Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.

===================================================================
RCS file: /ftp/cvs/cvsroot/src/sys/arch/i386/i386/locore.S,v
rcsdiff: /ftp/cvs/cvsroot/src/sys/arch/i386/i386/locore.S,v: warning: Unknown phrases like `commitid ...;' are present.
retrieving revision 1.128
retrieving revision 1.153
diff -u -p -r1.128 -r1.153
--- src/sys/arch/i386/i386/locore.S	2016/06/04 09:52:41	1.128
+++ src/sys/arch/i386/i386/locore.S	2017/09/30 11:43:57	1.153
@@ -1,4 +1,4 @@
-/*	$NetBSD: locore.S,v 1.128 2016/06/04 09:52:41 maxv Exp $	*/
+/*	$NetBSD: locore.S,v 1.153 2017/09/30 11:43:57 maxv Exp $	*/
 
 /*
  * Copyright-o-rama!
@@ -128,15 +128,13 @@
  */
 
 #include <machine/asm.h>
-__KERNEL_RCSID(0, "$NetBSD: locore.S,v 1.128 2016/06/04 09:52:41 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: locore.S,v 1.153 2017/09/30 11:43:57 maxv Exp $");
 
-#include "opt_compat_oldboot.h"
 #include "opt_copy_symtab.h"
 #include "opt_ddb.h"
 #include "opt_modular.h"
 #include "opt_multiboot.h"
 #include "opt_realmem.h"
-#include "opt_vm86.h"
 #include "opt_xen.h"
 
 #include "assym.h"
@@ -219,6 +217,20 @@ __KERNEL_RCSID(0, "$NetBSD: locore.S,v 1
 2:					;
 
 /*
+ * fillkpt_blank - Fill in a kernel page table with blank entries
+ *	ebx = page table address
+ *	ecx = number of pages to map
+ */
+#define fillkpt_blank	\
+	cmpl	$0,%ecx			;	/* zero-sized? */	\
+	je 	2f			; \
+1:	movl	$0,(PDE_SIZE-4)(%ebx)	;	/* upper 32 bits: 0 */	\
+	movl	$0,(%ebx)		;	/* lower 32 bits: 0 */	\
+	addl	$PDE_SIZE,%ebx		;	/* next PTE/PDE */	\
+	loop	1b			; \
+2:					;
+
+/*
  * killkpt - Destroy a kernel page table
  *	ebx = page table address
  *	ecx = number of pages to destroy
@@ -256,45 +268,6 @@ __KERNEL_RCSID(0, "$NetBSD: locore.S,v 1
  */
 	.data
 
-	.globl	_C_LABEL(lapic_tpr)
-
-#if NLAPIC > 0
-#ifdef __ELF__
-	.align	PAGE_SIZE
-#else
-	.align	12
-#endif
-	.globl	_C_LABEL(local_apic)
-	.globl	_C_LABEL(lapic_id)
-
-	.type	_C_LABEL(local_apic), @object
-LABEL(local_apic)
-	.space	LAPIC_ID
-END(local_apic)
-	.type	_C_LABEL(lapic_id), @object
-LABEL(lapic_id)
-	.long	0x00000000
-	.space	LAPIC_TPRI-(LAPIC_ID+4)
-END(lapic_id)
-	.type	_C_LABEL(lapic_tpr), @object
-LABEL(lapic_tpr)
-	.space	LAPIC_PPRI-LAPIC_TPRI
-END(lapic_tpr)
-	.type	_C_LABEL(lapic_ppr), @object
-_C_LABEL(lapic_ppr):
-	.space	LAPIC_ISR-LAPIC_PPRI
-END(lapic_ppr)
-	.type	_C_LABEL(lapic_isr), @object
-_C_LABEL(lapic_isr):
-	.space	PAGE_SIZE-LAPIC_ISR
-END(lapic_isr)
-#else
-	.type	_C_LABEL(lapic_tpr), @object
-LABEL(lapic_tpr)
-	.long 0
-END(lapic_tpr)
-#endif
-
 	.globl	_C_LABEL(tablesize)
 	.globl	_C_LABEL(nox_flag)
 	.globl	_C_LABEL(cputype)
@@ -368,11 +341,13 @@ _C_LABEL(Multiboot_Header):
 	jne	1f
 
 	/*
-	 * Indeed, a multiboot-compliant boot loader executed us.  We copy
-	 * the received Multiboot information structure into kernel's data
-	 * space to process it later -- after we are relocated.  It will
-	 * be safer to run complex C code than doing it at this point.
+	 * Indeed, a multiboot-compliant boot loader executed us. We switch
+	 * to the temporary stack, and copy the received Multiboot information
+	 * structure into kernel's data space to process it later -- after we
+	 * are relocated. It will be safer to run complex C code than doing it
+	 * at this point.
 	 */
+	movl	$_RELOC(tmpstk),%esp
 	pushl	%ebx		/* Address of Multiboot information */
 	call	_C_LABEL(multiboot_pre_reloc)
 	addl	$4,%esp
@@ -613,9 +588,9 @@ no_NOX:
  * +------+--------+------+-----+--------+---------------------+-----------
  *                             (1)      (2)                   (3)
  *
- * -------+------------+
- * TABLES | ISA IO MEM |
- * -------+------------+
+ * -------+-------------+
+ * TABLES | ISA I/O MEM |
+ * -------+-------------+
  *       (4)
  *
  * PROC0 STK is obviously not linked as a page level. It just happens to be
@@ -626,7 +601,7 @@ no_NOX:
  */
 
 	/* Find end of kernel image; brings us on (1). */
-	movl	$RELOC(end),%edi
+	movl	$RELOC(__kernel_end),%edi
 
 #if (NKSYMS || defined(DDB) || defined(MODULAR)) && !defined(makeoptions_COPY_SYMTAB)
 	/* Save the symbols (if loaded); brinds us on (2). */
@@ -658,7 +633,7 @@ no_NOX:
 	incl	%eax		/* one more PTP for VAs stolen by bootstrap */
 1:	movl	%eax,RELOC(nkptp)+1*4
 
-	/* tablesize = (PDP_SIZE + UPAGES + nkptp) << PGSHIFT; */
+	/* tablesize = (PDP_SIZE + UPAGES + nkptp[1]) << PGSHIFT; */
 	addl	$(PDP_SIZE+UPAGES),%eax
 #ifdef PAE
 	incl	%eax 		/* one more page for L3 */
@@ -668,7 +643,8 @@ no_NOX:
 #endif
 	movl	%eax,RELOC(tablesize)
 
-	/* Ensure that nkptp covers BOOTSTRAP TABLES. */
+	/* Ensure that nkptp[1] covers BOOTSTRAP TABLES, ie:
+	 * (esi + tablesize) >> L2_SHIFT + 1 < nkptp[1] */
 	addl	%esi,%eax
 	addl	$~L2_FRAME,%eax
 	shrl	$L2_SHIFT,%eax
@@ -695,16 +671,13 @@ no_NOX:
 	 */
 	leal	(PROC0_PTP1_OFF)(%esi),%ebx
 
-	/* Skip the first MB. */
-	movl	$(KERNTEXTOFF - KERNBASE),%eax
-	movl	%eax,%ecx
-	shrl	$(PGSHIFT-2),%ecx	/* ((n >> PGSHIFT) << 2) for # PDEs */
-#ifdef PAE
-	shll	$1,%ecx			/* PDEs are twice larger with PAE */
-#endif
-	addl	%ecx,%ebx
+	/* Skip the area below the kernel text. */
+	movl	$(KERNTEXTOFF - KERNBASE),%ecx
+	shrl	$PGSHIFT,%ecx
+	fillkpt_blank
 
 	/* Map the kernel text RX. */
+	movl	$(KERNTEXTOFF - KERNBASE),%eax	/* start of TEXT */
 	movl	$RELOC(__rodata_start),%ecx
 	subl	%eax,%ecx
 	shrl	$PGSHIFT,%ecx
@@ -727,13 +700,13 @@ no_NOX:
 	orl	$(PG_V|PG_KW),%eax
 	fillkpt_nox
 
-	/* Map [SYMS]+[PRELOADED MODULES] RWX. */
+	/* Map [SYMS]+[PRELOADED MODULES] RW. */
 	movl	$RELOC(__kernel_end),%eax
 	movl	%esi,%ecx		/* start of BOOTSTRAP TABLES */
 	subl	%eax,%ecx
 	shrl	$PGSHIFT,%ecx
 	orl	$(PG_V|PG_KW),%eax
-	fillkpt
+	fillkpt_nox
 
 	/* Map the BOOTSTRAP TABLES RW. */
 	movl	%esi,%eax		/* start of BOOTSTRAP TABLES */
@@ -742,10 +715,12 @@ no_NOX:
 	orl	$(PG_V|PG_KW),%eax
 	fillkpt_nox
 
-	/* We are on (4). Map ISA I/O mem (later atdevbase) RWX. */
-	movl	$(IOM_BEGIN|PG_V|PG_KW/*|PG_N*/),%eax
-	movl	$(IOM_SIZE>>PGSHIFT),%ecx
-	fillkpt
+	/* We are on (4). Map ISA I/O MEM RW. */
+	movl	$IOM_BEGIN,%eax
+	movl	$IOM_SIZE,%ecx	/* size of ISA I/O MEM */
+	shrl	$PGSHIFT,%ecx
+	orl	$(PG_V|PG_KW/*|PG_N*/),%eax
+	fillkpt_nox
 
 	/*
 	 * Build L2 for identity mapping. Linked to L1.
@@ -763,12 +738,12 @@ no_NOX:
 	movl	RELOC(nkptp)+1*4,%ecx
 	fillkpt
 
-	/* Install a PDE recursively mapping page directory as a page table! */
+	/* Install recursive top level PDE */
 	leal	(PROC0_PDIR_OFF + PDIR_SLOT_PTE * PDE_SIZE)(%esi),%ebx
 	leal	(PROC0_PDIR_OFF)(%esi),%eax
 	orl	$(PG_V|PG_KW),%eax
 	movl	$PDP_SIZE,%ecx
-	fillkpt
+	fillkpt_nox
 
 #ifdef PAE
 	/*
@@ -863,28 +838,11 @@ begin:
 	pushl	$0	/* init386() expects a 64 bits paddr_t with PAE */
 #endif
 	pushl	%eax
-	call	_C_LABEL(init386)	/* wire 386 chip for unix operation */
+	call	_C_LABEL(init_bootspace)
+	call	_C_LABEL(init386)
 	addl	$PDE_SIZE,%esp		/* pop paddr_t */
 	addl	$NGDT*8,%esp		/* pop temporary gdt */
 
-#ifdef SAFARI_FIFO_HACK
-	movb	$5,%al
-	movw	$0x37b,%dx
-	outb	%al,%dx
-	movw	$0x37f,%dx
-	inb	%dx,%al
-	movb	%al,%cl
-
-	orb	$1,%cl
-
-	movb	$5,%al
-	movw	$0x37b,%dx
-	outb	%al,%dx
-	movw	$0x37f,%dx
-	movb	%cl,%al
-	outb	%al,%dx
-#endif /* SAFARI_FIFO_HACK */
-
 	call 	_C_LABEL(main)
 #else /* XEN */
 	/* First, reset the PSL. */
@@ -892,62 +850,67 @@ begin:
 	popfl
 
 	cld
-	movl	%esp, %ebx		/* save start of available space */
-	movl	$_RELOC(tmpstk),%esp	/* bootstrap stack end location */
+
+	/*
+	 * Xen info:
+	 * - %esp -> stack, *theoretically* the last used page by Xen bootstrap
+	 */
+	movl	%esp,%ebx
+	movl	$_RELOC(tmpstk),%esp
 
 	/* Clear BSS. */
 	xorl	%eax,%eax
 	movl	$RELOC(__bss_start),%edi
 	movl	$RELOC(_end),%ecx
 	subl	%edi,%ecx
-	rep stosb
+	rep
+	stosb
 
 	/* Copy the necessary stuff from start_info structure. */
 	/* We need to copy shared_info early, so that sti/cli work */
 	movl	$RELOC(start_info_union),%edi
 	movl	$128,%ecx
-	rep movsl
+	rep
+	movsl
 
-	/* Clear segment registers; always null in proc0. */
+	/* Clear segment registers. */
 	xorl	%eax,%eax
 	movw	%ax,%fs
 	movw	%ax,%gs
-	decl	%eax
-	movl	%eax,RELOC(cpuid_level)
 
 	xorl	%eax,%eax
 	cpuid
 	movl	%eax,RELOC(cpuid_level)
 
 	/*
-	 * Use a temp page. We'll re- add it to uvm(9) once we're
-	 * done using it.
+	 * Use a temporary GDT page. We'll re-add it to uvm(9) once we're done
+	 * using it.
 	 */
-	movl	$RELOC(tmpgdt), %eax
+	movl	$RELOC(tmpgdt),%eax
 	pushl	%eax		/* start of temporary gdt */
 	call	_C_LABEL(initgdt)
 	addl	$4,%esp
 
-	call	xen_pmap_bootstrap
+	call	xen_locore
 
 	/*
-	 * First avail returned by xen_pmap_bootstrap in %eax
+	 * The first VA available is returned by xen_locore in %eax. We
+	 * use it as the UAREA, and set up the stack here.
 	 */
-	movl	%eax, %esi;
-	movl	%esi, _C_LABEL(lwp0uarea)
-
-	/* Set up bootstrap stack. */
+	movl	%eax,%esi
+	movl	%esi,_C_LABEL(lwp0uarea)
 	leal	(USPACE-FRAMESIZE)(%eax),%esp
 	xorl	%ebp,%ebp		/* mark end of frames */
 
-	addl	$USPACE, %esi
-	subl	$KERNBASE, %esi		/* init386 wants a physical address */
+	/* Set first_avail after the DUMMY PAGE (see xen_locore). */
+	addl	$(USPACE+PAGE_SIZE),%esi
+	subl	$KERNBASE,%esi		/* init386 wants a physical address */
 
 #ifdef PAE
 	pushl	$0	/* init386() expects a 64 bits paddr_t with PAE */
 #endif
 	pushl	%esi
-	call	_C_LABEL(init386)	/* wire 386 chip for unix operation */
+	call	_C_LABEL(init386)
 	addl	$PDE_SIZE,%esp		/* pop paddr_t */
 	call 	_C_LABEL(main)
 #endif /* XEN */
@@ -1087,7 +1050,7 @@ END(dumpsys)
 
 /*
  * struct lwp *cpu_switchto(struct lwp *oldlwp, struct lwp *newlwp,
- *			    bool returning)
+ *     bool returning)
  *
  *	1. if (oldlwp != NULL), save its context.
  *	2. then, restore context of newlwp.
@@ -1115,16 +1078,18 @@ ENTRY(cpu_switchto)
 	movl	16(%esp),%esi		/* oldlwp */
 	movl	20(%esp),%edi		/* newlwp */
 	movl	24(%esp),%edx		/* returning */
-	testl	%esi,%esi
-	jz	1f
+
+	testl	%esi,%esi		/* oldlwp = NULL ? */
+	jz	skip_save
 
 	/* Save old context. */
 	movl	L_PCB(%esi),%eax
 	movl	%esp,PCB_ESP(%eax)
 	movl	%ebp,PCB_EBP(%eax)
+skip_save:
 
 	/* Switch to newlwp's stack. */
-1:	movl	L_PCB(%edi),%ebx
+	movl	L_PCB(%edi),%ebx
 	movl	PCB_EBP(%ebx),%ebp
 	movl	PCB_ESP(%ebx),%esp
 
@@ -1137,33 +1102,33 @@ ENTRY(cpu_switchto)
 
 	/* Skip the rest if returning to a pinned LWP. */
 	testl	%edx,%edx
-	jnz	4f
+	jnz	switch_return
 
+	/* Switch ring0 stack */
 #ifdef XEN
 	pushl	%edi
 	call	_C_LABEL(i386_switch_context)
 	addl	$4,%esp
-#else /* !XEN */
-	/* Switch ring0 esp */
+#else
 	movl	PCB_ESP0(%ebx),%eax
 	movl	%eax,CPUVAR(ESP0)
-#endif /* !XEN */
+#endif
 
 	/* Don't bother with the rest if switching to a system process. */
 	testl	$LW_SYSTEM,L_FLAG(%edi)
-	jnz	4f
+	jnz	switch_return
 
 #ifndef XEN
 	/* Restore thread-private %fs/%gs descriptors. */
 	movl	CPUVAR(GDT),%ecx
-	movl	PCB_FSD(%ebx), %eax
-	movl	PCB_FSD+4(%ebx), %edx
-	movl	%eax, (GUFS_SEL*8)(%ecx)
-	movl	%edx, (GUFS_SEL*8+4)(%ecx)
-	movl	PCB_GSD(%ebx), %eax
-	movl	PCB_GSD+4(%ebx), %edx
-	movl	%eax, (GUGS_SEL*8)(%ecx)
-	movl	%edx, (GUGS_SEL*8+4)(%ecx)
+	movl	PCB_FSD(%ebx),%eax
+	movl	PCB_FSD+4(%ebx),%edx
+	movl	%eax,(GUFS_SEL*8)(%ecx)
+	movl	%edx,(GUFS_SEL*8+4)(%ecx)
+	movl	PCB_GSD(%ebx),%eax
+	movl	PCB_GSD+4(%ebx),%edx
+	movl	%eax,(GUGS_SEL*8)(%ecx)
+	movl	%edx,(GUGS_SEL*8+4)(%ecx)
 #endif /* !XEN */
 
 	/* Switch I/O bitmap */
@@ -1176,14 +1141,25 @@ ENTRY(cpu_switchto)
 	/* Is this process using RAS (restartable atomic sequences)? */
 	movl	L_PROC(%edi),%eax
 	cmpl	$0,P_RASLIST(%eax)
-	jne	5f
+	je	no_RAS
+
+	/* Handle restartable atomic sequences (RAS). */
+	movl	L_MD_REGS(%edi),%ecx
+	pushl	TF_EIP(%ecx)
+	pushl	%eax
+	call	_C_LABEL(ras_lookup)
+	addl	$8,%esp
+	cmpl	$-1,%eax
+	je	no_RAS
+	movl	L_MD_REGS(%edi),%ecx
+	movl	%eax,TF_EIP(%ecx)
+no_RAS:
 
 	/*
 	 * Restore cr0 (including FPU state).  Raise the IPL to IPL_HIGH.
 	 * FPU IPIs can alter the LWP's saved cr0.  Dropping the priority
 	 * is deferred until mi_switch(), when cpu_switchto() returns.
 	 */
-2:
 #ifdef XEN
 	pushl	%edi
 	call	_C_LABEL(i386_tls_switch)
@@ -1198,34 +1174,24 @@ ENTRY(cpu_switchto)
 	 * set CR0_TS so we'll trap rather than reuse bogus state.
 	 */
 	cmpl	CPUVAR(FPCURLWP),%edi
-	je	3f
+	je	skip_TS
 	orl	$CR0_TS,%ecx
+skip_TS:
 
 	/* Reloading CR0 is very expensive - avoid if possible. */
-3:	cmpl	%edx,%ecx
-	je	4f
+	cmpl	%edx,%ecx
+	je	switch_return
 	movl	%ecx,%cr0
 #endif /* !XEN */
 
+switch_return:
 	/* Return to the new LWP, returning 'oldlwp' in %eax. */
-4:	movl	%esi,%eax
+	movl	%esi,%eax
 	popl	%edi
 	popl	%esi
 	popl	%ebx
 	ret
 
-	/* Check for restartable atomic sequences (RAS). */
-5:	movl	L_MD_REGS(%edi),%ecx
-	pushl	TF_EIP(%ecx)
-	pushl	%eax
-	call	_C_LABEL(ras_lookup)
-	addl	$8,%esp
-	cmpl	$-1,%eax
-	je	2b
-	movl	L_MD_REGS(%edi),%ecx
-	movl	%eax,TF_EIP(%ecx)
-	jmp	2b
-
 .Lcopy_iobitmap:
 	/* Copy I/O bitmap. */
 	incl	_C_LABEL(pmap_iobmp_evcnt)+EV_COUNT
@@ -1256,30 +1222,12 @@ ENTRY(savectx)
 END(savectx)
 
 /*
- * osyscall()
- *
- * Old call gate entry for syscall
- */
-IDTVEC(osyscall)
-#ifndef XEN
-	/* XXX we are in trouble! interrupts be off here. */
-	cli			/* must be first instruction */
-#endif
-	pushfl			/* set eflags in trap frame */
-	popl	8(%esp)
-	orl	$PSL_I,8(%esp)	/* re-enable ints on return to user */
-	pushl	$7		/* size of instruction for restart */
-	jmp	syscall1
-IDTVEC_END(osyscall)
-
-/*
  * syscall()
  *
  * Trap gate entry for syscall
  */
 IDTVEC(syscall)
 	pushl	$2		/* size of instruction for restart */
-syscall1:
 	pushl	$T_ASTFLT	/* trap # for doing ASTs */
 	INTRENTRY
 	STI(%eax)
@@ -1366,39 +1314,6 @@ syscall1:
 	jmp	.Lsyscall_checkast	/* re-check ASTs */
 IDTVEC_END(syscall)
 
-IDTVEC(svr4_fasttrap)
-	pushl	$2		/* size of instruction for restart */
-	pushl	$T_ASTFLT	/* trap # for doing ASTs */
-	INTRENTRY
-	STI(%eax)
-	pushl	$RW_READER
-	pushl	$_C_LABEL(svr4_fasttrap_lock)
-	call	_C_LABEL(rw_enter)
-	addl	$8,%esp
-	call	*_C_LABEL(svr4_fasttrap_vec)
-	pushl	$_C_LABEL(svr4_fasttrap_lock)
-	call	_C_LABEL(rw_exit)
-	addl	$4,%esp
-2:	/* Check for ASTs on exit to user mode. */
-	cli
-	CHECK_ASTPENDING(%eax)
-	je	1f
-	/* Always returning to user mode here. */
-	CLEAR_ASTPENDING(%eax)
-	sti
-	/* Pushed T_ASTFLT into tf_trapno on entry. */
-	pushl	%esp
-	call	_C_LABEL(trap)
-	addl	$4,%esp
-	jmp	2b
-1:	CHECK_DEFERRED_SWITCH
-	jnz	9f
-	INTRFASTEXIT
-9:	sti
-	call	_C_LABEL(pmap_load)
-	cli
-	jmp	2b
-
 /*
  * int npx586bug1(int a, int b)
  * Used when checking for the FDIV bug on first generations pentiums.
@@ -1453,3 +1368,20 @@ ENTRY(sse2_idlezero_page)
 	popl	%ebp
 	ret
 END(sse2_idlezero_page)
+
+ENTRY(intrfastexit)
+	movw	TF_GS(%esp),%gs
+	movw	TF_FS(%esp),%fs
+	movw	TF_ES(%esp),%es
+	movw	TF_DS(%esp),%ds
+	movl	TF_EDI(%esp),%edi
+	movl	TF_ESI(%esp),%esi
+	movl	TF_EBP(%esp),%ebp
+	movl	TF_EBX(%esp),%ebx
+	movl	TF_EDX(%esp),%edx
+	movl	TF_ECX(%esp),%ecx
+	movl	TF_EAX(%esp),%eax
+	addl	$(TF_PUSHSIZE+8),%esp
+	iret
+END(intrfastexit)
+