exec: Split softmmu_defs.h

The _cmmu helpers can be moved to exec-all.h.  The helpers that are
used from TCG will shortly need access to tcg_target_long so move
their declarations into tcg.h.

This requires minor include adjustments to all TCG backends.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

target: Include softmmu_exec.h where forgotten

Several targets forgot to include softmmu_exec.h, which would
break them with a header cleanup to follow.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

exec: Rename USUFFIX to LSUFFIX

In a following patch, there will be confusion between multiple "unsigned"
suffixes; rename this one so as to imply "load".

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Don't perform GETPC adjustment in TCG code

Since we now perform it inside the helper, no need to do it here.
This also lets us perform a tail-call from the store slow path to
the helper.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

exec: Reorganize the GETRA/GETPC macros

Always define GETRA; use __builtin_extract_return_addr, rather than
having a special case for s390.  Split GETPC_ADJ out of GETPC; use 2
universally, rather than having a special case for arm.

Rename GETPC_LDST to GETRA_LDST to indicate that it does not
contain the GETPC_ADJ value.  Likewise with GETPC_EXT to GETRA_EXT.

Perform the GETPC_ADJ adjustment inside helper_ret_ld/st.  This will
allow backends to pass along the "true" return address rather than
the massaged GETPC value.  In the meantime, double application of
GETPC_ADJ does not hurt, since the call insn in all ISAs is at least
4 bytes long.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

configure: Allow x32 as a host

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Adjust tcg_out_tlb_load for x32

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Use intptr_t appropriately

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Fix jit debug for x32

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Use appropriate types in tcg_reg_alloc_call

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change tcg_out_ld/st offset to intptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change tcg_gen_exit_tb argument to uintptr_t

And update all users.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Use uintptr_t in TCGHelperInfo

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change relocation offsets to intptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change memory offsets to intptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change frame pointer offsets to intptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Define TCG_ptr properly

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Define TCG_TYPE_PTR properly

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Allow TCG_TARGET_REG_BITS to be specified independantly

There are several hosts for which it would be useful to use the
available 64-bit registers in a 32-bit pointer environment.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Fix next_tb type in cpu_exec

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change tcg_qemu_tb_exec return to uintptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Change flush_icache_range arguments to uintptr_t

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

qtest: Fix FMT_timeval vs time_t

Since FMT_timeval unconditionally uses %ld for both tv_sec and tv_usec,
and already casts tv_usec to long, also cast tv_sec to long.

Cc: Andreas Färber <[email protected]>
Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Constant fold div, rem

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg-ppc64: Implement muluh, mulsh

Using these instead of mulu2 and muls2 lets us avoid having to argument
overlap analysis in the backend.  Normal register allocation will DTRT.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg-mips: Implement mulsh, muluh

With the optimization in tcg_liveness_analysis,
we can avoid the MFLO when it is unused.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Add muluh and mulsh opcodes

Use them in places where mulu2 and muls2 are used.
Optimize mulx2 with dead low part to mulxh.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

Merge remote-tracking branch 'luiz/queue/qmp' into staging

# By Wenchao Xia (15) and Stefan Weil (1)
# Via Luiz Capitulino
* luiz/queue/qmp:
  monitor: improve auto complete of "help" for single command in sub group
  monitor: allow "help" show message for single command in sub group
  monitor: support sub command in auto completion
  monitor: refine monitor_find_completion()
  monitor: support sub command in help
  monitor: refine parse_cmdline()
  monitor: code move for parse_cmdline()
  monitor: avoid direct use of global variable *mon_cmds
  monitor: split off monitor_data_init()
  monitor: call sortcmdlist() only one time
  monitor: avoid use of global *cur_mon in readline_completion()
  monitor: avoid use of global *cur_mon in monitor_find_completion()
  monitor: avoid use of global *cur_mon in block_completion_it()
  monitor: avoid use of global *cur_mon in file_completion()
  monitor: avoid use of global *cur_mon in cmd_completion()
  monitor: Add missing attributes to local function

Message-id: 1377865357[email protected]

Merge remote-tracking branch 'borntraeger/tags/kdump' into staging

This is a set of patches dealing with kdump support for s390x/kvm.
kdump on s390x uses subcode 1 of diagnose 0x308 to put the hardware
in a defined state. This is different from a full reset, since it
does not touch all CPU registers.
These patches define the cpu resets, the subsystem reset a load
function and also wires up the "nmi" command to issue a RESTART
interrupt as defined in the z/Architecture principles of operation.

This allows recent guest kernels with properly setup userspace
to trigger kdump:
- via guest crash
- via nmi from the host

# gpg: Signature made Fri 30 Aug 2013 07:19:18 AM CDT using RSA key ID B5A61C7C
# gpg: Can't check signature: public key not found

# By Christian Borntraeger (5) and Eugene (jno) Dvurechenski (2)
# Via Christian Borntraeger
* borntraeger/tags/kdump:
  s390: wire up nmi command to raise a RESTART interrupt on S390
  s390: Implement load normal reset
  s390/cpu: split CPU reset into architectured functions
  s390: provide a cpu load normal function
  s390: provide I/O subsystem reset
  s390/kvm: basic implementation of diagnose 308 subcode 6
  s390x/kvm: Fix switch/case indentation for handle_diag

Message-id: 1377810649[email protected]

s390: wire up nmi command to raise a RESTART interrupt on S390

There is the 'nmi' command that is used to trigger a guest dump via kdump feature on x86.
s390 uses RESTART interrupt to trigger kdump.
So, this patch provides a mean to use 'nmi' command on s390 to raise RESTART interrupt.

The CPU to receive the RESTART interrupt is the "default" one.

There is an infrastructure to select the "default" CPU using 'cpu' command.
The 'info cpus' command can be used to see which one is the "default".

In order to wire up the RESTART to 'nmi' command we had to:
1. implement the kvm_s390_cpu_restart function by exporting the existing code
2. implement s390_cpu_restart function as kvm-aware wrapper
3. modify the qmp_inject_nmi function to enable (for s390) the scan for
   "default" CPU and call s390_cpu_restart for it;
3. fix some messages.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
Acked-by: Alexander Graf <[email protected]>

s390: Implement load normal reset

kdump on s390 uses a load normal reset to bring the system in a defined
state by doing a subsystem reset. The issuing CPUs will have an initial
CPU reset, all other CPUs will have a CPU reset as defined in POP (no
register content will change).

Implement this as architectured.

Signed-off-by: Christian Borntraeger <[email protected]>

s390/cpu: split CPU reset into architectured functions

s390 provides several CPU resets:
- CPU reset, clears interrupts, stop processing, clears TLB, but does
  not touch registers
- initial CPU reset, like CPU reset, but also clears PSW, prefix, FPC,
  timer and control registers. It does not touch gprs, fprs and acrs (!)
- Power on reset: the full monty

wire up CPUClass reset to the full monty, but provide the lesser resets
as part of S390CPUClass.

Signed-off-by: Christian Borntraeger <[email protected]>

monitor: improve auto complete of "help" for single command in sub group

Now special case "help *" in auto completion can work with sub commands,
such as "help info u*".

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: allow "help" show message for single command in sub group

A new parameter type 'S' is introduced to allow user input any string.
"help info block" works normal now.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: support sub command in auto completion

This patch allows auto completion work normal for sub command case,
"info block [DEVICE]" can auto complete now, by re-enter the completion
function. In original code "info" is treated as a special case, now it
is treated as a sub command group, global variable info_cmds is not used
any more.

"help" command is still treated as a special case, since it is not a sub
command group but want to auto complete command in root command table.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: refine monitor_find_completion()

In order to support sub command in auto completion, a reentrant function
is needed, so monitor_find_completion() is split into two parts. The
first part does parsing of user input which need to be done only once,
the second part does the auto completion job according to the parsing
result, which contains the necessary code to support sub command and
works as the reentrant function. The global "info_cmds" is still used
in second part, which will be replaced by sub command code later.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: support sub command in help

The old code in help_cmd() uses global 'info_cmds' and treats it as a
special case. Actually 'info_cmds' is a sub command group of 'mon_cmds',
in order to avoid direct use of it, help_cmd() needs to change its work
mechanism to support sub command and not treat it as a special case
any more.

To support sub command, help_cmd() will first parse the input and then call
help_cmd_dump(), which works as a reentrant function. When it meets a sub
command, it simply enters the function again. Since help dumping needs to
know whole input to printf full help message include prefix, for example,
"help info block" need to printf prefix "info", so help_cmd_dump() takes all
args from input and extra parameter arg_index to identify the progress.
Another function help_cmd_dump_one() is introduced to printf the prefix
and command's help message.

Now help supports sub command, so later if another sub command group is
added in any depth, help will automatically work for it. Still "help info
block" will show error since command parser reject additional parameter,
which can be improved later. "log" is still treated as a special case.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: refine parse_cmdline()

Since this function will be used by help_cmd() later, so improve
it to make it more generic and easier to use. free_cmdline_args()
is added too as paired function to free the result.

One change of this function is that, when the valid args in input
exceed the limit of MAX_ARGS, it fails now, instead of return with
MAX_ARGS of parsed args in old code. This should not impact much
since it is rare that user input many args in monitor's "help" and
auto complete scenario.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: code move for parse_cmdline()

help_cmd() need this function later, so move it. get_str() is called by
parse_cmdline() so it is moved also. Some code style error reported by
check script, is also fixed.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid direct use of global variable *mon_cmds

New member *cmd_table is added in structure Monitor to avoid direct usage of
*mon_cmds. Now monitor have an associated command table, when global variable
*info_cmds is also discarded, structure Monitor would gain full control about
how to deal with user input.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: split off monitor_data_init()

In qmp_human_monitor_command(), the monitor need to initialized for
basic functionalities, and later more init code will be added, so
split off this function. Note that it is different with QMP mode
monitor which accept json string from monitor's input,
qmp_human_monitor_command() retrieve the human style command from
QMP input, then send the command to a normal mode monitor.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: call sortcmdlist() only one time

It doesn't need to be done for every monitor, so change it.

Signed-off-by: Wenchao Xia <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid use of global *cur_mon in readline_completion()

Now all completion functions do not use *cur_mon any more, instead
they use rs->mon. In short, structure ReadLineState decide where
the complete action would be taken now.

Tested with the case that qemu have two telnet monitors, auto
completion function works normal.

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid use of global *cur_mon in monitor_find_completion()

Parameter *mon is added, and local variable *mon added in previous patch
is removed. The caller readline_completion(), pass rs->mon as value, which
should be initialized in readline_init() called by monitor_init().

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid use of global *cur_mon in block_completion_it()

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid use of global *cur_mon in file_completion()

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: avoid use of global *cur_mon in cmd_completion()

A new local variable *mon is added in monitor_find_completion()
to make compile pass, which will be removed later in
conversion patch for monitor_find_completion().

Signed-off-by: Wenchao Xia <[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

monitor: Add missing attributes to local function

Function expr_error gets a format string and variable arguments like printf.
It also never returns. Add the necessary attributes.

Signed-off-by: Stefan Weil <[email protected]>
Signed-off-by: Luiz Capitulino <[email protected]>

s390: provide a cpu load normal function

Some code needs to perform an IPL-like bootup that mimics the
ESA (31bit) restart. Provide a cpu class method that does so.

Signed-off-by: Christian Borntraeger <[email protected]>
Acked-by: Alexander Graf <[email protected]>

s390: provide I/O subsystem reset

Provide a function that resets the I/O subsystem.

Signed-off-by: Christian Borntraeger <[email protected]>
Acked-by: Alexander Graf <[email protected]>

s390/kvm: basic implementation of diagnose 308 subcode 6

Linux uses a check for subcode 6 to decide if other subcodes are
available. Provide a minimal implementation for subcode 6, as well
as for subcode 5.

Signed-off-by: Eugene (jno) Dvurechenski <[email protected]>
Reviewed-by: Cornelia Huck <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
[Move code from kvm.c into misc_helper.c]

s390x/kvm: Fix switch/case indentation for handle_diag

This alignes case statements to switch statements in the handle_diag
function as mandated by coding style.

Signed-off-by: Christian Borntraeger <[email protected]>

Merge remote-tracking branch 'qemu-kvm/uq/master' into stable-1.5

* qemu-kvm/uq/master:
  kvm-stub: fix compilation
  kvm: shorten the parameter list for get_real_device()
  kvm: i386: fix LAPIC TSC deadline timer save/restore
  kvm-all.c: max_cpus should not exceed KVM vcpu limit
  kvm: Simplify kvm_handle_io
  kvm: x86: fix setting IA32_FEATURE_CONTROL with nested VMX disabled
  kvm: add KVM_IRQFD_FLAG_RESAMPLE support
  kvm: migrate vPMU state
  target-i386: remove tabs from target-i386/cpu.h
  Initialize IA32_FEATURE_CONTROL MSR in reset and migration

Conflicts:
target-i386/cpu.h
target-i386/kvm.c

aliguori: fixup trivial conflicts due to whitespace and added cpu
          argument

Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'sweil/mingw' into stable-1.5

# By Stefan Weil
# Via Stefan Weil
* sweil/mingw:
  gtk: Remove unused include statements which are not portable
  w32: Add an icon resource
  w32: Fix broken out-of-tree builds (missing version.o)

Message-id: 1377607132[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'mst/tags/for_anthony' into stable-1.5

pc,pci,virtio fixes and cleanups

This includes pc and pci cleanups, future-proofing of ROM files,
and a virtio bugfix correcting splice on virtio console.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Mon 26 Aug 2013 01:34:20 AM CDT using RSA key ID D28D5469
# gpg: Can't check signature: public key not found

# By Markus Armbruster (5) and others
# Via Michael S. Tsirkin
* mst/tags/for_anthony:
  virtio: virtqueue_get_avail_bytes: fix desc_pa when loop over the indirect descriptor table
  pc_piix: Kill pc_init1() memory region args
  pc: pc_compat_1_4() now can call pc_compat_1_5()
  pc: Create pc_compat_*() functions
  pc: Kill pc_init_pci_1_0()
  pc: Don't explode QEMUMachineInitArgs into local variables needlessly
  pc: Don't prematurely explode QEMUMachineInitArgs
  ppc: Don't duplicate QEMUMachineInitArgs in PPCE500Params
  ppc: Don't explode QEMUMachineInitArgs into local variables needlessly
  sun4: Don't prematurely explode QEMUMachineInitArgs
  q35: Add PCIe switch to example q35 configuration
  loader: store FW CFG ROM files in RAM
  arch_init: align MR size to target page size
  pc: cleanup 1.4 compat support

Message-id: 1377535318[email protected]

tcg-i386: Remove abort from GETPC_LDST

Indeed, remove it entirely and remove the is_tcg_gen_code check
from GETPC_EXT.

Fixes https://bugs.launchpad.net/qemu/+bug/1218098 wherein a call
to a "normal" helper function performed a sequence of tail calls
all the way into the memory helper functions, leading to a stack
frame in which the memory helper function appeared to be called
directly from tcg.

Signed-off-by: Richard Henderson <[email protected]>

target-mips: fix get_physical_address() #if 0 build error

In get_physical_address() is a qemu_log() call inside an #if 0 block.
When enabled the following build error is hit:

target-mips/helper.c In function ‘get_physical_address’:
target-mips/helper.c:220:13: error: format ‘%x’ expects argument of type ‘unsigned int’, but argument 5 has type ‘hwaddr’ [-Werror=format]

Fix the *physical (hwaddr) formatting by using "%"HWADDR_PRIx instead of
TARGET_FMT_lx.

Signed-off-by: James Hogan <[email protected]>
Cc: Aurelien Jarno <[email protected]>
Signed-off-by: Yongbok Kim <[email protected]>
Signed-off-by: Aurelien Jarno <[email protected]>

kvm-stub: fix compilation

Non-KVM targets fail compilation on the uq/master branch.
Fix the prototype of kvm_irqchip_add_irqfd_notifier to match
the one in kvm-all.c.

Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Gleb Natapov <[email protected]>

gtk: Remove unused include statements which are not portable

These include files don't exist for MinGW and are not needed for Linux
(and hopefully for other hosts as well), so remove them.

Signed-off-by: Stefan Weil <[email protected]>

w32: Add an icon resource

The QEMU mascot which was already used for the NSIS installer
is now used for all QEMU executables.

Signed-off-by: Stefan Weil <[email protected]>

w32: Fix broken out-of-tree builds (missing version.o)

Commit 0b516ef0dfad9a7b34c675c98e8ec92ab4d38466 added version.o to all
executables, but broke out-of-tree builds: for those builds the pattern
rule %.o: %.rc from rules.mak does not match, so version.o was no longer
built.

Adding explicit build rules fixes this.

Reported-by: Michael Roth <[email protected]>
Signed-off-by: Stefan Weil <[email protected]>
Tested-by: Michael Roth <[email protected]>

tcg-i386: Use new return-argument ld/st helpers

Discontinue the jump-around-jump-to-jump scheme, trading it for a single
immediate move instruction.  The two extra jumps always consume 7 bytes,
whereas the immediate move is either 5 or 7 bytes depending on where the
code_gen_buffer gets located.

Signed-off-by: Richard Henderson <[email protected]>

tcg: Tidy softmmu_template.h

Avoid a loop in the tlb_fill path; the fill will either succeed or
generate an exception.

Inline the slow_ld/st function; it was a complete copy of the main
helper except for the actual cross-page unaligned code, and the
compiler was inlining it anyway.

Add unlikely markers optimizing for the most common case of simple
tlb miss.

Make sure the compiler can optimize away the unaligned paths for a
1 byte access.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Add mmu helpers that take a return address argument

Allow the code that tcg generates to be less obtuse, passing in
the return address directly instead of computing it in the helper.

Maintain the old entrance point unchanged as an alternate entry point.

Delete the helper_st*_cmmu prototypes; the implementations did not exist.

Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Tidy qemu_ld/st slow path

Use existing stack space for arguments; don't push/pop.
Use less ifdefs and more C ifs.

Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Try pc-relative lea for constant formation

Use a 7 byte lea before the ultimate 10 byte movq.

Signed-off-by: Richard Henderson <[email protected]>

tcg-i386: Add and use tcg_out64

No point in splitting the write into 32-bit pieces.

Reviewed-by: Aurelien Jarno <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>

tcg: Tidy generated code for tcg_outN

Aliasing was forcing s->code_ptr to be re-read after the store.
Keep the pointer in a local variable to help the compiler.

Signed-off-by: Richard Henderson <[email protected]>

Merge remote-tracking branch 'stefanha/block' into staging

# By Alex Bligh (32) and others
# Via Stefan Hajnoczi
* stefanha/block: (42 commits)
  win32-aio: drop win32_aio_flush_cb()
  aio-win32: replace incorrect AioHandler->opaque usage with ->e
  aio / timers: remove dummy_io_handler_flush from tests/test-aio.c
  aio / timers: Remove legacy interface
  aio / timers: Switch entire codebase to the new timer API
  aio / timers: Add scripts/switch-timer-api
  aio / timers: Add test harness for AioContext timers
  aio / timers: convert block_job_sleep_ns and co_sleep_ns to new API
  aio / timers: Convert rtc_clock to be a QEMUClockType
  aio / timers: Remove main_loop_timerlist
  aio / timers: Rearrange timer.h & make legacy functions call non-legacy
  aio / timers: Add qemu_clock_get_ms and qemu_clock_get_ms
  aio / timers: Remove legacy qemu_clock_deadline & qemu_timerlist_deadline
  aio / timers: Remove alarm timers
  aio / timers: Add documentation and new format calls
  aio / timers: Use all timerlists in icount warp calculations
  aio / timers: Introduce new API timer_new and friends
  aio / timers: On timer modification, qemu_notify or aio_notify
  aio / timers: Convert mainloop to use timeout
  aio / timers: Convert aio_poll to use AioContext timers' deadline
  ...

Message-id: 1377202298[email protected]
Signed-off-by: Anthony Liguori <[email protected]>

Merge remote-tracking branch 'afaerber/tags/0.15-maintainer-for-anthony' into staging

MAINTAINERS update for stable-0.15

# gpg: Signature made Thu 22 Aug 2013 10:59:31 AM CDT using RSA key ID 3E7E013F
# gpg: Can't check signature: public key not found

# By Andreas Färber
# Via Andreas Färber
* afaerber/tags/0.15-maintainer-for-anthony:
  MAINTAINERS: Take over 0.15 maintenance

virtio: virtqueue_get_avail_bytes: fix desc_pa when loop over the indirect descriptor table

virtqueue_get_avail_bytes: when found a indirect desc, we need loop over it.
           /* loop over the indirect descriptor table */
           indirect = 1;
           max = vring_desc_len(desc_pa, i) / sizeof(VRingDesc);
           num_bufs = i = 0;
           desc_pa = vring_desc_addr(desc_pa, i);
But, It init i to 0, then use i to update desc_pa. so we will always get:
desc_pa = vring_desc_addr(desc_pa, 0);
the last two line should swap.

Cc: [email protected]
Signed-off-by: Yin Yin <[email protected]>
Reviewed-by: Stefan Hajnoczi <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

disas-objdump: Pass --adjust-vma to objdump

This gives the dumped blob its correct address during disassembly,
which makes pc-relative insns much easier to interpret.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

disas: Add disas-objdump.pl

The script massages the output produced for architectures that are
not supported internally by qemu though an external objdump program
for disassembly.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

disas: Implement fallback to dump object code as hex

The OBJD-[HT] tags will be used by a script to run the hex blob
through objdump --disassemble.

Signed-off-by: Richard Henderson <[email protected]>
Signed-off-by: Edgar E. Iglesias <[email protected]>

kvm: shorten the parameter list for get_real_device()

get_real_device() has 5 parameters with the last 4 is contained in the first
structure.

This patch removes the last 4 parameters and directly use them from the first
parameter.

Acked-by: Alex Williamson <[email protected]>
Signed-off-by: Wei Yang <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

win32-aio: drop win32_aio_flush_cb()

The io_flush argument to qemu_aio_set_event_notifier() has been removed
since the block layer learnt to drain requests by itself.  Fix the
Windows build for win32-aio.o by updating the
qemu_aio_set_event_notifier() call and dropping win32_aio_flush_cb().

Reviewed-by: Stefan Weil <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio-win32: replace incorrect AioHandler->opaque usage with ->e

The AioHandler->opaque field does not exist in aio-win32.c.  The code
that uses it was incorrectly copied from aio-posix.c.  For Windows we
can use AioHandler->e to match against AioContext->notifier.

This patch fixes the Windows build for aio-win32.o.

Reviewed-by: Stefan Weil <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: remove dummy_io_handler_flush from tests/test-aio.c

Remove dummy_io_handler_flush from tests/test-aio.c as it does
nothing now.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Remove legacy interface

Remove the legacy interface from include/qemu/timers.h.

Ensure struct QEMUClock is not exposed at all.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Switch entire codebase to the new timer API

This is an autogenerated patch using scripts/switch-timer-api.

Switch the entire code base to using the new timer API.

Note this patch may introduce some line length issues.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add scripts/switch-timer-api

Add scripts/switch-timer-api to programatically rewrite source
files to use the new timer system.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add test harness for AioContext timers

Add a test harness for AioContext timers. The g_source equivalent is
unsatisfactory as it suffers from false wakeups.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: convert block_job_sleep_ns and co_sleep_ns to new API

Convert block_job_sleep_ns and co_sleep_ns to use the new timer
API.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Convert rtc_clock to be a QEMUClockType

Convert rtc_clock to be a QEMUClockType

Move rtc_clock users to use the new API

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Remove main_loop_timerlist

Now we have timerlistgroups implemented and main_loop_tlg, we
no longer need the concept of a default timer list associated
with each clock. Remove it and simplify initialisation of
clocks and timer lists.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Rearrange timer.h & make legacy functions call non-legacy

Rearrange timer.h so it is in order by function type.

Make legacy functions call non-legacy functions rather than vice-versa.

Convert cpus.c to use new API.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add qemu_clock_get_ms and qemu_clock_get_ms

Add utility functions qemu_clock_get_ms and qemu_clock_get_us

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Remove legacy qemu_clock_deadline & qemu_timerlist_deadline

Remove qemu_clock_deadline and qemu_timerlist_deadline now we are using
the ns functions throughout.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Remove alarm timers

Remove alarm timers from qemu-timers.c now we use g_poll / ppoll
instead.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add documentation and new format calls

Add documentation for existing qemu timer calls. Add new format
calls of the format timer_XXX rather than qemu_XXX_timer
for consistency.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Use all timerlists in icount warp calculations

Notify all timerlists derived from vm_clock in icount warp
calculations.

When calculating timer delay based on vm_clock deadline, use
all timerlists.

For compatibility, maintain an apparent bug where when using
icount, if no vm_clock timer was set, qemu_clock_deadline
would return INT32_MAX and always set an icount clock expiry
about 2 seconds ahead.

NB: thread safety - when different timerlists sit on different
threads, this will need some locking.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Introduce new API timer_new and friends

Introduce new API for creating timers - timer_new and
_ns, _ms, _us derivatives.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: On timer modification, qemu_notify or aio_notify

On qemu_mod_timer_ns, ensure qemu_notify or aio_notify is called to
end the appropriate poll(), irrespective of use_icount value.

On qemu_clock_enable, ensure qemu_notify or aio_notify is called for
all QEMUTimerLists attached to the QEMUClock.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Convert mainloop to use timeout

Convert mainloop to use timeout from default timerlist group
(i.e. the current 3 static timers)

main-loop.c produces a (possibly spurious) warning about
multiple iterations. Adapt the way this works for a signed
timeout and make the warning a bit safer.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Convert aio_poll to use AioContext timers' deadline

Convert aio_poll to use deadline based on AioContext's timers.

aio_poll has been changed to return accurately whether progress
has occurred. Prior to this commit, aio_poll always returned
true if g_poll was entered, whether or not any progress was
made. This required a change to tests/test-aio.c where an
assert was backwards.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add aio_timer_init & aio_timer_new wrappers

Add aio_timer_init and aio_timer_new wrapper functions.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: aio_ctx_prepare sets timeout from AioContext timers

Calculate the timeout in aio_ctx_prepare taking into account
the timers attached to the AioContext.

Alter aio_ctx_check similarly.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add a notify callback to QEMUTimerList

Add a notify pointer to QEMUTimerList so it knows what to notify
on a timer change.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add QEMUTimerListGroup to AioContext

Add a QEMUTimerListGroup each AioContext (meaning a QEMUTimerList
associated with each clock is added) and delete it when the
AioContext is freed.

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>

aio / timers: Add QEMUTimerListGroup and helper functions

Add QEMUTimerListGroup and helper functions, to represent
a QEMUTimerList associated with each clock. Add a default
QEMUTimerListGroup representing the default timer lists
which are not associated with any other object (e.g.
an AioContext as added by future patches).

Signed-off-by: Alex Bligh <[email protected]>
Signed-off-by: Stefan Hajnoczi <[email protected]>