[RFC] Attempt to clean up sigsuspend et al
From: David Woodhouse <dwmw2@infradead.org>
Date: 2005-11-11 21:36:39
We currently play tricks with assembly wrappers for various syscalls, in case a signal is delivered. We need to make sure the registers are all saved for the benefit of ptrace, and we need to return to userspace without clobbering r4 and r5, as the normal syscall exit path would. As usual, these tricks differ subtly between ppc32 and ppc64 too. This is an attempt to get rid of it all, motivated mostly because I want to add two _new_ functions which would need such treatment (pselect, ppoll). We define a new thread flag TIF_RESTOREALL which is set by the syscall which needs this treatment, and make the syscall return path honour it. This seems to work as far as I've tested it (i.e. the system boots) for ppc64, but I haven't even attempted to compile the 32-bit version yet. I've also removed the explicit calls to ptrace_notify() which were in #ifdef CONFIG_PPC64 -- if we're now stopping on the way back to userspace, we don't need to do that any more. I'm going to need to go through some of the sigsuspend/signal cases with a fine-tooth comb and make sure we're still getting a ptrace stop (and only _one_ ptrace stop) in all cases where that's appropriate. I also tried to clean up entry_64.S so it only checks the thread flags in the fast path _once_. This means that for ptrace we'll disable interrupts, then check the flags and enable interrupts again before calling do_syscall_trace_leave(). That makes the slow path match the PPC32 code and makes the fast path smaller. Comments?
diff --git a/arch/powerpc/kernel/entry_32.S b/arch/powerpc/kernel/entry_32.S
index 2e99ae4..59dc2e7 100644
--- a/arch/powerpc/kernel/entry_32.S
+++ b/arch/powerpc/kernel/entry_32.S@@ -222,25 +222,25 @@ ret_from_syscall: bl do_show_syscall_exit #endif mr r6,r3 - li r11,-_LAST_ERRNO - cmplw 0,r3,r11 + lwz r11,_CCR(r1) /* Load CR */ + li r10,-_LAST_ERRNO + cmplw 0,r3,r10 rlwinm r12,r1,0,0,(31-THREAD_SHIFT) /* current_thread_info() */ blt+ 30f - lbz r11,TI_SC_NOERR(r12) - cmpwi r11,0 + lbz r10,TI_SC_NOERR(r12) + cmpwi r10,0 bne 30f neg r3,r3 - lwz r10,_CCR(r1) /* Set SO bit in CR */ - oris r10,r10,0x1000 - stw r10,_CCR(r1) + oris r11,r11,0x1000 /* Set SO bit in CR */ /* disable interrupts so current_thread_info()->flags can't change */ 30: LOAD_MSR_KERNEL(r10,MSR_KERNEL) /* doesn't include MSR_EE */ SYNC MTMSRD(r10) lwz r9,TI_FLAGS(r12) - andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SIGPENDING|_TIF_NEED_RESCHED) + andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_RESTOREALL) bne- syscall_exit_work + stw r11,_CCR(r1) syscall_exit_cont: #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE) /* If the process has its own DBCR0 value, load it up. The single
@@ -292,9 +292,13 @@ syscall_dotrace: b syscall_dotrace_cont syscall_exit_work: + andi. r0,r9,_TIF_RESTOREALL + bne- 3f + stw r6,RESULT(r1) /* Save result */ stw r3,GPR3(r1) /* Update return value */ - andi. r0,r9,_TIF_SYSCALL_T_OR_A + stw r11,_CCR(r1) +3: andi. r0,r9,_TIF_SYSCALL_T_OR_A beq 5f ori r10,r10,MSR_EE SYNC
@@ -407,22 +411,6 @@ show_syscalls_task: * to save all the nonvolatile registers (r13 - r31) before calling * the C code. */ - .globl ppc_sigsuspend -ppc_sigsuspend: - SAVE_NVGPRS(r1) - lwz r0,_TRAP(r1) - rlwinm r0,r0,0,0,30 /* clear LSB to indicate full */ - stw r0,_TRAP(r1) /* register set saved */ - b sys_sigsuspend - - .globl ppc_rt_sigsuspend -ppc_rt_sigsuspend: - SAVE_NVGPRS(r1) - lwz r0,_TRAP(r1) - rlwinm r0,r0,0,0,30 - stw r0,_TRAP(r1) - b sys_rt_sigsuspend - .globl ppc_fork ppc_fork: SAVE_NVGPRS(r1)
@@ -447,14 +435,6 @@ ppc_clone: stw r0,_TRAP(r1) /* register set saved */ b sys_clone - .globl ppc_swapcontext -ppc_swapcontext: - SAVE_NVGPRS(r1) - lwz r0,_TRAP(r1) - rlwinm r0,r0,0,0,30 /* clear LSB to indicate full */ - stw r0,_TRAP(r1) /* register set saved */ - b sys_swapcontext - /* * Top-level page fault handling. * This is in assembler because if do_page_fault tells us that
@@ -626,16 +606,6 @@ END_FTR_SECTION_IFSET(CPU_FTR_601) .long ret_from_except #endif - .globl sigreturn_exit -sigreturn_exit: - subi r1,r3,STACK_FRAME_OVERHEAD - rlwinm r12,r1,0,0,(31-THREAD_SHIFT) /* current_thread_info() */ - lwz r9,TI_FLAGS(r12) - andi. r0,r9,_TIF_SYSCALL_T_OR_A - beq+ ret_from_except_full - bl do_syscall_trace_leave - /* fall through */ - .globl ret_from_except_full ret_from_except_full: REST_NVGPRS(r1)
@@ -658,7 +628,7 @@ user_exc_return: /* r10 contains MSR_KE /* Check current_thread_info()->flags */ rlwinm r9,r1,0,0,(31-THREAD_SHIFT) lwz r9,TI_FLAGS(r9) - andi. r0,r9,(_TIF_SIGPENDING|_TIF_NEED_RESCHED) + andi. r0,r9,(_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_RESTOREALL) bne do_work restore_user:
@@ -876,6 +846,8 @@ load_dbcr0: #endif /* !(CONFIG_4xx || CONFIG_BOOKE) */ do_work: /* r10 contains MSR_KERNEL here */ + andi. r0,r9,_TIF_RESTOREALL + bne- do_clear_restoreall andi. r0,r9,_TIF_NEED_RESCHED beq do_user_signal
@@ -890,6 +862,8 @@ recheck: MTMSRD(r10) /* disable interrupts */ rlwinm r9,r1,0,0,(31-THREAD_SHIFT) lwz r9,TI_FLAGS(r9) + andi. r0,r9,_TIF_RESTOREALL + bne- do_clear_restoreall andi. r0,r9,_TIF_NEED_RESCHED bne- do_resched andi. r0,r9,_TIF_SIGPENDING
@@ -911,6 +885,9 @@ do_user_signal: /* r10 contains MSR_KE REST_NVGPRS(r1) b recheck +do_clear_restoreall: + bl do_clear_tif_restoreall + b recheck /* * We come here when we are at the end of handling an exception * that occurred at a place where taking an exception will lose
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 2d22bf0..756663e 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S@@ -157,12 +157,6 @@ syscall_exit: bge- syscall_error syscall_error_cont: - /* check for syscall tracing or audit */ - ld r9,TI_FLAGS(r12) - andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP) - bne- syscall_exit_trace -syscall_exit_trace_cont: - /* disable interrupts so current_thread_info()->flags can't change, and so that we don't get interrupted after loading SRR0/1. */ ld r8,_MSR(r1)
@@ -173,8 +167,9 @@ syscall_exit_trace_cont: rotldi r10,r10,16 mtmsrd r10,1 ld r9,TI_FLAGS(r12) - andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SIGPENDING|_TIF_NEED_RESCHED) + andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP|_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_RESTOREALL) bne- syscall_exit_work + std r5,_CCR(r1) ld r7,_NIP(r1) stdcx. r0,0,r1 /* to clear the reservation */ andi. r6,r8,MSR_PR
@@ -205,7 +200,6 @@ syscall_error: bne- syscall_error_cont neg r3,r3 oris r5,r5,0x1000 /* Set SO bit in CR */ - std r5,_CCR(r1) b syscall_error_cont /* Traced system call support */
@@ -225,8 +219,22 @@ syscall_dotrace: ld r10,TI_FLAGS(r10) b syscall_dotrace_cont -syscall_exit_trace: +syscall_exit_work: + /* Avoid saving r3 and ccr if TIF_RESTOREALL flag is set */ + andi. r0,r9,_TIF_RESTOREALL + bne- 1f + std r3,GPR3(r1) + std r5,_CCR(r1) + + /* If tracing, re-enable interrupts and do it */ +1: andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP) + beq .ret_from_except_lite + + mfmsr r10 + ori r10,r10,MSR_EE + mtmsrd r10,1 + bl .save_nvgprs addi r3,r1,STACK_FRAME_OVERHEAD bl .do_syscall_trace_leave
@@ -234,12 +242,14 @@ syscall_exit_trace: ld r3,GPR3(r1) ld r5,_CCR(r1) clrrdi r12,r1,THREAD_SHIFT - b syscall_exit_trace_cont -/* Stuff to do on exit from a system call. */ -syscall_exit_work: - std r3,GPR3(r1) - std r5,_CCR(r1) + /* Disable interrupts again and handle other work if any */ + mfmsr r10 + rldicl r10,r10,48,1 + rotldi r10,r10,16 + mtmsrd r10,1 + + ld r9,TI_FLAGS(r12) b .ret_from_except_lite /* Save non-volatile GPRs, if not already saved. */
@@ -260,35 +270,6 @@ _GLOBAL(save_nvgprs) * the C code. Similarly, fork, vfork and clone need the full * register state on the stack so that it can be copied to the child. */ -_GLOBAL(ppc32_sigsuspend) - bl .save_nvgprs - bl .compat_sys_sigsuspend - b 70f - -_GLOBAL(ppc64_rt_sigsuspend) - bl .save_nvgprs - bl .sys_rt_sigsuspend - b 70f - -_GLOBAL(ppc32_rt_sigsuspend) - bl .save_nvgprs - bl .compat_sys_rt_sigsuspend -70: cmpdi 0,r3,0 - /* If it returned an error, we need to return via syscall_exit to set - the SO bit in cr0 and potentially stop for ptrace. */ - bne syscall_exit - /* If sigsuspend() returns zero, we are going into a signal handler. We - may need to call audit_syscall_exit() to mark the exit from sigsuspend() */ -#ifdef CONFIG_AUDITSYSCALL - ld r3,PACACURRENT(r13) - ld r4,AUDITCONTEXT(r3) - cmpdi 0,r4,0 - beq .ret_from_except /* No audit_context: Leave immediately. */ - li r4, 2 /* AUDITSC_FAILURE */ - li r5,-4 /* It's always -EINTR */ - bl .audit_syscall_exit -#endif - b .ret_from_except _GLOBAL(ppc_fork) bl .save_nvgprs
@@ -305,37 +286,6 @@ _GLOBAL(ppc_clone) bl .sys_clone b syscall_exit -_GLOBAL(ppc32_swapcontext) - bl .save_nvgprs - bl .compat_sys_swapcontext - b 80f - -_GLOBAL(ppc64_swapcontext) - bl .save_nvgprs - bl .sys_swapcontext - b 80f - -_GLOBAL(ppc32_sigreturn) - bl .compat_sys_sigreturn - b 80f - -_GLOBAL(ppc32_rt_sigreturn) - bl .compat_sys_rt_sigreturn - b 80f - -_GLOBAL(ppc64_rt_sigreturn) - bl .sys_rt_sigreturn - -80: cmpdi 0,r3,0 - blt syscall_exit - clrrdi r4,r1,THREAD_SHIFT - ld r4,TI_FLAGS(r4) - andi. r4,r4,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP) - beq+ 81f - addi r3,r1,STACK_FRAME_OVERHEAD - bl .do_syscall_trace_leave -81: b .ret_from_except - _GLOBAL(ret_from_fork) bl .schedule_tail REST_NVGPRS(r1)
@@ -602,7 +552,12 @@ user_work: bl .schedule b .ret_from_except_lite -1: bl .save_nvgprs +1: andi. r0,r4,_TIF_RESTOREALL + beq 2f + bl .do_clear_tif_restoreall + b .ret_from_except_lite + +2: bl .save_nvgprs li r3,0 addi r4,r1,STACK_FRAME_OVERHEAD bl .do_signal
diff --git a/arch/powerpc/kernel/signal_32.c b/arch/powerpc/kernel/signal_32.c
index 081d931..d18aebd 100644
--- a/arch/powerpc/kernel/signal_32.c
+++ b/arch/powerpc/kernel/signal_32.c@@ -75,7 +75,6 @@ * registers from *regs. This is what we need * to do when a signal has been delivered. */ -#define sigreturn_exit(regs) return 0 #define GP_REGS_SIZE min(sizeof(elf_gregset_t32), sizeof(struct pt_regs32)) #undef __SIGNAL_FRAMESIZE
@@ -178,8 +177,6 @@ static inline int restore_general_regs(s #else /* CONFIG_PPC64 */ -extern void sigreturn_exit(struct pt_regs *); - #define GP_REGS_SIZE min(sizeof(elf_gregset_t), sizeof(struct pt_regs)) static inline int put_sigset_t(sigset_t __user *uset, sigset_t *set)
@@ -255,8 +252,10 @@ long sys_sigsuspend(old_sigset_t mask, i while (1) { current->state = TASK_INTERRUPTIBLE; schedule(); - if (do_signal(&saveset, regs)) - sigreturn_exit(regs); + if (do_signal(&saveset, regs)) { + set_thread_flag(TIF_RESTOREALL); + return 0; + } } }
@@ -291,8 +290,10 @@ long sys_rt_sigsuspend( while (1) { current->state = TASK_INTERRUPTIBLE; schedule(); - if (do_signal(&saveset, regs)) - sigreturn_exit(regs); + if (do_signal(&saveset, regs)) { + set_thread_flag(TIF_RESTOREALL); + return 0; + } } }
@@ -829,12 +830,6 @@ static int handle_rt_signal(unsigned lon regs->gpr[6] = (unsigned long) rt_sf; regs->nip = (unsigned long) ka->sa.sa_handler; regs->trap = 0; -#ifdef CONFIG_PPC64 - regs->result = 0; - - if (test_thread_flag(TIF_SINGLESTEP)) - ptrace_notify(SIGTRAP); -#endif return 1; badframe:
@@ -912,8 +907,8 @@ long sys_swapcontext(struct ucontext __u */ if (do_setcontext(new_ctx, regs, 0)) do_exit(SIGSEGV); - sigreturn_exit(regs); - /* doesn't actually return back to here */ + + set_thread_flag(TIF_RESTOREALL); return 0; }
@@ -946,12 +941,11 @@ long sys_rt_sigreturn(int r3, int r4, in * nobody does any... */ compat_sys_sigaltstack((u32)(u64)&rt_sf->uc.uc_stack, 0, 0, 0, 0, 0, regs); - return (int)regs->result; #else do_sigaltstack(&rt_sf->uc.uc_stack, NULL, regs->gpr[1]); - sigreturn_exit(regs); /* doesn't return here */ - return 0; #endif + set_thread_flag(TIF_RESTOREALL); + return 0; bad: force_sig(SIGSEGV, current);
@@ -1042,9 +1036,7 @@ int sys_debug_setcontext(struct ucontext */ do_sigaltstack(&ctx->uc_stack, NULL, regs->gpr[1]); - sigreturn_exit(regs); - /* doesn't actually return back to here */ - + set_thread_flag(TIF_RESTOREALL); out: return 0; }
@@ -1109,12 +1101,6 @@ static int handle_signal(unsigned long s regs->gpr[4] = (unsigned long) sc; regs->nip = (unsigned long) ka->sa.sa_handler; regs->trap = 0; -#ifdef CONFIG_PPC64 - regs->result = 0; - - if (test_thread_flag(TIF_SINGLESTEP)) - ptrace_notify(SIGTRAP); -#endif return 1;
@@ -1162,12 +1148,8 @@ long sys_sigreturn(int r3, int r4, int r || restore_user_regs(regs, sr, 1)) goto badframe; -#ifdef CONFIG_PPC64 - return (int)regs->result; -#else - sigreturn_exit(regs); /* doesn't return */ + set_thread_flag(TIF_RESTOREALL); return 0; -#endif badframe: force_sig(SIGSEGV, current);
diff --git a/arch/powerpc/kernel/signal_64.c b/arch/powerpc/kernel/signal_64.c
index 58194e1..7520ad4 100644
--- a/arch/powerpc/kernel/signal_64.c
+++ b/arch/powerpc/kernel/signal_64.c@@ -96,8 +96,10 @@ long sys_rt_sigsuspend(sigset_t __user * while (1) { current->state = TASK_INTERRUPTIBLE; schedule(); - if (do_signal(&saveset, regs)) + if (do_signal(&saveset, regs)) { + set_thread_flag(TIF_RESTOREALL); return 0; + } } }
@@ -343,6 +345,7 @@ int sys_swapcontext(struct ucontext __us do_exit(SIGSEGV); /* This returns like rt_sigreturn */ + set_thread_flag(TIF_RESTOREALL); return 0; }
@@ -375,7 +378,8 @@ int sys_rt_sigreturn(unsigned long r3, u */ do_sigaltstack(&uc->uc_stack, NULL, regs->gpr[1]); - return regs->result; + set_thread_flag(TIF_RESTOREALL); + return 0; badframe: #if DEBUG_SIG
@@ -454,9 +458,6 @@ static int setup_rt_frame(int signr, str if (err) goto badframe; - if (test_thread_flag(TIF_SINGLESTEP)) - ptrace_notify(SIGTRAP); - return 1; badframe:
diff --git a/arch/powerpc/kernel/syscalls.c b/arch/powerpc/kernel/syscalls.c
index f72ced1..f8ec816 100644
--- a/arch/powerpc/kernel/syscalls.c
+++ b/arch/powerpc/kernel/syscalls.c@@ -356,3 +356,8 @@ void do_show_syscall_exit(unsigned long { printk(" -> %lx, current=%p cpu=%d\n", r3, current, smp_processor_id()); } + +void do_clear_tif_restoreall(void) +{ + clear_thread_flag(TIF_RESTOREALL); +}
diff --git a/arch/powerpc/kernel/systbl.S b/arch/powerpc/kernel/systbl.S
index 65eaea9..4bb3650 100644
--- a/arch/powerpc/kernel/systbl.S
+++ b/arch/powerpc/kernel/systbl.S@@ -113,7 +113,7 @@ SYSCALL(sgetmask) COMPAT_SYS(ssetmask) SYSCALL(setreuid) SYSCALL(setregid) -SYSX(sys_ni_syscall,ppc32_sigsuspend,ppc_sigsuspend) +SYS32ONLY(sigsuspend) COMPAT_SYS(sigpending) COMPAT_SYS(sethostname) COMPAT_SYS(setrlimit)
@@ -160,7 +160,7 @@ SYSCALL(swapoff) COMPAT_SYS(sysinfo) COMPAT_SYS(ipc) SYSCALL(fsync) -SYSX(sys_ni_syscall,ppc32_sigreturn,sys_sigreturn) +SYS32ONLY(sigreturn) PPC_SYS(clone) COMPAT_SYS(setdomainname) PPC_SYS(newuname)
@@ -213,13 +213,13 @@ COMPAT_SYS(nfsservctl) SYSCALL(setresgid) SYSCALL(getresgid) COMPAT_SYS(prctl) -SYSX(ppc64_rt_sigreturn,ppc32_rt_sigreturn,sys_rt_sigreturn) +COMPAT_SYS(rt_sigreturn) COMPAT_SYS(rt_sigaction) COMPAT_SYS(rt_sigprocmask) COMPAT_SYS(rt_sigpending) COMPAT_SYS(rt_sigtimedwait) COMPAT_SYS(rt_sigqueueinfo) -SYSX(ppc64_rt_sigsuspend,ppc32_rt_sigsuspend,ppc_rt_sigsuspend) +COMPAT_SYS(rt_sigsuspend) COMPAT_SYS(pread64) COMPAT_SYS(pwrite64) SYSCALL(chown)
@@ -290,7 +290,7 @@ COMPAT_SYS(clock_settime) COMPAT_SYS(clock_gettime) COMPAT_SYS(clock_getres) COMPAT_SYS(clock_nanosleep) -SYSX(ppc64_swapcontext,ppc32_swapcontext,ppc_swapcontext) +COMPAT_SYS(swapcontext) COMPAT_SYS(tgkill) COMPAT_SYS(utimes) COMPAT_SYS(statfs64)
diff --git a/include/asm-powerpc/thread_info.h b/include/asm-powerpc/thread_info.h
index e525f49..d10ec17 100644
--- a/include/asm-powerpc/thread_info.h
+++ b/include/asm-powerpc/thread_info.h@@ -123,6 +123,7 @@ static inline struct thread_info *curren #define TIF_SINGLESTEP 9 /* singlestepping active */ #define TIF_MEMDIE 10 #define TIF_SECCOMP 11 /* secure computing */ +#define TIF_RESTOREALL 12 /* signal handler invoked */ /* as above, but as bit values */ #define _TIF_SYSCALL_TRACE (1<<TIF_SYSCALL_TRACE)
@@ -136,10 +137,11 @@ static inline struct thread_info *curren #define _TIF_SYSCALL_AUDIT (1<<TIF_SYSCALL_AUDIT) #define _TIF_SINGLESTEP (1<<TIF_SINGLESTEP) #define _TIF_SECCOMP (1<<TIF_SECCOMP) +#define _TIF_RESTOREALL (1<<TIF_RESTOREALL) #define _TIF_SYSCALL_T_OR_A (_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP) #define _TIF_USER_WORK_MASK (_TIF_NOTIFY_RESUME | _TIF_SIGPENDING | \ - _TIF_NEED_RESCHED) + _TIF_NEED_RESCHED | _TIF_RESTOREALL) #endif /* __KERNEL__ */
--
dwmw2