Zhiyi reported an infinite loop issue in VFIO use case. The cause of that
was a separate discussion, however during that I found a regression of
dirty sync slowness when profiling.
Each KVMMemoryListerner maintains an array of kvm memslots. Currently it's
statically allocated to be the max supported by the kernel. However after
Linux commit 4fc096a99e ("KVM: Raise the maximum number of user memslots"),
the max supported memslots reported now grows to some number large enough
so that it may not be wise to always statically allocate with the max
reported.
What's worse, QEMU kvm code still walks all the allocated memslots entries
to do any form of lookups. It can drastically slow down all memslot
operations because each of such loop can run over 32K times on the new
kernels.
Fix this issue by making the memslots to be allocated dynamically.
Here the initial size was set to 16 because it should cover the basic VM
usages, so that the hope is the majority VM use case may not even need to
grow at all (e.g. if one starts a VM with ./qemu-system-x86_64 by default
it'll consume 9 memslots), however not too large to waste memory.
There can also be even better way to address this, but so far this is the
simplest and should be already better even than before we grow the max
supported memslots. For example, in the case of above issue when VFIO was
attached on a 32GB system, there are only ~10 memslots used. So it could
be good enough as of now.
In the above VFIO context, measurement shows that the precopy dirty sync
shrinked from ~86ms to ~3ms after this patch applied. It should also apply
to any KVM enabled VM even without VFIO.
NOTE: we don't have a FIXES tag for this patch because there's no real
commit that regressed this in QEMU. Such behavior existed for a long time,
but only start to be a problem when the kernel reports very large
nr_slots_max value. However that's pretty common now (the kernel change
was merged in 2021) so we attached cc:stable because we'll want this change
to be backported to stable branches.
Cc: qemu-stable <qemu-stable@nongnu.org>
Reported-by: Zhiyi Guo <zhguo@redhat.com>
Tested-by: Zhiyi Guo <zhguo@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240917163835.194664-2-peterx@redhat.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Change the data type of the ioctl _request_ argument from 'int' to
'unsigned long' for the various accel/kvm functions which are
essentially wrappers around the ioctl() syscall.
The correct type for ioctl()'s 'request' argument is confused:
* POSIX defines the request argument as 'int'
* glibc uses 'unsigned long' in the prototype in sys/ioctl.h
* the glibc info documentation uses 'int'
* the Linux manpage uses 'unsigned long'
* the Linux implementation of the syscall uses 'unsigned int'
If we wrap ioctl() with another function which uses 'int' as the
type for the request argument, then requests with the 0x8000_0000
bit set will be sign-extended when the 'int' is cast to
'unsigned long' for the call to ioctl().
On x86_64 one such example is the KVM_IRQ_LINE_STATUS request.
Bit requests with the _IOC_READ direction bit set, will have the high
bit set.
Fortunately the Linux Kernel truncates the upper 32bit of the request
on 64bit machines (because it uses 'unsigned int', and see also Linus
Torvalds' comments in
https://sourceware.org/bugzilla/show_bug.cgi?id=14362 )
so this doesn't cause active problems for us. However it is more
consistent to follow the glibc ioctl() prototype when we define
functions that are essentially wrappers around ioctl().
This resolves a Coverity issue where it points out that in
kvm_get_xsave() we assign a value (KVM_GET_XSAVE or KVM_GET_XSAVE2)
to an 'int' variable which can't hold it without overflow.
Resolves: Coverity CID 1547759
Signed-off-by: Johannes Stoelp <johannes.stoelp@gmail.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20240815122747.3053871-1-peter.maydell@linaro.org
[PMM: Rebased patch, adjusted commit message, included note about
Coverity fix, updated the type of the local var in kvm_get_xsave,
updated the comment in the KVMState struct definition]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
KVM vCPU creation is done once during the vCPU realization when Qemu vCPU thread
is spawned. This is common to all the architectures as of now.
Hot-unplug of vCPU results in destruction of the vCPU object in QOM but the
corresponding KVM vCPU object in the Host KVM is not destroyed as KVM doesn't
support vCPU removal. Therefore, its representative KVM vCPU object/context in
Qemu is parked.
Refactor architecture common logic so that some APIs could be reused by vCPU
Hotplug code of some architectures likes ARM, Loongson etc. Update new/old APIs
with trace events. New APIs qemu_{create,park,unpark}_vcpu() can be externally
called. No functional change is intended here.
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Reviewed-by: Gavin Shan <gshan@redhat.com>
Tested-by: Vishnu Pajjuri <vishnu@os.amperecomputing.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Tested-by: Xianglai Li <lixianglai@loongson.cn>
Tested-by: Miguel Luis <miguel.luis@oracle.com>
Reviewed-by: Shaoqin Huang <shahuang@redhat.com>
Reviewed-by: Vishnu Pajjuri <vishnu@os.amperecomputing.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Tested-by: Zhao Liu <zhao1.liu@intel.com>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Reviewed-by: Harsh Prateek Bora <harshpb@linux.ibm.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20240716111502.202344-2-salil.mehta@huawei.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Upon an KVM_EXIT_MEMORY_FAULT exit, userspace needs to do the memory
conversion on the RAMBlock to turn the memory into desired attribute,
switching between private and shared.
Currently only KVM_MEMORY_EXIT_FLAG_PRIVATE in flags is valid when
KVM_EXIT_MEMORY_FAULT happens.
Note, KVM_EXIT_MEMORY_FAULT makes sense only when the RAMBlock has
guest_memfd memory backend.
Note, KVM_EXIT_MEMORY_FAULT returns with -EFAULT, so special handling is
added.
When page is converted from shared to private, the original shared
memory can be discarded via ram_block_discard_range(). Note, shared
memory can be discarded only when it's not back'ed by hugetlb because
hugetlb is supposed to be pre-allocated and no need for discarding.
Signed-off-by: Chao Peng <chao.p.peng@linux.intel.com>
Co-developed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-ID: <20240320083945.991426-13-michael.roth@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Switch to KVM_SET_USER_MEMORY_REGION2 when supported by KVM.
With KVM_SET_USER_MEMORY_REGION2, QEMU can set up memory region that
backend'ed both by hva-based shared memory and guest memfd based private
memory.
Signed-off-by: Chao Peng <chao.p.peng@linux.intel.com>
Co-developed-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-ID: <20240320083945.991426-10-michael.roth@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The upper 16 bits of kvm_userspace_memory_region::slot are
address space id. Parse it separately in trace_kvm_set_user_memory().
Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Message-ID: <20240229063726.610065-5-xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Patch removes DPRINTF macro and adds multiple tracepoints
to capture different kvm events.
We also drop the DPRINTFs that don't add any additional
information than trace_kvm_run_exit already does.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1827
Signed-off-by: Jai Arora <arorajai2798@gmail.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Commit e50caf4a5c ("tracing: convert documentation to rST")
converted docs/devel/tracing.txt to docs/devel/tracing.rst.
We still have several references to the old file, so let's fix them
with the following command:
sed -i s/tracing.txt/tracing.rst/ $(git grep -l docs/devel/tracing.txt)
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210517151702.109066-2-sgarzare@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
KVM dirty ring is a new interface to pass over dirty bits from kernel to the
userspace. Instead of using a bitmap for each memory region, the dirty ring
contains an array of dirtied GPAs to fetch (in the form of offset in slots).
For each vcpu there will be one dirty ring that binds to it.
kvm_dirty_ring_reap() is the major function to collect dirty rings. It can be
called either by a standalone reaper thread that runs in the background,
collecting dirty pages for the whole VM. It can also be called directly by any
thread that has BQL taken.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20210506160549.130416-11-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Clean up the error handling in kvm_init_vcpu so we can see what went
wrong more easily.
Make it take an Error ** and fill it out with what failed, including
the cpu id, so you can tell if it only fails at a given ID.
Replace the remaining DPRINTF by a trace.
This turns a:
kvm_init_vcpu failed: Invalid argument
into:
kvm_init_vcpu: kvm_get_vcpu failed (256): Invalid argument
and with the trace you then get to see:
19049@1595520414.310107:kvm_init_vcpu index: 169 id: 212
19050@1595520414.310635:kvm_init_vcpu index: 170 id: 256
qemu-system-x86_64: kvm_init_vcpu: kvm_get_vcpu failed (256): Invalid argument
which makes stuff a lot more obvious.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200723160915.129069-1-dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This is majorly only for X86 because that's the only one that supports
split irqchip for now.
When the irqchip is split, we face a dilemma that KVM irqfd will be
enabled, however the slow irqchip is still running in the userspace.
It means that the resamplefd in the kernel irqfds won't take any
effect and it will miss to ack INTx interrupts on EOIs.
One example is split irqchip with VFIO INTx, which will break if we
use the VFIO INTx fast path.
This patch can potentially supports the VFIO fast path again for INTx,
that the IRQ delivery will still use the fast path, while we don't
need to trap MMIOs in QEMU for the device to emulate the EIOs (see the
callers of vfio_eoi() hook). However the EOI of the INTx will still
need to be done from the userspace by caching all the resamplefds in
QEMU and kick properly for IOAPIC EOI broadcast.
This is tricky because in this case the userspace ioapic irr &
remote-irr will be bypassed. However such a change will greatly boost
performance for assigned devices using INTx irqs (TCP_RR boosts 46%
after this patch applied).
When the userspace is responsible for the resamplefd kickup, don't
register it on the kvm_irqfd anymore, because on newer kernels (after
commit 654f1f13ea56, 5.2+) the KVM_IRQFD will fail if with both split
irqchip and resamplefd. This will make sure that the fast path will
work for all supported kernels.
https://patchwork.kernel.org/patch/10738541/#22609933
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20200318145204.74483-5-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Firstly detect the interface using KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
and mark it. When failed to enable the new feature we'll fall back to
the old sync.
Provide the log_clear() hook for the memory listeners for both address
spaces of KVM (normal system memory, and SMM) and deliever the clear
message to kernel.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20190603065056.25211-11-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Almost all trace-events point to docs/devel/tracing.txt in a comment
right at the beginning. Touch up the ones that don't.
[Updated with Markus' new commit description wording.
--Stefan]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190314180929.27722-2-armbru@redhat.com
Message-Id: <20190314180929.27722-2-armbru@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Add a couple of traces around the kvm_set_ioeventfd* calls.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20190212134758.10514-4-dgilbert@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This adds a tracepoint to trace the KVM_SET_USER_MEMORY_REGION ioctl
parameters which is quite useful for debugging VFIO memory regions
being actually registered with KVM.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20171215052326.21386-1-aik@ozlabs.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
move kvm related accelerator files into accel/ subdirectory, also
create one stub subdirectory, which will include accelerator's stub
files.
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <1496383606-18060-5-git-send-email-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Peter Xu