rseq: Switch to fast path processing on exit to user

Now that all bits and pieces are in place, hook the RSEQ handling fast path function into exit_to_user_mode_prepare() after the TIF work bits have been handled. If case of fast path failure, TIF_NOTIFY_RESUME has been raised and the caller needs to take another turn through the TIF handling slow path. This only works for architectures which use the generic entry code. Architectures who still have their own incomplete hacks are not supported and won't be. This results in the following improvements: Kernel build Before After Reduction exit to user 80692981 80514451 signal checks: 32581 121 99% slowpath runs: 1201408 1.49% 198 0.00% 100% fastpath runs: 675941 0.84% N/A id updates: 1233989 1.53% 50541 0.06% 96% cs checks: 1125366 1.39% 0 0.00% 100% cs cleared: 1125366 100% 0 100% cs fixup: 0 0% 0 RSEQ selftests Before After Reduction exit to user: 386281778 387373750 signal checks: 35661203 0 100% slowpath runs: 140542396 36.38% 100 0.00% 100% fastpath runs: 9509789 2.51% N/A id updates: 176203599 45.62% 9087994 2.35% 95% cs checks: 175587856 45.46% 4728394 1.22% 98% cs cleared: 172359544 98.16% 1319307 27.90% 99% cs fixup: 3228312 1.84% 3409087 72.10% The 'cs cleared' and 'cs fixup' percentages are not relative to the exit to user invocations, they are relative to the actual 'cs check' invocations. While some of this could have been avoided in the original code, like the obvious clearing of CS when it's already clear, the main problem of going through TIF_NOTIFY_RESUME cannot be solved. In some workloads the RSEQ notify handler is invoked more than once before going out to user space. Doing this once when everything has stabilized is the only solution to avoid this. The initial attempt to completely decouple it from the TIF work turned out to be suboptimal for workloads, which do a lot of quick and short system calls. Even if the fast path decision is only 4 instructions (including a conditional branch), this adds up quickly and becomes measurable when the rate for actually having to handle rseq is in the low single digit percentage range of user/kernel transitions.

Thomas Gleixner committed 1 year ago

Commit 3db6b381 parent 05b44ae6 files · +41 −22Browse files

Unified · Split

6 files

include/linux/irq-entry-common.h +2 −5
include/linux/resume_user_mode.h +1 −1
include/linux/rseq.h +12 −6
init/Kconfig +1 −1
kernel/entry/common.c +19 −7
kernel/rseq.c +6 −2

include/linux/irq-entry-common.h+2−5

@@ -197,11 +197,8 @@ static __always_inline void arch_exit_to_user_mode(void) { } · 196 unmodified lines
197	197	*/
198	198	void arch_do_signal_or_restart(struct pt_regs *regs);
199	199
200		−/**
201		− * exit_to_user_mode_loop - do any pending work before leaving to user space
202		− */
203		−unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
204		− unsigned long ti_work);
	200	+/* Handle pending TIF work */
	201	+unsigned long exit_to_user_mode_loop(struct pt_regs *regs, unsigned long ti_work);
205	202
206	203	/**
207	204	* exit_to_user_mode_prepare - call exit_to_user_mode_loop() if required

include/linux/resume_user_mode.h+1−1

@@ -59,7 +59,7 @@ static inline void resume_user_mode_work(struct pt_regs *regs) · 58 unmodified lines
59	59	mem_cgroup_handle_over_high(GFP_KERNEL);
60	60	blkcg_maybe_throttle_current();
61	61
62		− rseq_handle_notify_resume(regs);
	62	+ rseq_handle_slowpath(regs);
63	63	}
64	64
65	65	#endif /* LINUX_RESUME_USER_MODE_H */

include/linux/rseq.h+12−6

@@ -7,13 +7,19 @@ · 6 unmodified lines
7	7
8	8	#include <uapi/linux/rseq.h>
9	9
10		−void __rseq_handle_notify_resume(struct pt_regs *regs);
	10	+void __rseq_handle_slowpath(struct pt_regs *regs);
11	11
12		−static inline void rseq_handle_notify_resume(struct pt_regs *regs)
	12	+/* Invoked from resume_user_mode_work() */
	13	+static inline void rseq_handle_slowpath(struct pt_regs *regs)
13	14	{
14		− /* '&' is intentional to spare one conditional branch */
15		− if (current->rseq.event.sched_switch & current->rseq.event.has_rseq)
16		− __rseq_handle_notify_resume(regs);
	15	+ if (IS_ENABLED(CONFIG_GENERIC_ENTRY)) {
	16	+ if (current->rseq.event.slowpath)
	17	+ __rseq_handle_slowpath(regs);
	18	+ } else {
	19	+ /* '&' is intentional to spare one conditional branch */
	20	+ if (current->rseq.event.sched_switch & current->rseq.event.has_rseq)
	21	+ __rseq_handle_slowpath(regs);
	22	+ }
17	23	}
18	24
19	25	void __rseq_signal_deliver(int sig, struct pt_regs *regs);
@@ -152,7 +158,7 @@ static inline void rseq_fork(struct task_struct *t, u64 clone_flags) · 132 unmodified lines
152	158	}
153	159
154	160	#else /* CONFIG_RSEQ */
155		−static inline void rseq_handle_notify_resume(struct pt_regs *regs) { }
	161	+static inline void rseq_handle_slowpath(struct pt_regs *regs) { }
156	162	static inline void rseq_signal_deliver(struct ksignal ksig, struct pt_regs regs) { }
157	163	static inline void rseq_sched_switch_event(struct task_struct *t) { }
158	164	static inline void rseq_sched_set_task_cpu(struct task_struct *t, unsigned int cpu) { }

init/Kconfig+1−1

@@ -1941,7 +1941,7 @@ config RSEQ_DEBUG_DEFAULT_ENABLE · 1940 unmodified lines
1941	1941	config DEBUG_RSEQ
1942	1942	default n
1943	1943	bool "Enable debugging of rseq() system call" if EXPERT
1944		− depends on RSEQ && DEBUG_KERNEL
	1944	+ depends on RSEQ && DEBUG_KERNEL && !GENERIC_ENTRY
1945	1945	select RSEQ_DEBUG_DEFAULT_ENABLE
1946	1946	help
1947	1947	Enable extra debugging checks for the rseq system call.

kernel/entry/common.c+19−7

@@ -11,13 +11,8 @@ · 10 unmodified lines
11	11	/* Workaround to allow gradual conversion of architecture code */
12	12	void __weak arch_do_signal_or_restart(struct pt_regs *regs) { }
13	13
14		−/**
15		− * exit_to_user_mode_loop - do any pending work before leaving to user space
16		− * @regs: Pointer to pt_regs on entry stack
17		− * @ti_work: TIF work flags as read by the caller
18		− */
19		−__always_inline unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
20		− unsigned long ti_work)
	14	+static __always_inline unsigned long __exit_to_user_mode_loop(struct pt_regs *regs,
	15	+ unsigned long ti_work)
21	16	{
22	17	/*
23	18	* Before returning to user space ensure that all pending work
@@ -62,6 +57,23 @@ __always_inline unsigned long exit_to_user_mode_loop(struct pt_regs *regs, · 38 unmodified lines
62	57	return ti_work;
63	58	}
64	59
	60	+/**
	61	+ * exit_to_user_mode_loop - do any pending work before leaving to user space
	62	+ * @regs: Pointer to pt_regs on entry stack
	63	+ * @ti_work: TIF work flags as read by the caller
	64	+ */
	65	+__always_inline unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
	66	+ unsigned long ti_work)
	67	+{
	68	+ for (;;) {
	69	+ ti_work = __exit_to_user_mode_loop(regs, ti_work);
	70	+
	71	+ if (likely(!rseq_exit_to_user_mode_restart(regs)))
	72	+ return ti_work;
	73	+ ti_work = read_thread_flags();
	74	+ }
	75	+}
	76	+
65	77	noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs)
66	78	{
67	79	irqentry_state_t ret = {

kernel/rseq.c+6−2

@@ -237,7 +237,11 @@ efault: · 236 unmodified lines
237	237
238	238	static void rseq_slowpath_update_usr(struct pt_regs *regs)
239	239	{
240		− /* Preserve rseq state and user_irq state for exit to user */
	240	+ /*
	241	+ * Preserve rseq state and user_irq state. The generic entry code
	242	+ * clears user_irq on the way out, the non-generic entry
	243	+ * architectures are not having user_irq.
	244	+ */
241	245	const struct rseq_event evt_mask = { .has_rseq = true, .user_irq = true, };
242	246	struct task_struct *t = current;
243	247	struct rseq_ids ids;
@@ -289,7 +293,7 @@ static void rseq_slowpath_update_usr(struct pt_regs *regs) · 45 unmodified lines
289	293	}
290	294	}
291	295
292		−void __rseq_handle_notify_resume(struct pt_regs *regs)
	296	+void __rseq_handle_slowpath(struct pt_regs *regs)
293	297	{
294	298	/*
295	299	* If invoked from hypervisors before entering the guest via