| Message ID | 20240913142239.2949727-2-adhemerval.zanella@linaro.org |
|---|---|
| State | Accepted |
| Commit | 2c1903cbbac0022153a67776f474c221250ad6ed |
| Headers | show |
| Series | [1/2] support: Make support_process_state_wait return the found state | expand |
Adhemerval Zanella <adhemerval.zanella@linaro.org> writes: > The commit 'sparc: Use Linux kABI for syscall return' > (86c5d2cf0ce046279baddc7faa27da71f1a89fde) did not take into consideration > a subtle sparc syscall kABI constraint. For syscalls that might block > indefinitely, on an interrupt (like SIGCONT) the kernel will set the > instruction pointer to just before the syscall: > > arch/sparc/kernel/signal_64.c > 476 static void do_signal(struct pt_regs *regs, unsigned long orig_i0) > 477 { > [...] > 525 if (restart_syscall) { > 526 switch (regs->u_regs[UREG_I0]) { > 527 case ERESTARTNOHAND: > 528 case ERESTARTSYS: > 529 case ERESTARTNOINTR: > 530 /* replay the system call when we are done */ > 531 regs->u_regs[UREG_I0] = orig_i0; > 532 regs->tpc -= 4; > 533 regs->tnpc -= 4; > 534 pt_regs_clear_syscall(regs); > 535 fallthrough; > 536 case ERESTART_RESTARTBLOCK: > 537 regs->u_regs[UREG_G1] = __NR_restart_syscall; > 538 regs->tpc -= 4; > 539 regs->tnpc -= 4; > 540 pt_regs_clear_syscall(regs); > 541 } > > However, on a SIGCONT it seems that 'g1' register is being clobbered after the > syscall returns. Before 86c5d2cf0ce046279, the 'g1' was always placed jus > before the 'ta' instruction which then reloads the syscall number and restarts > the syscall. > > On master, where 'g1' might be placed before 'ta': > > $ cat test.c > #include <unistd.h> > > int main () > { > pause (); > } > $ gcc test.c -o test > $ strace -f ./t > [...] > ppoll(NULL, 0, NULL, NULL, 0 > > On another terminal > > $ kill -STOP 2262828 > > $ strace -f ./t > [...] > --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- > --- stopped by SIGSTOP --- > > And then > > $ kill -CONT 2262828 > > Results in: > > --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- > restart_syscall(<... resuming interrupted ppoll ...>) = -1 EINTR (Interrupted system call) > > Where the expected behaviour would be: > > $ strace -f ./t > [...] > ppoll(NULL, 0, NULL, NULL, 0) = ? ERESTARTNOHAND (To be restarted if no handler) > --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- > --- stopped by SIGSTOP --- > --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- > ppoll(NULL, 0, NULL, NULL, 0 > > Just moving the 'g1' setting near the syscall asm is not suffice, > the compiler might optimize it away (as I saw on cancellation.c by > trying this fix). Instead, I have change the inline asm to put the > 'g1' setup in ithe asm block. This would require to change the asm > constraint for INTERNAL_SYSCALL_NCS, since the syscall number is not > constant. > > Checked on sparc64-linux-gnu. > > Reported-by: René Rebe <rene@exactcode.de> Tested-by: Sam James <sam@gentoo.org> and with much thanks! I don't feel qualified to review the patch here -- I'm happy to try push myself in some cases where it'd be useful (feel free to ask) but I don't think I can offer much quickly in this instance. Cheers. > [...] thanks, sam
On 01/10/24 19:07, Sam James wrote: > Adhemerval Zanella <adhemerval.zanella@linaro.org> writes: > >> The commit 'sparc: Use Linux kABI for syscall return' >> (86c5d2cf0ce046279baddc7faa27da71f1a89fde) did not take into consideration >> a subtle sparc syscall kABI constraint. For syscalls that might block >> indefinitely, on an interrupt (like SIGCONT) the kernel will set the >> instruction pointer to just before the syscall: >> >> arch/sparc/kernel/signal_64.c >> 476 static void do_signal(struct pt_regs *regs, unsigned long orig_i0) >> 477 { >> [...] >> 525 if (restart_syscall) { >> 526 switch (regs->u_regs[UREG_I0]) { >> 527 case ERESTARTNOHAND: >> 528 case ERESTARTSYS: >> 529 case ERESTARTNOINTR: >> 530 /* replay the system call when we are done */ >> 531 regs->u_regs[UREG_I0] = orig_i0; >> 532 regs->tpc -= 4; >> 533 regs->tnpc -= 4; >> 534 pt_regs_clear_syscall(regs); >> 535 fallthrough; >> 536 case ERESTART_RESTARTBLOCK: >> 537 regs->u_regs[UREG_G1] = __NR_restart_syscall; >> 538 regs->tpc -= 4; >> 539 regs->tnpc -= 4; >> 540 pt_regs_clear_syscall(regs); >> 541 } >> >> However, on a SIGCONT it seems that 'g1' register is being clobbered after the >> syscall returns. Before 86c5d2cf0ce046279, the 'g1' was always placed jus >> before the 'ta' instruction which then reloads the syscall number and restarts >> the syscall. >> >> On master, where 'g1' might be placed before 'ta': >> >> $ cat test.c >> #include <unistd.h> >> >> int main () >> { >> pause (); >> } >> $ gcc test.c -o test >> $ strace -f ./t >> [...] >> ppoll(NULL, 0, NULL, NULL, 0 >> >> On another terminal >> >> $ kill -STOP 2262828 >> >> $ strace -f ./t >> [...] >> --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >> --- stopped by SIGSTOP --- >> >> And then >> >> $ kill -CONT 2262828 >> >> Results in: >> >> --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >> restart_syscall(<... resuming interrupted ppoll ...>) = -1 EINTR (Interrupted system call) >> >> Where the expected behaviour would be: >> >> $ strace -f ./t >> [...] >> ppoll(NULL, 0, NULL, NULL, 0) = ? ERESTARTNOHAND (To be restarted if no handler) >> --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >> --- stopped by SIGSTOP --- >> --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >> ppoll(NULL, 0, NULL, NULL, 0 >> >> Just moving the 'g1' setting near the syscall asm is not suffice, >> the compiler might optimize it away (as I saw on cancellation.c by >> trying this fix). Instead, I have change the inline asm to put the >> 'g1' setup in ithe asm block. This would require to change the asm >> constraint for INTERNAL_SYSCALL_NCS, since the syscall number is not >> constant. >> >> Checked on sparc64-linux-gnu. >> >> Reported-by: René Rebe <rene@exactcode.de> > > Tested-by: Sam James <sam@gentoo.org> > > and with much thanks! I don't feel qualified to review the patch here -- > I'm happy to try push myself in some cases where it'd be useful (feel > free to ask) but I don't think I can offer much quickly in this > instance. Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since I think no one really will pay much attention to sparc). The testcase should be straightforward enough, but it would be good to have some review at least for it. It is sad that I haven't heard anything from Rene, since he was the one that complained about this issue and was really vocal about it.
Hi Adhemerval, On Wed, 2024-10-09 at 11:59 -0300, Adhemerval Zanella Netto wrote: > Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since > I think no one really will pay much attention to sparc). The testcase should be > straightforward enough, but it would be good to have some review at least for it. I agree. Let's get it merged. If there are any regressions, we will find and report them. But since Sam tested the patch already, I'm very confident that it works as expected and fixes the issue. > It is sad that I haven't heard anything from Rene, since he was the one that > complained about this issue and was really vocal about it. Yes, indeed. But since Sam tested it, I think it's safe to merge it. Adrian
Adhemerval Zanella Netto <adhemerval.zanella@linaro.org> writes: > On 01/10/24 19:07, Sam James wrote: >> Adhemerval Zanella <adhemerval.zanella@linaro.org> writes: >> >>> The commit 'sparc: Use Linux kABI for syscall return' >>> (86c5d2cf0ce046279baddc7faa27da71f1a89fde) did not take into consideration >>> a subtle sparc syscall kABI constraint. For syscalls that might block >>> indefinitely, on an interrupt (like SIGCONT) the kernel will set the >>> instruction pointer to just before the syscall: >>> >>> arch/sparc/kernel/signal_64.c >>> 476 static void do_signal(struct pt_regs *regs, unsigned long orig_i0) >>> 477 { >>> [...] >>> 525 if (restart_syscall) { >>> 526 switch (regs->u_regs[UREG_I0]) { >>> 527 case ERESTARTNOHAND: >>> 528 case ERESTARTSYS: >>> 529 case ERESTARTNOINTR: >>> 530 /* replay the system call when we are done */ >>> 531 regs->u_regs[UREG_I0] = orig_i0; >>> 532 regs->tpc -= 4; >>> 533 regs->tnpc -= 4; >>> 534 pt_regs_clear_syscall(regs); >>> 535 fallthrough; >>> 536 case ERESTART_RESTARTBLOCK: >>> 537 regs->u_regs[UREG_G1] = __NR_restart_syscall; >>> 538 regs->tpc -= 4; >>> 539 regs->tnpc -= 4; >>> 540 pt_regs_clear_syscall(regs); >>> 541 } >>> >>> However, on a SIGCONT it seems that 'g1' register is being clobbered after the >>> syscall returns. Before 86c5d2cf0ce046279, the 'g1' was always placed jus >>> before the 'ta' instruction which then reloads the syscall number and restarts >>> the syscall. >>> >>> On master, where 'g1' might be placed before 'ta': >>> >>> $ cat test.c >>> #include <unistd.h> >>> >>> int main () >>> { >>> pause (); >>> } >>> $ gcc test.c -o test >>> $ strace -f ./t >>> [...] >>> ppoll(NULL, 0, NULL, NULL, 0 >>> >>> On another terminal >>> >>> $ kill -STOP 2262828 >>> >>> $ strace -f ./t >>> [...] >>> --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >>> --- stopped by SIGSTOP --- >>> >>> And then >>> >>> $ kill -CONT 2262828 >>> >>> Results in: >>> >>> --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >>> restart_syscall(<... resuming interrupted ppoll ...>) = -1 EINTR (Interrupted system call) >>> >>> Where the expected behaviour would be: >>> >>> $ strace -f ./t >>> [...] >>> ppoll(NULL, 0, NULL, NULL, 0) = ? ERESTARTNOHAND (To be restarted if no handler) >>> --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >>> --- stopped by SIGSTOP --- >>> --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- >>> ppoll(NULL, 0, NULL, NULL, 0 >>> >>> Just moving the 'g1' setting near the syscall asm is not suffice, >>> the compiler might optimize it away (as I saw on cancellation.c by >>> trying this fix). Instead, I have change the inline asm to put the >>> 'g1' setup in ithe asm block. This would require to change the asm >>> constraint for INTERNAL_SYSCALL_NCS, since the syscall number is not >>> constant. >>> >>> Checked on sparc64-linux-gnu. >>> >>> Reported-by: René Rebe <rene@exactcode.de> >> >> Tested-by: Sam James <sam@gentoo.org> >> >> and with much thanks! I don't feel qualified to review the patch here -- >> I'm happy to try push myself in some cases where it'd be useful (feel >> free to ask) but I don't think I can offer much quickly in this >> instance. > > Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since > I think no one really will pay much attention to sparc). The testcase should be > straightforward enough, but it would be good to have some review at least for it. > The testcase is Reviewed-by: Sam James <sam@gentoo.org>. Thanks. > [...]
John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> writes: > Hi Adhemerval, > > On Wed, 2024-10-09 at 11:59 -0300, Adhemerval Zanella Netto wrote: >> Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since >> I think no one really will pay much attention to sparc). The testcase should be >> straightforward enough, but it would be good to have some review at least for it. > > I agree. Let's get it merged. If there are any regressions, we will find and report > them. But since Sam tested the patch already, I'm very confident that it works > as expected and fixes the issue. > >> It is sad that I haven't heard anything from Rene, since he was the one that >> complained about this issue and was really vocal about it. > > Yes, indeed. But since Sam tested it, I think it's safe to merge it. We can also pull it in early on our side if you give me a list of prereqs. IIRC I hackily applied some commits from master to get it to apply to 2.39. Or if you'd be so kind to put a branch somewhere based on 2.39, but not necessary (just some advice on what is best for 2.39 downstream would be great). > > Adrian
Sam James <sam@gentoo.org> writes: > John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> writes: > >> Hi Adhemerval, >> >> On Wed, 2024-10-09 at 11:59 -0300, Adhemerval Zanella Netto wrote: >>> Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since >>> I think no one really will pay much attention to sparc). The testcase should be >>> straightforward enough, but it would be good to have some review at least for it. >> >> I agree. Let's get it merged. If there are any regressions, we will find and report >> them. But since Sam tested the patch already, I'm very confident that it works >> as expected and fixes the issue. >> >>> It is sad that I haven't heard anything from Rene, since he was the one that >>> complained about this issue and was really vocal about it. >> >> Yes, indeed. But since Sam tested it, I think it's safe to merge it. > > We can also pull it in early on our side if you give me a list of > prereqs. IIRC I hackily applied some commits from master to get it to > apply to 2.39. Or if you'd be so kind to put a branch somewhere based on > 2.39, but not necessary (just some advice on what is best for 2.39 > downstream would be great). .. 2.40, I meant > >> >> Adrian
On 10/10/24 15:14, Sam James wrote: > John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> writes: > >> Hi Adhemerval, >> >> On Wed, 2024-10-09 at 11:59 -0300, Adhemerval Zanella Netto wrote: >>> Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since >>> I think no one really will pay much attention to sparc). The testcase should be >>> straightforward enough, but it would be good to have some review at least for it. >> >> I agree. Let's get it merged. If there are any regressions, we will find and report >> them. But since Sam tested the patch already, I'm very confident that it works >> as expected and fixes the issue. >> >>> It is sad that I haven't heard anything from Rene, since he was the one that >>> complained about this issue and was really vocal about it. >> >> Yes, indeed. But since Sam tested it, I think it's safe to merge it. > > We can also pull it in early on our side if you give me a list of > prereqs. IIRC I hackily applied some commits from master to get it to > apply to 2.39. Or if you'd be so kind to put a branch somewhere based on > 2.39, but not necessary (just some advice on what is best for 2.39 > downstream would be great). Ok, I will push this upstream and create a backport for 2.39/2.40.
On 16/10/24 14:32, Adhemerval Zanella Netto wrote: > > > On 10/10/24 15:14, Sam James wrote: >> John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> writes: >> >>> Hi Adhemerval, >>> >>> On Wed, 2024-10-09 at 11:59 -0300, Adhemerval Zanella Netto wrote: >>>> Thanks, I am re-testing it on a sparc64 box but I am inclined to install it (since >>>> I think no one really will pay much attention to sparc). The testcase should be >>>> straightforward enough, but it would be good to have some review at least for it. >>> >>> I agree. Let's get it merged. If there are any regressions, we will find and report >>> them. But since Sam tested the patch already, I'm very confident that it works >>> as expected and fixes the issue. >>> >>>> It is sad that I haven't heard anything from Rene, since he was the one that >>>> complained about this issue and was really vocal about it. >>> >>> Yes, indeed. But since Sam tested it, I think it's safe to merge it. >> >> We can also pull it in early on our side if you give me a list of >> prereqs. IIRC I hackily applied some commits from master to get it to >> apply to 2.39. Or if you'd be so kind to put a branch somewhere based on >> 2.39, but not necessary (just some advice on what is best for 2.39 >> downstream would be great). > > Ok, I will push this upstream and create a backport for 2.39/2.40. There here is: https://sourceware.org/git/?p=glibc.git;a=shortlog;h=refs/heads/azanella/2.39/master https://sourceware.org/git/?p=glibc.git;a=shortlog;h=refs/heads/azanella/2.40/master
diff --git a/sysdeps/unix/sysv/linux/Makefile b/sysdeps/unix/sysv/linux/Makefile index 7df51a325c..527c7a5ae8 100644 --- a/sysdeps/unix/sysv/linux/Makefile +++ b/sysdeps/unix/sysv/linux/Makefile @@ -230,6 +230,7 @@ tests += \ tst-scm_rights \ tst-sigtimedwait \ tst-sync_file_range \ + tst-syscall-restart \ tst-sysconf-iov_max \ tst-sysvmsg-linux \ tst-sysvsem-linux \ diff --git a/sysdeps/unix/sysv/linux/sparc/sparc32/syscall_cancel.S b/sysdeps/unix/sysv/linux/sparc/sparc32/syscall_cancel.S index aa5c658ce1..20f665d861 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc32/syscall_cancel.S +++ b/sysdeps/unix/sysv/linux/sparc/sparc32/syscall_cancel.S @@ -44,13 +44,13 @@ __syscall_cancel_arch_start: andcc %g2, TCB_CANCELED_BITMASK, %g0 bne,pn %icc, 2f /* Issue a 6 argument syscall. */ - mov %i1, %g1 - mov %i2, %o0 + mov %i2, %o0 mov %i3, %o1 mov %i4, %o2 mov %i5, %o3 ld [%fp+92], %o4 ld [%fp+96], %o5 + mov %i1, %g1 ta 0x10 .globl __syscall_cancel_arch_end diff --git a/sysdeps/unix/sysv/linux/sparc/sparc32/sysdep.h b/sysdeps/unix/sysv/linux/sparc/sparc32/sysdep.h index d2d68f5312..c2ffbb5c8f 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc32/sysdep.h +++ b/sysdeps/unix/sysv/linux/sparc/sparc32/sysdep.h @@ -107,6 +107,7 @@ ENTRY(name); \ #else /* __ASSEMBLER__ */ #define __SYSCALL_STRING \ + "mov %[scn], %%g1;" \ "ta 0x10;" \ "bcc 1f;" \ " nop;" \ @@ -114,7 +115,7 @@ ENTRY(name); \ "1:" #define __SYSCALL_CLOBBERS \ - "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ + "g1", "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", \ "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", \ "f24", "f25", "f26", "f27", "f28", "f29", "f30", "f31", \ diff --git a/sysdeps/unix/sysv/linux/sparc/sparc64/syscall_cancel.S b/sysdeps/unix/sysv/linux/sparc/sparc64/syscall_cancel.S index 21b0728d5a..6c8d1330cb 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc64/syscall_cancel.S +++ b/sysdeps/unix/sysv/linux/sparc/sparc64/syscall_cancel.S @@ -46,13 +46,13 @@ __syscall_cancel_arch_start: andcc %g2, TCB_CANCELED_BITMASK, %g0 bne,pn %xcc, 2f /* Issue a 6 argument syscall. */ - mov %i1, %g1 - mov %i2, %o0 + mov %i2, %o0 mov %i3, %o1 mov %i4, %o2 mov %i5, %o3 ldx [%fp + STACK_BIAS + 176], %o4 ldx [%fp + STACK_BIAS + 184], %o5 + mov %i1, %g1 ta 0x6d .global __syscall_cancel_arch_end diff --git a/sysdeps/unix/sysv/linux/sparc/sparc64/sysdep.h b/sysdeps/unix/sysv/linux/sparc/sparc64/sysdep.h index 96047424e9..5598fab08a 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc64/sysdep.h +++ b/sysdeps/unix/sysv/linux/sparc/sparc64/sysdep.h @@ -106,6 +106,7 @@ ENTRY(name); \ #else /* __ASSEMBLER__ */ #define __SYSCALL_STRING \ + "mov %[scn], %%g1;" \ "ta 0x6d;" \ "bcc,pt %%xcc, 1f;" \ " nop;" \ @@ -113,7 +114,7 @@ ENTRY(name); \ "1:" #define __SYSCALL_CLOBBERS \ - "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ + "g1", "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", \ "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", \ "f24", "f25", "f26", "f27", "f28", "f29", "f30", "f31", \ diff --git a/sysdeps/unix/sysv/linux/sparc/sysdep.h b/sysdeps/unix/sysv/linux/sparc/sysdep.h index dcabb57fe2..c287740a8c 100644 --- a/sysdeps/unix/sysv/linux/sparc/sysdep.h +++ b/sysdeps/unix/sysv/linux/sparc/sysdep.h @@ -50,97 +50,109 @@ #undef INTERNAL_SYSCALL_NCS #define INTERNAL_SYSCALL_NCS(name, nr, args...) \ - internal_syscall##nr(__SYSCALL_STRING, name, args) + _internal_syscall##nr(__SYSCALL_STRING, "p", name, args) -#define internal_syscall0(string,name,dummy...) \ +#define _internal_syscall0(string,nc,name,dummy...) \ ({ \ - register long int __g1 __asm__ ("g1") = (name); \ register long __o0 __asm__ ("o0"); \ + long int _name = (long int) (name); \ __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1) : \ + [scn] nc (_name) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall0(string,name,args...) \ + _internal_syscall0(string, "i", name, args) -#define internal_syscall1(string,name,arg1) \ +#define _internal_syscall1(string,nc,name,arg1) \ ({ \ long int _arg1 = (long int) (arg1); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0) : \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall1(string,name,args...) \ + _internal_syscall1(string, "i", name, args) -#define internal_syscall2(string,name,arg1,arg2) \ +#define _internal_syscall2(string,nc,name,arg1,arg2) \ ({ \ long int _arg1 = (long int) (arg1); \ long int _arg2 = (long int) (arg2); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0), "r" (__o1) : \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name), "r" (__o1) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall2(string,name,args...) \ + _internal_syscall2(string, "i", name, args) -#define internal_syscall3(string,name,arg1,arg2,arg3) \ +#define _internal_syscall3(string,nc,name,arg1,arg2,arg3) \ ({ \ long int _arg1 = (long int) (arg1); \ long int _arg2 = (long int) (arg2); \ long int _arg3 = (long int) (arg3); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ register long int __o2 __asm__ ("o2") = _arg3; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0), "r" (__o1), \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name), "r" (__o1), \ "r" (__o2) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall3(string,name,args...) \ + _internal_syscall3(string, "i", name, args) -#define internal_syscall4(string,name,arg1,arg2,arg3,arg4) \ +#define _internal_syscall4(string,nc,name,arg1,arg2,arg3,arg4) \ ({ \ long int _arg1 = (long int) (arg1); \ long int _arg2 = (long int) (arg2); \ long int _arg3 = (long int) (arg3); \ long int _arg4 = (long int) (arg4); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ register long int __o2 __asm__ ("o2") = _arg3; \ register long int __o3 __asm__ ("o3") = _arg4; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0), "r" (__o1), \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name), "r" (__o1), \ "r" (__o2), "r" (__o3) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall4(string,name,args...) \ + _internal_syscall4(string, "i", name, args) -#define internal_syscall5(string,name,arg1,arg2,arg3,arg4,arg5) \ +#define _internal_syscall5(string,nc,name,arg1,arg2,arg3,arg4,arg5) \ ({ \ long int _arg1 = (long int) (arg1); \ long int _arg2 = (long int) (arg2); \ long int _arg3 = (long int) (arg3); \ long int _arg4 = (long int) (arg4); \ long int _arg5 = (long int) (arg5); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ register long int __o2 __asm__ ("o2") = _arg3; \ register long int __o3 __asm__ ("o3") = _arg4; \ register long int __o4 __asm__ ("o4") = _arg5; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0), "r" (__o1), \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name), "r" (__o1), \ "r" (__o2), "r" (__o3), "r" (__o4) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall5(string,name,args...) \ + _internal_syscall5(string, "i", name, args) -#define internal_syscall6(string,name,arg1,arg2,arg3,arg4,arg5,arg6) \ +#define _internal_syscall6(string,nc,name,arg1,arg2,arg3,arg4,arg5,arg6)\ ({ \ long int _arg1 = (long int) (arg1); \ long int _arg2 = (long int) (arg2); \ @@ -148,20 +160,22 @@ long int _arg4 = (long int) (arg4); \ long int _arg5 = (long int) (arg5); \ long int _arg6 = (long int) (arg6); \ - register long int __g1 __asm__("g1") = (name); \ + long int _name = (long int) (name); \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ register long int __o2 __asm__ ("o2") = _arg3; \ register long int __o3 __asm__ ("o3") = _arg4; \ register long int __o4 __asm__ ("o4") = _arg5; \ register long int __o5 __asm__ ("o5") = _arg6; \ - __asm __volatile (string : "=r" (__o0) : \ - "r" (__g1), "0" (__o0), "r" (__o1), \ + __asm __volatile (string : "+r" (__o0) : \ + [scn] nc (_name), "r" (__o1), \ "r" (__o2), "r" (__o3), "r" (__o4), \ "r" (__o5) : \ __SYSCALL_CLOBBERS); \ __o0; \ }) +#define internal_syscall6(string,name,args...) \ + _internal_syscall6(string, "i", name, args) #define INLINE_CLONE_SYSCALL(arg1,arg2,arg3,arg4,arg5) \ ({ \ @@ -170,15 +184,15 @@ long int _arg3 = (long int) (arg3); \ long int _arg4 = (long int) (arg4); \ long int _arg5 = (long int) (arg5); \ + long int _name = __NR_clone; \ register long int __o0 __asm__ ("o0") = _arg1; \ register long int __o1 __asm__ ("o1") = _arg2; \ register long int __o2 __asm__ ("o2") = _arg3; \ register long int __o3 __asm__ ("o3") = _arg4; \ register long int __o4 __asm__ ("o4") = _arg5; \ - register long int __g1 __asm__ ("g1") = __NR_clone; \ __asm __volatile (__SYSCALL_STRING : \ "=r" (__o0), "=r" (__o1) : \ - "r" (__g1), "0" (__o0), "1" (__o1), \ + [scn] "i" (_name), "0" (__o0), "1" (__o1), \ "r" (__o2), "r" (__o3), "r" (__o4) : \ __SYSCALL_CLOBBERS); \ if (__glibc_unlikely ((unsigned long int) (__o0) > -4096UL)) \ diff --git a/sysdeps/unix/sysv/linux/tst-syscall-restart.c b/sysdeps/unix/sysv/linux/tst-syscall-restart.c new file mode 100644 index 0000000000..f18443abe3 --- /dev/null +++ b/sysdeps/unix/sysv/linux/tst-syscall-restart.c @@ -0,0 +1,112 @@ +/* Test if a syscall is correctly restarted. + Copyright (C) 2024 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <https://www.gnu.org/licenses/>. */ + +#include <support/xsignal.h> +#include <support/check.h> +#include <support/process_state.h> +#include <support/xunistd.h> +#include <support/xthread.h> +#include <sys/wait.h> + +static int +check_pid (pid_t pid) +{ + /* Wait until child has called pause and it blocking on kernel. */ + support_process_state_wait (pid, support_process_state_sleeping); + + TEST_COMPARE (kill (pid, SIGSTOP), 0); + + /* Adding process_state_tracing_stop ('t') allows the test to work under + trace programs such as ptrace. */ + support_process_state_wait (pid, support_process_state_stopped + | support_process_state_tracing_stop); + + TEST_COMPARE (kill (pid, SIGCONT), 0); + + enum support_process_state state + = support_process_state_wait (pid, support_process_state_sleeping + | support_process_state_zombie); + + TEST_COMPARE (state, support_process_state_sleeping); + + TEST_COMPARE (kill (pid, SIGTERM), 0); + + siginfo_t info; + TEST_COMPARE (waitid (P_PID, pid, &info, WEXITED), 0); + TEST_COMPARE (info.si_signo, SIGCHLD); + TEST_COMPARE (info.si_code, CLD_KILLED); + TEST_COMPARE (info.si_status, SIGTERM); + TEST_COMPARE (info.si_pid, pid); + + return 0; +} + +static void * +tf (void *) +{ + pause (); + return NULL; +} + +static void +child_mt (void) +{ + /* Let only the created thread to handle signals. */ + sigset_t set; + sigfillset (&set); + xpthread_sigmask (SIG_BLOCK, &set, NULL); + + sigdelset (&set, SIGSTOP); + sigdelset (&set, SIGCONT); + sigdelset (&set, SIGTERM); + + pthread_attr_t attr; + xpthread_attr_init (&attr); + TEST_COMPARE (pthread_attr_setsigmask_np (&attr, &set), 0); + + xpthread_join (xpthread_create (&attr, tf, NULL)); +} + +static void +do_test_syscall (bool multithread) +{ + pid_t pid = xfork (); + if (pid == 0) + { + if (multithread) + child_mt (); + else + pause (); + _exit (127); + } + + check_pid (pid); +} + +static int +do_test (void) +{ + /* Check for both single and multi thread, since they use different syscall + mechanisms. */ + do_test_syscall (false); + do_test_syscall (true); + + return 0; +} + +#include <support/test-driver.c>
The commit 'sparc: Use Linux kABI for syscall return' (86c5d2cf0ce046279baddc7faa27da71f1a89fde) did not take into consideration a subtle sparc syscall kABI constraint. For syscalls that might block indefinitely, on an interrupt (like SIGCONT) the kernel will set the instruction pointer to just before the syscall: arch/sparc/kernel/signal_64.c 476 static void do_signal(struct pt_regs *regs, unsigned long orig_i0) 477 { [...] 525 if (restart_syscall) { 526 switch (regs->u_regs[UREG_I0]) { 527 case ERESTARTNOHAND: 528 case ERESTARTSYS: 529 case ERESTARTNOINTR: 530 /* replay the system call when we are done */ 531 regs->u_regs[UREG_I0] = orig_i0; 532 regs->tpc -= 4; 533 regs->tnpc -= 4; 534 pt_regs_clear_syscall(regs); 535 fallthrough; 536 case ERESTART_RESTARTBLOCK: 537 regs->u_regs[UREG_G1] = __NR_restart_syscall; 538 regs->tpc -= 4; 539 regs->tnpc -= 4; 540 pt_regs_clear_syscall(regs); 541 } However, on a SIGCONT it seems that 'g1' register is being clobbered after the syscall returns. Before 86c5d2cf0ce046279, the 'g1' was always placed jus before the 'ta' instruction which then reloads the syscall number and restarts the syscall. On master, where 'g1' might be placed before 'ta': $ cat test.c #include <unistd.h> int main () { pause (); } $ gcc test.c -o test $ strace -f ./t [...] ppoll(NULL, 0, NULL, NULL, 0 On another terminal $ kill -STOP 2262828 $ strace -f ./t [...] --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- --- stopped by SIGSTOP --- And then $ kill -CONT 2262828 Results in: --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- restart_syscall(<... resuming interrupted ppoll ...>) = -1 EINTR (Interrupted system call) Where the expected behaviour would be: $ strace -f ./t [...] ppoll(NULL, 0, NULL, NULL, 0) = ? ERESTARTNOHAND (To be restarted if no handler) --- SIGSTOP {si_signo=SIGSTOP, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- --- stopped by SIGSTOP --- --- SIGCONT {si_signo=SIGCONT, si_code=SI_USER, si_pid=2521813, si_uid=8289} --- ppoll(NULL, 0, NULL, NULL, 0 Just moving the 'g1' setting near the syscall asm is not suffice, the compiler might optimize it away (as I saw on cancellation.c by trying this fix). Instead, I have change the inline asm to put the 'g1' setup in ithe asm block. This would require to change the asm constraint for INTERNAL_SYSCALL_NCS, since the syscall number is not constant. Checked on sparc64-linux-gnu. Reported-by: René Rebe <rene@exactcode.de> --- sysdeps/unix/sysv/linux/Makefile | 1 + .../sysv/linux/sparc/sparc32/syscall_cancel.S | 4 +- .../unix/sysv/linux/sparc/sparc32/sysdep.h | 3 +- .../sysv/linux/sparc/sparc64/syscall_cancel.S | 4 +- .../unix/sysv/linux/sparc/sparc64/sysdep.h | 3 +- sysdeps/unix/sysv/linux/sparc/sysdep.h | 74 +++++++----- sysdeps/unix/sysv/linux/tst-syscall-restart.c | 112 ++++++++++++++++++ 7 files changed, 165 insertions(+), 36 deletions(-) create mode 100644 sysdeps/unix/sysv/linux/tst-syscall-restart.c