patchwork.kernel.org/patch/10766219/
VM runtime like kata uses this feature to boot microVM,
so we must fix it up. We hack into the pcie native hotplug
patch so that hotplug/unplug will work under this circumstance.
Signed-off-by: Ying Fang <fangying1@huawei.com>
Signed-off-by: Yan Wang <wangyan122@huawei.com>
scsi_block_realize() use scsi_realize() to init some props, but
these props is not defined in scsi_block_disk_properties, so they will
not be freed.
This patch defines these prop in scsi_block_disk_properties to avoid memleaks.
Signed-off-by: Pan Nengyuan <pannengyuan@huawei.com>
Signed-off-by: Yan Wang <wangyan122@huawei.com>
We can enable ACPI when AArch64 Linux is booted with QEMU and UEFI (AAVMF).
When VM is booting and the SBSA driver has not initialized, writting data
that exceds 32 bytes will cause the read FIFO full and proceeding data will
be lost. The searil port appears to be stuck in this abnormal situation.
A hack to reset read FIFO when UARTTIMSC=0 & UARTICR=0xffff appears to
resolve the issue.
The question is fully discussed at
https://www.spinics.net/lists/linux-serial/msg23163.html
Signed-off-by: Haibin Wang <wanghaibin.wang@huawei.com>
Reviewed-by: Shannon Zhao <shannon.zhaosl@gmail.com>
Reviewed-by: Ying Fang <fangying1@huawei.com>
Signed-off-by: Yan Wang <wangyan122@huawei.com>
acpi/madt: Factor out the building of MADT GICC struct
hw/arm/virt: Assign virt_madt_cpu_entry to acpi_ged madt_cpu hook
arm/virt/acpi: Factor out CPPC building from DSDT CPU aml
acpi/cpu: Prepare build_cpus_aml for arm virt
acpi/ged: Extend ACPI GED to support CPU hotplug
arm/cpu: assign arm_get_arch_id handler to get_arch_id hook
tests/acpi/bios-tables-test: Allow changes to virt/DSDT file
arm/virt: Attach ACPI CPU hotplug support to virt
tests/acpi/bios-table-test: Update expected virt/DSDT file
arm/virt: Add CPU hotplug framework
arm/virt: Add CPU topology support
test/numa: Adjust aarch64 numa test
hw/arm/virt: Factor out some CPU init codes to pre_plug hook
hw/arm/boot: Add manually register and trigger of CPU reset
arm/virt/gic: Construct irqs connection from create_gic
intc/gicv3_common: Factor out arm_gicv3_common_cpu_realize
intc/gicv3_cpuif: Factor out gicv3_init_one_cpuif
intc/kvm_gicv3: Factor out kvm_arm_gicv3_cpu_realize
hw/intc/gicv3: Add CPU hotplug realize hook
accel/kvm: Add pre-park vCPU support
intc/gicv3: Add pre-sizing capability to GICv3
acpi/madt: Add pre-sizing capability to MADT GICC struct
arm/virt: Add cpu_hotplug_enabled field
arm/virt/acpi: Extend cpufreq to support max_cpus
arm/virt: Pre-sizing MADT-GICC GICv3 and Pre-park KVM vCPU
arm/virt: Start up CPU hot-plug and cold-plug
Signed-off-by: Chen Qun<kuhn.chenqun@huawei.com>
(cherry picked from commit 9390aa07e450024769ceb3bb3ab481af8911a77b)
All the CPU hotplug facilities are ready. Assemble them
to start up CPU hot-plug capability for arm/virt.
This also adds CPU cold plug support to arm virt machine
board. CPU cold plug means adding CPU by using "-device
xx-arm-cpu" when we bring up Qemu.
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
(cherry picked from commit d5d118431a64c4202cb28c27bdea0f8667beb5dd)
Establish all pre-sizing facilities based on cpu_hotplug_enabled
field.
Note: PPTT is constructed for possible_cpus, so it does not need
to pre-sizing it.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit c01cf3e56182eea0b0a469a91ccbfcaf37c4c5e6)
We will support CPU hotplug soon, so extend memory region size to
allow hotplugged CPU access cpufreq space.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
(cherry picked from commit 53c60486a60b76be9e2afe04e53759cf5475dd57)
Some conditions must be satisfied to support CPU hotplug, including
ACPI, GED, 64bit CPU, GICv3.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 7550d4ab290a6c114a528b867aacdf80207e211d)
The count of possible CPUs is exposed to guest through the count
of MADT GICC struct, so we should pre-sizing MADT GICC too.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 101bb4ff32325cdc374719f006b6c5b24ff63d0d)
Currently GICv3 supports fixed smp_cpus CPUs, and all CPUs are
present always. Now we want to pre-sizing GICv3 to support max_cpus
CPUs and not all of them are present always, so some sizing codes
should be concerned.
GIC irqs, GICR and GICC are pre-created for all possible CPUs at
start, but only smp_cpus CPUs are realize and irqs of smp_cpus CPUs
are connected.
Other code changes are mainly for arm_gicv3, and we do little about
kvm_arm_gicv3 becasue KVM will deal with the sizing information properly.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit cd6ddaa7a22579bd3caaaf323201cad55e1d1407)
For that KVM do not support dynamic adjustment of vCPU count,
we must pre-park all possible vCPU at start.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 000c4a3d545146ee38ea4334fee3a93669c11b28)
GICv3 exposes individual CPU realization capability through
this hook. It will be used for hotplugged CPU.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit e376844ee4fefdb6ffabb751d4d08363bc1a4faf)
The CPU object of hotplugged CPU will be defer-created (during
hotplug session), so we must factor out realization code to let
it can be applied to individual CPU.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 7b0dc0a49e5ea43bd4b4d85d3630db2efcbb493b)
The CPU object of hotplugged CPU will be defer-created (during
hotplug session), so we must factor out some code to let it can
be applied to individual CPU.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 0583d886a2bfa994a8be64d678a54c333cc5377d)
The CPU object of hotplugged CPU will be defer-created (during
hotplug session), so we must factor out realization code to let
it can be applied to individual CPU.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 7e5f5b7ff26bc48205f47d8f29477a8587c71446)
Make the irqs can be connected to for individual CPU.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 6161210e01ad1c757e44b0a31c3424f770c9281a)
We need to register and trigger CPU reset manually for hotplugged
CPU. Besides, we gather CPU reset handlers of all CPUs because CPU
reset should happen before GIC reset.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit e38ab50a87d6f581d282e606ac12db5e8238c1e1)
The init path of hotplugged CPU is pre_plug/realize/plug, so we
must move these init code in machvirt_init to pre_plug hook, to
let them be shared by all CPUs.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 3587e6e0654962090fdfbae30485c0355f1a9faf)
We have supported topology for arm/virt in previous patch, which
changes the meaning of "thread-id", so we must modify test case.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
(cherry picked from commit 61bc7026130c63f054fa97f59b48e068f0210b79)
The CPU topology specified by user (through -smp options) is used in
ACPI PPTT. Now we will use this information to locate which CPU to
plug or unplug.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit f53ccf3efe6b1894a4a38dd5acae724b9036cbda)
Establish the CPU hotplug framework for arm/virt, we will add
necessary code legs to this framework gradually to realize CPU
hotplug finally.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 9dfe0cc4e9970f93b817068efd0bd6e0d98ca0b2)
Attach cpus aml building and GED support for CPU hotplug to
arm/virt, but currently we make it diabled by not add CPU
hotplug event to GED.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit dc9cea17499b378bde70f3854786ba56daf6a09a)
Let virt/DSDT as the expected file allowed to be changed.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
(cherry picked from commit d160f3e1e3737c47558c1f7908ddbd6cf188b3ff)
This hook will be called in get_cpu_status, which is called
during cpu hotplug.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit dc4b675a897d6094b60ef462482e03b2905e1737)
This adds a new GED event called ACPI_GED_CPU_HOTPLUG_EVT.
The basic workflow is that: GED sends this event to guest,
then ACPI driver in guest will call _EVT method of GED aml,
then _EVT will call CSCN method in cpus aml to get status of
all cpus.
The status of cpus is maintained by CPUHotplugState in GED and
is made accessable to guest through memory region.
This also adds migration support to CPUHotplugState.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit a072cbe5beab53e9328c2fe6576e755e88b2c40a)
We will reuse build_cpus_aml to build DSDT cpus aml in arm/virt
ACPI to realize cpu hotplug. Three points are added.
1. Make ACPI IO address space configurable, because ARM64 platforms
don't use port IO for ACPI IO space.
2. Add GICC struct building support in _MAT of cpu aml.
3. Let the hotplug method parameter can be NULL, because ACPI GED
will realize it.
Besides, CPU CPPC building is injected.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 4d7c437b4d1a8ac175e7b9224a9e402b619d6cd3)
When CPU hotplug is enabled, we will use build_cpus_aml instead of
acpi_dsdt_add_cpus, so factor out CPPC building and we can reuse it
in build_cpus_aml.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
(cherry picked from commit c6b2e91699e31e3473e94d86d9c8db767f0ebb59)
In build_cpus_aml, we will invoke this hook to build _MAT
aml mehtod for cpus.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit 80bcd0313ba67d5c4439030c96c26aee1d9288f9)
To realize CPU hotplug, the cpus aml within ACPI DSDT should contain
_MAT mathod, which is equal to the GICC struct in ACPI MADT. Factor
out the GICC building code from ACPI MADT and reuse it in build_cpus_aml.
Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
(cherry picked from commit f2b68e8adc98f8db20b1edc84b8f485975ed987c)
