s390/mm: Reimplement lazy ASCE handling

Reduce system call overhead time (round trip time for invoking a
non-existent system call) by 25%.

With the removal of set_fs() [1] lazy control register handling was removed
in order to keep kernel entry and exit simple. However this made system
calls slower.

With the conversion to generic entry [2] and numerous follow up changes
which simplified the entry code significantly, adding support for lazy asce
handling doesn't add much complexity to the entry code anymore.

In particular this means:

- On kernel entry the primary asce is not modified and contains the user
asce

- Kernel accesses which require secondary-space mode (for example futex
operations) are surrounded by enable_sacf_uaccess() and
disable_sacf_uaccess() calls. enable_sacf_uaccess() sets the primary asce
to kernel asce so that the sacf instruction can be used to switch to
secondary-space mode. The primary asce is changed back to user asce with
disable_sacf_uaccess().

The state of the control register which contains the primary asce is
reflected with a new TIF_ASCE_PRIMARY bit. This is required on context
switch so that the correct asce is restored for the scheduled in process.

In result address spaces are now setup like this:

CPU running in | %cr1 ASCE | %cr7 ASCE | %cr13 ASCE
-----------------------------|-----------|-----------|-----------
user space | user | user | kernel
kernel (no sacf) | user | user | kernel
kernel (during sacf uaccess) | kernel | user | kernel
kernel (kvm guest execution) | guest | user | kernel

In result cr1 control register content is not changed except for:
- futex system calls
- legacy s390 PCI system calls
- the kvm specific cmpxchg_user_key() uaccess helper

This leads to faster system call execution.

[1] 87d598634521 ("s390/mm: remove set_fs / rework address space handling")
[2] 56e62a737028 ("s390: convert to generic entry")

Reviewed-by: Alexander Gordeev <[e-mail]>
Signed-off-by: Heiko Carstens <[e-mail]>

arch/s390/include/asm/asce.h

···
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef _ASM_S390_ASCE_H
+#define _ASM_S390_ASCE_H
+
+#include <linux/thread_info.h>
+#include <linux/irqflags.h>
+#include <asm/lowcore.h>
+#include <asm/ctlreg.h>
+
+static inline bool enable_sacf_uaccess(void)
+{
+	unsigned long flags;
+
+	if (test_thread_flag(TIF_ASCE_PRIMARY))
+		return true;
+	local_irq_save(flags);
+	local_ctl_load(1, &get_lowcore()->kernel_asce);
+	set_thread_flag(TIF_ASCE_PRIMARY);
+	local_irq_restore(flags);
+	return false;
+}
+
+static inline void disable_sacf_uaccess(bool previous)
+{
+	unsigned long flags;
+
+	if (previous)
+		return;
+	local_irq_save(flags);
+	local_ctl_load(1, &get_lowcore()->user_asce);
+	clear_thread_flag(TIF_ASCE_PRIMARY);
+	local_irq_restore(flags);
+}
+
+#endif /* _ASM_S390_ASCE_H */

arch/s390/include/asm/futex.h

···
 static uaccess_kmsan_or_inline int						\
 __futex_atomic_##name(int oparg, int *old, u32 __user *uaddr)			\
 {										\
+	bool sacf_flag;								\
 	int rc, new;								\
 										\
 	instrument_copy_from_user_before(old, uaddr, sizeof(*old));		\
+	sacf_flag = enable_sacf_uaccess();					\
 	asm_inline volatile(							\
 		"	sacf	256\n"						\
 		"0:	l	%[old],%[uaddr]\n"				\
···
 		  [new] "=&d" (new), [uaddr] "+Q" (*uaddr)			\
 		: [oparg] "d" (oparg)						\
 		: "cc");							\
+	disable_sacf_uaccess(sacf_flag);					\
 	if (!rc)								\
 		instrument_copy_from_user_after(old, uaddr, sizeof(*old), 0);	\
 	return rc;								\
···
 static uaccess_kmsan_or_inline
 int futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr, u32 oldval, u32 newval)
 {
+	bool sacf_flag;
 	int rc;
 
 	instrument_copy_from_user_before(uval, uaddr, sizeof(*uval));
+	sacf_flag = enable_sacf_uaccess();
 	asm_inline volatile(
 		"	sacf	256\n"
 		"0:	cs	%[old],%[new],%[uaddr]\n"
···
 		: [rc] "=d" (rc), [old] "+d" (oldval), [uaddr] "+Q" (*uaddr)
 		: [new] "d" (newval)
 		: "cc", "memory");
+	disable_sacf_uaccess(sacf_flag);
 	*uval = oldval;
 	instrument_copy_from_user_after(uval, uaddr, sizeof(*uval), 0);
 	return rc;

arch/s390/include/asm/mmu_context.h

···
 #include <linux/mm_types.h>
 #include <asm/tlbflush.h>
 #include <asm/ctlreg.h>
+#include <asm/asce.h>
 #include <asm-generic/mm_hooks.h>
 
 #define init_new_context init_new_context
···
 	else
 		get_lowcore()->user_asce.val = next->context.asce;
 	cpumask_set_cpu(cpu, &next->context.cpu_attach_mask);
-	/* Clear previous user-ASCE from CR7 */
+	/* Clear previous user-ASCE from CR1 and CR7 */
+	local_ctl_load(1, &s390_invalid_asce);
 	local_ctl_load(7, &s390_invalid_asce);
 	if (prev != next)
 		cpumask_clear_cpu(cpu, &prev->context.cpu_attach_mask);
···
 {
 	struct task_struct *tsk = current;
 	struct mm_struct *mm = tsk->mm;
+	unsigned long flags;
 
 	if (mm) {
 		preempt_disable();
···
 		__tlb_flush_mm_lazy(mm);
 		preempt_enable();
 	}
+	local_irq_save(flags);
+	if (test_thread_flag(TIF_ASCE_PRIMARY))
+		local_ctl_load(1, &get_lowcore()->kernel_asce);
+	else
+		local_ctl_load(1, &get_lowcore()->user_asce);
 	local_ctl_load(7, &get_lowcore()->user_asce);
+	local_irq_restore(flags);
 }
 
 #define activate_mm activate_mm
 static inline void activate_mm(struct mm_struct *prev,
                                struct mm_struct *next)
 {
+	unsigned long flags;
+
 	switch_mm(prev, next, current);
 	cpumask_set_cpu(smp_processor_id(), mm_cpumask(next));
+	local_irq_save(flags);
+	if (test_thread_flag(TIF_ASCE_PRIMARY))
+		local_ctl_load(1, &get_lowcore()->kernel_asce);
+	else
+		local_ctl_load(1, &get_lowcore()->user_asce);
 	local_ctl_load(7, &get_lowcore()->user_asce);
+	local_irq_restore(flags);
 }
 
 #include <asm-generic/mmu_context.h>

arch/s390/include/asm/ptrace.h

···
 		struct tpi_info tpi_info;
 	};
 	unsigned long flags;
-	unsigned long cr1;
 	unsigned long last_break;
 };
 

arch/s390/include/asm/thread_info.h

···
 #define TIF_NEED_RESCHED_LAZY	3	/* lazy rescheduling needed */
 #define TIF_UPROBE		4	/* breakpointed or single-stepping */
 #define TIF_PATCH_PENDING	5	/* pending live patching update */
+#define TIF_ASCE_PRIMARY	6	/* primary asce is kernel asce */
 #define TIF_NOTIFY_SIGNAL	7	/* signal notifications exist */
 #define TIF_GUARDED_STORAGE	8	/* load guarded storage control block */
 #define TIF_ISOLATE_BP_GUEST	9	/* Run KVM guests with isolated BP */
···
 #define _TIF_NEED_RESCHED_LAZY	BIT(TIF_NEED_RESCHED_LAZY)
 #define _TIF_UPROBE		BIT(TIF_UPROBE)
 #define _TIF_PATCH_PENDING	BIT(TIF_PATCH_PENDING)
+#define _TIF_ASCE_PRIMARY	BIT(TIF_ASCE_PRIMARY)
 #define _TIF_NOTIFY_SIGNAL	BIT(TIF_NOTIFY_SIGNAL)
 #define _TIF_GUARDED_STORAGE	BIT(TIF_GUARDED_STORAGE)
 #define _TIF_ISOLATE_BP_GUEST	BIT(TIF_ISOLATE_BP_GUEST)

arch/s390/include/asm/uaccess.h

···
 #include <asm/extable.h>
 #include <asm/facility.h>
 #include <asm-generic/access_ok.h>
+#include <asm/asce.h>
 #include <linux/instrumented.h>
 
 void debug_user_asce(int exit);
···
 					      __uint128_t old, __uint128_t new,
 					      unsigned long key, int size)
 {
+	bool sacf_flag;
 	int rc = 0;
 
 	switch (size) {
···
 		_old = ((unsigned int)old & 0xff) << shift;
 		_new = ((unsigned int)new & 0xff) << shift;
 		mask = ~(0xff << shift);
+		sacf_flag = enable_sacf_uaccess();
 		asm_inline volatile(
 			"	spka	0(%[key])\n"
 			"	sacf	256\n"
···
 			  [default_key] "J" (PAGE_DEFAULT_KEY),
 			  [max_loops] "J" (CMPXCHG_USER_KEY_MAX_LOOPS)
 			: "memory", "cc");
+		disable_sacf_uaccess(sacf_flag);
 		*(unsigned char *)uval = prev >> shift;
 		if (!count)
 			rc = -EAGAIN;
···
 		_old = ((unsigned int)old & 0xffff) << shift;
 		_new = ((unsigned int)new & 0xffff) << shift;
 		mask = ~(0xffff << shift);
+		sacf_flag = enable_sacf_uaccess();
 		asm_inline volatile(
 			"	spka	0(%[key])\n"
 			"	sacf	256\n"
···
 			  [default_key] "J" (PAGE_DEFAULT_KEY),
 			  [max_loops] "J" (CMPXCHG_USER_KEY_MAX_LOOPS)
 			: "memory", "cc");
+		disable_sacf_uaccess(sacf_flag);
 		*(unsigned short *)uval = prev >> shift;
 		if (!count)
 			rc = -EAGAIN;
···
 	case 4:	{
 		unsigned int prev = old;
 
+		sacf_flag = enable_sacf_uaccess();
 		asm_inline volatile(
 			"	spka	0(%[key])\n"
 			"	sacf	256\n"
···
 			  [key] "a" (key << 4),
 			  [default_key] "J" (PAGE_DEFAULT_KEY)
 			: "memory", "cc");
+		disable_sacf_uaccess(sacf_flag);
 		*(unsigned int *)uval = prev;
 		return rc;
 	}
 	case 8: {
 		unsigned long prev = old;
 
+		sacf_flag = enable_sacf_uaccess();
 		asm_inline volatile(
 			"	spka	0(%[key])\n"
 			"	sacf	256\n"
···
 			  [key] "a" (key << 4),
 			  [default_key] "J" (PAGE_DEFAULT_KEY)
 			: "memory", "cc");
+		disable_sacf_uaccess(sacf_flag);
 		*(unsigned long *)uval = prev;
 		return rc;
 	}
 	case 16: {
 		__uint128_t prev = old;
 
+		sacf_flag = enable_sacf_uaccess();
 		asm_inline volatile(
 			"	spka	0(%[key])\n"
 			"	sacf	256\n"
···
 			  [key] "a" (key << 4),
 			  [default_key] "J" (PAGE_DEFAULT_KEY)
 			: "memory", "cc");
+		disable_sacf_uaccess(sacf_flag);
 		*(__uint128_t *)uval = prev;
 		return rc;
 	}

arch/s390/kernel/asm-offsets.c

···
 	OFFSET(__PT_ORIG_GPR2, pt_regs, orig_gpr2);
 	OFFSET(__PT_INT_CODE, pt_regs, int_code);
 	OFFSET(__PT_FLAGS, pt_regs, flags);
-	OFFSET(__PT_CR1, pt_regs, cr1);
 	OFFSET(__PT_LAST_BREAK, pt_regs, last_break);
 	DEFINE(__PT_SIZE, sizeof(struct pt_regs));
 	BLANK();

arch/s390/kernel/entry.S

···
 	.macro SIEEXIT sie_control,lowcore
 	lg	%r9,\sie_control			# get control block pointer
 	ni	__SIE_PROG0C+3(%r9),0xfe		# no longer in SIE
-	lctlg	%c1,%c1,__LC_KERNEL_ASCE(\lowcore)	# load primary asce
+	lctlg	%c1,%c1,__LC_USER_ASCE(\lowcore)	# load primary asce
 	lg	%r9,__LC_CURRENT(\lowcore)
 	mvi	__TI_sie(%r9),0
 	larl	%r9,sie_exit			# skip forward to sie_exit
···
 	lg	%r14,__SF_SIE_CONTROL(%r15)	# get control block pointer
 	ni	__SIE_PROG0C+3(%r14),0xfe	# no longer in SIE
 	GET_LC	%r14
-	lctlg	%c1,%c1,__LC_KERNEL_ASCE(%r14)	# load primary asce
+	lctlg	%c1,%c1,__LC_USER_ASCE(%r14)	# load primary asce
 	lg	%r14,__LC_CURRENT(%r14)
 	mvi	__TI_sie(%r14),0
 SYM_INNER_LABEL(sie_exit, SYM_L_GLOBAL)
···
 	lghi	%r14,0
 .Lsysc_per:
 	STBEAR	__LC_LAST_BREAK(%r13)
-	lctlg	%c1,%c1,__LC_KERNEL_ASCE(%r13)
 	lg	%r15,__LC_KERNEL_STACK(%r13)
 	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
 	stmg	%r0,%r7,STACK_FRAME_OVERHEAD+__PT_R0(%r15)
···
 	lgr	%r3,%r14
 	brasl	%r14,__do_syscall
 	STACKLEAK_ERASE
-	lctlg	%c1,%c1,__LC_USER_ASCE(%r13)
 	mvc	__LC_RETURN_PSW(16,%r13),STACK_FRAME_OVERHEAD+__PT_PSW(%r15)
 	BPON
 	LBEAR	STACK_FRAME_OVERHEAD+__PT_LAST_BREAK(%r15)
···
 	brasl	%r14,__ret_from_fork
 	STACKLEAK_ERASE
 	GET_LC	%r13
-	lctlg	%c1,%c1,__LC_USER_ASCE(%r13)
 	mvc	__LC_RETURN_PSW(16,%r13),STACK_FRAME_OVERHEAD+__PT_PSW(%r15)
 	BPON
 	LBEAR	STACK_FRAME_OVERHEAD+__PT_LAST_BREAK(%r15)
···
 	lmg	%r8,%r9,__LC_PGM_OLD_PSW(%r13)
 	xgr	%r10,%r10
 	tmhh	%r8,0x0001		# coming from user space?
-	jno	.Lpgm_skip_asce
-	lctlg	%c1,%c1,__LC_KERNEL_ASCE(%r13)
-	j	3f			# -> fault in user space
-.Lpgm_skip_asce:
+	jo	3f			# -> fault in user space
 #if IS_ENABLED(CONFIG_KVM)
 	lg	%r11,__LC_CURRENT(%r13)
 	tm	__TI_sie(%r11),0xff
···
 	tmhh	%r8,0x0001		# returning to user space?
 	jno	.Lpgm_exit_kernel
 	STACKLEAK_ERASE
-	lctlg	%c1,%c1,__LC_USER_ASCE(%r13)
 	BPON
 	stpt	__LC_EXIT_TIMER(%r13)
 .Lpgm_exit_kernel:
···
 #endif
 0:	aghi	%r15,-(STACK_FRAME_OVERHEAD + __PT_SIZE)
 	j	2f
-1:	lctlg	%c1,%c1,__LC_KERNEL_ASCE(%r13)
-	lg	%r15,__LC_KERNEL_STACK(%r13)
+1:	lg	%r15,__LC_KERNEL_STACK(%r13)
 2:	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
 	la	%r11,STACK_FRAME_OVERHEAD(%r15)
 	stmg	%r0,%r7,__PT_R0(%r11)
···
 	tmhh	%r8,0x0001		# returning to user ?
 	jno	2f
 	STACKLEAK_ERASE
-	lctlg	%c1,%c1,__LC_USER_ASCE(%r13)
 	BPON
 	stpt	__LC_EXIT_TIMER(%r13)
 2:	LBEAR	__PT_LAST_BREAK(%r11)
···
 .Lmcck_user:
 	lg	%r15,__LC_MCCK_STACK(%r13)
 	la	%r11,STACK_FRAME_OVERHEAD(%r15)
-	stctg	%c1,%c1,__PT_CR1(%r11)
-	lctlg	%c1,%c1,__LC_KERNEL_ASCE(%r13)
 	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
 	lay	%r14,__LC_GPREGS_SAVE_AREA(%r13)
 	mvc	__PT_R0(128,%r11),0(%r14)
···
 	xc	__SF_BACKCHAIN(8,%r15),__SF_BACKCHAIN(%r15)
 	lgr	%r2,%r11		# pass pointer to pt_regs
 	brasl	%r14,s390_do_machine_check
-	lctlg	%c1,%c1,__PT_CR1(%r11)
 	lmg	%r0,%r10,__PT_R0(%r11)
 	mvc	__LC_RETURN_MCCK_PSW(16,%r13),__PT_PSW(%r11) # move return PSW
 	tm	__LC_RETURN_MCCK_PSW+1(%r13),0x01 # returning to user ?

arch/s390/kernel/smp.c

···
 	abs_lc = get_abs_lowcore();
 	memcpy(lc->cregs_save_area, abs_lc->cregs_save_area, sizeof(lc->cregs_save_area));
 	put_abs_lowcore(abs_lc);
-	lc->cregs_save_area[1] = lc->kernel_asce;
+	lc->cregs_save_area[1] = lc->user_asce;
 	lc->cregs_save_area[7] = lc->user_asce;
 	save_access_regs((unsigned int *) lc->access_regs_save_area);
 	arch_spin_lock_setup(cpu);

arch/s390/lib/uaccess.c

···
 #ifdef CONFIG_DEBUG_ENTRY
 void debug_user_asce(int exit)
 {
+	struct lowcore *lc = get_lowcore();
 	struct ctlreg cr1, cr7;
 
 	local_ctl_store(1, &cr1);
 	local_ctl_store(7, &cr7);
-	if (cr1.val == get_lowcore()->kernel_asce.val && cr7.val == get_lowcore()->user_asce.val)
+	if (cr1.val == lc->user_asce.val && cr7.val == lc->user_asce.val)
 		return;
 	panic("incorrect ASCE on kernel %s\n"
 	      "cr1:    %016lx cr7:  %016lx\n"
 	      "kernel: %016lx user: %016lx\n",
 	      exit ? "exit" : "entry", cr1.val, cr7.val,
-	      get_lowcore()->kernel_asce.val, get_lowcore()->user_asce.val);
+	      lc->kernel_asce.val, lc->user_asce.val);
 }
 #endif /*CONFIG_DEBUG_ENTRY */
 

arch/s390/mm/pgalloc.c

···
 static void __crst_table_upgrade(void *arg)
 {
 	struct mm_struct *mm = arg;
+	struct ctlreg asce;
 
+	asce.val = mm->context.asce;
 	/* change all active ASCEs to avoid the creation of new TLBs */
 	if (current->active_mm == mm) {
-		get_lowcore()->user_asce.val = mm->context.asce;
-		local_ctl_load(7, &get_lowcore()->user_asce);
+		get_lowcore()->user_asce = asce;
+		local_ctl_load(7, &asce);
+		if (!test_thread_flag(TIF_ASCE_PRIMARY))
+			local_ctl_load(1, &asce);
 	}
 	__tlb_flush_local();
 }

arch/s390/pci/pci_mmio.c

···
 		u64 len, u8 *status)
 {
 	int cc, exception;
+	bool sacf_flag;
 
 	exception = 1;
+	sacf_flag = enable_sacf_uaccess();
 	asm_inline volatile (
 		"	sacf	256\n"
 		"0:	.insn	rsy,0xeb00000000d4,%[len],%[ioaddr],%[src]\n"
···
 		: CC_OUT(cc, cc), [len] "+d" (len), [exc] "+d" (exception)
 		: [ioaddr] "a" (ioaddr), [src] "Q" (*((u8 __force *)src))
 		: CC_CLOBBER_LIST("memory"));
+	disable_sacf_uaccess(sacf_flag);
 	*status = len >> 24 & 0xff;
 	return exception ? -ENXIO : CC_TRANSFORM(cc);
 }
···
 {
 	union register_pair ioaddr_len = {.even = (u64 __force)ioaddr, .odd = ulen};
 	int cc, exception;
+	bool sacf_flag;
 	u64 val = 0;
 	u64 cnt = ulen;
 	u8 tmp;
···
 	 * address space. pcistg then uses the user mappings.
 	 */
 	exception = 1;
+	sacf_flag = enable_sacf_uaccess();
 	asm_inline volatile (
 		"	sacf	256\n"
 		"0:	llgc	%[tmp],0(%[src])\n"
···
 		  CC_OUT(cc, cc), [ioaddr_len] "+&d" (ioaddr_len.pair)
 		:
 		: CC_CLOBBER_LIST("memory"));
+	disable_sacf_uaccess(sacf_flag);
 	*status = ioaddr_len.odd >> 24 & 0xff;
 
 	cc = exception ? -ENXIO : CC_TRANSFORM(cc);
···
 		u64 ulen, u8 *status)
 {
 	union register_pair ioaddr_len = {.even = (u64 __force)ioaddr, .odd = ulen};
+	bool sacf_flag;
 	u64 cnt = ulen;
 	int shift = ulen * 8;
 	int cc, exception;
···
 	 * user address @dst
 	 */
 	exception = 1;
+	sacf_flag = enable_sacf_uaccess();
 	asm_inline volatile (
 		"	sacf	256\n"
 		"0:	.insn	rre,0xb9d60000,%[val],%[ioaddr_len]\n"
···
 		  [shift] "+d" (shift)
 		:
 		: CC_CLOBBER_LIST("memory"));
-
+	disable_sacf_uaccess(sacf_flag);
 	cc = exception ? -ENXIO : CC_TRANSFORM(cc);
 	/* did we write everything to the user space buffer? */
 	if (!cc && cnt != 0)