[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20220107170302.8325-1-mathieu.desnoyers@efficios.com>
Date: Fri, 7 Jan 2022 12:03:02 -0500
From: Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: linux-kernel@...r.kernel.org, Thomas Gleixner <tglx@...utronix.de>,
"Paul E . McKenney" <paulmck@...nel.org>,
Boqun Feng <boqun.feng@...il.com>,
"H . Peter Anvin" <hpa@...or.com>, Paul Turner <pjt@...gle.com>,
linux-api@...r.kernel.org,
Christian Brauner <christian.brauner@...ntu.com>,
Florian Weimer <fw@...eb.enyo.de>, carlos@...hat.com,
Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
Subject: [RFC PATCH] rseq: x86: implement abort-at-ip extension
Allow rseq critical section abort handlers to optionally figure out at
which instruction pointer the rseq critical section was aborted.
This allows implementing rseq critical sections containing loops, in
which case the commit side-effect cannot be the last instruction. This
is useful to implement adaptative mutexes aware of preemption in
user-space. (see [1])
This also allows implementing rseq critical sections with multiple
commit steps, and use the abort-at-ip information to figure out what
needs to be undone in the abort handler.
Introduce the RSEQ_FLAG_QUERY_ABORT_AT_IP rseq system call flag. This
lets userspace query whether the kernel and architecture supports the
abort-at-ip rseq extension.
Only provide this information for rseq critical sections for which the
rseq_cs descriptor has the RSEQ_CS_FLAG_ABORT_AT_IP flag set. Those
critical sections need to expect those ecx/rcx registers to be
clobbered on abort.
For x86-32, provide the abort-at-ip information in the %ecx register.
For x86-64, provide the abort-at-ip information in the %rcx register.
[1] https://github.com/compudj/rseq-test/blob/adapt-lock-abort-at-ip/test-rseq-adaptative-lock.c#L80
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
---
arch/x86/include/asm/ptrace.h | 12 ++++++++++++
include/uapi/linux/rseq.h | 4 ++++
kernel/rseq.c | 15 +++++++++++++++
3 files changed, 31 insertions(+)
diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index b94f615600d5..0a50e7f14c64 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -387,5 +387,17 @@ extern int do_set_thread_area(struct task_struct *p, int idx,
# define do_set_thread_area_64(p, s, t) (0)
#endif
+#ifdef CONFIG_RSEQ
+
+#define RSEQ_ARCH_HAS_ABORT_AT_IP
+
+/* Use ecx/rcx as placeholder for abort-at ip. */
+static inline void rseq_abort_at_ip(struct pt_regs *regs, unsigned long ip)
+{
+ regs->cx = ip;
+}
+
+#endif
+
#endif /* !__ASSEMBLY__ */
#endif /* _ASM_X86_PTRACE_H */
diff --git a/include/uapi/linux/rseq.h b/include/uapi/linux/rseq.h
index 9a402fdb60e9..3130232e6d0c 100644
--- a/include/uapi/linux/rseq.h
+++ b/include/uapi/linux/rseq.h
@@ -20,12 +20,14 @@ enum rseq_cpu_id_state {
enum rseq_flags {
RSEQ_FLAG_UNREGISTER = (1 << 0),
+ RSEQ_FLAG_QUERY_ABORT_AT_IP = (1 << 1),
};
enum rseq_cs_flags_bit {
RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT = 0,
RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT = 1,
RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT = 2,
+ RSEQ_CS_FLAG_ABORT_AT_IP_BIT = 3,
};
enum rseq_cs_flags {
@@ -35,6 +37,8 @@ enum rseq_cs_flags {
(1U << RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT),
RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE =
(1U << RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT),
+ RSEQ_CS_FLAG_ABORT_AT_IP =
+ (1U << RSEQ_CS_FLAG_ABORT_AT_IP_BIT),
};
/*
diff --git a/kernel/rseq.c b/kernel/rseq.c
index 6d45ac3dae7f..6612136412c8 100644
--- a/kernel/rseq.c
+++ b/kernel/rseq.c
@@ -21,6 +21,13 @@
#define RSEQ_CS_PREEMPT_MIGRATE_FLAGS (RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE | \
RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT)
+#ifdef RSEQ_ARCH_HAS_ABORT_AT_IP
+static bool rseq_has_abort_at_ip(void) { return true; }
+#else
+static bool rseq_has_abort_at_ip(void) { return false; }
+static void rseq_abort_at_ip(struct pt_regs *regs, unsigned long ip) { }
+#endif
+
/*
*
* Restartable sequences are a lightweight interface that allows
@@ -261,6 +268,8 @@ static int rseq_ip_fixup(struct pt_regs *regs)
trace_rseq_ip_fixup(ip, rseq_cs.start_ip, rseq_cs.post_commit_offset,
rseq_cs.abort_ip);
instruction_pointer_set(regs, (unsigned long)rseq_cs.abort_ip);
+ if (rseq_cs.flags & RSEQ_CS_FLAG_ABORT_AT_IP)
+ rseq_abort_at_ip(regs, ip);
return 0;
}
@@ -330,6 +339,12 @@ SYSCALL_DEFINE4(rseq, struct rseq __user *, rseq, u32, rseq_len,
{
int ret;
+ if (flags & RSEQ_FLAG_QUERY_ABORT_AT_IP) {
+ if (flags & ~RSEQ_FLAG_QUERY_ABORT_AT_IP)
+ return -EINVAL;
+ return rseq_has_abort_at_ip() ? 0 : -EINVAL;
+ }
+
if (flags & RSEQ_FLAG_UNREGISTER) {
if (flags & ~RSEQ_FLAG_UNREGISTER)
return -EINVAL;
--
2.17.1
Powered by blists - more mailing lists