[PATCH] s390: system call speedup part 2.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

System call speedup part 2.

include/asm-s390/elf.h

@@ -155,9 +155,31 @@ typedef s390_regs elf_gregset_t;
 /* regs is struct pt_regs, pr_reg is elf_gregset_t (which is
    now struct_user_regs, they are different) */
 
-#define ELF_CORE_COPY_REGS(pr_reg, regs)	\
-	memcpy(&pr_reg,regs,sizeof(elf_gregset_t)); \
-
+static inline int dump_regs(struct pt_regs *ptregs, elf_gregset_t *regs)
+{
+	memcpy(&regs->psw, &ptregs->psw, sizeof(regs->psw)+sizeof(regs->gprs));
+	regs->orig_gpr2 = ptregs->orig_gpr2;
+	return 1;
+}
+
+#define ELF_CORE_COPY_REGS(pr_reg, regs) dump_regs(regs, &pr_reg);
+
+static inline int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
+{
+	dump_regs(__KSTK_PTREGS(tsk), regs);
+	memcpy(regs->acrs, tsk->thread.acrs, sizeof(regs->acrs));
+	return 1;
+}
+
+#define ELF_CORE_COPY_TASK_REGS(tsk, regs) dump_task_regs(tsk, regs)
+
+static inline int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
+{
+	memcpy(fpregs, &tsk->thread.fp_regs, sizeof(elf_fpregset_t));
+	return 1;
+}
+
+#define ELF_CORE_COPY_FPREGS(tsk, fpregs) dump_task_fpu(tsk, fpregs)
 
 
 /* This yields a mask that user programs can use to figure out what

include/asm-s390/lowcore.h

@@ -64,7 +64,10 @@
 
 #ifndef __s390x__
 #define __LC_KERNEL_STACK               0xC40
-#define __LC_ASYNC_STACK                0xC44
+#define __LC_THREAD_INFO		0xC44
+#define __LC_ASYNC_STACK                0xC48
+#define __LC_KERNEL_ASCE		0xC4C
+#define __LC_USER_ASCE			0xC50
 #define __LC_CPUID                      0xC60
 #define __LC_CPUADDR                    0xC68
 #define __LC_IPLDEV                     0xC7C
@@ -73,13 +76,16 @@
 #define __LC_INT_CLOCK			0xC98
 #else /* __s390x__ */
 #define __LC_KERNEL_STACK               0xD40
-#define __LC_ASYNC_STACK                0xD48
+#define __LC_THREAD_INFO		0xD48
+#define __LC_ASYNC_STACK                0xD50
+#define __LC_KERNEL_ASCE		0xD58
+#define __LC_USER_ASCE			0xD60
 #define __LC_CPUID                      0xD90
 #define __LC_CPUADDR                    0xD98
 #define __LC_IPLDEV                     0xDB8
 #define __LC_JIFFY_TIMER		0xDC0
 #define __LC_CURRENT			0xDD8
-#define __LC_INT_CLOCK			0xDE8
+#define __LC_INT_CLOCK			0xDe8
 #endif /* __s390x__ */
 
 #define __LC_PANIC_MAGIC                0xE00
@@ -99,7 +105,6 @@
 #include <linux/config.h>
 #include <asm/processor.h>
 #include <linux/types.h>
-#include <asm/atomic.h>
 #include <asm/sigp.h>
 
 void restart_int_handler(void);
@@ -167,9 +172,12 @@ struct _lowcore
         /* System info area */
 	__u32        save_area[16];            /* 0xc00 */
 	__u32        kernel_stack;             /* 0xc40 */
-	__u32        async_stack;              /* 0xc44 */
+	__u32        thread_info;              /* 0xc44 */
+	__u32        async_stack;              /* 0xc48 */
+	__u32        kernel_asce;              /* 0xc4c */
+	__u32        user_asce;                /* 0xc50 */
+	__u8         pad10[0xc60-0xc54];       /* 0xc54 */
 	/* entry.S sensitive area start */
-	__u8         pad10[0xc60-0xc48];       /* 0xc5c */
 	struct       cpuinfo_S390 cpu_data;    /* 0xc60 */
 	__u32        ipl_device;               /* 0xc7c */
 	/* entry.S sensitive area end */
@@ -245,9 +253,12 @@ struct _lowcore
 	__u64        save_area[16];            /* 0xc00 */
         __u8         pad9[0xd40-0xc80];        /* 0xc80 */
  	__u64        kernel_stack;             /* 0xd40 */
-	__u64        async_stack;              /* 0xd48 */
+	__u64        thread_info;              /* 0xd48 */
+	__u64        async_stack;              /* 0xd50 */
+	__u64        kernel_asce;              /* 0xd58 */
+	__u64        user_asce;                /* 0xd60 */
+	__u8         pad10[0xd80-0xd68];       /* 0xd68 */
 	/* entry.S sensitive area start */
-	__u8         pad10[0xd80-0xd50];       /* 0xd64 */
 	struct       cpuinfo_S390 cpu_data;    /* 0xd80 */
 	__u32        ipl_device;               /* 0xdb8 */
 	__u32        pad11;                    /* 0xdbc */

include/asm-s390/mmu_context.h

@@ -24,22 +24,19 @@ static inline void enter_lazy_tlb(struct mm_struct *mm,
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
                              struct task_struct *tsk)
 {
-        unsigned long pgd;
-
         if (prev != next) {
 #ifndef __s390x__
-	        pgd = (__pa(next->pgd)&PAGE_MASK) | 
+	        S390_lowcore.user_asce = (__pa(next->pgd)&PAGE_MASK) |
                       (_SEGMENT_TABLE|USER_STD_MASK);
-                /* Load page tables */
-                asm volatile("    lctl  7,7,%0\n"   /* secondary space */
-                             "    lctl  13,13,%0\n" /* home space */
-                             : : "m" (pgd) );
+                /* Load home space page table origin. */
+                asm volatile("lctl  13,13,%0"
+			     : : "m" (S390_lowcore.user_asce) );
 #else /* __s390x__ */
-                pgd = (__pa(next->pgd)&PAGE_MASK) | (_REGION_TABLE|USER_STD_MASK);
-                /* Load page tables */
-                asm volatile("    lctlg 7,7,%0\n"   /* secondary space */
-                             "    lctlg 13,13,%0\n" /* home space */
-                             : : "m" (pgd) );
+                S390_lowcore.user_asce = (__pa(next->pgd) & PAGE_MASK) |
+			(_REGION_TABLE|USER_STD_MASK);
+		/* Load home space page table origin. */
+		asm volatile("lctlg  13,13,%0"
+			     : : "m" (S390_lowcore.user_asce) );
 #endif /* __s390x__ */
         }
 	cpu_set(smp_processor_id(), next->cpu_vm_mask);
@@ -51,6 +48,7 @@ extern inline void activate_mm(struct mm_struct *prev,
                                struct mm_struct *next)
 {
         switch_mm(prev, next, current);
+	set_fs(current->thread.mm_segment);
 }
 
 #endif

include/asm-s390/processor.h

@@ -82,10 +82,10 @@ typedef struct {
  */
 struct thread_struct {
 	s390_fp_regs fp_regs;
-	unsigned int ar2;		/* kernel access register 2         */
-        unsigned int ar4;               /* kernel access register 4         */
+	unsigned int  acrs[NUM_ACRS];
         unsigned long ksp;              /* kernel stack pointer             */
         unsigned long user_seg;         /* HSTD                             */
+	mm_segment_t mm_segment;
         unsigned long prot_addr;        /* address of protection-excep.     */
         unsigned int error_code;        /* error-code of last prog-excep.   */
         unsigned int trap_no;
@@ -106,9 +106,10 @@ typedef struct thread_struct thread_struct;
 
 #define INIT_THREAD {{0,{{0},{0},{0},{0},{0},{0},{0},{0},{0},{0},	       \
 			    {0},{0},{0},{0},{0},{0}}},			       \
-		     0, 0,						       \
+		     {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},	       \
 		     sizeof(init_stack) + (unsigned long) &init_stack,	       \
 		     __SWAPPER_PG_DIR,					       \
+		     {0},						       \
 		     0,0,0,						       \
 		     (per_struct) {{{{0,}}},0,0,0,0,{{0,}}},		       \
 		     0, 0						       \
@@ -167,7 +168,7 @@ extern void show_trace(struct task_struct *task, unsigned long *sp);
 
 unsigned long get_wchan(struct task_struct *p);
 #define __KSTK_PTREGS(tsk) ((struct pt_regs *) \
-        (((unsigned long) tsk->thread_info + THREAD_SIZE - sizeof(struct pt_regs)) & -8L))
+        ((unsigned long) tsk->thread_info + THREAD_SIZE - sizeof(struct pt_regs)))
 #define KSTK_EIP(tsk)	(__KSTK_PTREGS(tsk)->psw.addr)
 #define KSTK_ESP(tsk)	(__KSTK_PTREGS(tsk)->gprs[15])
 

include/asm-s390/ptrace.h

@@ -286,10 +286,7 @@ typedef struct
 	 ((NEW) & (PSW_MASK_CC|PSW_MASK_PM)))
 
 /*
- * The first entries in pt_regs and user_regs_struct
- * are common for the two structures. The s390_regs structure
- * covers the common parts. It simplifies copying the common part
- * between the three structures.
+ * The s390_regs structure is used to define the elf_gregset_t.
  */
 typedef struct
 {
@@ -299,6 +296,7 @@ typedef struct
 	unsigned long orig_gpr2;
 } s390_regs;
 
+#ifdef __KERNEL__
 /*
  * The pt_regs struct defines the way the registers are stored on
  * the stack during a system call.
@@ -307,11 +305,11 @@ struct pt_regs
 {
 	psw_t psw;
 	unsigned long gprs[NUM_GPRS];
-	unsigned int  acrs[NUM_ACRS];
 	unsigned long orig_gpr2;
 	unsigned short ilc;
 	unsigned short trap;
-} __attribute__ ((packed));
+};
+#endif
 
 /*
  * Now for the program event recording (trace) definitions.

include/asm-s390/system.h

@@ -83,14 +83,33 @@ static inline void restore_fp_regs(s390_fp_regs *fpregs)
 		: : "a" (fpregs), "m" (*fpregs) );
 }
 
+static inline void save_access_regs(unsigned int *acrs)
+{
+	asm volatile ("stam 0,15,0(%0)" : : "a" (acrs) : "memory" );
+}
+
+static inline void restore_access_regs(unsigned int *acrs)
+{
+	asm volatile ("lam 0,15,0(%0)" : : "a" (acrs) );
+}
+
 #define switch_to(prev,next,last) do {					     \
 	if (prev == next)						     \
 		break;							     \
 	save_fp_regs(&prev->thread.fp_regs);				     \
 	restore_fp_regs(&next->thread.fp_regs);				     \
+	save_access_regs(&prev->thread.acrs[0]);			     \
+	restore_access_regs(&next->thread.acrs[0]);			     \
 	prev = __switch_to(prev,next);					     \
 } while (0)
 
+#define prepare_arch_switch(rq, next)	do { } while(0)
+#define task_running(rq, p)		((rq)->curr == (p))
+#define finish_arch_switch(rq, prev) do {				     \
+	set_fs(current->thread.mm_segment);				     \
+	spin_unlock_irq(&(rq)->lock);					     \
+} while (0)
+
 #define nop() __asm__ __volatile__ ("nop")
 
 #define xchg(ptr,x) \