The SH4 FPSCR rounding-mode (RM) and denormal (DN) bits are not held
only in env->fpscr: they are also reflected into the derived
env->fp_status via set_float_rounding_mode()/set_flush_to_zero(). The
guest keeps the two in sync by routing every write to FPSCR through
helper_ld_fpscr().
restore_sigcontext() wrote the saved value straight into env->fpscr and
never touched env->fp_status, so on sigreturn the interrupted code
resumed with whatever FP rounding mode and flush-to-zero setting the
signal handler last installed. (regs->flags = 0 forces the FR/SZ/PR TB
flags to be recomputed, but fp_status is runtime float state, not a TB
flag, so it was left stale.) This is the FP analogue of the T/M/Q bit
problem just fixed for the integer status register.
Factor the FPSCR -> fp_status synchronisation out of helper_ld_fpscr()
into cpu_load_fpscr() and use it from restore_sigcontext() so the
rounding mode round-trips correctly across signal delivery.
Fixes: c3b5bc8ab3 ("SH4: Signal handling for the user space emulator, by Magnus Damm.")
Cc: qemu-stable@nongnu.org
Reviewed-by: Yoshinori Sato <yoshinori.sato@nifty.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Helge Deller <deller@gmx.de>
QEMU keeps the SH4 T, M and Q status-register bits outside env->sr, in
the dedicated env->sr_t, env->sr_m and env->sr_q fields; cpu_read_sr()
folds them back into the architectural SR value and cpu_write_sr()
splits them back out.
setup_sigcontext() saved the bare env->sr (so the T/M/Q bits were always
zero in the signal frame) and restore_sigcontext() wrote the value
straight back into env->sr without updating sr_t/sr_m/sr_q. As a result
the T bit was never preserved across signal delivery: on sigreturn the
interrupted code resumed with whatever T value the signal handler last
left behind. Any conditional branch (or addc/subc/rotcl/div1, etc.)
immediately following the interrupted instruction could then take the
wrong path.
This is the cause of the long-standing intermittent failures of the
tests/tcg/multiarch/signals.c test on sh4, which was marked BROKEN. With
a SIGRTMIN timer firing every millisecond across many threads, the race
was hit a few percent of the time and corrupted the guest heap, surfacing
as a SIGSEGV in memset, a malloc assertion, or an rseq registration abort.
Traced on a deterministic rr recording: a cmp/hi set T=0, the timer
signal interrupted the very next instruction (a bf), the handler left
T=1, and the resumed bf took glibc calloc's MORECORE_CLEARS branch,
using the old top-chunk size as the clear length for a freshly split
small chunk and running memset off the end of the heap.
Fix setup_sigcontext()/restore_sigcontext() to use cpu_read_sr() and
cpu_write_sr() so the T, M and Q bits round-trip correctly, and drop the
BROKEN annotation on the sh4 signals test.
Fixes: c3b5bc8ab3 ("SH4: Signal handling for the user space emulator, by Magnus Damm.")
Cc: qemu-stable@nongnu.org
Reviewed-by: Yoshinori Sato <yoshinori.sato@nifty.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Helge Deller <deller@gmx.de>
Provides replacement VDSO with sigreturn trampolines
(__kernel_sigreturn, __kernel_rt_sigreturn) and syscall stubs
(clock_gettime, clock_gettime64, clock_getres, gettimeofday).
Both LE and BE blobs are committed and selected at compile time via
TARGET_BIG_ENDIAN. The BE variant requires an sh4eb-unknown-linux-gnu
toolchain; sh4-unknown-linux-gnu does not support -mb.
CFI register numbers follow GCC's SH_DEBUGGER_REGNO:
PR=17, GBR=18, MACH=20, MACL=21, FPUL=23, FPSCR=24, FR0-15=25-40.
Signed-off-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Helge Deller <deller@gmx.de>
QEMU used MOVW(2) (0x9300), which loads the syscall number from PC+4,
instead of the kernel's MOVW(7) (0x9305), which loads from PC+14. The
kernel uses five "or r0,r0" nop pads between TRAP_NOARG and the syscall
number word to reach that offset. libunwind's unw_is_signal_frame checks
for the exact kernel byte pattern 0xc3109305 at the frame PC, so QEMU's
compact layout was not detected, breaking unwinding through signal frames.
Expand each trampoline from 6 to 16 bytes matching the kernel layout
defined in arch/sh/kernel/signal_32.c:
#define MOVW(n) (0x9300|((n)-2)) /* Move mem word at PC+n to R3 */
#define TRAP_NOARG 0xc310 /* Syscall w/no args (NR in R3) */
#define OR_R0_R0 0x200b /* or r0,r0 (insert to avoid hardware bug) */
__put_user(MOVW(7), &frame->retcode[0]); /* 0x9305 */
__put_user(TRAP_NOARG, &frame->retcode[1]); /* 0xc310 */
__put_user(OR_R0_R0, &frame->retcode[2]); /* 0x200b */
__put_user(OR_R0_R0, &frame->retcode[3]); /* 0x200b */
__put_user(OR_R0_R0, &frame->retcode[4]); /* 0x200b */
__put_user(OR_R0_R0, &frame->retcode[5]); /* 0x200b */
__put_user(OR_R0_R0, &frame->retcode[6]); /* 0x200b */
__put_user((__NR_sigreturn), &frame->retcode[7]);
The first two halfwords (MOVW(7) || TRAP_NOARG = 0xc3109305) form the
32-bit value libunwind checks at the frame PC, followed by two
OR_R0_R0 halfwords (0x200b200b) at PC+4. The same layout applies to
the rt_sigreturn trampoline (lines 366-373 of signal_32.c).
Neither this fix nor the companion tuc_link fix is independently
sufficient: this fix makes signal frames detectable but register reads
remain garbage without the correct ucontext layout; that fix corrects the
ucontext layout but libunwind still cannot detect the frame without the
correct trampoline pattern. Together they fix the following libunwind
tests on a 64-bit host:
Gtest-sig-context, Gtest-trace, Ltest-init-local-signal,
Ltest-sig-context, Ltest-trace
Signed-off-by: Matt Turner <mattst88@gmail.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Helge Deller <deller@gmx.de>
tuc_link is declared as 'struct target_ucontext *', which is a HOST
pointer. On a 64-bit host running a 32-bit SH4 target, this is 8 bytes
instead of the 4 bytes the target expects, padding pushes tuc_mcontext
8 bytes past its correct offset.
When a signal handler receives ucontext_t *, every field accessed through
uc_mcontext (gregs[], pc, pr, ...) is read from the wrong address. In
particular the saved PC comes back as a garbage stack value, which breaks
any code that initialises a libunwind cursor from the signal context.
Fix it by using abi_ulong, which is always sized to the target ABI (4
bytes for SH4), matching the layout the kernel and glibc agree on. This
is the same pattern used by arm/signal.c.
Also remove the (unsigned long *) cast from the __put_user that zeros
tuc_link. The cast was harmless when tuc_link was pointer-sized (8
bytes matching unsigned long on a 64-bit host), but after the type
change __put_user's sizeof dispatch would select stq_le_p (8-byte write)
for a now-4-byte field, silently overwriting the start of tuc_stack.
Neither this fix nor the companion setup_sigtramp fix is independently
sufficient: this fix corrects register values read from the signal context
but libunwind still cannot detect the frame without the correct trampoline
pattern; that fix makes the frame detectable but register reads remain
garbage without the correct ucontext layout. Together they fix the
following libunwind tests on a 64-bit host:
Gtest-sig-context, Gtest-trace, Ltest-init-local-signal,
Ltest-sig-context, Ltest-trace
Signed-off-by: Matt Turner <mattst88@gmail.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Helge Deller <deller@gmx.de>
The syscall emulation code previously wasn't interruptible via
cpu_loop_exit(), as this construct relies on a longjmp target that is not
live anymore in the syscall handling code. Consequently, longjmp() would
operate on a (potentially overwritten) stale jump buffer. This patch adds an additional
setjmp and the necessary handling around it to make longjmp() (and by
proxy cpu_loop_exit() safe to call even within a syscall context.
Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Florian Hofhammer <florian.hofhammer@epfl.ch>
Link: https://lore.kernel.org/qemu-devel/20260305-setpc-v5-v7-3-4c3adba52403@epfl.ch
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Add a user-mode emulation version of the function. More will be
added later, for now it is just process_queued_cpu_work.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
PowerPC was the one outlier that defined both ELF_ARCH and
ELF_MACHINE; ELF_ARCH was defined incorrectly, necessitating
the definition of elf_check_arch.
However, the elf file header field in question is called
e_machine, so ELF_MACHINE is in fact the better name.
Mechanically change most target/target_elf.h files,
then adjust ppc/target_elf.h manually.
Do not provide a default for ELF_MACHINE.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Make use of the fact that target_elf_gregset_t is a proper structure.
Drop ELF_NREG, target_elf_greg_t, and tswapreg.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Move elf_core_copy_regs to elfload.c.
Move HAVE_ELF_CORE_DUMP, ELF_NREGS, target_elf_gregset_t to target_elf.h.
For now, duplicate the definitions of target_elf_greg_t and tswapreg.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Merge init_thread and target_cpu_copy_regs.
There's no point going through a target_pt_regs intermediate.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Rename from cpu_get_model to emphasize that this is an elf-specific
function. Declare the function once in loader.h.
This frees up target_elf.h for other uses.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Prepare to split the main linux-user/elfload.c.
Create empty files for each target, and add the common build rule.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
SA_RESTORER and the associated sa_restorer field of struct sigaction are
an obsolete feature, not expected to be used by future architectures.
They are also absent on RISC-V, LoongArch, Hexagon and OpenRISC, but
defined due to their use of generic/signal.h. This leads to corrupted
data and out-of-bounds accesses.
Move the definition of TARGET_SA_RESTORER out of generic/signal.h into the
target_signal.h files that need it. Note that m68k has the sa_restorer
field, but does not use it and does not define SA_RESTORER.
Reported-by: Thomas Weißschuh <thomas@t-8ch.de>
Signed-off-by: Andreas Schwab <schwab@suse.de>
Reviewed-by: Thomas Weißschuh <thomas@t-8ch.de>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <mvmed060xc9.fsf@suse.de>
"linux-user/cpu_loop-common.h" is generic enough to be used by
bsd-user, so rename it as "user/cpu_loop.h".
Mechanical change running:
$ sed -i -e 's,cpu_loop-common.h,user/cpu_loop.h,' \
$(git grep -l cpu_loop-common.h)
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20241212185341.2857-17-philmd@linaro.org>
sh4 uses gUSA (general UserSpace Atomicity) to provide atomicity on CPUs
that don't have atomic instructions. A gUSA region that adds 1 to an
atomic variable stored in @R2 looks like this:
4004b6: 03 c7 mova 4004c4 <gusa+0x10>,r0
4004b8: f3 61 mov r15,r1
4004ba: 09 00 nop
4004bc: fa ef mov #-6,r15
4004be: 22 63 mov.l @r2,r3
4004c0: 01 73 add #1,r3
4004c2: 32 22 mov.l r3,@r2
4004c4: 13 6f mov r1,r15
R0 contains a pointer to the end of the gUSA region
R1 contains the saved stack pointer
R15 contains negative length of the gUSA region
When this region is interrupted by a signal, the kernel detects if
R15 >= -128U. If yes, the kernel rolls back PC to the beginning of the
region and restores SP by copying R1 to R15.
The problem happens if we are interrupted by a signal at address 4004c4.
R15 still holds the value -6, but the atomic value was already written by
an instruction at address 4004c2. In this situation we can't undo the
gUSA. The function unwind_gusa does nothing, the signal handler attempts
to push a signal frame to the address -6 and crashes.
This patch fixes it, so that if we are interrupted at the last instruction
in a gUSA region, we copy R1 to R15 to restore the correct stack pointer
and avoid crashing.
There's another bug: if we are interrupted in a delay slot, we save the
address of the instruction in the delay slot. We must save the address of
the previous instruction.
Cc: qemu-stable@nongnu.org
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Reviewed-by: Yoshinori Sato <ysato@users.sourcefoege.jp>
Message-Id: <b16389f7-6c62-70b7-59b3-87533c0bcc@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Move the various open_cpuinfo functions into new files.
Move the m68k open_hardware function as well.
All other guest architectures get a boilerplate empty file.
Tested-by: Helge Deller <deller@gmx.de>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Copy each guest kernel's default value, then bound it
against reserved_va or the host address space.
Tested-by: Helge Deller <deller@gmx.de>
Reviewed-by: Helge Deller <deller@gmx.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This makes target_flat.h behave like every other target_xxx.h header.
It also makes it actually work -- while the current header says adding
a header to the target subdir overrides the common one, it doesn't.
This is for two reasons:
* meson.build adds -Ilinux-user before -Ilinux-user/$arch
* the compiler search path for "target_flat.h" looks in the same dir
as the source file before searching -I paths.
This can be seen with the xtensa port -- the subdir settings aren't
used which breaks stack setup.
Move it to the generic/ subdir and add include stubs like every
other target_xxx.h header is handled.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230129004625.11228-1-vapier@gentoo.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
The value previously chosen overlaps GUSA_MASK.
Rename all DELAY_SLOT_* and GUSA_* defines to emphasize
that they are included in TB_FLAGs. Add aliases for the
FPSCR and SR bits that are included in TB_FLAGS, so that
we don't accidentally reassign those bits.
Fixes: 4da06fb306 ("target/sh4: Implement prctl_unalign_sigbus")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/856
Reviewed-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Seen while compiling on Alpine:
In file included from ../linux-user/strace.c:17:
In file included from ../linux-user/qemu.h:11:
In file included from ../linux-user/syscall_defs.h:1247:
../linux-user/sh4/termbits.h:276:10: warning: 'TIOCSER_TEMT' macro redefined
[-Wmacro-redefined]
# define TIOCSER_TEMT 0x01 /* Transmitter physically empty */
^
/usr/include/sys/ioctl.h:50:9: note: previous definition is here
#define TIOCSER_TEMT 1
^
1 warning generated.
Add the TARGET_ prefix here, too, like we do it on the other architectures.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Message-Id: <20220330134302.979686-1-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Leave TARGET_ALIGNED_ONLY set, but use the new CPUState
flag to set MO_UNALN for the instructions that the kernel
handles in the unaligned trap.
The Linux kernel does not handle all memory operations: no
floating-point and no MAC.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20211227150127.2659293-7-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Since the prctl constants are supposed to be generic, supply
any that are not provided by the host.
Split out subroutines for PR_GET_FP_MODE, PR_SET_FP_MODE,
PR_GET_VL, PR_SET_VL, PR_RESET_KEYS, PR_SET_TAGGED_ADDR_CTRL,
PR_GET_TAGGED_ADDR_CTRL. Return EINVAL for guests that do
not support these options rather than pass them on to the host.
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20211227150127.2659293-2-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
This value is fully internal to qemu, and so is not a TARGET define.
Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This value is fully internal to qemu, and so is not a TARGET define.
We use this as an extra marker for both host and target errno.
Reviewed-by: Warner Losh <imp@bsdimp.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
The fallback code in cpu_loop_exit_sigsegv is sufficient
for sh4 linux-user.
Remove the code from cpu_loop that raised SIGSEGV.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
qemu.h is included in various non-linux-user files (which
mostly want the TaskState struct and the functions for
doing usermode access to guest addresses like lock_user(),
unlock_user(), get_user*(), etc).
Split out the parts that are only used in linux-user itself
into a new user-internals.h. This leaves qemu.h with basically
three things:
* the definition of the TaskState struct
* the user-access functions and macros
* do_brk()
all of which are needed by code outside linux-user that
includes qemu.h.
The addition of all the extra #include lines was done with
sed -i '/include.*qemu\.h/a #include "user-internals.h"' $(git grep -l 'include.*qemu\.h' linux-user)
(and then undoing the change to fpa11.h).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210908154405.15417-8-peter.maydell@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
We want to access the target errno indepently of the rest of the
linux-user code. Move the header containing the generic errno
definitions ('errno_defs.h') to 'generic/target_errno_defs.h',
create a new 'target_errno_defs.h' in each target which itself
includes 'generic/target_errno_defs.h'.
Suggested-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20210708170550.1846343-4-f4bug@amsat.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
