qemu/linux-user/hexagon/signal.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

294 lines
9.7 KiB
C
Raw Normal View History

/*
* Emulation of Linux signals
*
* Copyright (c) 2003 Fabrice Bellard
* Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
#include "qemu/osdep.h"
#include "qemu.h"
#include "user-internals.h"
#include "signal-common.h"
#include "linux-user/trace.h"
struct target_sigcontext {
target_ulong r0, r1, r2, r3;
target_ulong r4, r5, r6, r7;
target_ulong r8, r9, r10, r11;
target_ulong r12, r13, r14, r15;
target_ulong r16, r17, r18, r19;
target_ulong r20, r21, r22, r23;
target_ulong r24, r25, r26, r27;
target_ulong r28, r29, r30, r31;
target_ulong sa0;
target_ulong lc0;
target_ulong sa1;
target_ulong lc1;
target_ulong m0;
target_ulong m1;
target_ulong usr;
target_ulong gp;
target_ulong ugp;
target_ulong pc;
target_ulong cause;
target_ulong badva;
linux-user/hexagon: fix signal context save & restore This patch fixes the issue originally reported in this thread: https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg01102.html The root cause of the issue is a bug in the hexagon specific logic for saving & restoring context during signal delivery. The CPU state has two different representations for the predicate registers. The current logic saves & restores only the aliased HEX_REG_P3_O register, which is part of env->gpr[] field in the CPU state, but not the individual byte-level predicate registers (pO, p1, p2, p3) backed by env->pred[]. Since all predicated instructions refer only to the indiviual registers, switching to and back from a signal handler can clobber these registers if the signal handler writes to them causing the normal application code to behave unpredictably when context is restored. In the reported issue with the 'signals' test, since the updated hexagon toolchain had built musl with -O2, the functions called from non_trivial_free were inlined. This meant that the code emitted reused predicate P0 computed in the entry translation block of the function non_trivial_free in one of the child TB as part of an assertion. Since P0 is clobbered by the signal handler in the signals test, the assertion in non_trivial_free fails incorectly. Since musl for hexagon implements the 'abort' function by deliberately writing to memory via null pointer, this causes the test to fail with segmentation fault. This patch modifies the signal context save & restore logic to include the individual p0, p1, p2, p3 and excludes the 32b p3_0 register since its value is derived from the former registers. It also adds a new test case that reliabily reproduces the issue for all four predicate registers. Buglink: https://github.com/quic/toolchain_for_hexagon/issues/6 Signed-off-by: Mukilan Thiyagarajan <quic_mthiyaga@quicinc.com> Signed-off-by: Taylor Simpson <tsimpson@quicinc.com> Reviewed-by: Taylor Simpson <tsimpson@quicinc.com> Message-Id: <20221229092006.10709-2-quic_mthiyaga@quicinc.com>
2022-12-29 12:20:05 +03:00
target_ulong pred[NUM_PREGS];
};
struct target_ucontext {
unsigned long uc_flags;
target_ulong uc_link; /* target pointer */
target_stack_t uc_stack;
struct target_sigcontext uc_mcontext;
target_sigset_t uc_sigmask;
};
struct target_rt_sigframe {
uint32_t tramp[2];
struct target_siginfo info;
struct target_ucontext uc;
};
static abi_ulong get_sigframe(struct target_sigaction *ka,
CPUHexagonState *regs, size_t framesize)
{
abi_ulong sp = get_sp_from_cpustate(regs);
/* This is the X/Open sanctioned signal stack switching. */
sp = target_sigsp(sp, ka) - framesize;
sp = QEMU_ALIGN_DOWN(sp, 8);
return sp;
}
static void setup_sigcontext(struct target_sigcontext *sc, CPUHexagonState *env)
{
__put_user(env->gpr[HEX_REG_R00], &sc->r0);
__put_user(env->gpr[HEX_REG_R01], &sc->r1);
__put_user(env->gpr[HEX_REG_R02], &sc->r2);
__put_user(env->gpr[HEX_REG_R03], &sc->r3);
__put_user(env->gpr[HEX_REG_R04], &sc->r4);
__put_user(env->gpr[HEX_REG_R05], &sc->r5);
__put_user(env->gpr[HEX_REG_R06], &sc->r6);
__put_user(env->gpr[HEX_REG_R07], &sc->r7);
__put_user(env->gpr[HEX_REG_R08], &sc->r8);
__put_user(env->gpr[HEX_REG_R09], &sc->r9);
__put_user(env->gpr[HEX_REG_R10], &sc->r10);
__put_user(env->gpr[HEX_REG_R11], &sc->r11);
__put_user(env->gpr[HEX_REG_R12], &sc->r12);
__put_user(env->gpr[HEX_REG_R13], &sc->r13);
__put_user(env->gpr[HEX_REG_R14], &sc->r14);
__put_user(env->gpr[HEX_REG_R15], &sc->r15);
__put_user(env->gpr[HEX_REG_R16], &sc->r16);
__put_user(env->gpr[HEX_REG_R17], &sc->r17);
__put_user(env->gpr[HEX_REG_R18], &sc->r18);
__put_user(env->gpr[HEX_REG_R19], &sc->r19);
__put_user(env->gpr[HEX_REG_R20], &sc->r20);
__put_user(env->gpr[HEX_REG_R21], &sc->r21);
__put_user(env->gpr[HEX_REG_R22], &sc->r22);
__put_user(env->gpr[HEX_REG_R23], &sc->r23);
__put_user(env->gpr[HEX_REG_R24], &sc->r24);
__put_user(env->gpr[HEX_REG_R25], &sc->r25);
__put_user(env->gpr[HEX_REG_R26], &sc->r26);
__put_user(env->gpr[HEX_REG_R27], &sc->r27);
__put_user(env->gpr[HEX_REG_R28], &sc->r28);
__put_user(env->gpr[HEX_REG_R29], &sc->r29);
__put_user(env->gpr[HEX_REG_R30], &sc->r30);
__put_user(env->gpr[HEX_REG_R31], &sc->r31);
__put_user(env->gpr[HEX_REG_SA0], &sc->sa0);
__put_user(env->gpr[HEX_REG_LC0], &sc->lc0);
__put_user(env->gpr[HEX_REG_SA1], &sc->sa1);
__put_user(env->gpr[HEX_REG_LC1], &sc->lc1);
__put_user(env->gpr[HEX_REG_M0], &sc->m0);
__put_user(env->gpr[HEX_REG_M1], &sc->m1);
__put_user(env->gpr[HEX_REG_USR], &sc->usr);
__put_user(env->gpr[HEX_REG_GP], &sc->gp);
__put_user(env->gpr[HEX_REG_UGP], &sc->ugp);
__put_user(env->gpr[HEX_REG_PC], &sc->pc);
linux-user/hexagon: fix signal context save & restore This patch fixes the issue originally reported in this thread: https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg01102.html The root cause of the issue is a bug in the hexagon specific logic for saving & restoring context during signal delivery. The CPU state has two different representations for the predicate registers. The current logic saves & restores only the aliased HEX_REG_P3_O register, which is part of env->gpr[] field in the CPU state, but not the individual byte-level predicate registers (pO, p1, p2, p3) backed by env->pred[]. Since all predicated instructions refer only to the indiviual registers, switching to and back from a signal handler can clobber these registers if the signal handler writes to them causing the normal application code to behave unpredictably when context is restored. In the reported issue with the 'signals' test, since the updated hexagon toolchain had built musl with -O2, the functions called from non_trivial_free were inlined. This meant that the code emitted reused predicate P0 computed in the entry translation block of the function non_trivial_free in one of the child TB as part of an assertion. Since P0 is clobbered by the signal handler in the signals test, the assertion in non_trivial_free fails incorectly. Since musl for hexagon implements the 'abort' function by deliberately writing to memory via null pointer, this causes the test to fail with segmentation fault. This patch modifies the signal context save & restore logic to include the individual p0, p1, p2, p3 and excludes the 32b p3_0 register since its value is derived from the former registers. It also adds a new test case that reliabily reproduces the issue for all four predicate registers. Buglink: https://github.com/quic/toolchain_for_hexagon/issues/6 Signed-off-by: Mukilan Thiyagarajan <quic_mthiyaga@quicinc.com> Signed-off-by: Taylor Simpson <tsimpson@quicinc.com> Reviewed-by: Taylor Simpson <tsimpson@quicinc.com> Message-Id: <20221229092006.10709-2-quic_mthiyaga@quicinc.com>
2022-12-29 12:20:05 +03:00
int i;
for (i = 0; i < NUM_PREGS; i++) {
__put_user(env->pred[i], &(sc->pred[i]));
}
}
static void setup_ucontext(struct target_ucontext *uc,
CPUHexagonState *env, target_sigset_t *set)
{
__put_user(0, &(uc->uc_flags));
__put_user(0, &(uc->uc_link));
target_save_altstack(&uc->uc_stack, env);
int i;
for (i = 0; i < TARGET_NSIG_WORDS; i++) {
__put_user(set->sig[i], &(uc->uc_sigmask.sig[i]));
}
setup_sigcontext(&uc->uc_mcontext, env);
}
static inline void install_sigtramp(uint32_t *tramp)
{
__put_user(0x7800d166, tramp + 0); /* { r6=#__NR_rt_sigreturn } */
__put_user(0x5400c004, tramp + 1); /* { trap0(#1) } */
}
void setup_rt_frame(int sig, struct target_sigaction *ka,
target_siginfo_t *info,
target_sigset_t *set, CPUHexagonState *env)
{
abi_ulong frame_addr;
struct target_rt_sigframe *frame;
frame_addr = get_sigframe(ka, env, sizeof(*frame));
trace_user_setup_rt_frame(env, frame_addr);
if (!lock_user_struct(VERIFY_WRITE, frame, frame_addr, 0)) {
goto badframe;
}
setup_ucontext(&frame->uc, env, set);
tswap_siginfo(&frame->info, info);
/*
* The on-stack signal trampoline is no longer executed;
* however, the libgcc signal frame unwinding code checks
* for the presence of these two numeric magic values.
*/
install_sigtramp(frame->tramp);
env->gpr[HEX_REG_PC] = ka->_sa_handler;
env->gpr[HEX_REG_SP] = frame_addr;
env->gpr[HEX_REG_R00] = sig;
env->gpr[HEX_REG_R01] =
frame_addr + offsetof(struct target_rt_sigframe, info);
env->gpr[HEX_REG_R02] =
frame_addr + offsetof(struct target_rt_sigframe, uc);
env->gpr[HEX_REG_LR] = default_rt_sigreturn;
return;
badframe:
unlock_user_struct(frame, frame_addr, 1);
if (sig == TARGET_SIGSEGV) {
ka->_sa_handler = TARGET_SIG_DFL;
}
force_sig(TARGET_SIGSEGV);
}
static void restore_sigcontext(CPUHexagonState *env,
struct target_sigcontext *sc)
{
__get_user(env->gpr[HEX_REG_R00], &sc->r0);
__get_user(env->gpr[HEX_REG_R01], &sc->r1);
__get_user(env->gpr[HEX_REG_R02], &sc->r2);
__get_user(env->gpr[HEX_REG_R03], &sc->r3);
__get_user(env->gpr[HEX_REG_R04], &sc->r4);
__get_user(env->gpr[HEX_REG_R05], &sc->r5);
__get_user(env->gpr[HEX_REG_R06], &sc->r6);
__get_user(env->gpr[HEX_REG_R07], &sc->r7);
__get_user(env->gpr[HEX_REG_R08], &sc->r8);
__get_user(env->gpr[HEX_REG_R09], &sc->r9);
__get_user(env->gpr[HEX_REG_R10], &sc->r10);
__get_user(env->gpr[HEX_REG_R11], &sc->r11);
__get_user(env->gpr[HEX_REG_R12], &sc->r12);
__get_user(env->gpr[HEX_REG_R13], &sc->r13);
__get_user(env->gpr[HEX_REG_R14], &sc->r14);
__get_user(env->gpr[HEX_REG_R15], &sc->r15);
__get_user(env->gpr[HEX_REG_R16], &sc->r16);
__get_user(env->gpr[HEX_REG_R17], &sc->r17);
__get_user(env->gpr[HEX_REG_R18], &sc->r18);
__get_user(env->gpr[HEX_REG_R19], &sc->r19);
__get_user(env->gpr[HEX_REG_R20], &sc->r20);
__get_user(env->gpr[HEX_REG_R21], &sc->r21);
__get_user(env->gpr[HEX_REG_R22], &sc->r22);
__get_user(env->gpr[HEX_REG_R23], &sc->r23);
__get_user(env->gpr[HEX_REG_R24], &sc->r24);
__get_user(env->gpr[HEX_REG_R25], &sc->r25);
__get_user(env->gpr[HEX_REG_R26], &sc->r26);
__get_user(env->gpr[HEX_REG_R27], &sc->r27);
__get_user(env->gpr[HEX_REG_R28], &sc->r28);
__get_user(env->gpr[HEX_REG_R29], &sc->r29);
__get_user(env->gpr[HEX_REG_R30], &sc->r30);
__get_user(env->gpr[HEX_REG_R31], &sc->r31);
__get_user(env->gpr[HEX_REG_SA0], &sc->sa0);
__get_user(env->gpr[HEX_REG_LC0], &sc->lc0);
__get_user(env->gpr[HEX_REG_SA1], &sc->sa1);
__get_user(env->gpr[HEX_REG_LC1], &sc->lc1);
__get_user(env->gpr[HEX_REG_M0], &sc->m0);
__get_user(env->gpr[HEX_REG_M1], &sc->m1);
__get_user(env->gpr[HEX_REG_USR], &sc->usr);
__get_user(env->gpr[HEX_REG_GP], &sc->gp);
__get_user(env->gpr[HEX_REG_UGP], &sc->ugp);
__get_user(env->gpr[HEX_REG_PC], &sc->pc);
linux-user/hexagon: fix signal context save & restore This patch fixes the issue originally reported in this thread: https://lists.gnu.org/archive/html/qemu-devel/2021-11/msg01102.html The root cause of the issue is a bug in the hexagon specific logic for saving & restoring context during signal delivery. The CPU state has two different representations for the predicate registers. The current logic saves & restores only the aliased HEX_REG_P3_O register, which is part of env->gpr[] field in the CPU state, but not the individual byte-level predicate registers (pO, p1, p2, p3) backed by env->pred[]. Since all predicated instructions refer only to the indiviual registers, switching to and back from a signal handler can clobber these registers if the signal handler writes to them causing the normal application code to behave unpredictably when context is restored. In the reported issue with the 'signals' test, since the updated hexagon toolchain had built musl with -O2, the functions called from non_trivial_free were inlined. This meant that the code emitted reused predicate P0 computed in the entry translation block of the function non_trivial_free in one of the child TB as part of an assertion. Since P0 is clobbered by the signal handler in the signals test, the assertion in non_trivial_free fails incorectly. Since musl for hexagon implements the 'abort' function by deliberately writing to memory via null pointer, this causes the test to fail with segmentation fault. This patch modifies the signal context save & restore logic to include the individual p0, p1, p2, p3 and excludes the 32b p3_0 register since its value is derived from the former registers. It also adds a new test case that reliabily reproduces the issue for all four predicate registers. Buglink: https://github.com/quic/toolchain_for_hexagon/issues/6 Signed-off-by: Mukilan Thiyagarajan <quic_mthiyaga@quicinc.com> Signed-off-by: Taylor Simpson <tsimpson@quicinc.com> Reviewed-by: Taylor Simpson <tsimpson@quicinc.com> Message-Id: <20221229092006.10709-2-quic_mthiyaga@quicinc.com>
2022-12-29 12:20:05 +03:00
int i;
for (i = 0; i < NUM_PREGS; i++) {
__get_user(env->pred[i], &(sc->pred[i]));
}
}
static void restore_ucontext(CPUHexagonState *env, struct target_ucontext *uc)
{
sigset_t blocked;
target_sigset_t target_set;
int i;
target_sigemptyset(&target_set);
for (i = 0; i < TARGET_NSIG_WORDS; i++) {
__get_user(target_set.sig[i], &(uc->uc_sigmask.sig[i]));
}
target_to_host_sigset_internal(&blocked, &target_set);
set_sigmask(&blocked);
restore_sigcontext(env, &uc->uc_mcontext);
}
long do_rt_sigreturn(CPUHexagonState *env)
{
struct target_rt_sigframe *frame;
abi_ulong frame_addr;
frame_addr = env->gpr[HEX_REG_SP];
trace_user_do_sigreturn(env, frame_addr);
if (!lock_user_struct(VERIFY_READ, frame, frame_addr, 1)) {
goto badframe;
}
restore_ucontext(env, &frame->uc);
target_restore_altstack(&frame->uc.uc_stack, env);
unlock_user_struct(frame, frame_addr, 0);
return -QEMU_ESIGRETURN;
badframe:
unlock_user_struct(frame, frame_addr, 0);
force_sig(TARGET_SIGSEGV);
return 0;
}
void setup_sigtramp(abi_ulong sigtramp_page)
{
uint32_t *tramp = lock_user(VERIFY_WRITE, sigtramp_page, 4 * 2, 0);
assert(tramp != NULL);
default_rt_sigreturn = sigtramp_page;
install_sigtramp(tramp);
unlock_user(tramp, sigtramp_page, 4 * 2);
}