Peter Maydell [Mon, 6 Jun 2016 18:58:19 +0000 (19:58 +0100)]
linux-user: Special-case ERESTARTSYS in target_strerror()
Since TARGET_ERESTARTSYS and TARGET_ESIGRETURN are internal-to-QEMU
error numbers, handle them specially in target_strerror(), to avoid
confusing strace output like:
Peter Maydell [Mon, 6 Jun 2016 18:58:16 +0000 (19:58 +0100)]
linux-user: Correct signedness of target_flock l_start and l_len fields
The l_start and l_len fields in the various target_flock structures are
supposed to be '__kernel_off_t' or '__kernel_loff_t', which means they
should be signed, not unsigned. Correcting the structure definitions means
that __get_user() and __put_user() will correctly sign extend them if
the guest is using 32 bit offsets and the host is using 64 bit offsets.
This fixes failures in the LTP 'fcntl14' tests where it checks that
negative seek offsets work correctly.
We reindent the structures to drop hard tabs since we're touching 40%
of the fields anyway.
Peter Maydell [Mon, 6 Jun 2016 18:58:12 +0000 (19:58 +0100)]
linux-user: Use safe_syscall wrapper for semop
Use the safe_syscall wrapper for the semop syscall or IPC operation.
(We implement via the semtimedop syscall to make it easier to
implement the guest semtimedop syscall later.)
Peter Maydell [Mon, 6 Jun 2016 18:58:11 +0000 (19:58 +0100)]
linux-user: Use safe_syscall wrapper for epoll_wait syscalls
Use the safe_syscall wrapper for epoll_wait and epoll_pwait syscalls.
Since we now directly use the host epoll_pwait syscall for both
epoll_wait and epoll_pwait, we don't need the configure machinery
to check whether glibc supports epoll_pwait(). (The kernel has
supported the syscall since 2.6.19 so we can assume it's always there.)
Peter Maydell [Mon, 6 Jun 2016 18:58:10 +0000 (19:58 +0100)]
linux-user: Use safe_syscall wrapper for poll and ppoll syscalls
Use the safe_syscall wrapper for the poll and ppoll syscalls.
Since not all host architectures will have a poll syscall, we
have to rewrite the TARGET_NR_poll handling to use ppoll instead
(we can assume everywhere has ppoll by now).
We take the opportunity to switch to the code structure
already used in the implementation of epoll_wait and epoll_pwait,
which uses a switch() to avoid interleaving #if and if (),
and to stop using a variable with a leading '_' which is in
the implementation's namespace.
Peter Maydell [Mon, 6 Jun 2016 18:58:05 +0000 (19:58 +0100)]
linux-user: Use safe_syscall wrapper for msgsnd and msgrcv
Use the safe_syscall wrapper for msgsnd and msgrcv syscalls.
This is made slightly awkward by some host architectures providing
only a single 'ipc' syscall rather than separate syscalls per
operation; we provide safe_msgsnd() and safe_msgrcv() as wrappers
around safe_ipc() to handle this if needed.
Peter Maydell [Tue, 31 May 2016 14:45:10 +0000 (15:45 +0100)]
linux-user: Fix NR_fadvise64 and NR_fadvise64_64 for 32-bit guests
Fix errors in the implementation of NR_fadvise64 and NR_fadvise64_64
for 32-bit guests, which pass their off_t values in register pairs.
We can't use the 64-bit code path for this, so split out the 32-bit
cases, so that we can correctly handle the "only offset is 64-bit"
and "both offset and length are 64-bit" syscall flavours, and
"uses aligned register pairs" and "does not" flavours of target.
Peter Maydell [Tue, 31 May 2016 14:45:09 +0000 (15:45 +0100)]
linux-user: Fix handling of arm_fadvise64_64 syscall
32-bit ARM has an odd variant of the fadvise syscall which has
rearranged arguments, which we try to implement. Unfortunately we got
the rearrangement wrong.
This is a six-argument syscall whose arguments are:
* fd
* advise parameter
* offset high half
* offset low half
* len high half
* len low half
Stop trying to share code with the standard fadvise syscalls,
and just implement the syscall with the correct argument order.
Peter Maydell [Mon, 6 Jun 2016 18:56:54 +0000 (19:56 +0100)]
linux-user: provide frame information in x86-64 safe_syscall
Use cfi directives in the x86-64 safe_syscall to allow gdb to get
backtraces right from within it. (In particular this will be
quite a common situation if the user interrupts QEMU while it's
in a blocked safe-syscall: at the point of the syscall insn RBP
is in use for something else, and so gdb can't find the frame then
without assistance.)
Peter Maydell [Fri, 27 May 2016 14:51:59 +0000 (15:51 +0100)]
linux-user: Use both si_code and si_signo when converting siginfo_t
The siginfo_t struct includes a union. The correct way to identify
which fields of the union are relevant is complicated, because we
have to use a combination of the si_code and si_signo to figure out
which of the union's members are valid. (Within the host kernel it
is always possible to tell, but the kernel carefully avoids giving
userspace the high 16 bits of si_code, so we don't have the
information to do this the easy way...) We therefore make our best
guess, bearing in mind that a guest can spoof most of the si_codes
via rt_sigqueueinfo() if it likes. Once we have made our guess, we
record it in the top 16 bits of the si_code, so that tswap_siginfo()
later can use it. tswap_siginfo() then strips these top bits out
before writing si_code to the guest (sign-extending the lower bits).
This fixes a bug where fields were sometimes wrong; in particular
the LTP kill10 test went into an infinite loop because its signal
handler got a si_pid value of 0 rather than the pid of the sending
process.
As part of this change, we switch to using __put_user() in the
tswap_siginfo code which writes out the byteswapped values to
the target memory, in case the target memory pointer is not
sufficiently aligned for the host CPU's requirements.
If there is a signal pending during fork() the signal handler will
erroneously be called in both the parent and child, so handle any
pending signals first.
Peter Maydell [Fri, 27 May 2016 14:51:57 +0000 (15:51 +0100)]
linux-user: Use safe_syscall for kill, tkill and tgkill syscalls
Use the safe_syscall wrapper for the kill, tkill and tgkill syscalls.
Without this, if a thread sent a SIGKILL to itself it could kill the
thread before we had a chance to process a signal that arrived just
before the SIGKILL, and that signal would get lost.
We drop all the ifdeffery for tkill and tgkill, because every guest
architecture we support implements them, and they've been in Linux
since 2003 so we can assume the host headers define the __NR_tkill
and __NR_tgkill constants.
linux-user: pause() should not pause if signal pending
Fix races between signal handling and the pause syscall by
reimplementing it using block_signals() and sigsuspend().
(Using safe_syscall(pause) would also work, except that the
pause syscall doesn't exist on all architectures.)
linux-user: Block signals during sigaction() handling
Block signals while emulating sigaction. This is a non-interruptible
syscall, and using block_signals() avoids races where the host
signal handler is invoked and tries to examine the signal handler
data structures while we are updating them.
If a synchronous signal and an asynchronous signal arrive near simultaneously,
and the signal number of the asynchronous signal is lower than that of the
synchronous signal the the handler for the asynchronous would be called first,
and then the handler for the synchronous signal would be called within or
after the first handler with an incorrect context.
This is fixed by queuing synchronous signals separately. Note that this does
risk delaying a asynchronous signal until the synchronous signal handler
returns rather than handling the signal on another thread, but this seems
unlikely to cause problems for real guest programs and is unavoidable unless
we could guarantee to roll back and reexecute whatever guest instruction
caused the synchronous signal (which would be a bit odd if we've already
logged its execution, for instance, and would require careful analysis of
all guest CPUs to check it was possible in all cases).
As host signals are now blocked whenever guest signals are blocked, the
queue of realtime signals is now in Linux. The QEMU queue is now
redundant and can be removed. (We already did not queue non-RT signals, and
none of the calls to queue_signal() except the one in host_signal_handler()
pass an RT signal number.)
linux-user: Remove redundant default action check in queue_signal()
Both queue_signal() and process_pending_signals() did check for default
actions of signals, this is redundant and also causes fatal and stopping
signals to incorrectly cause guest system calls to be interrupted.
Peter Maydell [Fri, 27 May 2016 14:51:49 +0000 (15:51 +0100)]
linux-user: Fix race between multiple signals
If multiple host signals are received in quick succession they would
be queued in TaskState then delivered to the guest in spite of
signals being supposed to be blocked by the guest signal handler's
sa_mask. Fix this by decoupling the guest signal mask from the
host signal mask, so we can have protected sections where all
host signals are blocked. In particular we block signals from
when host_signal_handler() queues a signal from the guest until
process_pending_signals() has unqueued it. We also block signals
while we are manipulating the guest signal mask in emulation of
sigprocmask and similar syscalls.
Blocking host signals also ensures the correct behaviour with respect
to multiple threads and the overrun count of timer related signals.
Alas blocking and queuing in qemu is still needed because of virtual
processor exceptions, SIGSEGV and SIGBUS.
Blocking signals inside process_pending_signals() protects against
concurrency problems that would otherwise happen if host_signal_handler()
ran and accessed the signal data structures while process_pending_signals()
was manipulating them.
Since we now track the guest signal mask separately from that
of the host, the sigsuspend system calls must track the signal
mask passed to them, because when we process signals as we leave
the sigsuspend the guest signal mask in force is that passed to
sigsuspend.
Signed-off-by: Timothy Edward Baldwin <[email protected]>
Message-id: 1441497448[email protected]
[PMM: make signal_pending a simple flag rather than a word with two flag bits;
ensure we don't call block_signals() twice in sigreturn codepaths;
document and assert() the guarantee that using do_sigprocmask() to
get the current mask never fails; use the qemu atomics.h functions
rather than raw volatile variable access; add extra commentary and
documentation; block SIGSEGV/SIGBUS in block_signals() and in
process_pending_signals() because they can't occur synchronously here;
check the right do_sigprocmask() call for errors in ssetmask syscall;
expand commit message; fixed sigsuspend() hanging] Reviewed-by: Peter Maydell <[email protected]> Signed-off-by: Peter Maydell <[email protected]> Signed-off-by: Riku Voipio <[email protected]>
Peter Maydell [Fri, 27 May 2016 14:51:48 +0000 (15:51 +0100)]
linux-user: Use safe_syscall for sigsuspend syscalls
Use the safe_syscall wrapper for sigsuspend syscalls. This
means that we will definitely deliver a signal that arrives
before we do the sigsuspend call, rather than blocking first
and delivering afterwards.
Peter Maydell [Fri, 27 May 2016 14:51:47 +0000 (15:51 +0100)]
linux-user: Define macro for size of host kernel sigset_t
Some host syscalls take an argument specifying the size of a
host kernel's sigset_t (which isn't necessarily the same as
that of the host libc's type of that name). Instead of hardcoding
_NSIG / 8 where we do this, define and use a SIGSET_T_SIZE macro.
Peter Maydell [Fri, 27 May 2016 14:51:46 +0000 (15:51 +0100)]
linux-user: Factor out uses of do_sigprocmask() from sigreturn code
All the architecture specific handlers for sigreturn include calls
to do_sigprocmask(SIGSETMASK, &set, NULL) to set the signal mask
from the uc_sigmask in the context being restored. Factor these
out into calls to a set_sigmask() function. The next patch will
want to add code which is not run when setting the signal mask
via do_sigreturn, and this change allows us to separate the two
cases.
Peter Maydell [Fri, 27 May 2016 14:51:43 +0000 (15:51 +0100)]
linux-user: Factor out handle_signal code from process_pending_signals()
Factor out the code to handle a single signal from the
process_pending_signals() function. The use of goto for flow control
is OK currently, but would get significantly uglier if extended to
allow running the handle_signal code multiple times.
Laurent Vivier [Thu, 2 Jun 2016 20:14:15 +0000 (22:14 +0200)]
linux-user: check if NETLINK_ROUTE is available
Some IFLA_* symbols can be missing in the host linux/if_link.h,
but as they are enums and not "#defines", check in "configure" if
last known (IFLA_PROTO_DOWN) is available and if not, disable
management of NETLINK_ROUTE protocol.
Laurent Vivier [Sun, 22 May 2016 16:56:19 +0000 (18:56 +0200)]
linux-user: add rtnetlink(7) support
rtnetlink is needed to use iproute package (ip addr, ip route)
and dhcp client.
Examples:
Without this patch:
# ip link
Cannot open netlink socket: Address family not supported by protocol
# ip addr
Cannot open netlink socket: Address family not supported by protocol
# ip route
Cannot open netlink socket: Address family not supported by protocol
# dhclient eth0
Cannot open netlink socket: Address family not supported by protocol
Cannot open netlink socket: Address family not supported by protocol
With this patch:
# ip link
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN mode DEFAULT
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
51: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP mode DEFAULT qlen 1000
link/ether 00:16:3e:89:6b:d7 brd ff:ff:ff:ff:ff:ff
# ip addr show eth0
51: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP qlen 1000
link/ether 00:16:3e:89:6b:d7 brd ff:ff:ff:ff:ff:ff
inet 192.168.122.197/24 brd 192.168.122.255 scope global eth0
valid_lft forever preferred_lft forever
inet6 fe80::216:3eff:fe89:6bd7/64 scope link
valid_lft forever preferred_lft forever
# ip route
default via 192.168.122.1 dev eth0
192.168.122.0/24 dev eth0 proto kernel scope link src 192.168.122.197
# ip addr flush eth0
# ip addr add 192.168.122.10 dev eth0
# ip addr show eth0
51: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP qlen 1000
link/ether 00:16:3e:89:6b:d7 brd ff:ff:ff:ff:ff:ff
inet 192.168.122.10/32 scope global eth0
valid_lft forever preferred_lft forever
# ip route add 192.168.122.0/24 via 192.168.122.10
# ip route
192.168.122.0/24 via 192.168.122.10 dev eth0
Laurent Vivier [Fri, 29 Jan 2016 16:07:31 +0000 (17:07 +0100)]
linux-user: Fix qemu-binfmt-conf.sh to store config across reboot
Original qemu-binfmt-conf.sh is only able to write configuration
into /proc/sys/fs/binfmt_misc, and the configuration is lost on reboot.
This script can configure debian and systemd services to restore
configuration on reboot. Moreover, it is able to manage binfmt
credential and to configure the path of the interpreter.
--help: display this usage
--qemu-path: set path to qemu interpreter (/usr/local/bin)
--debian: don't write into /proc,
instead generate update-binfmts templates
--systemd: don't write into /proc,
instead generate file for systemd-binfmt.service
for the given CPU
--exportdir: define where to write configuration files
(default: /etc/binfmt.d or /usr/share/binfmts)
--credential: if yes, credential an security tokens are
calculated according to the binary to interpret
Peter Maydell [Mon, 6 Jun 2016 16:02:42 +0000 (17:02 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20160606-1' into staging
target-arm queue:
* support instruction syndrome info for data aborts from A64 to EL2
* add HSTR_EL2 register
* fix incorrect ESR IL bits in various syndrome register cases
* virt: fix limit of 64-bit ACPI/ECAM PCI MMIO range
* gicv2: RAZ/WI non-sec access to sec interrupts
* i2c: add aspeed i2c controller
* virt: Reject gic-version=host for non-KVM (don't segv on aarch64 host)
* xlnx-zynqmp: Add a secure prop to en/disable ARM Security Extensions
* xlnx-zynqmp: Support KVM on AArch64 hosts
* ptimer: Various fixes for awkward corner cases
* char: QOMify various ARM UART models
* char: get rid of qemu_char_get_next_serial
* target-arm: Fix TTBR selecting logic on AArch32 Stage 2 translation
* zynqmp: Add the ZCU102 board
# gpg: Signature made Mon 06 Jun 2016 17:01:11 BST
# gpg: using RSA key 0x3C2525ED14360CDE
# gpg: Good signature from "Peter Maydell <[email protected]>"
# gpg: aka "Peter Maydell <[email protected]>"
# gpg: aka "Peter Maydell <[email protected]>"
* remotes/pmaydell/tags/pull-target-arm-20160606-1: (25 commits)
zynqmp: Add the ZCU102 board
target-arm: Fix TTBR selecting logic on AArch32 Stage 2 translation
char: get rid of qemu_char_get_next_serial
hw/char: QOM'ify xilinx_uartlite model
hw/char: QOM'ify stm32f2xx_usart model
hw/char: QOM'ify digic-uart model
hw/char: QOM'ify cadence_uart model
hw/char: QOM'ify pl011 model
hw/ptimer: Introduce ptimer_get_limit
hw/ptimer: Support "on the fly" timer mode switch
hw/ptimer: Update .delta on period/freq change
hw/ptimer: Perform counter wrap around if timer already expired
hw/ptimer: Fix issues caused by the adjusted timer limit value
xlnx-zynqmp: Use the in kernel GIC model for KVM runs
xlnx-zynqmp: Delay realization of GIC until post CPU realization
xlnx-zynqmp: Make the RPU subsystem optional
xlnx-zynqmp: Add a secure prop to en/disable ARM Security Extensions
hw/arm/virt: Reject gic-version=host for non-KVM
i2c: add aspeed i2c controller
hw/intc/gic: RAZ/WI non-sec access to sec interrupts
...
Most Zynq UltraScale+ users will be targetting and using the ZCU102
board instead of the development focused EP108. To make our QEMU machine
names clearer add a ZCU102 machine model.
Sergey Sorokin [Mon, 6 Jun 2016 15:59:32 +0000 (16:59 +0100)]
target-arm: Fix TTBR selecting logic on AArch32 Stage 2 translation
Address size is 40-bit for the AArch32 stage 2 translation,
and t0sz can be negative (from -8 to 7),
so we need to adjust it to use the existing TTBR selecting logic.
xiaoqiang zhao [Mon, 6 Jun 2016 15:59:32 +0000 (16:59 +0100)]
hw/char: QOM'ify xilinx_uartlite model
* drop qemu_char_get_next_serial and use chardev prop
* create xilinx_uartlite_create wrapper function to create
xilinx_uartlite device
* change affected board code to use the new way
xiaoqiang zhao [Mon, 6 Jun 2016 15:59:31 +0000 (16:59 +0100)]
hw/char: QOM'ify cadence_uart model
* drop qemu_char_get_next_serial and use chardev prop
* create cadence_uart_create wrapper function to create
cadence_uart_device
* change affected board code to use the new way
xiaoqiang zhao [Mon, 6 Jun 2016 15:59:31 +0000 (16:59 +0100)]
hw/char: QOM'ify pl011 model
* drop qemu_char_get_next_serial and use chardev prop
* add pl011_create wrapper function to create pl011 uart device
* change affected board code to use the new way
Dmitry Osipenko [Mon, 6 Jun 2016 15:59:30 +0000 (16:59 +0100)]
hw/ptimer: Update .delta on period/freq change
Delta value must be updated on period/freq change, otherwise running timer
would be restarted (counter reloaded with old delta). Only m68k/mcf520x
and arm/arm_timer devices are currently doing freq change correctly, i.e.
stopping the timer. Perform delta update to fix affected devices and
eliminate potential further mistakes.
Dmitry Osipenko [Mon, 6 Jun 2016 15:59:30 +0000 (16:59 +0100)]
hw/ptimer: Perform counter wrap around if timer already expired
ptimer_get_count() might be called while QEMU timer already been expired.
In that case ptimer would return counter = 0, which might be undesirable
in case of polled timer. Do counter wrap around for periodic timer to keep
it distributed. In order to achieve more accurate emulation behaviour of
certain hardware, don't perform wrap around when in icount mode and return
counter = 0 in that case (that doesn't affect polled counter distribution).
Dmitry Osipenko [Mon, 6 Jun 2016 15:59:30 +0000 (16:59 +0100)]
hw/ptimer: Fix issues caused by the adjusted timer limit value
Multiple issues here related to the timer with a adjusted .limit value:
1) ptimer_get_count() returns incorrect counter value for the disabled
timer after loading the counter with a small value, because adjusted limit
value is used instead of the original.
For instance:
1) ptimer_stop(t)
2) ptimer_set_period(t, 1)
3) ptimer_set_limit(t, 0, 1)
4) ptimer_get_count(t) <-- would return 10000 instead of 0
2) ptimer_get_count() might return incorrect value for the timer running
with a adjusted limit value.
3) Neither ptimer_set_period() nor ptimer_set_freq() are adjusting the
limit value, so it is still possible to make timer timeout value
arbitrary small.
Fix all of the above issues by adjusting timer period instead of the limit.
Perform the adjustment for periodic timer only. Use the delta value instead
of the limit to make decision whether adjustment is required, as limit could
be altered while timer is running, resulting in incorrect value returned by
ptimer_get_count.
xlnx-zynqmp: Delay realization of GIC until post CPU realization
Delay the realization of the GIC until after CPUs are
realized. This is needed for KVM as the in-kernel GIC
model will fail if it is realized with no available CPUs.
The Aspeed AST2400 integrates a set of 14 I2C/SMBus bus controllers
directly connected to the APB bus. They can be programmed as master or
slave but the propopsed model only supports the master mode.
On the TODO list, we also have :
- improve and harden the state machine.
- bus recovery support (used by the Linux driver).
- transfer mode state machine bits. this is not strictly necessary as
it is mostly used for debug. The bus busy bit is deducted from the
I2C core engine of qemu.
- support of the pool buffer: 2048 bytes of internal SRAM (not used
by the Linux driver).
Peter Maydell [Mon, 6 Jun 2016 15:59:28 +0000 (16:59 +0100)]
target-arm: Don't try to set ESR IL bit in arm_cpu_do_interrupt_aarch64()
Remove some incorrect code from arm_cpu_do_interrupt_aarch64()
which attempts to set the IL bit in the syndrome register based
on the value of env->thumb. This is wrong in several ways:
* IL doesn't indicate Thumb-vs-ARM, it indicates instruction
length (which may be 16 or 32 for Thumb and is always 32 for ARM)
* not every syndrome format uses IL like this -- for some IL is
always set, and for some it is always clear
* the code is changing esr_el[new_el] even for interrupt entry,
which is not supposed to modify ESR_ELx at all
Delete the code, and instead rely on the syndrome value in
env->exception.syndrome having already been set up with the
correct value of IL.
Peter Maydell [Mon, 6 Jun 2016 15:59:28 +0000 (16:59 +0100)]
target-arm: Set IL bit in syndromes for insn abort, watchpoint, swstep
For some exception syndrome types, the IL bit should always be set.
This includes the instruction abort, watchpoint and software step
syndrome types; add the missing ARM_EL_IL bit to the syndrome
values returned by syn_insn_abort(), syn_swstep() and syn_watchpoint().
target-arm: A64: Create Instruction Syndromes for Data Aborts
Add support for generating the ISS (Instruction Specific Syndrome) for
Data Abort exceptions taken from AArch64.
These syndromes are used by hypervisors for example to trap and emulate
memory accesses.
We save the decoded data out-of-band with the TBs at translation time.
When exceptions hit, the extra data attached to the TB is used to
recreate the state needed to encode instruction syndromes.
This avoids the need to emit moves with every load/store.
Peter Maydell [Mon, 6 Jun 2016 14:17:52 +0000 (15:17 +0100)]
Merge remote-tracking branch 'remotes/gkurz/tags/for-upstream' into staging
readdir_r() to readdir() conversion, various minor cleanups
# gpg: Signature made Mon 06 Jun 2016 10:52:52 BST
# gpg: using DSA key 0x02FC3AEB0101DBC2
# gpg: Good signature from "Greg Kurz <[email protected]>"
# gpg: aka "Greg Kurz <[email protected]>"
# gpg: aka "Greg Kurz <[email protected]>"
# gpg: aka "Gregory Kurz (Groug) <[email protected]>"
# gpg: aka "Gregory Kurz (Cimai Technology) <[email protected]>"
# gpg: aka "Gregory Kurz (Meiosys Technology) <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 2BD4 3B44 535E C0A7 9894 DBA2 02FC 3AEB 0101 DBC2
* remotes/gkurz/tags/for-upstream:
9p: switch back to readdir()
9p: add locking to V9fsDir
9p: introduce the V9fsDir type
9p: drop useless out: label
9p: drop useless inclusion of hw/i386/pc.h
9p/fsdev: remove obsolete references to virtio
9p: some more cleanup in #include directives
Peter Maydell [Mon, 6 Jun 2016 12:58:24 +0000 (13:58 +0100)]
Merge remote-tracking branch 'remotes/kraxel/tags/pull-vga-20160606-1' into staging
virtio-gpu: scanout fix, live migration support
vmsvga: security fixes
# gpg: Signature made Mon 06 Jun 2016 08:05:00 BST
# gpg: using RSA key 0x4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>"
# gpg: aka "Gerd Hoffmann <[email protected]>"
# gpg: aka "Gerd Hoffmann (private) <[email protected]>"
* remotes/kraxel/tags/pull-vga-20160606-1:
virtio-gpu: add live migration support
vmsvga: don't process more than 1024 fifo commands at once
vmsvga: shadow fifo registers
vmsvga: add more fifo checks
vmsvga: move fifo sanity checks to vmsvga_fifo_length
virtio-gpu: fix scanout rectangles
Commit fcaafb1001b9c42817714dd3b2aadcfdb997b53d accidentally broke reads from
scsi-disk devices when being updated from its original form to use the new
byte-based block functions. Add the extra missing sector to offset conversion
in order to restore read functionality.
Peter Wu [Sun, 5 Jun 2016 14:35:48 +0000 (16:35 +0200)]
gdbstub: avoid busy loop while waiting for gdb
While waiting for a gdb response, or while sending an acknowledgement
there is not much to do, so do not mark the socket as non-blocking to
avoid a busy loop while paused at gdb. This only affects the user-mode
emulation (qemu-arm -g 1234 ./a.out).
Note that this issue was reported before at
https://lists.nongnu.org/archive/html/qemu-devel/2013-02/msg02277.html.
While at it, close the gdb client fd on EOF or error while reading.
Greg Kurz [Mon, 6 Jun 2016 09:52:34 +0000 (11:52 +0200)]
9p: add locking to V9fsDir
If several threads concurrently call readdir() with the same directory
stream pointer, it is possible that they all get a pointer to the same
dirent structure, whose content is overwritten each time readdir() is
called.
We must thus serialize accesses to the dirent structure.
This may be achieved with a mutex like below:
lock_mutex();
readdir();
// work with the dirent
unlock_mutex();
This patch adds all the locking, to prepare the switch to readdir().
Dmitry Fleytman [Sat, 4 Jun 2016 07:02:43 +0000 (10:02 +0300)]
e1000e: Fix build with gcc 4.6.3 and ust tracing
This patch fixes used-uninitialized false
positive while compiling with ust tracing
backend plus gcc 4.6.3:
hw/net/e1000e.c: In function ‘e1000e_io_write’:
hw/net/e1000e.c:170:39: error: ‘idx’ may be used uninitialized in this function [-Werror=uninitialized]
hw/net/e1000e.c: In function ‘e1000e_io_read’:
hw/net/e1000e.c:145:35: error: ‘idx’ may be used uninitialized in this function [-Werror=uninitialized]
cc1: all warnings being treated as errors
make: *** [hw/net/e1000e.o] Error 1
Gerd Hoffmann [Mon, 23 May 2016 13:22:07 +0000 (15:22 +0200)]
virtio-gpu: add live migration support
Store some additional state for cursor and resource backing storage,
so we can write out and reload things. Implement vmsave+vmload for
2d mode. Continue blocking live migration in 3d/virgl mode.
Gerd Hoffmann [Mon, 30 May 2016 07:09:21 +0000 (09:09 +0200)]
vmsvga: don't process more than 1024 fifo commands at once
vmsvga_fifo_run is called in regular intervals (on each display update)
and will resume where it left off. So we can simply exit the loop,
without having to worry about how processing will continue.
Gerd Hoffmann [Mon, 30 May 2016 07:09:20 +0000 (09:09 +0200)]
vmsvga: shadow fifo registers
The fifo is normal ram. So kvm vcpu threads and qemu iothread can
access the fifo in parallel without syncronization. Which in turn
implies we can't use the fifo pointers in-place because the guest
can try changing them underneath us. So add shadows for them, to
make sure the guest can't modify them after we've applied sanity
checks.
Gerd Hoffmann [Mon, 30 May 2016 07:09:18 +0000 (09:09 +0200)]
vmsvga: move fifo sanity checks to vmsvga_fifo_length
Sanity checks are applied when the fifo is enabled by the guest
(SVGA_REG_CONFIG_DONE write). Which doesn't help much if the guest
changes the fifo registers afterwards. Move the checks to
vmsvga_fifo_length so they are done each time qemu is about to read
from the fifo.
The arm target was handled by 06486077, but other targets
were ignored. This handles all the rest which actually support
disassembly (that is, skipping moxie and tilegx).
Dmitry Fleytman [Thu, 2 Jun 2016 19:12:28 +0000 (22:12 +0300)]
e1000e: Fix build with ust trace backend
ust trace backend has limitation of maximum 10
arguments per event. Traces with more arguments
cannot be compiled for this backend.
Trace e1000e_rx_rss_ip6 introduced by previous
commits has 11 arguments and fails to compile with
ust trace backend.
This patch fixes the problem by splitting this
tracepoint into two successive tracepoints with
smaller number of arguments.
For more information see comment regarding TP_ARGS
in lttng/tracepoint.h:
/*
* TP_ARGS takes tuples of type, argument separated by a comma.
* It can take up to 10 tuples (which means that less than 10 tuples is
* fine too).
* Each tuple is also separated by a comma.
*/
Build log generated by this problem:
In file included from ./trace/generated-tracers.h:9:0,
from /home/travis/build/qemu/qemu/include/trace.h:4,
from util/oslib-posix.c:36:
./trace/generated-ust-provider.h:16556:3: error: unknown type name ‘_TP_EXPROTO_Bool’
In file included from /home/travis/build/qemu/qemu/include/trace.h:4:0,
from util/oslib-posix.c:36:
./trace/generated-tracers.h: In function ‘trace_e1000e_rx_rss_ip6’:
./trace/generated-tracers.h:8379:431: error: expected string literal before ‘_SDT_ASM_OPERANDS_ipv6_enabled’
./trace/generated-tracers.h:8379:431: error: implicit declaration of function ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=implicit-function-declaration]
./trace/generated-tracers.h:8379:431: error: nested extern declaration of ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=nested-externs]
cc1: all warnings being treated as errors
make: *** [util/oslib-posix.o] Error 1
make: *** Waiting for unfinished jobs....
In file included from ./trace/generated-tracers.h:9:0,
from /home/travis/build/qemu/qemu/include/trace.h:4,
from util/hbitmap.c:16:
./trace/generated-ust-provider.h:16556:3: error: unknown type name ‘_TP_EXPROTO_Bool’
In file included from /home/travis/build/qemu/qemu/include/trace.h:4:0,
from util/hbitmap.c:16:
./trace/generated-tracers.h: In function ‘trace_e1000e_rx_rss_ip6’:
./trace/generated-tracers.h:8379:431: error: expected string literal before ‘_SDT_ASM_OPERANDS_ipv6_enabled’
./trace/generated-tracers.h:8379:431: error: implicit declaration of function ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=implicit-function-declaration]
./trace/generated-tracers.h:8379:431: error: nested extern declaration of ‘__tracepoint_cb_qemu___e1000e_rx_rss_ip6’ [-Werror=nested-externs]
cc1: all warnings being treated as errors
make: *** [util/hbitmap.o] Error 1
Peter Krempa [Wed, 11 May 2016 10:31:04 +0000 (12:31 +0200)]
audio: pa: Set volume of recording stream instead of recording device
Since pulseaudio 1.0 it's possible to set the individual stream volume
rather than setting the device volume. With this, setting hardware mixer
of a emulated sound card doesn't mess up the volume configuration of the
host.
A side effect is that this limits compatible pulseaudio version to 1.0
which was released on 2011-09-27.
Gerd Hoffmann [Mon, 30 May 2016 08:40:55 +0000 (10:40 +0200)]
virtio-gpu: fix scanout rectangles
Commit "ca58b45 ui/virtio-gpu: add and use qemu_create_displaysurface_pixman"
breaks scanouts which use a region of the underlying resource only.
So, we need another way to handle the underlying issue. Lets create a
new pixman image, grab a reference on the pixman providing the
underlying storage, hook up a destroy callback which releases the
reference. That way regions work again and releasing the backing
storage should still be impossible thanks to the extra reference we are
holding.
Gerd Hoffmann [Wed, 1 Jun 2016 06:22:30 +0000 (08:22 +0200)]
vnc: add configurable keyboard delay
Limits the rate kbd events from the vnc server are forwarded to the
guest, so input devices which are typically low-bandwidth can keep
up even on bulky input.
Alexander Graf [Tue, 24 May 2016 14:19:19 +0000 (17:19 +0300)]
vnc: Add support for color map
Our current VNC code does not handle color maps (aka non-true-color) at all
and aborts if a client requests them. There are 2 major issues with this:
1) A VNC viewer on an 8-bit X11 system may request color maps
2) RealVNC _always_ starts requesting color maps, then moves on to full color
In order to support these 2 use cases, let's just create a fake color map
that covers exactly our normal true color 8 bit color space. That way we don't
lose anything over a client that wants true color.
Actually this is a very old patch originally submitted in 2013 by
Alexander. The situation is still the same with RealVNC, it does not
connect by default to QEMU VNC. The problem is that this client is
really popular. This is better to be kludged.
projects / qemu.git / log