| Message ID | 1583733592-22873-3-git-send-email-vincent.chen@sifive.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | riscv: add support for restartable sequence | expand |
----- On Mar 9, 2020, at 1:59 AM, Vincent Chen vincent.chen@sifive.com wrote: [...] > --- a/arch/riscv/kernel/signal.c > +++ b/arch/riscv/kernel/signal.c > @@ -234,6 +234,7 @@ static void handle_signal(struct ksignal *ksig, struct > pt_regs *regs) > sigset_t *oldset = sigmask_to_save(); > int ret; > > + rseq_signal_deliver(ksig, regs); > /* Are we from a system call? */ > if (regs->cause == EXC_SYSCALL) { [...] As Al Viro pointed out on IRC, the rseq_signal_deliver() should go after syscall restart handling, similarly to what is done on every other supported architecture. Note that there is already an upstream commit derived on this non-upstream patch: commit 9866d141a097 ("csky: Add support for restartable sequence") which is broken in the same way. I'm not sure why I was never CC'd on the csky patch. Considering that nobody bothered to implement the rseq selftests for csky, I don't see how any of this is tested. I would favor a revert of that commit until the testing glue is contributed. Unfortunately, the csky commit has been upstream since v5.7. Thanks, Mathieu
On Mon, Jul 19, 2021 at 10:43 PM Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote: > > ----- On Mar 9, 2020, at 1:59 AM, Vincent Chen vincent.chen@sifive.com wrote: > [...] > > --- a/arch/riscv/kernel/signal.c > > +++ b/arch/riscv/kernel/signal.c > > @@ -234,6 +234,7 @@ static void handle_signal(struct ksignal *ksig, struct > > pt_regs *regs) > > sigset_t *oldset = sigmask_to_save(); > > int ret; > > > > + rseq_signal_deliver(ksig, regs); > > /* Are we from a system call? */ > > if (regs->cause == EXC_SYSCALL) { > > [...] > > As Al Viro pointed out on IRC, the rseq_signal_deliver() should go after syscall > restart handling, similarly to what is done on every other supported architecture. Thanks for the notification. I will adjust the porting and try to send the patch again for review. > > Note that there is already an upstream commit derived on this non-upstream patch: > > commit 9866d141a097 ("csky: Add support for restartable sequence") > > which is broken in the same way. > > I'm not sure why I was never CC'd on the csky patch. Considering that nobody > bothered to implement the rseq selftests for csky, I don't see how any of > this is tested. I would favor a revert of that commit until the testing glue > is contributed. Unfortunately, the csky commit has been upstream since v5.7. > > Thanks, > > Mathieu > > > -- > Mathieu Desnoyers > EfficiOS Inc. > http://www.efficios.com
diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index fa98b279257e..ded32979d33d 100644 --- a/arch/riscv/Kconfig +++ b/arch/riscv/Kconfig @@ -67,6 +67,7 @@ config RISCV select HAVE_COPY_THREAD_TLS select HAVE_ARCH_KASAN if MMU && 64BIT select HAVE_REGS_AND_STACK_ACCESS_API + select HAVE_RSEQ config ARCH_MMAP_RND_BITS_MIN default 18 if 64BIT diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S index bad4d85b5e91..89d3713b0aef 100644 --- a/arch/riscv/kernel/entry.S +++ b/arch/riscv/kernel/entry.S @@ -214,6 +214,10 @@ ENTRY(handle_exception) handle_syscall: /* save the initial A0 value (needed in signal handlers) */ REG_S a0, PT_ORIG_A0(sp) +#ifdef CONFIG_RSEQ_DEBUG + move a0, sp + call rseq_syscall +#endif /* * Advance SEPC to avoid executing the original * scall instruction on sret diff --git a/arch/riscv/kernel/signal.c b/arch/riscv/kernel/signal.c index 17ba190e84a5..d939c5de41c4 100644 --- a/arch/riscv/kernel/signal.c +++ b/arch/riscv/kernel/signal.c @@ -234,6 +234,7 @@ static void handle_signal(struct ksignal *ksig, struct pt_regs *regs) sigset_t *oldset = sigmask_to_save(); int ret; + rseq_signal_deliver(ksig, regs); /* Are we from a system call? */ if (regs->cause == EXC_SYSCALL) { /* Avoid additional syscall restarting via ret_from_exception */ @@ -316,5 +317,6 @@ asmlinkage __visible void do_notify_resume(struct pt_regs *regs, if (thread_info_flags & _TIF_NOTIFY_RESUME) { clear_thread_flag(TIF_NOTIFY_RESUME); tracehook_notify_resume(regs); + rseq_handle_notify_resume(NULL, regs); } }
Add calls to rseq_signal_deliver(), rseq_handle_notify_resume() and rseq_syscall() to introduce RSEQ support. 1. Call the rseq_handle_notify_resume() function on return to userspace if TIF_NOTIFY_RESUME thread flag is set. 2. Call the rseq_signal_deliver() function to fixup on the pre-signal frame when a signal is delivered on top of a restartable sequence critical section. 3. Check that system calls are not invoked from within rseq critical sections by invoking rseq_signal() from ret_from_syscall(). With CONFIG_DEBUG_RSEQ, such behavior results in termination of the process with SIGSEGV. Signed-off-by: Vincent Chen <vincent.chen@sifive.com> --- arch/riscv/Kconfig | 1 + arch/riscv/kernel/entry.S | 4 ++++ arch/riscv/kernel/signal.c | 2 ++ 3 files changed, 7 insertions(+)