target/xtensa: Make xtensa_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for xtensa linux-user.

Remove the code from cpu_loop that raised SIGSEGV.

Acked-by: Max Filippov <jcmvbkbc@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/sparc: Make sparc_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for sparc linux-user.

This makes all of the code in mmu_helper.c sysemu only, so remove
the ifdefs and move the file to sparc_softmmu_ss.  Remove the code
from cpu_loop that handled TT_DFAULT and TT_TFAULT.

Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/sh4: Make sh4_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for sh4 linux-user.

Remove the code from cpu_loop that raised SIGSEGV.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/s390x: Implement s390_cpu_record_sigsegv

Move the masking of the address from cpu_loop into
s390_cpu_record_sigsegv -- this is governed by hw, not linux.
This does mean we have to raise our own exception, rather
than return to the fallback.

Use maperr to choose between PGM_PROTECTION and PGM_ADDRESSING.
Use the appropriate si_code for each in cpu_loop.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/s390x: Use probe_access_flags in s390_probe_access

Not sure why the user-only code wasn't rewritten to use
probe_access_flags at the same time that the sysemu code
was converted.  For the purpose of user-only, this is an
exact replacement.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/riscv: Make riscv_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for riscv linux-user.

Remove the code from cpu_loop that raised SIGSEGV.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/ppc: Implement ppc_cpu_record_sigsegv

Record DAR, DSISR, and exception_index.  That last means
that we must exit to cpu_loop ourselves, instead of letting
exception_index being overwritten.

This is exactly what the user-mode ppc_cpu_tlb_fill does,
so simply rename it as ppc_cpu_record_sigsegv.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/openrisc: Make openrisc_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient for
openrisc linux-user.

This makes all of the code in mmu.c sysemu only, so remove
the ifdefs and move the file to openrisc_softmmu_ss.
Remove the code from cpu_loop that handled EXCP_DPF.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/openrisc: Abort for EXCP_RANGE, EXCP_FPE

QEMU does not allow the system control bits for either exception to
be enabled in linux-user, therefore both exceptions are dead code.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/nios2: Implement nios2_cpu_record_sigsegv

Because the linux-user kuser page handling is currently implemented
by detecting magic addresses in the unnamed 0xaa trap, we cannot
simply remove nios2_cpu_tlb_fill and rely on the fallback code.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/mips: Make mips_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for mips linux-user.

This means we can remove tcg/user/tlb_helper.c entirely.
Remove the code from cpu_loop that raised SIGSEGV.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/microblaze: Make mb_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for microblaze linux-user.

Remove the code from cpu_loop that handled the unnamed 0xaa exception.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/m68k: Make m68k_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for m68k linux-user.

Remove the code from cpu_loop that handled EXCP_ACCESS.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/i386: Implement x86_cpu_record_sigsegv

Record cr2, error_code, and exception_index.  That last means
that we must exit to cpu_loop ourselves, instead of letting
exception_index being overwritten.

Use the maperr parameter to properly set PG_ERROR_P_MASK.

Reviewed by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/hppa: Make hppa_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for hppa linux-user.

Remove the code from cpu_loop that raised SIGSEGV.
This makes all of the code in mem_helper.c sysemu only,
so remove the ifdefs and move the file to hppa_softmmu_ss.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/hexagon: Remove hexagon_cpu_tlb_fill

The fallback code in cpu_loop_exit_sigsegv is sufficient
for hexagon linux-user.

Remove the code from cpu_loop that raises SIGSEGV.

Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/cris: Make cris_cpu_tlb_fill sysemu only

The fallback code in cpu_loop_exit_sigsegv is sufficient
for cris linux-user.

Remove the code from cpu_loop that handled the unnamed 0xaa exception.
This makes all of the code in helper.c sysemu only, so remove the
ifdefs and move the file to cris_softmmu_ss.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/arm: Implement arm_cpu_record_sigsegv

Because of the complexity of setting ESR, continue to use
arm_deliver_fault.  This means we cannot remove the code
within cpu_loop that decodes EXCP_DATA_ABORT and
EXCP_PREFETCH_ABORT.

But using the new hook means that we don't have to do the
page_get_flags check manually, and we'll be able to restrict
the tlb_fill hook to sysemu later.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/arm: Use cpu_loop_exit_sigsegv for mte tag lookup

Use the new os interface for raising the exception,
rather than calling arm_cpu_tlb_fill directly.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/alpha: Implement alpha_cpu_record_sigsegv

Record trap_arg{0,1,2} for the linux-user signal frame.

Fill in the stores to trap_arg{1,2} that were missing
from the previous user-only alpha_cpu_tlb_fill function.
Use maperr to simplify computation of trap_arg1.

Remove the code for EXCP_MMFAULT from cpu_loop, as
that part is now handled by cpu_loop_exit_sigsegv.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user: Add cpu_loop_exit_sigsegv

This is a new interface to be provided by the os emulator for
raising SIGSEGV on fault.  Use the new record_sigsegv target hook.

Reviewed by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

hw/core: Add TCGCPUOps.record_sigsegv

Add a new user-only interface for updating cpu state before
raising a signal.  This will replace tlb_fill for user-only
and should result in less boilerplate for each guest.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/signal: Drop HOST_SIGNAL_PLACEHOLDER

Now that all of the linux-user hosts have been converted
to host-signal.h, drop the compatibility code.

Reviewed by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/riscv: Improve host_signal_write

Do not read 4 bytes before we determine the size of the insn.
Simplify triple switches in favor of checking major opcodes.
Include the missing cases of compact fsd and fsdsp.

Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/arm: Fixup comment re handle_cpu_signal

The named function no longer exists.
Refer to host_signal_handler instead.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/riscv: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/mips: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/s390: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.

Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/aarch64: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.
Drop the *BSD code, to be re-created under bsd-user/ later.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/arm: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.
Drop the *BSD code, to be re-created under bsd-user/ later.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/sparc: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.
Drop the *BSD code, to be re-created under bsd-user/ later.
Drop the Solaris code as completely unused.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/alpha: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/ppc: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.
Drop the *BSD code, to be re-created under bsd-user/ later.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/host/x86: Populate host_signal.h

Split host_signal_pc and host_signal_write out of user-exec.c.
Drop the *BSD code, to be re-created under bsd-user/ later.

Reviewed-by: Warner Losh <imp@bsdimp.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Merge remote-tracking branch 'remotes/philmd/tags/machine-20211101' into staging

Machine core patches

- Move GPIO code out of qdev.c
- Move hotplug code out of qdev.c
- Restrict various files to sysemu
- Move SMP code out of machine.c
- Add SMP parsing unit tests
- Move dynamic sysbus device check earlier

# gpg: Signature made Mon 01 Nov 2021 02:44:32 PM EDT
# gpg:                using RSA key FAABE75E12917221DCFD6BB2E3E32C2CDEADC0DE
# gpg: Good signature from "Philippe Mathieu-Daudé (F4BUG) <f4bug@amsat.org>" [full]

* remotes/philmd/tags/machine-20211101:
  machine: remove the done notifier for dynamic sysbus device type check
  qdev-monitor: Check sysbus device type before creating it
  machine: add device_type_is_dynamic_sysbus function
  tests/unit: Add an unit test for smp parsing
  hw/core/machine: Split out the smp parsing code
  hw/core: Restrict hotplug to system emulation
  hw/core: Extract hotplug-related functions to qdev-hotplug.c
  hw/core: Declare meson source set
  hw/core: Restrict sysemu specific files
  machine: Move gpio code to hw/core/gpio.c

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

hw/i386: fix vmmouse registration

According to the logic of vmmouse_update_handler function,
vmmouse should be registered as an event handler when
it's status is zero.
vmmouse_read_id resets the status but does not register
the handler.
This patch adds vmmouse registration and activation when
status is reset.

Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Message-Id: <163524204515.1914131.16465061981774791228.stgit@pasha-ThinkPad-X280>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

pci: Export pci_for_each_device_under_bus*()

They're actually more commonly used than the helper without _under_bus, because
most callers do have the pci bus on hand.  After exporting we can switch a lot
of the call sites to use these two helpers.

Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20211028043129.38871-3-peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>

pci: Define pci_bus_dev_fn/pci_bus_fn/pci_bus_ret_fn

They're used in quite a few places of pci.[ch] and also in the rest of the code
base.  Define them so that it doesn't need to be defined all over the places.

The pci_bus_fn is similar to pci_bus_dev_fn that only takes a PCIBus* and an
opaque.  The pci_bus_ret_fn is similar to pci_bus_fn but it allows to return a
void* pointer.

Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20211028043129.38871-2-peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/i386/pc: Allow instantiating a virtio-iommu device

Allow instantiating a virtio-iommu device by adding an ACPI Virtual I/O
Translation table (VIOT), which describes the relation between the
virtio-iommu and the endpoints it manages.

Add a hotplug handler for virtio-iommu on x86 and set the necessary
reserved region property. On x86, the [0xfee00000, 0xfeefffff] DMA
region is reserved for MSIs. DMA transactions to this range either
trigger IRQ remapping in the IOMMU or bypasses IOMMU translation.

Although virtio-iommu does not support IRQ remapping it must be informed
of the reserved region so that it can forward DMA transactions targeting
this region.

Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Message-Id: <20211026182024.2642038-5-jean-philippe@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/i386/pc: Move IOMMU singleton into PCMachineState

We're about to support a third vIOMMU for x86, virtio-iommu which
doesn't inherit X86IOMMUState. Move the IOMMU singleton into
PCMachineState, so it can be shared between all three vIOMMUs.

The x86_iommu_get_default() helper is still needed by KVM and IOAPIC to
fetch the default IRQ-remapping IOMMU. Since virtio-iommu doesn't
support IRQ remapping, this interface doesn't need to change for the
moment. We could later replace X86IOMMUState with an "IRQ remapping
IOMMU" interface if necessary.

Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Message-Id: <20211026182024.2642038-4-jean-philippe@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/i386/pc: Remove x86_iommu_get_type()

To generate the IOMMU ACPI table, acpi-build.c can use base QEMU types
instead of a special IommuType value.

Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Message-Id: <20211026182024.2642038-3-jean-philippe@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

hw/acpi: Add VIOT table

Add a function that generates a Virtual I/O Translation table (VIOT),
describing the topology of paravirtual IOMMUs. The table is created if a
virtio-iommu device is present. It contains a virtio-iommu node and PCI
Range nodes for endpoints managed by the IOMMU. By default, a single
node describes all PCI devices. When passing the
"default_bus_bypass_iommu" machine option and "bypass_iommu" PXB option,
only buses that do not bypass the IOMMU are described by PCI Range
nodes.

Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org>
Message-Id: <20211026182024.2642038-2-jean-philippe@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vhost-vdpa: Set discarding of RAM broken when initializing the backend

Similar to VFIO, vDPA will go ahead an map+pin all guest memory. Memory
that used to be discarded will get re-populated and if we
discard+re-access memory after mapping+pinning, the pages mapped into the
vDPA IOMMU will go out of sync with the actual pages mapped into the user
space page tables.

Set discarding of RAM broken such that:
- virtio-mem and vhost-vdpa run mutually exclusive
- virtio-balloon is inhibited and no memory discards will get issued

In the future, we might be able to support coordinated discarding of RAM
as used by virtio-mem and already supported by vfio via the
RamDiscardManager.

Acked-by: Jason Wang <jasowang@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Cindy Lu <lulu@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20211027130324.59791-1-david@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>

qtest: fix 'expression is always false' build failure in qtest_has_accel()

If KVM is disabled or not present, qtest library build
may fail with:
   libqtest.c: In function 'qtest_has_accel':
      comparison of unsigned expression < 0 is always false
      [-Werror=type-limits]
         for (i = 0; i < ARRAY_SIZE(targets); i++) {

due to empty 'targets' array.
Fix it by making sure that CONFIG_KVM_TARGETS isn't empty.

Fixes: e741aff0f43343 ("tests: qtest: add qtest_has_accel() to check if tested binary supports accelerator")
Reported-by: Jason Andryuk <jandryuk@gmail.com>
Suggested-by: "Michael S. Tsirkin" <mst@redhat.com>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20211027151012.2639284-1-imammedo@redhat.com>
Tested-by: Jason Andryuk <jandryuk@gmail.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

migration/dirtyrate: implement dirty-bitmap dirtyrate calculation

introduce dirty-bitmap mode as the third method of calc-dirty-rate.
implement dirty-bitmap dirtyrate calculation, which can be used
to measuring dirtyrate in the absence of dirty-ring.

introduce "dirty_bitmap:-b" option in hmp calc_dirty_rate to
indicate dirty bitmap method should be used for calculation.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

memory: introduce total_dirty_pages to stat dirty pages

introduce global var total_dirty_pages to stat dirty pages
along with memory_global_dirty_log_sync.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/ram: Handle RAMBlocks with a RamDiscardManager on background snapshots

We already don't ever migrate memory that corresponds to discarded ranges
as managed by a RamDiscardManager responsible for the mapped memory region
of the RAMBlock.

virtio-mem uses this mechanism to logically unplug parts of a RAMBlock.
Right now, we still populate zeropages for the whole usable part of the
RAMBlock, which is undesired because:

1. Even populating the shared zeropage will result in memory getting
   consumed for page tables.
2. Memory backends without a shared zeropage (like hugetlbfs and shmem)
   will populate an actual, fresh page, resulting in an unintended
   memory consumption.

Discarded ("logically unplugged") parts have to remain discarded. As
these pages are never part of the migration stream, there is no need to
track modifications via userfaultfd WP reliably for these parts.

Further, any writes to these ranges by the VM are invalid and the
behavior is undefined.

Note that Linux only supports userfaultfd WP on private anonymous memory
for now, which usually results in the shared zeropage getting populated.
The issue will become more relevant once userfaultfd WP supports shmem
and hugetlb.

Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/ram: Factor out populating pages readable in ram_block_populate_pages()

Let's factor out prefaulting/populating to make further changes easier to
review and add a comment what we are actually expecting to happen. While at
it, use the actual page size of the ramblock, which defaults to
qemu_real_host_page_size for anonymous memory. Further, rename
ram_block_populate_pages() to ram_block_populate_read() as well, to make
it clearer what we are doing.

In the future, we might want to use MADV_POPULATE_READ to speed up
population.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: Simplify alignment and alignment checks

Let's use QEMU_ALIGN_DOWN() and friends to make the code a bit easier to
read.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/postcopy: Handle RAMBlocks with a RamDiscardManager on the destination

Currently, when someone (i.e., the VM) accesses discarded parts inside a
RAMBlock with a RamDiscardManager managing the corresponding mapped memory
region, postcopy will request migration of the corresponding page from the
source. The source, however, will never answer, because it refuses to
migrate such pages with undefined content ("logically unplugged"): the
pages are never dirty, and get_queued_page() will consequently skip
processing these postcopy requests.

Especially reading discarded ("logically unplugged") ranges is supposed to
work in some setups (for example with current virtio-mem), although it
barely ever happens: still, not placing a page would currently stall the
VM, as it cannot make forward progress.

Let's check the state via the RamDiscardManager (the state e.g.,
of virtio-mem is migrated during precopy) and avoid sending a request
that will never get answered. Place a fresh zero page instead to keep
the VM working. This is the same behavior that would happen
automatically without userfaultfd being active, when accessing virtual
memory regions without populated pages -- "populate on demand".

For now, there are valid cases (as documented in the virtio-mem spec) where
a VM might read discarded memory; in the future, we will disallow that.
Then, we might want to handle that case differently, e.g., warning the
user that the VM seems to be mis-behaving.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

virtio-mem: Drop precopy notifier

Migration code now properly handles RAMBlocks which are indirectly managed
by a RamDiscardManager. No need for manual handling via the free page
optimization interface, let's get rid of it.

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/ram: Handle RAMBlocks with a RamDiscardManager on the migration source

We don't want to migrate memory that corresponds to discarded ranges as
managed by a RamDiscardManager responsible for the mapped memory region of
the RAMBlock. The content of these pages is essentially stale and
without any guarantees for the VM ("logically unplugged").

Depending on the underlying memory type, even reading memory might populate
memory on the source, resulting in an undesired memory consumption. Of
course, on the destination, even writing a zeropage consumes memory,
which we also want to avoid (similar to free page hinting).

Currently, virtio-mem tries achieving that goal (not migrating "unplugged"
memory that was discarded) by going via qemu_guest_free_page_hint() - but
it's hackish and incomplete.

For example, background snapshots still end up reading all memory, as
they don't do bitmap syncs. Postcopy recovery code will re-add
previously cleared bits to the dirty bitmap and migrate them.

Let's consult the RamDiscardManager after setting up our dirty bitmap
initially and when postcopy recovery code reinitializes it: clear
corresponding bits in the dirty bitmaps (e.g., of the RAMBlock and inside
KVM). It's important to fixup the dirty bitmap *after* our initial bitmap
sync, such that the corresponding dirty bits in KVM are actually cleared.

As colo is incompatible with discarding of RAM and inhibits it, we don't
have to bother.

Note: if a misbehaving guest would use discarded ranges after migration
started we would still migrate that memory: however, then we already
populated that memory on the migration source.

Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

virtio-mem: Implement replay_discarded RamDiscardManager callback

Implement it similar to the replay_populated callback.

Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

memory: Introduce replay_discarded callback for RamDiscardManager

Introduce replay_discarded callback similar to our existing
replay_populated callback, to be used my migration code to never migrate
discarded memory.

Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

dump-guest-memory: Block live migration

Both dump-guest-memory and live migration caches vm state at the beginning.
Either of them entering the other one will cause race on the vm state, and even
more severe on that (please refer to the crash report in the bug link).

Let's block live migration in dump-guest-memory, and that'll also block
dump-guest-memory if it detected that we're during a live migration.

Side note: migrate_del_blocker() can be called even if the blocker is not
inserted yet, so it's safe to unconditionally delete that blocker in
dump_cleanup (g_slist_remove allows no-entry-found case).

Suggested-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1996609
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: Add migrate_add_blocker_internal()

An internal version that removes -only-migratable implications.  It can be used
for temporary migration blockers like dump-guest-memory.

Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: Make migration blocker work for snapshots too

save_snapshot() checks migration blocker, which looks sane.  At the meantime we
should also teach the blocker add helper to fail if during a snapshot, just
like for migrations.

Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/dirtyrate: implement dirty-ring dirtyrate calculation

use dirty ring feature to implement dirtyrate calculation.

introduce mode option in qmp calc_dirty_rate to specify what
method should be used when calculating dirtyrate, either
page-sampling or dirty-ring should be passed.

introduce "dirty_ring:-r" option in hmp calc_dirty_rate to
indicate dirty ring method should be used for calculation.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <7db445109bd18125ce8ec86816d14f6ab5de6a7d.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/dirtyrate: move init step of calculation to main thread

since main thread may "query dirty rate" at any time, it's better
to move init step into main thead so that synchronization overhead
between "main" and "get_dirtyrate" can be reduced.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <109f8077518ed2f13068e3bfb10e625e964780f1.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/dirtyrate: adjust order of registering thread

registering get_dirtyrate thread in advance so that both
page-sampling and dirty-ring mode can be covered.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <d7727581a8e86d4a42fc3eacf7f310419b9ebf7e.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/dirtyrate: introduce struct and adjust DirtyRateStat

introduce "DirtyRateMeasureMode" to specify what method should be
used to calculate dirty rate, introduce "DirtyRateVcpu" to store
dirty rate for each vcpu.

use union to store stat data of specific mode

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <661c98c40f40e163aa58334337af8f3ddf41316a.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

memory: make global_dirty_tracking a bitmask

since dirty ring has been introduced, there are two methods
to track dirty pages of vm. it seems that "logging" has
a hint on the method, so rename the global_dirty_log to
global_dirty_tracking would make description more accurate.

dirty rate measurement may start or stop dirty tracking during
calculation. this conflict with migration because stop dirty
tracking make migration leave dirty pages out then that'll be
a problem.

make global_dirty_tracking a bitmask can let both migration and
dirty rate measurement work fine. introduce GLOBAL_DIRTY_MIGRATION
and GLOBAL_DIRTY_DIRTY_RATE to distinguish what current dirty
tracking aims for, migration or dirty rate.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <9c9388657cfa0301bd2c1cfa36e7cf6da4aeca19.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

KVM: introduce dirty_pages and kvm_dirty_ring_enabled

dirty_pages is used to calculate dirtyrate via dirty ring, when
enabled, kvm-reaper will increase the dirty pages after gfns
being dirtied.

kvm_dirty_ring_enabled shows if kvm-reaper is working. dirtyrate
thread could use it to check if measurement can base on dirty
ring feature.

Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn>
Message-Id: <fee5fb2ab17ec2159405fc54a3cff8e02322f816.1624040308.git.huangy81@chinatelecom.cn>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

roms/openbios: update OpenBIOS images to b9062dea built from submodule

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>

machine: remove the done notifier for dynamic sysbus device type check

Now that we check sysbus device types during device creation, we
can remove the check in the machine init done notifier.
This was the only thing done by this notifier, so we remove the
whole sysbus_notifier structure of the MachineState.

Note: This notifier was checking all /peripheral and /peripheral-anon
sysbus devices. Now we only check those added by -device cli option or
device_add qmp command when handling the command/option. So if there
are some devices added in one of these containers manually (eg in
machine C code), these will not be checked anymore.
This use case does not seem to appear apart from
hw/xen/xen-legacy-backend.c (it uses qdev_set_id() and in this case,
not for a sysbus device, so it's ok).

Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211029142258.484907-4-damien.hedde@greensocs.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

qdev-monitor: Check sysbus device type before creating it

Add an early check to test if the requested sysbus device type
is allowed by the current machine before creating the device. This
impacts both -device cli option and device_add qmp command.

Before this patch, the check was done well after the device has
been created (in a machine init done notifier). We can now report
the error right away.

Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211029142258.484907-3-damien.hedde@greensocs.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

machine: add device_type_is_dynamic_sysbus function

Right now the allowance check for adding a sysbus device using
-device cli option (or device_add qmp command) is done well after
the device has been created. It is done during the machine init done
notifier: machine_init_notify() in hw/core/machine.c

This new function will allow us to do the check at the right time and
issue an error if it fails.

Also make device_is_dynamic_sysbus() use the new function.

Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211029142258.484907-2-damien.hedde@greensocs.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

tests/unit: Add an unit test for smp parsing

Now that we have a generic parser smp_parse(), let's add an unit
test for the code. All possible valid/invalid SMP configurations
that the user can specify are covered.

Signed-off-by: Yanan Wang <wangyanan55@huawei.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20211026034659.22040-3-wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <bfed7144-af86-7098-e7a6-731ff13c2cf7@huawei.com>
[PMD: Squashed format string fixup from Yanan Wang]
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

hw/core/machine: Split out the smp parsing code

We are going to introduce an unit test for the parser smp_parse()
in hw/core/machine.c, but now machine.c is only built in softmmu.

In order to solve the build dependency on the smp parsing code and
avoid building unrelated stuff for the unit tests, move the tested
code from machine.c into a separate file, i.e., machine-smp.c and
build it in common field.

Signed-off-by: Yanan Wang <wangyanan55@huawei.com>
Reviewed-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20211026034659.22040-2-wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

hw/core: Restrict hotplug to system emulation

Restrict hotplug to system emulation, add stubs for the other uses.

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Yanan Wang <wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211028150521.1973821-5-philmd@redhat.com>

hw/core: Extract hotplug-related functions to qdev-hotplug.c

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Yanan Wang <wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211028150521.1973821-4-philmd@redhat.com>

hw/core: Declare meson source set

As we want to be able to conditionally add files to the hw/core
file list, use a source set.

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Yanan Wang <wangyanan55@huawei.com>
Tested-by: Yanan Wang <wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211028150521.1973821-3-philmd@redhat.com>

hw/core: Restrict sysemu specific files

All these files don't make sense for tools and user emulation,
restrict them to system emulation.

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Yanan Wang <wangyanan55@huawei.com>
Tested-by: Yanan Wang <wangyanan55@huawei.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20211028150521.1973821-2-philmd@redhat.com>

machine: Move gpio code to hw/core/gpio.c

Only softmmu code uses gpio, so move gpio code from qdev.c to
gpio.c and compile it only on softmmu mode.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190425200051.19906-2-ehabkost@redhat.com>
[PMD: Rebased]
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Merge remote-tracking branch 'remotes/jsnow/tags/python-pull-request' into staging

Pull request

# gpg: Signature made Mon 01 Nov 2021 01:29:49 PM EDT
# gpg:                using RSA key F9B7ABDBBCACDF95BE76CBD07DEF8106AAFC390E
# gpg: Good signature from "John Snow (John Huston) <jsnow@redhat.com>" [full]

* remotes/jsnow/tags/python-pull-request: (22 commits)
  python, iotests: replace qmp with aqmp
  python/aqmp: Create sync QMP wrapper for iotests
  iotests/300: avoid abnormal shutdown race condition
  iotests: Conditionally silence certain AQMP errors
  iotests: Accommodate async QMP Exception classes
  python/aqmp: Remove scary message
  python/machine: Handle QMP errors on close more meticulously
  python/machine: remove has_quit argument
  python: Add iotest linters to test suite
  iotests/linters: Add workaround for mypy bug #9852
  iotests/linters: Add entry point for linting via Python CI
  iotests: split linters.py out from 297
  iotests/297: split test into sub-cases
  iotests/297: update tool availability checks
  iotests/297: Change run_linter() to raise an exception on failure
  iotests/297: refactor run_[mypy|pylint] as generic execution shim
  iotests/297: Split run_linters apart into run_pylint and run_mypy
  iotests/297: Don't rely on distro-specific linter binaries
  iotests/297: Create main() function
  iotests/297: Add get_files() function
  ...

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

vfio/common: Add a trace point when a MMIO RAM section cannot be mapped

The MSI-X structures of some devices and other non-MSI-X structures
may be in the same BAR. They may share one host page, especially in
the case of large page granularity, such as 64K.

For example, MSIX-Table size of 82599 NIC is 0x30 and the offset in
Bar 3(size 64KB) is 0x0. vfio_listener_region_add() will be called
to map the remaining range (0x30-0xffff). If host page size is 64KB,
it will return early at 'int128_ge((int128_make64(iova), llend))'
without any message. Let's add a trace point to inform users like commit
5c08600547c0 ("vfio: Use a trace point when a RAM section cannot be DMA mapped")
did.

Signed-off-by: Kunkun Jiang <jiangkunkun@huawei.com>
Link: https://lore.kernel.org/r/20211027090406.761-3-jiangkunkun@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>

vfio/pci: Add support for mmapping sub-page MMIO BARs after live migration

We can expand MemoryRegions of sub-page MMIO BARs in
vfio_pci_write_config() to improve IO performance for some
devices. However, the MemoryRegions of destination VM are
not expanded any more after live migration. Because their
addresses have been updated in vmstate_load_state()
(vfio_pci_load_config) and vfio_sub_page_bar_update_mapping()
will not be called.

This may result in poor performance after live migration.
So iterate BARs in vfio_pci_load_config() and try to update
sub-page BARs.

Reported-by: Nianyao Tang <tangnianyao@huawei.com>
Reported-by: Qixin Gan <ganqixin@huawei.com>
Signed-off-by: Kunkun Jiang <jiangkunkun@huawei.com>
Link: https://lore.kernel.org/r/20211027090406.761-2-jiangkunkun@huawei.com
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>

python, iotests: replace qmp with aqmp

Swap out the synchronous QEMUMonitorProtocol from qemu.qmp with the sync
wrapper from qemu.aqmp instead.

Add an escape hatch in the form of the environment variable
QEMU_PYTHON_LEGACY_QMP which allows you to cajole QEMUMachine into using
the old implementation, proving that both implementations work
concurrently.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211026175612.4127598-9-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

python/aqmp: Create sync QMP wrapper for iotests

This is a wrapper around the async QMPClient that mimics the old,
synchronous QEMUMonitorProtocol class. It is designed to be
interchangeable with the old implementation.

It does not, however, attempt to mimic Exception compatibility.

Signed-off-by: John Snow <jsnow@redhat.com>
Acked-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211026175612.4127598-8-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/300: avoid abnormal shutdown race condition

Wait for the destination VM to close itself instead of racing to shut it
down first, which produces different error log messages from AQMP
depending on precisely when we tried to shut it down.

(For example: We may try to issue 'quit' immediately prior to the target
VM closing its QMP socket, which will cause an ECONNRESET error to be
logged. Waiting for the VM to exit itself avoids the race on shutdown
behavior.)

Reported-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211026175612.4127598-7-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests: Conditionally silence certain AQMP errors

AQMP likes to be very chatty about errors it encounters. In general,
this is good because it allows us to get good diagnostic information for
otherwise complex async failures.

For example, during a failed QMP connection attempt, we might see:

+ERROR:qemu.aqmp.qmp_client.qemub-2536319:Negotiation failed: EOFError
+ERROR:qemu.aqmp.qmp_client.qemub-2536319:Failed to establish session: EOFError

This might be nice in iotests output, because failure scenarios
involving the new QMP library will be spelled out plainly in the output
diffs.

For tests that are intentionally causing this scenario though, filtering
that log output could be a hassle. For now, add a context manager that
simply lets us toggle this output off during a critical region.

(Additionally, a forthcoming patch allows the use of either legacy or
async QMP to be toggled with an environment variable. In this
circumstance, we can't amend the iotest output to just always expect the
error message, either. Just suppress it for now. More rigorous log
filtering can be investigated later if/when it is deemed safe to
permanently replace the legacy QMP library.)

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20211026175612.4127598-6-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests: Accommodate async QMP Exception classes

(But continue to support the old ones for now, too.)

There are very few cases of any user of QEMUMachine or a subclass
thereof relying on a QMP Exception type. If you'd like to check for
yourself, you want to grep for all of the derivatives of QMPError,
excluding 'AQMPError' and its derivatives. That'd be these:

- QMPError
- QMPConnectError
- QMPCapabilitiesError
- QMPTimeoutError
- QMPProtocolError
- QMPResponseError
- QMPBadPortError

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20211026175612.4127598-5-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

python/aqmp: Remove scary message

The scary message interferes with the iotests output. Coincidentally, if
iotests works by removing this, then it's good evidence that we don't
really need to scare people away from using it.

Signed-off-by: John Snow <jsnow@redhat.com>
Acked-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211026175612.4127598-4-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

python/machine: Handle QMP errors on close more meticulously

To use the AQMP backend, Machine just needs to be a little more diligent
about what happens when closing a QMP connection. The operation is no
longer a freebie in the async world; it may return errors encountered in
the async bottom half on incoming message receipt, etc.

(AQMP's disconnect, ultimately, serves as the quiescence point where all
async contexts are gathered together, and any final errors reported at
that point.)

Because async QMP continues to check for messages asynchronously, it's
almost certainly likely that the loop will have exited due to EOF after
issuing the last 'quit' command. That error will ultimately be bubbled
up when attempting to close the QMP connection. The manager class here
then is free to discard it -- if it was expected.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20211026175612.4127598-3-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

python/machine: remove has_quit argument

If we spy on the QMP commands instead, we don't need callers to remember
to pass it. Seems like a fair trade-off.

The one slightly weird bit is overloading this instance variable for
wait(), where we use it to mean "don't issue the qmp 'quit'
command". This means that wait() will "fail" if the QEMU process does
not terminate of its own accord.

In most cases, we probably did already actually issue quit -- some
iotests do this -- but in some others, we may be waiting for QEMU to
terminate for some other reason, such as a test wherein we tell the
guest (directly) to shut down.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20211026175612.4127598-2-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

python: Add iotest linters to test suite

Run mypy and pylint on the iotests files directly from the Python CI
test infrastructure. This ensures that any accidental breakages to the
qemu.[qmp|aqmp|machine|utils] packages will be caught by that test
suite.

It also ensures that these linters are run with well-known versions and
test against a wide variety of python versions, which helps to find
accidental cross-version python compatibility issues.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-15-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/linters: Add workaround for mypy bug #9852

This one is insidious: if you write an import as "from {namespace}
import {subpackage}" as mirror-top-perms (now) does, mypy will fail on
every-other invocation *if* the package being imported is a typed,
installed, namespace-scoped package.

Upsettingly, that's exactly what 'qemu.[aqmp|qmp|machine]' et al are in
the context of Python CI tests.

Now, I could just edit mirror-top-perms to avoid this invocation, but
since I tripped on a landmine, I might as well head it off at the pass
and make sure nobody else trips on that same landmine.

It seems to have something to do with the order in which files are
checked as well, meaning the random order in which set(os.listdir())
produces the list of files to test will cause problems intermittently
and not just strictly "every other run".

This will be fixed in mypy >= 0.920, which is not released yet. The
workaround for now is to disable incremental checking, which avoids the
issue.

Note: This workaround is not applied when running iotest 297 directly,
because the bug does not surface there! Given the nature of CI jobs not
starting with any stale cache to begin with, this really only has a
half-second impact on manual runs of the Python test suite when executed
directly by a developer on their local machine. The workaround may be
removed when the Python package requirements can stipulate mypy 0.920 or
higher, which can happen as soon as it is released. (Barring any
unforseen compatibility issues that 0.920 may bring with it.)

See also:
 https://github.com/python/mypy/issues/11010
 https://github.com/python/mypy/issues/9852

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-14-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/linters: Add entry point for linting via Python CI

We need at least a tiny little shim here to join test file discovery
with test invocation. This logic could conceivably be hosted somewhere
in python/, but I felt it was strictly the least-rude thing to keep the
test logic here in iotests/, even if this small function isn't itself an
iotest.

Note that we don't actually even need the executable bit here, we'll be
relying on the ability to run this module as a script using Python CLI
arguments. No chance it gets misunderstood as an actual iotest that way.

(It's named, not in tests/, doesn't have the execute bit, and doesn't
have an execution shebang.)

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-13-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests: split linters.py out from 297

Now, 297 is just the iotests-specific incantations and linters.py is as
minimal as I can think to make it. The only remaining element in here
that ought to be configuration and not code is the list of skip files,
but they're still numerous enough that repeating them for mypy and
pylint configurations both would be ... a hassle.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-12-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: split test into sub-cases

Take iotest 297's main() test function and split it into two sub-cases
that can be skipped individually. We can also drop custom environment
setup from the pylint test as it isn't needed.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-11-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: update tool availability checks

As mentioned in 'iotests/297: Don't rely on distro-specific linter
binaries', these checks are overly strict. Update them to be in-line
with how we actually invoke the linters themselves.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-10-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Change run_linter() to raise an exception on failure

Instead of using a process return code as the python function return
value (or just not returning anything at all), allow run_linter() to
raise an exception instead.

The responsibility for printing output on error shifts from the function
itself to the caller, who will know best how to present/format that
information. (Also, "suppress_output" is now a lot more accurate of a
parameter name.)

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-9-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: refactor run_[mypy|pylint] as generic execution shim

There's virtually nothing special here anymore; we can combine these
into a single, rather generic function.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-8-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Split run_linters apart into run_pylint and run_mypy

Move environment setup into main(), and split the actual linter
execution into run_pylint and run_mypy, respectively.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-7-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Don't rely on distro-specific linter binaries

'pylint-3' is another Fedora-ism. Use "python3 -m pylint" or "python3 -m
mypy" to access these scripts instead. This style of invocation will
prefer the "correct" tool when run in a virtual environment.

Note that we still check for "pylint-3" before the test begins -- this
check is now "overly strict", but shouldn't cause anything that was
already running correctly to start failing. This is addressed by a
commit later in this series;
  'iotests/297: update tool availability checks'.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-6-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Create main() function

Instead of running "run_linters" directly, create a main() function that
will be responsible for environment setup, leaving run_linters()
responsible only for execution of the linters.

(That environment setup will be moved over in forthcoming commits.)

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-5-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Add get_files() function

Split out file discovery into its own method to begin separating out
configuration/setup and test execution.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-4-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Split mypy configuration out into mypy.ini

More separation of code and configuration.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-3-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

iotests/297: Move pylint config into pylintrc

Move --score=n and --notes=XXX,FIXME into pylintrc. This pulls
configuration out of code, which I think is probably a good thing in
general.

Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-id: 20211019144918.3159078-2-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>

migration/rdma: Fix out of order wrid

destination:
../qemu/build/qemu-system-x86_64 -enable-kvm -netdev tap,id=hn0,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown -device e1000,netdev=hn0,mac=50:52:54:00:11:22 -boot c -drive if=none,file=./Fedora-rdma-server-migration.qcow2,id=drive-virtio-disk0 -device virtio-blk-pci,bus=pci.0,addr=0x4,drive=drive-virtio-disk0,id=virtio-disk0 -m 2048 -smp 2 -device piix3-usb-uhci -device usb-tablet -monitor stdio -vga qxl -spice streaming-video=filter,port=5902,disable-ticketing -incoming rdma:192.168.22.23:8888
qemu-system-x86_64: -spice streaming-video=filter,port=5902,disable-ticketing: warning: short-form boolean option 'disable-ticketing' deprecated
Please use disable-ticketing=on instead
QEMU 6.0.50 monitor - type 'help' for more information
(qemu) trace-event qemu_rdma_block_for_wrid_miss on
(qemu) dest_init RDMA Device opened: kernel name rxe_eth0 uverbs device name uverbs2, infiniband_verbs class device path /sys/class/infiniband_verbs/uverbs2, infiniband class device path /sys/class/infiniband/rxe_eth0, transport: (2) Ethernet
qemu_rdma_block_for_wrid_miss A Wanted wrid CONTROL SEND (2000) but got CONTROL RECV (4000)

source:
../qemu/build/qemu-system-x86_64 -enable-kvm -netdev tap,id=hn0,script=/etc/qemu-ifup,downscript=/etc/qemu-ifdown -device e1000,netdev=hn0,mac=50:52:54:00:11:22 -boot c -drive if=none,file=./Fedora-rdma-server.qcow2,id=drive-virtio-disk0 -device virtio-blk-pci,bus=pci.0,addr=0x4,drive=drive-virtio-disk0,id=virtio-disk0 -m 2048 -smp 2 -device piix3-usb-uhci -device usb-tablet -monitor stdio -vga qxl -spice streaming-video=filter,port=5901,disable-ticketing -S
qemu-system-x86_64: -spice streaming-video=filter,port=5901,disable-ticketing: warning: short-form boolean option 'disable-ticketing' deprecated
Please use disable-ticketing=on instead
QEMU 6.0.50 monitor - type 'help' for more information
(qemu)
(qemu) trace-event qemu_rdma_block_for_wrid_miss on
(qemu) migrate -d rdma:192.168.22.23:8888
source_resolve_host RDMA Device opened: kernel name rxe_eth0 uverbs device name uverbs2, infiniband_verbs class device path /sys/class/infiniband_verbs/uverbs2, infiniband class device path /sys/class/infiniband/rxe_eth0, transport: (2) Ethernet
(qemu) qemu_rdma_block_for_wrid_miss A Wanted wrid WRITE RDMA (1) but got CONTROL RECV (4000)

NOTE: we use soft RoCE as the rdma device.
[root@iaas-rpma images]# rdma link show rxe_eth0/1
link rxe_eth0/1 state ACTIVE physical_state LINK_UP netdev eth0

This migration could not be completed when out of order(OOO) CQ event occurs.
The send queue and receive queue shared a same completion queue, and
qemu_rdma_block_for_wrid() will drop the CQs it's not interested in. But
the dropped CQs by qemu_rdma_block_for_wrid() could be later CQs it wants.
So in this case, qemu_rdma_block_for_wrid() will block forever.

OOO cases will occur in both source side and destination side. And a
forever blocking happens on only SEND and RECV are out of order. OOO between
'WRITE RDMA' and 'RECV' doesn't matter.

below the OOO sequence:
       source                             destination
      rdma_write_one()                   qemu_rdma_registration_handle()
1.    S1: post_recv X                    D1: post_recv Y
2.    wait for recv CQ event X
3.                                       D2: post_send X     ---------------+
4.                                       wait for send CQ send event X (D2) |
5.    recv CQ event X reaches (D2)                                          |
6.  +-S2: post_send Y                                                       |
7.  | wait for send CQ event Y                                              |
8.  |                                    recv CQ event Y (S2) (drop it)     |
9.  +-send CQ event Y reaches (S2)                                          |
10.                                      send CQ event X reaches (D2)  -----+
11.                                      wait recv CQ event Y (dropped by (8))

Although a hardware IB works fine in my a hundred of runs, the IB specification
doesn't guaratee the CQ order in such case.

Here we introduce a independent send completion queue to distinguish
ibv_post_send completion queue from the original mixed completion queue.
It helps us to poll the specific CQE we are really interested in.

Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>