The kernel and libc have different ideas about what a sigset_t
is -- for the kernel it is only _NSIG / 8 bytes in size (usually
8 bytes), but for libc it is much larger, 128 bytes. In most
situations the difference doesn't matter, because if you pass a
pointer to a libc sigset_t to the kernel it just acts on the first
8 bytes of it, but for the ucontext_t* argument to a signal handler
it trips us up. The kernel allocates this ucontext_t on the stack
according to its idea of the sigset_t type, but the type of the
ucontext_t defined by the libc headers uses the libc type, and
so do the manipulator functions like sigfillset(). This means that
(1) sizeof(uc->uc_sigmask) is much larger than the actual
space used on the stack
(2) sigfillset(&uc->uc_sigmask) will write garbage 0xff bytes
off the end of the structure, which can trash data that
was on the stack before the signal handler was invoked,
and may result in a crash after the handler returns
To avoid this, we use a memset() of the correct size to fill
the signal mask rather than using the libc function.
This fixes a problem where we would crash at least some of the
time on an i386 host when a signal was taken.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Make target_strerror() return 'const char *' rather than just 'char *';
this will allow us to return constant strings from it for some special
cases.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
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).
Signed-off-by: Timothy Edward Baldwin <T.E.Baldwin99@members.leeds.ac.uk>
Message-id: 1441497448-32489-24-git-send-email-T.E.Baldwin99@members.leeds.ac.uk
[PMM: added a comment]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
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.)
Signed-off-by: Timothy Edward Baldwin <T.E.Baldwin99@members.leeds.ac.uk>
Message-id: 1441497448-32489-23-git-send-email-T.E.Baldwin99@members.leeds.ac.uk
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
[PMM: minor commit message tweak]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
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 <T.E.Baldwin99@members.leeds.ac.uk>
Message-id: 1441497448-32489-19-git-send-email-T.E.Baldwin99@members.leeds.ac.uk
[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 <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
If a signal is delivered immediately before a blocking system call the
handler will only be called after the system call returns, which may be a
long time later or never.
This is fixed by using a function (safe_syscall) that checks if a guest
signal is pending prior to making a system call, and if so does not call the
system call and returns -TARGET_ERESTARTSYS. If a signal is received between
the check and the system call host_signal_handler() rewinds execution to
before the check. This rewinding has the effect of closing the race window
so that safe_syscall will reliably either (a) go into the host syscall
with no unprocessed guest signals pending or or (b) return
-TARGET_ERESTARTSYS so that the caller can deal with the signals.
Implementing this requires a per-host-architecture assembly language
fragment.
This will also resolve the mishandling of the SA_RESTART flag where
we would restart a host system call and not call the guest signal handler
until the syscall finally completed -- syscall restarting now always
happens at the guest syscall level so the guest signal handler will run.
(The host syscall will never be restarted because if the host kernel
rewinds the PC to point at the syscall insn for a restart then our
host_signal_handler() will see this and arrange the guest PC rewind.)
This commit contains the infrastructure for implementing safe_syscall
and the assembly language fragment for x86-64, but does not change any
syscalls to use it.
Signed-off-by: Timothy Edward Baldwin <T.E.Baldwin99@members.leeds.ac.uk>
Message-id: 1441497448-32489-14-git-send-email-T.E.Baldwin99@members.leeds.ac.uk
[PMM:
* Avoid having an architecture if-ladder in configure by putting
linux-user/host/$(ARCH) on the include path and including
safe-syscall.inc.S from it
* Avoid ifdef ladder in signal.c by creating new hostdep.h to hold
host-architecture-specific things
* Added copyright/license header to safe-syscall.inc.S
* Rewrote commit message
* Added comments to safe-syscall.inc.S
* Changed calling convention of safe_syscall() to match syscall()
(returns -1 and host error in errno on failure)
* Added a long comment in qemu.h about how to use safe_syscall()
to implement guest syscalls.
]
RV: squashed Peters "fixup! linux-user: compile on non-x86-64 hosts"
patch
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
exec-all.h contains TCG-specific definitions. It is not needed outside
TCG-specific files such as translate.c, exec.c or *helper.c.
One generic function had snuck into include/exec/exec-all.h; move it to
include/qom/cpu.h.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This fixes double-definitions in linux-user builds when using the UST
tracing backend (which indirectly includes the system's "syscall.h").
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Instead of creating a temporary copy for the whole environment and
the arguments, directly copy everything to the target stack.
For this to work, we have to change the order of stack creation and
copying the arguments.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Brüns <stefan.bruens@rwth-aachen.de>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
There is some iffy lock hierarchy going on in translate-all.c. To
fix it, we need to take the mmap_lock in cpu-exec.c. Make the
functions globally available.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
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.
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Cc: Richard Henderson <rth@twiddle.net>
This will collect all load and store helpers soon. For now
it is just a replacement for softmmu_exec.h, which this patch
stops including directly, but we also include it where this will
be necessary in order to simplify the next patch.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Don't allow the linux-user guest to block SIGSEGV -- QEMU needs this
signal to detect accesses to pages which it has marked read-only
because it has cached translated code from them.
We implement this by making the do_sigprocmask() wrapper suppress
SIGSEGV when doing the host process signal mask manipulation; instead
we store the current state of SIGSEGV in the TaskState struct.
If we get a SIGSEGV for the guest when the guest has blocked the
signal, we treat it as if the default SEGV handler was in place,
as the kernel does for forced SIGSEGV delivery.
This patch is based on an idea by Alex Barcelo, but rather than
simply lying to the guest about the SIGSEGV state we track it.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reported-by: Alex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Create a wrapper for signal mask changes initiated by the guest;
(this includes syscalls and also the sigreturns from signal.c)
this will give us a place to put code which prevents the guest
from changing the handling of signals used by QEMU itself
internally.
The wrapper is called from all the guest-initiated sigprocmask, but
is not called from internal qemu sigprocmask calls.
Signed-off-by: Alex Barcelo <abarcelo@ac.upc.edu>
[PMM: Added calls to wrapper for sigprocmask uses in signal.c
when setting the signal mask on entry and exit from signal
handlers, since these also are guest-provided signal masks.]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
The regs parameter is not used anywhere, so remove it.
Signed-off-by: Will Newton <will.newton@linaro.org>
Reviewed-by: Erik de Castro Lopo <erikd@mega-nerd.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Creating target_structs header in linux-user/$arch/ and making
target_ipc_perm and target_shmid_ds its first inhabitants.
The struct defintions may/should be further fine-tuned by arch maintainers.
Signed-off-by: Petar Jovanovic <petar.jovanovic@imgtec.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
The binfmt_misc module can calculate the credentials and security
token according to the binary instead of to the interpreter if the
'C' flag is enabled.
To be able to execute non-readable binaries, this flag implies 'O'
flag. When 'O' flag is enabled, bintfmt_misc opens the file for
reading and pass the file descriptor to the interpreter.
References:
linux/Documentation/binfmt_misc.txt ['O' and 'C' description]
linux/fs/binfmt_misc.c linux/fs/binfmt_elf.c [ AT_EXECFD usage ]
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Add a missing "function" and replace "and" by "any".
BSD and Linux use the same documentation here, so fix both.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
For newer target architectures, glibc can be picky about the kernel
version: for example, it will not run on an aarch64 system unless
the kernel reports itself as at least 3.8.0. Accommodate this by
enhancing the existing support for faking the kernel version so
that each target can optionally specify a minimum version: if
the user doesn't force a specific fake version then we will override
with the minimum required version only if the real host kernel
version is insufficient.
Use this facility to let aarch64 report a minimum of 3.8.0.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1378235544-22290-21-git-send-email-peter.maydell@linaro.org
On ARM, linux-user emulation includes NWFPE support for emulating the
ancient FPA floating point coprocessor. This has long since been
superseded by VFP and is only required for legacy binaries. The
AArch64 linux-user target doesn't compile in NWFPE support, so make
sure the relevant code is protected by suitable ifdefs.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1378235544-22290-18-git-send-email-peter.maydell@linaro.org
Now all linux-user targets support building with NPTL, we can make it
mandatory. This is a good idea because:
* NPTL is no longer new and experimental; it is completely standard
* in practice, linux-user without NPTL is nearly useless for
binaries built against non-ancient glibc
* it allows us to delete the rather untested code for handling
the non-NPTL configuration
Note that this patch leaves the CONFIG_USE_NPTL ifdefs in the
bsd-user codebase alone. This makes no change for bsd-user, since
our configure test for NPTL had a "#include <linux/futex.h>"
which means bsd-user would never have been compiled with
CONFIG_USE_NPTL defined, and it still is not.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
For m68k, per-thread data is a purely kernel construct with no
CPU level support. Implement it via a field in the TaskState structure,
used by cpu_set_tls() and the set_thread_area/get_thread_area
syscalls. This allows us to enable compilation with NPTL.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
The target-specific headers (target_cpu.h and target_signal.h)
might need to use the target-independent structure and function
definitions of qemu.h; so include them only at the bottom of
qemu.h, not the top.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
The functions cpu_clone_regs() and cpu_set_tls() are not purely CPU
related -- they are specific to the TLS ABI for a a particular OS.
Move them into the linux-user/ tree where they belong.
target-lm32 had entirely unused implementations, since it has no
linux-user target; just drop them.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Commit 658f2dc97 accidentally dropped the cast to the target type of
the value loaded by get_user(). The most visible effect of this would
be that the sequence "uint64_t v; get_user_u32(v, addr)" would sign
extend the 32 bit loaded value into v rather than zero extending as
would be expected for a _u32 accessor. Put the cast back again to
restore the old behaviour.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The previous formuation with multiple assignments to __typeof(*hptr) falls
down when hptr is qualified const. E.g. with const struct S *p, p->f is
also qualified const.
With this formulation, there's no assignment to any local variable.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
The header file is specific for *-user, but I plan to introduce a more
generic qemu-types.h file, so I'm renaming it.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Compare signal numbers in the proper domain.
Convert all of the fields for SIGIO and SIGCHLD.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
When host and target have differing alignment rules, using a cast
and direct memory operation can result in SIGBUS. Use memcpy instead,
which the compiler will happily optimize when alignment is satisfied.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Roll the code used to initialize the guest memory space when -R
or -B is used into 'init_guest_space' and then call 'init_guest_space'
from the driver. This way the reserved guest memory space can
be probed for. Calling 'mmap' just once as is currently done is not
guaranteed to succeed since the host address space validation might fail.
Signed-off-by: Meador Inge <meadori@codesourcery.com>
[PMM: Fixed minor whitespace errors.]
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Add support for ARM BE8 userspace binaries.
i.e. big-endian data and little-endian code.
In principle LE8 mode is also possible, but AFAIK has never actually
been implemented/used.
System emulation doesn't have any useable big-endian board models,
but should in principle work once you fix that.
Dynamic endianness switching requires messing with data accesses,
preferably with TCG cooperation, and is orthogonal to BE8 support.
Signed-off-by: Paul Brook <paul@codesourcery.com>
[PMM: various changes, mostly as per my suggestions in code review:
* rebase
* use EF_ defines rather than hardcoded constants
* make bswap_code a bool for future VMSTATE macro compatibility
* update comment in cpu.h about TB flags bit field usage
* factor out load-code-and-swap into arm_ld*_code functions and
get_user_code* macros
* fix stray trailing space at end of line
* added braces in disas.c to satisfy checkpatch
]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
After consulting with Paul Brook, we concluded that it's best to search
the VMA space downwards, so that we don't even get the chance to conflict
with the brk range.
This patch resolves a bunch of allocation conflicts when using -R.
Signed-off-by: Alexander Graf <agraf@suse.de>
[minor changes to get it to apply -- PMM]
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Scripted conversion:
for file in *.[hc] hw/*.[hc] hw/kvm/*.[hc] linux-user/*.[hc] linux-user/m68k/*.[hc] bsd-user/*.[hc] darwin-user/*.[hc] tcg/*/*.[hc] target-*/cpu.h; do
sed -i "s/CPUState/CPUArchState/g" $file
done
All occurrences of CPUArchState are expected to be replaced by QOM CPUState,
once all targets are QOM'ified and common fields have been extracted.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
We create our own AUXV segment on stack and save a pointer to it.
However we don't save the length of it, so any code that wants to
do anything useful with it later on has to walk it again.
Instead, let's remember the length of our AUXV segment. This
simplifies later uses by a lot.
(edited by Riku to apply to qemu HEAD)
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Riku Voipio <riku.voipio@iki.fi>
linux-user: Implement new ARM 64 bit cmpxchg kernel helper
Linux 3.1 will have a new kernel-page helper for ARM implementing
64 bit cmpxchg. Implement this helper in QEMU linux-user mode:
* Provide kernel helper emulation for 64bit cmpxchg
* Allow guest to object to guest offset to ensure it can map a page
* Populate page with kernel helper version
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Dr. David Alan Gilbert <david.gilbert@linaro.org>
On 32 bit MIPS a few syscalls have 7 arguments, and so to call
them via NR_syscall the guest needs to be able to pass 8 arguments
to do_syscall(). Raise the number of arguments do_syscall() takes
accordingly.
This fixes some gcc 4.6 compiler warnings about arg7 and arg8
variables being set and never used.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@iki.fi>
There are some bits in the code which were used to store the commandline for
the semihosting call. These bits are now write-only and can be removed.
Signed-off-by: Wolfgang Schildbach <wschi@dolby.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@iki.fi>
Remove an unnecessary local variable from the __get_user() and
__put_user() macros. This avoids confusing compilation failures
if the name of the local variable ('size') happens to be the
same as the variable the macro user is trying to read/write.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@nokia.com>
Running programs that create large numbers of threads, such as this
snippet from libstdc++'s pthread7-rope.cc:
const int max_thread_count = 4;
const int max_loop_count = 10000;
...
for (int j = 0; j < max_loop_count; j++)
{
...
for (int i = 0; i < max_thread_count; i++)
pthread_create (&tid[i], NULL, thread_main, 0);
for (int i = 0; i < max_thread_count; i++)
pthread_join (tid[i], NULL);
}
in user-mode emulation will quickly run out of memory. This is caused
by a failure to free memory in do_syscall prior to thread exit:
/* TODO: Free CPU state. */
pthread_exit(NULL);
The first step in fixing this is to make all TaskStates used by QEMU
dynamically allocated. The TaskState used by the initial thread was
not, as it was allocated on main's stack. So fix that, free the
cpu_env, free the TaskState, and we're home free, right?
Not exactly. When we create a thread, we do:
ts = qemu_mallocz(sizeof(TaskState) + NEW_STACK_SIZE);
...
new_stack = ts->stack;
...
ret = pthread_attr_setstack(&attr, new_stack, NEW_STACK_SIZE);
If we blindly free the TaskState, then, we yank the current (host)
thread's stack out from underneath it while it still has things to do,
like calling pthread_exit. That causes problems, as you might expect.
The solution adopted here is to let the C library allocate the thread's
stack (so the C library can properly clean it up at pthread_exit) and
provide a hint that we want NEW_STACK_SIZE bytes of stack.
With those two changes, we're done, right? Well, almost. You see,
we're creating all these host threads and their parent threads never
bother to check that their children are finished. There's no good place
for the parent threads to do so. Therefore, we need to create the
threads in a detached state so the parent thread doesn't have to call
pthread_join on the child to release the child's resources; the child
does so automatically.
With those three major changes, we can comfortably run programs like the
above without exhausting memory. We do need to delete 'stack' from the
TaskState structure.
Signed-off-by: Nathan Froyd <froydnj@codesourcery.com>
Signed-off-by: Riku Voipio <riku.voipio@nokia.com>
