sw64: activate CONFIG_THREAD_INFO_IN_TASK

Sunway inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I6ILGG

--------------------------------

This patch activates CONFIG_THREAD_INFO_IN_TASK which moves the
thread_info into task_struct.

This has the following consequences:
  - thread_info is now located at the beginning of task_struct.
  - thread_info doesn't have anymore the `task` field.

This patch:
  - Adds hmcall wrktp and rdktp to save/restore `current` task
    pointer.
  - Changes the current_thread_info() macro to point to `current`.
  - Selects CONFIG_THREAD_INFO_IN_TASK.
  - Modifies raw_smp_processor_id() to get ->cpu from `current`
    without including linux/sched.h to avoid circular inclusion and
    without including asm/asm-offsets.h to avoid symbol names
    duplication between ASM constants and C constants.
  - Modifies task_pt_regs() macro to avoid error which says that
    task_stack_page() is undefined. This error can not be fixed by
    including linux/sched.h which results in circular inclusion.
Signed-off-by: NHe Sheng <hesheng@wxiat.com>
Reviewed-by: NCui Wei <cuiwei@wxiat.com>
Signed-off-by: NGu Zitao <guzitao@wxiat.com>

arch/sw_64/Kconfig

@@ -107,6 +107,7 @@ config SW64
 	select SET_FS
 	select SPARSEMEM_EXTREME if SPARSEMEM
 	select SWIOTLB
+	select THREAD_INFO_IN_TASK
 
 config LOCKDEP_SUPPORT
 	def_bool y

arch/sw_64/include/asm/current.h

@@ -2,9 +2,18 @@
 #ifndef _ASM_SW64_CURRENT_H
 #define _ASM_SW64_CURRENT_H
 
-#include <linux/thread_info.h>
+#ifndef __ASSEMBLY__
+
+struct task_struct;
+static __always_inline struct task_struct *get_current(void)
+{
+	register struct task_struct *tp __asm__("$8");
+
+	return tp;
+}
 
-#define get_current()	(current_thread_info()->task)
 #define current get_current()
 
+#endif /* __ASSEMBLY__ */
+
 #endif /* _ASM_SW64_CURRENT_H */

arch/sw_64/include/asm/hmcall.h

@@ -13,6 +13,8 @@
 #define HMC_sleepen		0x05
 #define HMC_rdksp		0x06
 #define HMC_wrasid		0x08
+#define HMC_rdktp		0x09
+#define HMC_wrktp		0x0A
 #define HMC_rdptbr		0x0B
 #define HMC_wrptbr		0x0C
 #define HMC_wrksp		0x0E
@@ -150,6 +152,11 @@ __CALL_HMC_VOID(wrfen);
 __CALL_HMC_VOID(sleepen);
 __CALL_HMC_VOID(mtinten);
 
+__CALL_HMC_VOID(rdktp);
+#define restore_ktp()	rdktp()
+__CALL_HMC_VOID(wrktp);
+#define save_ktp()	wrktp()
+
 __CALL_HMC_R0(rdps, unsigned long);
 
 __CALL_HMC_R0(rdusp, unsigned long);

arch/sw_64/include/asm/processor.h

@@ -12,7 +12,7 @@
 #include <asm/ptrace.h>
 
 #define task_pt_regs(task) \
-	((struct pt_regs *) (task_stack_page(task) + 2 * PAGE_SIZE) - 1)
+	((struct pt_regs *) (task->stack + THREAD_SIZE) - 1)
 
 /*
  * Returns current instruction pointer ("program counter").

arch/sw_64/include/asm/ptrace.h

@@ -4,7 +4,6 @@
 
 #include <uapi/asm/ptrace.h>
 #include <asm/hmcall.h>
-#include <asm/thread_info.h>
 #include <asm/page.h>
 
 /*
@@ -58,11 +57,6 @@ struct pt_regs {
 #define kernel_stack_pointer(regs) ((unsigned long)((regs) + 1))
 #define instruction_pointer_set(regs, val) ((regs)->pc = val)
 
-
-#define current_pt_regs() \
-	((struct pt_regs *) ((char *)current_thread_info() + 2 * PAGE_SIZE) - 1)
-#define signal_pt_regs current_pt_regs
-
 #define force_successful_syscall_return() (current_pt_regs()->r0 = 0)
 
 #define MAX_REG_OFFSET (offsetof(struct pt_regs, r18))

arch/sw_64/include/asm/smp.h

@@ -2,14 +2,16 @@
 #ifndef _ASM_SW64_SMP_H
 #define _ASM_SW64_SMP_H
 
-#include <linux/threads.h>
-#include <linux/cpumask.h>
-#include <linux/bitops.h>
-#include <asm/hmcall.h>
-#include <asm/hcall.h>
 #include <asm/core.h>
+#include <asm/current.h>
+#include <asm/hcall.h>
+#include <asm/hmcall.h>
 #include <asm/hw_init.h>
 
+#include <linux/bitops.h>
+#include <linux/cpumask.h>
+#include <linux/threads.h>
+
 /* HACK: Cabrio WHAMI return value is bogus if more than 8 bits used.. :-( */
 
 extern cpumask_t core_start;
@@ -55,7 +57,13 @@ struct smp_rcb_struct {
 #define INIT_SMP_RCB ((struct smp_rcb_struct *) __va(0x820000UL))
 
 #define hard_smp_processor_id()	__hard_smp_processor_id()
-#define raw_smp_processor_id()	(current_thread_info()->cpu)
+
+#ifdef GENERATING_ASM_OFFSETS
+#define raw_smp_processor_id() (0)
+#else
+#include <asm/asm-offsets.h>
+#define raw_smp_processor_id() (*((unsigned int *)((void *)current + TASK_CPU)))
+#endif
 
 /* The map from sequential logical cpu number to hard cid.  */
 extern int __cpu_to_rcid[NR_CPUS];

arch/sw_64/include/asm/thread_info.h

@@ -5,7 +5,6 @@
 #ifdef __KERNEL__
 
 #ifndef __ASSEMBLY__
-#include <asm/processor.h>
 #include <asm/types.h>
 #include <asm/sysinfo.h>
 
@@ -25,7 +24,6 @@ struct pcb_struct {
 struct thread_info {
 	struct pcb_struct	pcb;		/* hmcode state */
 
-	struct task_struct	*task;		/* main task structure */
 	unsigned int		flags;		/* low level flags */
 	unsigned int		ieee_state;	/* see fpu.h */
 
@@ -58,14 +56,10 @@ static __always_inline u64 rtid(void)
  */
 #define INIT_THREAD_INFO(tsk)				\
 {							\
-	.task		= &tsk,				\
 	.addr_limit	= KERNEL_DS,			\
 	.preempt_count	= INIT_PREEMPT_COUNT,		\
 }
 
-/* How to get the thread information struct from C.  */
-register struct thread_info *__current_thread_info __asm__("$8");
-#define current_thread_info()	 __current_thread_info
 
 #endif /* __ASSEMBLY__ */
 

arch/sw_64/kernel/asm-offsets.c

@@ -5,6 +5,7 @@
  * and format the required data.
  */
 
+#define GENERATING_ASM_OFFSETS  /* asm/smp.h */
 #include <linux/stddef.h>
 #include <linux/sched.h>
 #include <linux/kbuild.h>
@@ -15,11 +16,11 @@
 
 #include "traps.c"
 
+
 void foo(void)
 {
-	DEFINE(TI_TASK, offsetof(struct thread_info, task));
+	DEFINE(ASM_THREAD_SIZE, THREAD_SIZE);
 	DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
-	DEFINE(TI_CPU, offsetof(struct thread_info, cpu));
 	BLANK();
 
 	DEFINE(TASK_BLOCKED, offsetof(struct task_struct, blocked));
@@ -27,6 +28,10 @@ void foo(void)
 	DEFINE(TASK_REAL_PARENT, offsetof(struct task_struct, real_parent));
 	DEFINE(TASK_GROUP_LEADER, offsetof(struct task_struct, group_leader));
 	DEFINE(TASK_TGID, offsetof(struct task_struct, tgid));
+	DEFINE(TASK_STACK, offsetof(struct task_struct, stack));
+#ifdef CONFIG_SMP
+	DEFINE(TASK_CPU, offsetof(struct task_struct, thread_info.cpu));
+#endif
 	BLANK();
 
 	OFFSET(PSTATE_REGS, processor_state, regs);

arch/sw_64/kernel/early_init.c

@@ -24,6 +24,7 @@ static void __init sw64_setup_platform_ops(void)
 asmlinkage __visible void __init sw64_start_kernel(void)
 {
 	fixup_hmcall();
+	save_ktp();
 	sw64_setup_chip_ops();
 	sw64_setup_platform_ops();
 	sw64_platform->ops_fixup();

arch/sw_64/kernel/entry.S

@@ -48,6 +48,7 @@
 	stl	$25, PT_REGS_R25($sp)
 	stl	$26, PT_REGS_R26($sp)
 	stl	$27, PT_REGS_R27($sp)
+	sys_call HMC_rdktp
 	.endm
 
 	.macro RESTORE_ALL
@@ -89,9 +90,7 @@
 	.ent entInt
 entInt:
 	SAVE_ALL
-	ldi	$8, 0x3fff
 	ldi	$26, ret_from_sys_call
-	bic	$sp, $8, $8
 	mov	$sp, $19
 	call	$31, do_entInt
 	.end entInt
@@ -101,9 +100,7 @@ entInt:
 	.ent entArith
 entArith:
 	SAVE_ALL
-	ldi	$8, 0x3fff
 	ldi	$26, ret_from_sys_call
-	bic	$sp, $8, $8
 	mov	$sp, $18
 	call	$31, do_entArith
 	.end entArith
@@ -113,9 +110,7 @@ entArith:
 	.ent entMM
 entMM:
 	SAVE_ALL
-	ldi	$8, 0x3fff
 	ldi	$26, ret_from_sys_call
-	bic	$sp, $8, $8
 	mov	$sp, $19
 	call	$31, do_page_fault
 	.end entMM
@@ -125,9 +120,7 @@ entMM:
 	.ent entIF
 entIF:
 	SAVE_ALL
-	ldi	$8, 0x3fff
 	ldi	$26, ret_from_sys_call
-	bic	$sp, $8, $8
 	mov	$sp, $17
 	call	$31, do_entIF
 	.end entIF
@@ -143,8 +136,6 @@ entIF:
 	.ent entUna
 entUna:
 	SAVE_ALL
-	ldi	$8, 0x3fff
-	bic	$sp, $8, $8
 	mov	$sp, $19
 	ldl	$0, PT_REGS_PS($sp)
 	and	$0, 8, $0		/* user mode ? */
@@ -176,8 +167,6 @@ entUna:
 entSys:
 
 	SAVE_ALL
-	ldi	$8, 0x3fff
-	bic	$sp, $8, $8
 	ldi	$4, NR_SYSCALLS($31)
 	stl	$16, PT_REGS_R16($sp)
 	ldi	$5, sys_call_table
@@ -395,8 +384,8 @@ __switch_to:
 	ldl	$13, TASK_THREAD_S4($17)
 	ldl	$14, TASK_THREAD_S5($17)
 	ldl	$15, TASK_THREAD_S6($17)
-	ldi	$8, 0x3fff
-	bic	$sp, $8, $8
+	mov	$17, $8
+	sys_call HMC_wrktp
 	mov	$16, $0
 	ret
 	.end __switch_to

arch/sw_64/kernel/head.S

@@ -22,9 +22,11 @@ __start:
 	br	$27, 1f
 1:	ldgp	$29, 0($27)
 	/* We need to get current_task_info loaded up...  */
-	ldi	$8, init_thread_union
+	ldi	$8, init_task
+	ldl	$30, TASK_STACK($8)
 	/* ... and find our stack ... */
-	ldi	$30, ASM_THREAD_SIZE($8)
+	ldi	$30, ASM_THREAD_SIZE($30)
+
 	/* ... and then we can clear bss data.  */
 	ldi	$16, __bss_start
 	ldi	$18, __bss_stop
@@ -68,11 +70,13 @@ __smp_callin:
 	s4addl	$0, $1, $1
 	ldw	$0, 0($1)	# Get logical cpu number
 
-	ldi	$2, tidle_ksp
+	ldi	$2, idle_task_pointer
 	s8addl	$0, $2, $2
-	ldl	$30, 0($2)	# Get ksp of idle thread
+	ldl	$8, 0($2)	# Get ksp of idle thread
+	sys_call HMC_wrktp
 
-	ldi	$8, -ASM_THREAD_SIZE($30) # Find "current"
+	ldl	$30, TASK_STACK($8)
+	ldi	$30, ASM_THREAD_SIZE($30)
 
 	call	$26, smp_callin
 	sys_call HMC_halt

arch/sw_64/kernel/hmcall.c

@@ -76,6 +76,22 @@ static inline void fixup_wrasid(void)
 	entry[9] = 0x1ef00000;	/* pri_ret/b p23 */
 }
 
+static inline void fixup_rdktp(void)
+{
+	unsigned int *entry = __va(HMCALL_ENTRY(rdktp));
+
+	entry[0] = 0x95161000;	/* pri_ldl/p $8, VC__KTP(vcpucb) */
+	entry[1] = 0x1ee00000;	/* pri_ret $23 */
+}
+
+static inline void fixup_wrktp(void)
+{
+	unsigned int *entry = __va(HMCALL_ENTRY(wrktp));
+
+	entry[0] = 0xb5161000;	/* pri_stl/p $8, VC__KTP(vcpucb) */
+	entry[1] = 0x1ee00000;	/* pri_ret $23 */
+}
+
 void __init fixup_hmcall(void)
 {
 #if defined(CONFIG_SUBARCH_C3B)
@@ -83,6 +99,8 @@ void __init fixup_hmcall(void)
 	fixup_wrtp();
 	fixup_tbiasid();
 	fixup_wrasid();
+	fixup_rdktp();
+	fixup_wrktp();
 #endif
 }
 

arch/sw_64/kernel/smp.c

@@ -34,7 +34,7 @@ EXPORT_SYMBOL(__cpu_to_rcid);
 int __rcid_to_cpu[NR_CPUS];		/* Map physical to logical */
 EXPORT_SYMBOL(__rcid_to_cpu);
 
-void *tidle_ksp[NR_CPUS];
+void *idle_task_pointer[NR_CPUS];
 
 /* State of each CPU */
 DEFINE_PER_CPU(int, cpu_state) = { 0 };
@@ -130,7 +130,7 @@ static int secondary_cpu_start(int cpuid, struct task_struct *idle)
 	/*
 	 * Precalculate the target ksp.
 	 */
-	tidle_ksp[cpuid] = idle->stack + THREAD_SIZE;
+	idle_task_pointer[cpuid] = idle;
 
 	DBGS("Starting secondary cpu %d: state 0x%lx\n", cpuid, idle->state);
 
@@ -237,7 +237,6 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
 	memset(ipi_data, 0, sizeof(ipi_data));
 
 	init_cpu_topology();
-	current_thread_info()->cpu = 0;
 	store_cpu_topology(smp_processor_id());
 	numa_add_cpu(smp_processor_id());
 

arch/sw_64/kvm/entry.S

@@ -18,22 +18,22 @@
  */
 ENTRY(__sw64_vcpu_run)
 	/* save host fpregs */
-	ldl	$1, TI_TASK($8)
 	rfpcr	$f0
-	fstd	$f0, TASK_THREAD_FPCR($1)
-	vstd	$f2, TASK_THREAD_F2($1)
-	vstd	$f3, TASK_THREAD_F3($1)
-	vstd	$f4, TASK_THREAD_F4($1)
-	vstd	$f5, TASK_THREAD_F5($1)
-	vstd	$f6, TASK_THREAD_F6($1)
-	vstd	$f7, TASK_THREAD_F7($1)
-	vstd	$f8, TASK_THREAD_F8($1)
-	vstd	$f9, TASK_THREAD_F9($1)
+	fstd	$f0, TASK_THREAD_FPCR($8)
+	vstd	$f2, TASK_THREAD_F2($8)
+	vstd	$f3, TASK_THREAD_F3($8)
+	vstd	$f4, TASK_THREAD_F4($8)
+	vstd	$f5, TASK_THREAD_F5($8)
+	vstd	$f6, TASK_THREAD_F6($8)
+	vstd	$f7, TASK_THREAD_F7($8)
+	vstd	$f8, TASK_THREAD_F8($8)
+	vstd	$f9, TASK_THREAD_F9($8)
 
 	ldi	sp, -VCPU_RET_SIZE(sp)
 	/* save host  pt_regs to current kernel  stack */
 	ldi	sp, -PT_REGS_SIZE(sp)
 	stl	$9, PT_REGS_R9(sp)
+	stl	$8, PT_REGS_R8(sp)
 	stl	$10, PT_REGS_R10(sp)
 	stl	$11, PT_REGS_R11(sp)
 	stl	$12, PT_REGS_R12(sp)
@@ -198,6 +198,7 @@ $g_setfpec_over:
 	stl	$28, KVM_REGS_R28($17)
 
 	/* restore host regs from host sp */
+	ldl	$8, PT_REGS_R8(sp)
 	ldl	$9, PT_REGS_R9(sp)
 	ldl	$10, PT_REGS_R10(sp)
 	ldl	$11, PT_REGS_R11(sp)
@@ -208,11 +209,8 @@ $g_setfpec_over:
 	ldl	$26, PT_REGS_R26(sp)
 	ldi	sp, PT_REGS_SIZE(sp)
 
-	ldi	$8, 0x3fff
-	bic	sp, $8, $8
 	/* restore host fpregs */
-	ldl	$1, TI_TASK($8)
-	fldd	$f0, TASK_THREAD_FPCR($1)
+	fldd	$f0, TASK_THREAD_FPCR($8)
 	wfpcr	$f0
 	fimovd	$f0, $2
 	and	$2, 0x3, $2
@@ -232,14 +230,14 @@ $setfpec_1:
 $setfpec_2:
 	setfpec2
 $setfpec_over:
-	vldd	$f2, TASK_THREAD_F2($1)
-	vldd	$f3, TASK_THREAD_F3($1)
-	vldd	$f4, TASK_THREAD_F4($1)
-	vldd	$f5, TASK_THREAD_F5($1)
-	vldd	$f6, TASK_THREAD_F6($1)
-	vldd	$f7, TASK_THREAD_F7($1)
-	vldd	$f8, TASK_THREAD_F8($1)
-	vldd	$f9, TASK_THREAD_F9($1)
+	vldd	$f2, TASK_THREAD_F2($8)
+	vldd	$f3, TASK_THREAD_F3($8)
+	vldd	$f4, TASK_THREAD_F4($8)
+	vldd	$f5, TASK_THREAD_F5($8)
+	vldd	$f6, TASK_THREAD_F6($8)
+	vldd	$f7, TASK_THREAD_F7($8)
+	vldd	$f8, TASK_THREAD_F8($8)
+	vldd	$f9, TASK_THREAD_F9($8)
 
 	/* if $0 > 0, handle hcall */
 	bgt	$0, $ret_to