Luiz Capitulino [Wed, 14 May 2014 09:43:10 +0000 (17:43 +0800)]
man: improve -numa doc
The -numa option documentation in qemu's manpage lacks the command-line
options and some information regarding how it relates to options -m and
-smp. This commit fills in the missing text.
Wanlong Gao [Wed, 14 May 2014 09:43:06 +0000 (17:43 +0800)]
NUMA: check if the total numa memory size is equal to ram_size
If the total number of the assigned numa nodes memory is not
equal to the assigned ram size, it will write the wrong data
to ACPI table, then the guest will ignore the wrong ACPI table
and recognize all memory to one node. It's buggy, we should
check it to ensure that we write the right data to ACPI table.
This test needs a bit more work: issues have been
found on legacy systems, disable it for now to
avoid false positives for people.
Will re-enable after issues are addressed.
Nikolay Nikolaev [Tue, 10 Jun 2014 10:03:23 +0000 (13:03 +0300)]
Add qtest for vhost-user
This test creates a 'server' chardev to listen for vhost-user messages.
Once VHOST_USER_SET_MEM_TABLE is received it mmaps each received region,
and read 1k bytes from it. The read data is compared to data from readl.
The test requires hugetlbfs to be already mounted and writable. The mount
point defaults to '/hugetlbfs' and can be specified via the environment
variable QTEST_HUGETLBFS_PATH.
The rom pc-bios/pxe-virtio.rom is used to instantiate a virtio pcicontroller.
This test needs a bit more work: issues have been
found on legacy systems, disable it for now to
avoid false positives for people.
Will re-enable after issues are addressed.
Nikolay Nikolaev [Tue, 27 May 2014 12:07:10 +0000 (15:07 +0300)]
libqemustub: add stubs to be able to use qemu-char.c
chardev depends on lots of external symbols that are not necessarily
needed to be able to use, for example, 'socket chardev'. So add stubs
for these functions:
Nikolay Nikolaev [Tue, 10 Jun 2014 10:02:57 +0000 (13:02 +0300)]
Add vhost-user protocol documentation
This document describes the basic message format used by vhost-user
for communication over a unix domain socket. The protocol is based
on the existing ioctl interface used for the kernel version of vhost.
Nikolay Nikolaev [Tue, 10 Jun 2014 10:02:16 +0000 (13:02 +0300)]
Add the vhost-user netdev backend to the command line
The supplied chardev id will be inspected for supported options. Only
a socket backend, with a set path (i.e. a Unix socket) and optionally
the server parameter set, will be allowed. Other options (nowait, telnet)
will make the chardev unusable and the netdev will not be initialised.
Nikolay Nikolaev [Tue, 27 May 2014 12:06:29 +0000 (15:06 +0300)]
Add new vhost-user netdev backend
Add a new QEMU netdev backend that is intended to invoke vhost_net with the
vhost-user backend. It uses an Unix socket chardev to establish a
communication with the 'slave' (client and server mode supported).
At runtime the netdev will handle OPEN/CLOSE events from the chardev. Upon
disconnection it will set link_down accordingly and notify virtio-net; the
virtio-net interface will go down.
Nikolay Nikolaev [Tue, 27 May 2014 12:06:16 +0000 (15:06 +0300)]
vhost-net: vhost-user feature bits support
Handle the feature bits negotiation when using vhost-user. Allow
the underlying implementation to have a finer control over all the
bits except the VIRTIO_NET_F_MAC.
Nikolay Nikolaev [Tue, 27 May 2014 12:06:02 +0000 (15:06 +0300)]
Add vhost-user as a vhost backend.
The initialization takes a chardev backed by a unix domain socket.
It should implement qemu_fe_set_msgfds in order to be able to pass
file descriptors to the remote process.
Each ioctl request of vhost-kernel has a vhost-user message equivalent,
which is sent over the control socket.
The general approach is to copy the data from the supplied argument
pointer to a designated field in the message. If a file descriptor is
to be passed it will be placed in the fds array for inclusion in
the sendmsg control header.
VHOST_SET_MEM_TABLE ignores the supplied vhost_memory structure and scans
the global ram_list for ram blocks with a valid fd field set. This would
be set when the '-object memory-file' option with share=on property is used.
Nikolay Nikolaev [Tue, 27 May 2014 12:05:49 +0000 (15:05 +0300)]
Add vhost-backend and VhostBackendType
Use vhost_set_backend_type to initialise a proper vhost_ops structure.
In vhost_net_init and vhost_net_start_one call conditionally TAP related
initialisation depending on the vhost backend type.
Nikolay Nikolaev [Tue, 27 May 2014 12:05:35 +0000 (15:05 +0300)]
Add vhost_ops to vhost_dev struct and replace all relevant ioctls
Decouple vhost from the Linux kernel by introducing vhost_ops. The
intention is to provide different backends - a 'kernel' backend based on
the ioctl interface, and an 'user' backend based on a UNIX domain socket
and shared memory interface.
Nikolay Nikolaev [Tue, 27 May 2014 12:05:22 +0000 (15:05 +0300)]
vhost_net_init will use VhostNetOptions to get all its arguments
vhost_dev_init will replace devfd and devpath with a single opaque argument.
This is initialised with a file descriptor. When TAP is used (through
vhost_net), open /dev/vhost-net and pass the fd as an opaque parameter in
VhostNetOptions. The same applies to vhost-scsi - open /dev/vhost-scsi and
pass the fd.
Nikolay Nikolaev [Tue, 27 May 2014 12:04:55 +0000 (15:04 +0300)]
vhost_net should call the poll callback only when it is set
The poll callback needs to be called when bringing up or down
the vhost_net instance. As it is not mandatory for an NetClient
to implement it, invoke it only when it is set.
Nikolay Nikolaev [Tue, 27 May 2014 12:04:42 +0000 (15:04 +0300)]
vhost: add vhost_get_features and vhost_ack_features
Generalize the features get/ack to be used for both vhost-net and vhost-scsi.
In vhost-net add vhost_net_get_feature_bits to select the feature bit set
depending on the NetClient kind.
Nikolay Nikolaev [Tue, 27 May 2014 12:04:15 +0000 (15:04 +0300)]
Add chardev API qemu_chr_fe_get_msgfds
This extends the existing qemu_chr_fe_get_msgfd by allowing to read a set
of fds. The function for receiving the fds - unix_process_msgfd is extended
to allocate the needed array size.
Nikolay Nikolaev [Tue, 27 May 2014 12:04:02 +0000 (15:04 +0300)]
Add chardev API qemu_chr_fe_set_msgfds
This will set an array of file descriptors to the internal structures.
The next time a message is send the array will be send as ancillary
data. This feature works on the UNIX domain socket backend only.
Nikolay Nikolaev [Tue, 27 May 2014 12:03:48 +0000 (15:03 +0300)]
Add chardev API qemu_chr_fe_read_all
This function will attempt to read data from the chardev trying
to fill the buffer up to the given length.
Add tcp_chr_disconnect to reuse disconnection code where needed.
Jason Wang [Tue, 20 May 2014 06:01:44 +0000 (14:01 +0800)]
virtio-net: announce self by guest
It's hard to track all mac addresses and their configurations (e.g
vlan or ipv6) in qemu. Without this information, it's impossible to
build proper garp packet after migration. The only possible solution
to this is let guest (who knows all configurations) to do this.
So, this patch introduces a new readonly config status bit of virtio-net,
VIRTIO_NET_S_ANNOUNCE which is used to notify guest to announce
presence of its link through config update interrupt.When guest has
done the announcement, it should ack the notification through
VIRTIO_NET_CTRL_ANNOUNCE_ACK cmd. This feature is negotiated by a new
feature bit VIRTIO_NET_F_ANNOUNCE (which has already been supported by
Linux guest).
During load, a counter of announcing rounds is set so that after the vm is
running it can trigger rounds of config interrupts to notify the guest to build
and send the correct garps.
Jason Wang [Tue, 20 May 2014 06:01:43 +0000 (14:01 +0800)]
migration: introduce self_announce_delay()
This patch introduces self_announce_delay() to calculate the delay for
the next announce round. This could be used by other device e.g
virtio-net who wants to do announcing by itself.
Igor Mammedov [Mon, 2 Jun 2014 13:25:28 +0000 (15:25 +0200)]
pc: ACPI BIOS: reserve SRAT entry for hotplug mem hole
Needed for Windows to use hotplugged memory device, otherwise
it complains that server is not configured for memory hotplug.
Tests shows that aftewards it uses dynamically provided
proximity value from _PXM() method if available.
Igor Mammedov [Mon, 2 Jun 2014 13:25:26 +0000 (15:25 +0200)]
pc: ACPI BIOS: implement memory hotplug interface
- provides static SSDT object for memory hotplug that can handle
upto 256 hotplugable memory slots
- SSDT template for memory devices and runtime generator
of them in SSDT table.
Igor Mammedov [Mon, 2 Jun 2014 13:25:21 +0000 (15:25 +0200)]
pc: ich9 lpc: make it work with global/compat properties
Propeties of object should be available after its instances_init()
callback is finished and not added in PCIDeviceClass.init which is
roughly corresponds to realize() method.
Moving properties adding into instances_init will fix missing
property error when global/compat property mechanism is used.
Igor Mammedov [Mon, 2 Jun 2014 13:25:18 +0000 (15:25 +0200)]
trace: pc: add PC_DIMM slot & address allocation
Add mhp_pc_dimm_assigned_slot & mhp_pc_dimm_assigned_address
events to trace which address and slot where assigned to
plugged in PC_DIMM device on target-i386 machine.
Igor Mammedov [Mon, 2 Jun 2014 13:25:16 +0000 (15:25 +0200)]
acpi: memory hotplug ACPI hardware implementation
- implements QEMU hardware part of memory hotplug protocol
described at "docs/specs/acpi_mem_hotplug.txt"
- handles only memory add notification event for now
Igor Mammedov [Mon, 2 Jun 2014 13:25:13 +0000 (15:25 +0200)]
pc-dimm: add busy address check and address auto-allocation
- if 'addr' property is not specified on -device/device_add command,
treat the default value as request for assigning PCDIMMDevice to
the first free memory region.
- if 'addr' is provided with -device/device_add command, attempt to
use it or fail command if it's already occupied or falls inside
of an existing PCDIMMDevice memory region.
Note:
GCompareFunc(a, b) used by g_slist_insert_sorted() returns 'gint',
however it might be too small to fit difference between
2 addresses. So use 128bit to calculate the difference and normalize
result to -1/0/1 return values.
Igor Mammedov [Mon, 2 Jun 2014 13:25:07 +0000 (15:25 +0200)]
pc-dimm: do not allow setting an in-use memdev
using the same memdev backend more than once will cause
assertion at MemoryRegion mapping time because it's already
mapped. Prevent it by checking that the associated MemoryRegion
is not mapped.
Each hotplug-able memory slot is a PCDIMMDevice.
A hot-add operation for a memory device:
- creates a new PCDIMMDevice and makes hotplug controller to map it into
guest address space
Hotplug operations are done through normal device_add commands.
For migration case, all hotplugged memory devices on source should be
specified on target's command line using '-device' option with
properties set to the same values as on source.
To simplify review, patch introduces only PCDIMMDevice QOM skeleton that
will be extended by following patches to implement actual memory hotplug
and related functions.
Igor Mammedov [Mon, 2 Jun 2014 13:25:03 +0000 (15:25 +0200)]
qdev: hotplug for bus-less devices
Add get_hotplug_handler() method to machine, and
make bus-less device use it during hotplug
as a means to discover a hotplug handler controller.
The returned controller is used to perform hotplug
actions.
Igor Mammedov [Fri, 6 Jun 2014 15:54:29 +0000 (17:54 +0200)]
add memdev backend infrastructure
Provides framework for splitting host RAM allocation/
policies into a separate backend that could be used
by devices.
Initially only legacy RAM backend is provided, which
uses memory_region_init_ram() allocator and compatible
with every CLI option that affects memory_region_init_ram().
Igor Mammedov [Mon, 2 Jun 2014 13:25:00 +0000 (15:25 +0200)]
vl.c: daemonize before guest memory allocation
memory allocated for guest before QEMU is daemonized and then mapped
later in guest's address space after it is daemonized, leads to EPT
violation and QEMU aborts.
To avoid this and similar issues switch to daemonized mode early
before applying/processing other options.
Igor Mammedov [Mon, 2 Jun 2014 13:24:59 +0000 (15:24 +0200)]
object_add: allow completion handler to get canonical path
Add object to /objects before calling user_creatable_complete()
handler, so that object might be able to call
object_get_canonical_path() in its completion handler.
Peter Maydell [Wed, 18 Jun 2014 14:08:38 +0000 (15:08 +0100)]
Merge remote-tracking branch 'remotes/bonzini/memory' into staging
* remotes/bonzini/memory:
memory: Don't call memory_region_update_coalesced_range if nothing changed
memory: MemoryRegion: rename parent to container
memory: MemoryRegion: factor out memory region re-adder
memory: MemoryRegion: factor out subregion add functionality
qtest: fix qtest_clock_warp() for no deadline case
exec: dummy_section: Pass address space through.
memory: Simplify mr_add_subregion() if-else
memory: Don't update all memory region when ioeventfd changed
unset RAMBlock idstr when unregister MemoryRegion
exec: introduce qemu_ram_unset_idstr() to unset RAMBlock idstr
MAINTAINERS: Add myself as Memory API maintainer
Fam Zheng [Fri, 13 Jun 2014 06:34:41 +0000 (14:34 +0800)]
memory: Don't call memory_region_update_coalesced_range if nothing changed
With huge number of PCI devices in the system (for example, 200
virtio-blk-pci), this unconditional call can slow down emulation of
irrelevant PCI operations drastically, such as a BAR update on a device
that has no coalescing region. So avoid it.
memory: MemoryRegion: factor out memory region re-adder
memory_region_set_address is mostly just a function that deletes and
re-adds a memory region. Factor this generic functionality out into a
re-usable function. This prepares support for further QOMification
of MemoryRegion.
memory: MemoryRegion: factor out subregion add functionality
Split off the core looping code that actually adds subregions into
it's own fn. This prepares support for Memory Region qomification
where setting the MR address or parent via QOM will back onto this more
minimal function.
This if else is not needed. The previous call to memory_region_add
(whether _overlap or not) will always set priority and may_overlap
to desired values. And its not possible to get here without having
called memory_region_add_subregion due to the null guard on parent.
So we can just directly call memory_region_add_subregion_common.
Gonglei [Thu, 8 May 2014 03:47:32 +0000 (11:47 +0800)]
memory: Don't update all memory region when ioeventfd changed
memory mappings don't rely on ioeventfds, there is no need
to destroy and rebuild them when manipulating ioeventfds,
otherwise it scarifies performance.
according to testing result, each ioeventfd deleing needs
about 5ms, within which memory mapping rebuilding needs
about 4ms. With many Nics and vmchannel in a VM doing migrating,
there can be many ioeventfds deleting which increasing
downtime remarkably.
Craig Heffner [Mon, 5 Dec 2011 19:14:27 +0000 (14:14 -0500)]
User mode support for Linux ELF files with no section header
In user mode Linux, Qemu currently refuses to load ELF files that do not
contain section headers (ehdr->e_shentsize == 0). Since section headers are not
required in order to load an ELF file, simply removing the e_shentsize check in
elf_check_ehdr() allows ELF binaries with no section headers to be run properly
in user mode:
Ed Swierk [Tue, 6 May 2014 03:04:45 +0000 (20:04 -0700)]
linux-user: Return correct errno for unsupported netlink socket
This fixes "Cannot open audit interface - aborting." when the
EAFNOSUPPORT errno differs between the target and host
architectures (e.g. mips target and x86_64 host).
Peter Maydell [Thu, 15 May 2014 13:40:23 +0000 (14:40 +0100)]
linux-user: Don't overrun guest buffer in sched_getaffinity
If the guest's "long" type is smaller than the host's, then
our sched_getaffinity wrapper needs to round the buffer size
up to a multiple of the host sizeof(long). This means that when
we copy the data back from the host buffer to the guest's
buffer there might be more than we can fit. Rather than
overflowing the guest's buffer, handle this case by returning
EINVAL or ignoring the unused extra space, as appropriate.
Note that only guests using the syscall interface directly might
run into this bug -- the glibc wrappers around it will always
use a buffer whose size is a multiple of 8 regardless of guest
architecture.
Peter Maydell [Sat, 10 May 2014 11:25:53 +0000 (12:25 +0100)]
linux-user/uname: Return correct uname string for x86_64
We were returning the incorrect uname string (with a hyphen, not
an underscore) for x86_64. Fix this by removing the x86_64 special
case, since the default "just use UNAME_MACHINE" behaviour suffices.
This leaves cpu_to_uname_machine() special cases for only those
architectures which need to vary the string based on runtime CPU
features.
linux-user: fix gcc-4.9 compiler error on __{get,put]}_user
gcc-4.9 finds unused operand:
linux-user/syscall.c: In function ‘host_to_target_stat64’:
linux-user/qemu.h:301:19: error: right-hand operand of comma expression
has no effect [-Werror=unused-value]
((hptr), (x)), 0)
Just removing the rh operand is no good, it will error in later:
linux-user/main.c: In function ‘arm_kernel_cmpxchg64_helper’:
linux-user/qemu.h:330:15: error: void value not ignored as it ought to be
__ret = __put_user((x), __hptr); \
Thus, remove setting __ret from __get_user and __put_user, as and
set the right hand operand to (void)0 to make it clear that these
return never nothing.
This commit depends on the signal.c cleanup, to ensure bisectable
version history.
Remove "if(__get_user" checks and their related error paths
for all architecture's do_sigreturn. Remove the unlock_user_struct
when the only way to end up there is from failed lock_user_struct.
v3: remove unneccesary sigsegv label as suggested by Peter
A function never called from anywhere, obviously half-complete.
Remove function and if someone wants to complete this, please
check the old version out of git history.
signal/all: remove return value from restore_sigcontext
make most implementations of restore_sigcontext void and
remove checking it's return value from functions calling
restore_sigcontext.
The exception is the X86 version of the function that is
too different from others to deal in this way, and arm
version, to keep possibility of erroring out from failed
valid_user_regs.
v3: keep arm valid_user_regs for filling in near future.
projects / qemu.git / log