[S390] hibernate: Do real CPU swap at resume time

Currently, when the physical resume CPU is not equal to the physical suspend
CPU, we swap the CPUs logically, by modifying the logical/physical CPU mapping.
This has two major drawbacks: First the change is visible from user space (e.g.
CPU sysfs files) and second it is hard to ensure that nowhere in the kernel
the physical CPU ID is stored before suspend.
To fix this, we now really swap the physical CPUs, if the resume CPU is not
the pysical suspend CPU. We restart the suspend CPU and stop the resume CPU
using SIGP restart and SIGP stop. If the suspend CPU is no longer available,
we write a message and load a disabled wait PSW.
Signed-off-by: Michael Holzheu <michael.holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/lowcore.h

@@ -86,6 +86,7 @@
 #define __LC_PGM_OLD_PSW		0x0150
 #define __LC_MCK_OLD_PSW		0x0160
 #define __LC_IO_OLD_PSW			0x0170
+#define __LC_RESTART_PSW		0x01a0
 #define __LC_EXT_NEW_PSW		0x01b0
 #define __LC_SVC_NEW_PSW		0x01c0
 #define __LC_PGM_NEW_PSW		0x01d0

arch/s390/kernel/asm-offsets.c

@@ -7,6 +7,7 @@
 #include <linux/sched.h>
 #include <linux/kbuild.h>
 #include <asm/vdso.h>
+#include <asm/sigp.h>
 
 int main(void)
 {
@@ -59,6 +60,10 @@ int main(void)
 	DEFINE(CLOCK_REALTIME, CLOCK_REALTIME);
 	DEFINE(CLOCK_MONOTONIC, CLOCK_MONOTONIC);
 	DEFINE(CLOCK_REALTIME_RES, MONOTONIC_RES_NSEC);
-
+	/* constants for SIGP */
+	DEFINE(__SIGP_STOP, sigp_stop);
+	DEFINE(__SIGP_RESTART, sigp_restart);
+	DEFINE(__SIGP_SENSE, sigp_sense);
+	DEFINE(__SIGP_INITIAL_CPU_RESET, sigp_initial_cpu_reset);
 	return 0;
 }

arch/s390/kernel/sclp.S

@@ -24,8 +24,6 @@ LC_EXT_INT_CODE		= 0x86			# addr of ext int code
 #   R3	= external interruption parameter if R2=0
 #
 
-.section ".init.text","ax"
-
 _sclp_wait_int:
 	stm	%r6,%r15,24(%r15)		# save registers
 	basr	%r13,0				# get base register
@@ -318,9 +316,8 @@ _sclp_print_early:
 	.long	_sclp_work_area
 .Lascebc:
 	.long	_ascebc
-.previous
 
-.section ".init.data","a"
+.section .data,"aw",@progbits
 	.balign 4096
 _sclp_work_area:
 	.fill	4096

arch/s390/kernel/smp.c

@@ -1041,42 +1041,6 @@ static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
 			 dispatching_store);
 
-/*
- * If the resume kernel runs on another cpu than the suspended kernel,
- * we have to switch the cpu IDs in the logical map.
- */
-void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
-				   struct _lowcore *suspend_lowcore)
-{
-	int cpu, suspend_cpu_id, resume_cpu_id;
-	u32 suspend_phys_cpu_id;
-
-	suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
-	suspend_cpu_id = suspend_lowcore->cpu_nr;
-
-	for_each_present_cpu(cpu) {
-		if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
-			resume_cpu_id = cpu;
-			goto found;
-		}
-	}
-	panic("Could not find resume cpu in logical map.\n");
-
-found:
-	printk("Resume  cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
-	printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
-
-	__cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
-	__cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
-
-	lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
-}
-
-u32 smp_get_phys_cpu_id(void)
-{
-	return __cpu_logical_map[smp_processor_id()];
-}
-
 static int __init topology_init(void)
 {
 	int cpu;

arch/s390/kernel/swsusp_asm64.S

@@ -9,6 +9,7 @@
 
 #include <asm/page.h>
 #include <asm/ptrace.h>
+#include <asm/thread_info.h>
 #include <asm/asm-offsets.h>
 
 /*
@@ -41,6 +42,9 @@ swsusp_arch_suspend:
 	/* Get pointer to save area */
 	lghi	%r1,0x1000
 
+	/* Save CPU address */
+	stap	__LC_CPU_ADDRESS(%r1)
+
 	/* Store registers */
 	mvc	0x318(4,%r1),__SF_EMPTY(%r15)	/* move prefix to lowcore */
 	stfpc	0x31c(%r1)			/* store fpu control */
@@ -105,12 +109,7 @@ swsusp_arch_resume:
 	/* Make all free pages stable */
 	lghi	%r2,1
 	brasl	%r14,arch_set_page_states
-#ifdef CONFIG_SMP
-	/* Save boot cpu number */
-	brasl	%r14,smp_get_phys_cpu_id
-	larl	%r1,saved_cpu_id
-	st	%r2,0(%r1)
-#endif
+
 	/* Deactivate DAT */
 	stnsm	__SF_EMPTY(%r15),0xfb
 
@@ -139,12 +138,10 @@ swsusp_arch_resume:
 
 	/* Reset System */
 	larl	%r1,restart_entry
-	larl	%r2,restart_psw
+	larl	%r2,.Lrestart_diag308_psw
 	og	%r1,0(%r2)
 	stg	%r1,0(%r0)
-	larl	%r1,saved_pgm_check_psw
-	mvc	0(16,%r1),__LC_PGM_NEW_PSW(%r0)
-	larl	%r1,new_pgm_check_psw
+	larl	%r1,.Lnew_pgm_check_psw
 	epsw	%r2,%r3
 	stm	%r2,%r3,0(%r1)
 	mvc	__LC_PGM_NEW_PSW(16,%r0),0(%r1)
@@ -154,12 +151,54 @@ restart_entry:
 	lhi	%r1,1
 	sigp	%r1,%r0,0x12
 	sam64
-	larl	%r1,new_pgm_check_psw
+	larl	%r1,.Lnew_pgm_check_psw
 	lpswe	0(%r1)
 pgm_check_entry:
-	larl	%r1,saved_pgm_check_psw
-	mvc	__LC_PGM_NEW_PSW(16,%r0),0(%r1)
 
+	/* Switch to original suspend CPU */
+	larl	%r1,.Lresume_cpu		/* Resume CPU address: r2 */
+	stap	0(%r1)
+	llgh	%r2,0(%r1)
+	lghi	%r3,0x1000
+	llgh	%r1,__LC_CPU_ADDRESS(%r3)	/* Suspend CPU address: r1 */
+	cgr	%r1,%r2
+	je	restore_registers		/* r1 = r2 -> nothing to do */
+	larl	%r4,.Lrestart_suspend_psw	/* Set new restart PSW */
+	mvc	__LC_RESTART_PSW(16,%r0),0(%r4)
+3:
+	sigp	%r9,%r1,__SIGP_INITIAL_CPU_RESET
+	brc	8,4f	/* accepted */
+	brc	2,3b	/* busy, try again */
+
+	/* Suspend CPU not available -> panic */
+	larl	%r15,init_thread_union
+	ahi	%r15,1<<(PAGE_SHIFT+THREAD_ORDER)
+	larl	%r2,.Lpanic_string
+	larl	%r3,_sclp_print_early
+	lghi	%r1,0
+	sam31
+	sigp	%r1,%r0,0x12
+	basr	%r14,%r3
+	larl	%r3,.Ldisabled_wait_31
+	lpsw	0(%r3)
+4:
+	/* Switch to suspend CPU */
+	sigp	%r9,%r1,__SIGP_RESTART	/* start suspend CPU */
+	brc	2,4b			/* busy, try again */
+5:
+	sigp	%r9,%r2,__SIGP_STOP	/* stop resume (current) CPU */
+6:	j	6b
+
+restart_suspend:
+	larl	%r1,.Lresume_cpu
+	llgh	%r2,0(%r1)
+7:
+	sigp	%r9,%r2,__SIGP_SENSE	/* Wait for resume CPU */
+	brc	2,7b			/* busy, try again */
+	tmll	%r9,0x40		/* Test if resume CPU is stopped */
+	jz	7b
+
+restore_registers:
 	/* Restore registers */
 	lghi	%r13,0x1000		/* %r1 = pointer to save arae */
 
@@ -193,13 +232,6 @@ pgm_check_entry:
 	/* Pointer to save area */
 	lghi	%r13,0x1000
 
-#ifdef CONFIG_SMP
-	/* Switch CPUs */
-	larl	%r1,saved_cpu_id
-	llgf	%r2,0(%r1)
-	llgf	%r3,0x318(%r13)
-	brasl	%r14,smp_switch_boot_cpu_in_resume
-#endif
 	/* Restore prefix register */
 	spx	0x318(%r13)
 
@@ -217,13 +249,16 @@ pgm_check_entry:
 
 	.section .data.nosave,"aw",@progbits
 	.align	8
-restart_psw:
+.Ldisabled_wait_31:
+	.long  0x000a0000,0x00000000
+.Lpanic_string:
+	.asciz	"Resume not possible because suspend CPU is no longer available"
+	.align	8
+.Lrestart_diag308_psw:
 	.long	0x00080000,0x80000000
-new_pgm_check_psw:
+.Lrestart_suspend_psw:
+	.quad	0x0000000180000000,restart_suspend
+.Lnew_pgm_check_psw:
 	.quad	0,pgm_check_entry
-saved_pgm_check_psw:
-	.quad	0,0
-#ifdef CONFIG_SMP
-saved_cpu_id:
-	.long	0
-#endif
+.Lresume_cpu:
+	.byte	0,0