i386: Add AP variants of descriptor tables

17 files changed, 165 insertions, 54 deletions

diff --git a/i386/i386/apic.c b/i386/i386/apic.c
index 2e0c1776..d30084e2 100644
--- a/i386/i386/apic.c
+++ b/i386/i386/apic.c
@@ -234,3 +234,9 @@ void apic_print_info(void)
         }
     }
 }
+
+void
+lapic_eoi(void)
+{
+    lapic->eoi.r = 0;
+}
diff --git a/i386/i386/gdt.c b/i386/i386/gdt.c
index fb18360e..ddda603b 100644
--- a/i386/i386/gdt.c
+++ b/i386/i386/gdt.c
@@ -39,6 +39,7 @@
 #include "vm_param.h"
 #include "seg.h"
 #include "gdt.h"
+#include "mp_desc.h"
 
 #ifdef	MACH_PV_DESCRIPTORS
 /* It is actually defined in xen_boothdr.S */
@@ -46,28 +47,28 @@ extern
 #endif	/* MACH_PV_DESCRIPTORS */
 struct real_descriptor gdt[GDTSZ];
 
-void
-gdt_init(void)
+static void
+gdt_fill(struct real_descriptor *mygdt)
 {
 	/* Initialize the kernel code and data segment descriptors.  */
 #ifdef __x86_64__
 	assert(LINEAR_MIN_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS == 0);
-	fill_gdt_descriptor(KERNEL_CS, 0, 0, ACC_PL_K|ACC_CODE_R, SZ_64);
-	fill_gdt_descriptor(KERNEL_DS, 0, 0, ACC_PL_K|ACC_DATA_W, SZ_64);
+	_fill_gdt_descriptor(mygdt, KERNEL_CS, 0, 0, ACC_PL_K|ACC_CODE_R, SZ_64);
+	_fill_gdt_descriptor(mygdt, KERNEL_DS, 0, 0, ACC_PL_K|ACC_DATA_W, SZ_64);
 #ifndef	MACH_PV_DESCRIPTORS
-	fill_gdt_descriptor(LINEAR_DS, 0, 0, ACC_PL_K|ACC_DATA_W, SZ_64);
+	_fill_gdt_descriptor(mygdt, LINEAR_DS, 0, 0, ACC_PL_K|ACC_DATA_W, SZ_64);
 #endif	/* MACH_PV_DESCRIPTORS */
 #else
-	fill_gdt_descriptor(KERNEL_CS,
+	_fill_gdt_descriptor(mygdt, KERNEL_CS,
 			    LINEAR_MIN_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
 			    LINEAR_MAX_KERNEL_ADDRESS - (LINEAR_MIN_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) - 1,
 			    ACC_PL_K|ACC_CODE_R, SZ_32);
-	fill_gdt_descriptor(KERNEL_DS,
+	_fill_gdt_descriptor(mygdt, KERNEL_DS,
 			    LINEAR_MIN_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
 			    LINEAR_MAX_KERNEL_ADDRESS - (LINEAR_MIN_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) - 1,
 			    ACC_PL_K|ACC_DATA_W, SZ_32);
 #ifndef	MACH_PV_DESCRIPTORS
-	fill_gdt_descriptor(LINEAR_DS,
+	_fill_gdt_descriptor(mygdt, LINEAR_DS,
 			    0,
 			    0xffffffff,
 			    ACC_PL_K|ACC_DATA_W, SZ_32);
@@ -75,8 +76,8 @@ gdt_init(void)
 #endif
 
 #ifdef	MACH_PV_DESCRIPTORS
-	unsigned long frame = kv_to_mfn(gdt);
-	pmap_set_page_readonly(gdt);
+	unsigned long frame = kv_to_mfn(mygdt);
+	pmap_set_page_readonly(mygdt);
 	if (hyp_set_gdt(kv_to_la(&frame), GDTSZ))
 		panic("couldn't set gdt\n");
 #endif
@@ -94,12 +95,16 @@ gdt_init(void)
 	{
 		struct pseudo_descriptor pdesc;
 
-		pdesc.limit = sizeof(gdt)-1;
-		pdesc.linear_base = kvtolin(&gdt);
+		pdesc.limit = (GDTSZ * sizeof(struct real_descriptor))-1;
+		pdesc.linear_base = kvtolin(mygdt);
 		lgdt(&pdesc);
 	}
 #endif	/* MACH_PV_DESCRIPTORS */
+}
 
+static void
+reload_segs(void)
+{
 	/* Reload all the segment registers from the new GDT.
 	   We must load ds and es with 0 before loading them with KERNEL_DS
 	   because some processors will "optimize out" the loads
@@ -117,6 +122,15 @@ gdt_init(void)
 		     "movw	%w1,%%ss\n"
 		     : : "i" (KERNEL_CS), "r" (KERNEL_DS), "r" (0));
 #endif
+}
+
+void
+gdt_init(void)
+{
+	gdt_fill(gdt);
+
+	reload_segs();
+
 #ifdef	MACH_PV_PAGETABLES
 #if VM_MIN_KERNEL_ADDRESS != LINEAR_MIN_KERNEL_ADDRESS
 	/* things now get shifted */
@@ -128,3 +142,12 @@ gdt_init(void)
 #endif	/* MACH_PV_PAGETABLES */
 }
 
+#if NCPUS > 1
+void
+ap_gdt_init(int cpu)
+{
+	gdt_fill(mp_gdt[cpu]);
+
+	reload_segs();
+}
+#endif
diff --git a/i386/i386/gdt.h b/i386/i386/gdt.h
index 9879ad3e..5def73cb 100644
--- a/i386/i386/gdt.h
+++ b/i386/i386/gdt.h
@@ -115,5 +115,6 @@ extern struct real_descriptor gdt[GDTSZ];
 #endif
 
 extern void gdt_init(void);
+extern void ap_gdt_init(int cpu);
 
 #endif /* _I386_GDT_ */
diff --git a/i386/i386/i386asm.sym b/i386/i386/i386asm.sym
index 9f4ebe55..85658b30 100644
--- a/i386/i386/i386asm.sym
+++ b/i386/i386/i386asm.sym
@@ -47,6 +47,7 @@
 #include <i386/mp_desc.h>
 #include <i386/xen.h>
 
+expr	CALL_SINGLE_FUNCTION_BASE
 
 offset	thread			th	pcb
 offset	thread			th	task
diff --git a/i386/i386/idt-gen.h b/i386/i386/idt-gen.h
index f86afb41..daa6aaf2 100644
--- a/i386/i386/idt-gen.h
+++ b/i386/i386/idt-gen.h
@@ -41,7 +41,7 @@
 extern struct real_gate idt[IDTSZ];
 
 /* Fill a gate in the IDT.  */
-#define fill_idt_gate(int_num, entry, selector, access, dword_count) \
-	fill_gate(&idt[int_num], entry, selector, access, dword_count)
+#define fill_idt_gate(_idt, int_num, entry, selector, access, dword_count) \
+	fill_gate(&_idt[int_num], entry, selector, access, dword_count)
 
 #endif /* _I386_IDT_ */
diff --git a/i386/i386/idt.c b/i386/i386/idt.c
index c6a778f1..cdfb9a88 100644
--- a/i386/i386/idt.c
+++ b/i386/i386/idt.c
@@ -25,6 +25,7 @@
 #include <i386/seg.h>
 #include <i386at/idt.h>
 #include <i386/gdt.h>
+#include <i386/mp_desc.h>
 
 struct real_gate idt[IDTSZ];
 
@@ -36,7 +37,8 @@ struct idt_init_entry
 };
 extern struct idt_init_entry idt_inittab[];
 
-void idt_init(void)
+static void
+idt_fill(struct real_gate *myidt)
 {
 #ifdef	MACH_PV_DESCRIPTORS
 	if (hyp_set_trap_table(kvtolin(idt_inittab)))
@@ -47,7 +49,7 @@ void idt_init(void)
 	/* Initialize the exception vectors from the idt_inittab.  */
 	while (iie->entrypoint)
 	{
-		fill_idt_gate(iie->vector, iie->entrypoint, KERNEL_CS, iie->type, 0);
+		fill_idt_gate(myidt, iie->vector, iie->entrypoint, KERNEL_CS, iie->type, 0);
 		iie++;
 	}
 
@@ -55,10 +57,21 @@ void idt_init(void)
 	{
 		struct pseudo_descriptor pdesc;
 
-		pdesc.limit = sizeof(idt)-1;
-		pdesc.linear_base = kvtolin(&idt);
+		pdesc.limit = (IDTSZ * sizeof(struct real_gate))-1;
+		pdesc.linear_base = kvtolin(myidt);
 		lidt(&pdesc);
 	}
 #endif	/* MACH_PV_DESCRIPTORS */
 }
 
+void idt_init(void)
+{
+	idt_fill(idt);
+}
+
+#if NCPUS > 1
+void ap_idt_init(int cpu)
+{
+	idt_fill(mp_desc_table[cpu]->idt);
+}
+#endif
diff --git a/i386/i386/ktss.c b/i386/i386/ktss.c
index 0d21d3eb..1d880167 100644
--- a/i386/i386/ktss.c
+++ b/i386/i386/ktss.c
@@ -35,14 +35,16 @@
 #include "seg.h"
 #include "gdt.h"
 #include "ktss.h"
+#include "mp_desc.h"
 
 /* A kernel TSS with a complete I/O bitmap.  */
 struct task_tss ktss;
 
 void
-ktss_init(void)
+ktss_fill(struct task_tss *myktss, struct real_descriptor *mygdt)
 {
 	/* XXX temporary exception stack */
+	/* FIXME: make it per-processor */
 	static int exception_stack[1024];
 
 #ifdef	MACH_RING1
@@ -52,24 +54,37 @@ ktss_init(void)
 		panic("couldn't register exception stack\n");
 #else	/* MACH_RING1 */
 	/* Initialize the master TSS descriptor.  */
-	fill_gdt_sys_descriptor(KERNEL_TSS,
-				kvtolin(&ktss), sizeof(struct task_tss) - 1,
+	_fill_gdt_sys_descriptor(mygdt, KERNEL_TSS,
+				kvtolin(myktss), sizeof(struct task_tss) - 1,
 				ACC_PL_K|ACC_TSS, 0);
 
 	/* Initialize the master TSS.  */
 #ifdef __x86_64__
-	ktss.tss.rsp0 = (unsigned long)(exception_stack+1024);
+	myktss->tss.rsp0 = (unsigned long)(exception_stack+1024);
 #else /* ! __x86_64__ */
-	ktss.tss.ss0 = KERNEL_DS;
-	ktss.tss.esp0 = (unsigned long)(exception_stack+1024);
+	myktss->tss.ss0 = KERNEL_DS;
+	myktss->tss.esp0 = (unsigned long)(exception_stack+1024);
 #endif /* __x86_64__ */
 
-	ktss.tss.io_bit_map_offset = IOPB_INVAL;
+	myktss->tss.io_bit_map_offset = IOPB_INVAL;
 	/* Set the last byte in the I/O bitmap to all 1's.  */
-	ktss.barrier = 0xff;
+	myktss->barrier = 0xff;
 
 	/* Load the TSS.  */
 	ltr(KERNEL_TSS);
 #endif	/* MACH_RING1 */
 }
 
+void
+ktss_init(void)
+{
+	ktss_fill(&ktss, gdt);
+}
+
+#if NCPUS > 1
+void
+ap_ktss_init(int cpu)
+{
+	ktss_fill(&mp_desc_table[cpu]->ktss, mp_gdt[cpu]);
+}
+#endif
diff --git a/i386/i386/ktss.h b/i386/i386/ktss.h
index 304a877a..171332da 100644
--- a/i386/i386/ktss.h
+++ b/i386/i386/ktss.h
@@ -28,5 +28,6 @@
 extern struct task_tss ktss;
 
 extern void ktss_init(void);
+extern void ap_ktss_init(int cpu);
 
 #endif /* _I386_KTSS_ */
diff --git a/i386/i386/ldt.c b/i386/i386/ldt.c
index 261df93a..3f9ac8ff 100644
--- a/i386/i386/ldt.c
+++ b/i386/i386/ldt.c
@@ -37,6 +37,7 @@
 #include "gdt.h"
 #include "ldt.h"
 #include "locore.h"
+#include "mp_desc.h"
 
 #ifdef	MACH_PV_DESCRIPTORS
 /* It is actually defined in xen_boothdr.S */
@@ -45,37 +46,51 @@ extern
 struct real_descriptor ldt[LDTSZ];
 
 void
-ldt_init(void)
+ldt_fill(struct real_descriptor *myldt, struct real_descriptor *mygdt)
 {
 #ifdef	MACH_PV_DESCRIPTORS
 #ifdef	MACH_PV_PAGETABLES
-	pmap_set_page_readwrite(ldt);
+	pmap_set_page_readwrite(myldt);
 #endif	/* MACH_PV_PAGETABLES */
 #else	/* MACH_PV_DESCRIPTORS */
 	/* Initialize the master LDT descriptor in the GDT.  */
-	fill_gdt_sys_descriptor(KERNEL_LDT,
-			        kvtolin(&ldt), sizeof(ldt)-1,
+	_fill_gdt_sys_descriptor(mygdt, KERNEL_LDT,
+			        kvtolin(myldt), (LDTSZ * sizeof(struct real_descriptor))-1,
 			        ACC_PL_K|ACC_LDT, 0);
 #endif	/* MACH_PV_DESCRIPTORS */
 
 	/* Initialize the 32bit LDT descriptors.  */
-	fill_ldt_gate(USER_SCALL,
+	fill_ldt_gate(myldt, USER_SCALL,
 		      (vm_offset_t)&syscall, KERNEL_CS,
 		      ACC_PL_U|ACC_CALL_GATE, 0);
-	fill_ldt_descriptor(USER_CS,
+	fill_ldt_descriptor(myldt, USER_CS,
 			    VM_MIN_ADDRESS,
 			    VM_MAX_ADDRESS-VM_MIN_ADDRESS-4096,
 			    /* XXX LINEAR_... */
 			    ACC_PL_U|ACC_CODE_R, SZ_32);
-	fill_ldt_descriptor(USER_DS,
+	fill_ldt_descriptor(myldt, USER_DS,
 			    VM_MIN_ADDRESS,
 			    VM_MAX_ADDRESS-VM_MIN_ADDRESS-4096,
 			    ACC_PL_U|ACC_DATA_W, SZ_32);
 
 	/* Activate the LDT.  */
 #ifdef	MACH_PV_DESCRIPTORS
-	hyp_set_ldt(&ldt, LDTSZ);
+	hyp_set_ldt(myldt, LDTSZ);
 #else	/* MACH_PV_DESCRIPTORS */
 	lldt(KERNEL_LDT);
 #endif	/* MACH_PV_DESCRIPTORS */
 }
+
+void
+ldt_init(void)
+{
+	ldt_fill(ldt, gdt);
+}
+
+#if NCPUS > 1
+void
+ap_ldt_init(int cpu)
+{
+	ldt_fill(mp_desc_table[cpu]->ldt, mp_gdt[cpu]);
+}
+#endif
diff --git a/i386/i386/ldt.h b/i386/i386/ldt.h
index 1f0d7014..b15f11a5 100644
--- a/i386/i386/ldt.h
+++ b/i386/i386/ldt.h
@@ -57,14 +57,15 @@
 extern struct real_descriptor ldt[LDTSZ];
 
 /* Fill a 32bit segment descriptor in the LDT.  */
-#define fill_ldt_descriptor(selector, base, limit, access, sizebits) \
-	fill_descriptor(&ldt[sel_idx(selector)], base, limit, access, sizebits)
+#define fill_ldt_descriptor(_ldt, selector, base, limit, access, sizebits) \
+	fill_descriptor(&_ldt[sel_idx(selector)], base, limit, access, sizebits)
 
-#define fill_ldt_gate(selector, offset, dest_selector, access, word_count) \
-	fill_gate((struct real_gate*)&ldt[sel_idx(selector)], \
+#define fill_ldt_gate(_ldt, selector, offset, dest_selector, access, word_count) \
+	fill_gate((struct real_gate*)&_ldt[sel_idx(selector)], \
 		  offset, dest_selector, access, word_count)
 
 void ldt_init(void);
+void ap_ldt_init(int cpu);
 
 #endif /* !__ASSEMBLER__ */
 
diff --git a/i386/i386/locore.S b/i386/i386/locore.S
index f50aca5d..ff78e80d 100644
--- a/i386/i386/locore.S
+++ b/i386/i386/locore.S
@@ -649,6 +649,10 @@ INTERRUPT(20)
 INTERRUPT(21)
 INTERRUPT(22)
 INTERRUPT(23)
+#endif
+/* Invalidate TLB IPI to call pmap_update_interrupt() on a specific cpu */
+INTERRUPT(CALL_SINGLE_FUNCTION_BASE)
+#ifdef APIC
 /* Spurious interrupt, set irq number to vect number */
 INTERRUPT(255)
 #endif
diff --git a/i386/i386/mp_desc.h b/i386/i386/mp_desc.h
index ebe1471d..ede8775f 100644
--- a/i386/i386/mp_desc.h
+++ b/i386/i386/mp_desc.h
@@ -27,6 +27,8 @@
 #ifndef	_I386_MP_DESC_H_
 #define	_I386_MP_DESC_H_
 
+#include <mach/kern_return.h>
+
 #if MULTIPROCESSOR
 
 /*
diff --git a/i386/i386at/idt.h b/i386/i386at/idt.h
index ac065aef..f080bb12 100644
--- a/i386/i386at/idt.h
+++ b/i386/i386at/idt.h
@@ -37,10 +37,14 @@
 /* IOAPIC spurious interrupt vector set to 0xff */
 #define IOAPIC_SPURIOUS_BASE 0xff
 
+/* Currently for TLB shootdowns */
+#define CALL_SINGLE_FUNCTION_BASE 0xfb
+
 #include <i386/idt-gen.h>
 
 #ifndef __ASSEMBLER__
 extern void idt_init (void);
+extern void ap_idt_init (int cpu);
 #endif /* __ASSEMBLER__ */
 
 #endif /* _I386AT_IDT_ */
diff --git a/i386/i386at/int_init.c b/i386/i386at/int_init.c
index 3fed4197..964ce1bb 100644
--- a/i386/i386at/int_init.c
+++ b/i386/i386at/int_init.c
@@ -23,29 +23,50 @@
 
 #include <i386at/idt.h>
 #include <i386at/int_init.h>
-#include <i386/gdt.h>
+#include <i386/mp_desc.h>
+#include <i386/i386asm.h>
 
 /* defined in locore.S */
 extern vm_offset_t int_entry_table[];
 
-void int_init(void)
+void
+int_fill(struct real_gate *myidt)
 {
 	int i;
 #ifndef APIC
-	for (i = 0; i < 16; i++) {
-		fill_idt_gate(PIC_INT_BASE + i,
+	int base = PIC_INT_BASE;
+	int nirq = 16;
+#else
+	int base = IOAPIC_INT_BASE;
+	int nirq = 24;
+#endif
+
+	for (i = 0; i < nirq; i++) {
+		fill_idt_gate(myidt, base + i,
 			      int_entry_table[i], KERNEL_CS,
 			      ACC_PL_K|ACC_INTR_GATE, 0);
 	}
-#else
-	for (i = 0; i < 24; i++) {
-		fill_idt_gate(IOAPIC_INT_BASE + i,
+	fill_idt_gate(myidt, CALL_SINGLE_FUNCTION_BASE,
 			      int_entry_table[i], KERNEL_CS,
 			      ACC_PL_K|ACC_INTR_GATE, 0);
-	}
-	fill_idt_gate(IOAPIC_SPURIOUS_BASE,
-			      int_entry_table[24], KERNEL_CS,
+	i++;
+#ifdef APIC
+	fill_idt_gate(myidt, IOAPIC_SPURIOUS_BASE,
+			      int_entry_table[i], KERNEL_CS,
 			      ACC_PL_K|ACC_INTR_GATE, 0);
+	i++;
 #endif
 }
 
+void
+int_init(void)
+{
+	int_fill(idt);
+}
+
+#if NCPUS > 1
+void ap_int_init(int cpu)
+{
+	int_fill(mp_desc_table[cpu]->idt);
+}
+#endif
diff --git a/i386/i386at/int_init.h b/i386/i386at/int_init.h
index f9b03b74..3c11ebce 100644
--- a/i386/i386at/int_init.h
+++ b/i386/i386at/int_init.h
@@ -29,6 +29,7 @@
 
 #ifndef __ASSEMBLER__
 extern void int_init (void);
+extern void ap_int_init (int cpu);
 #endif /* __ASSEMBLER__ */
 
 #endif /* _INT_INIT_H_ */
diff --git a/i386/i386at/interrupt.S b/i386/i386at/interrupt.S
index 16e0df10..9f1883ac 100644
--- a/i386/i386at/interrupt.S
+++ b/i386/i386at/interrupt.S
@@ -58,6 +58,9 @@ ENTRY(interrupt)
 	movl	%eax,12(%esp)		/* address of interrupted registers as 4th arg */
 
 	movl	S_IRQ,%eax		/* copy irq number */
+	cmpl	$CALL_SINGLE_FUNCTION_BASE,%eax		/* was this a SMP call single function request? */
+	je	_call_single
+
 	shll	$2,%eax			/* irq * 4 */
 	movl	EXT(iunit)(%eax),%edx	/* get device unit number */
 	movl	%edx,(%esp)		/* unit number as 1st arg */
@@ -117,4 +120,10 @@ _isa_eoi:
 	addl	$24,%esp		/* pop local variables */
 _no_eoi:
 	ret
+
+_call_single:
+	call	EXT(pmap_update_interrupt) /* TODO: Allow other functions */
+	call	EXT(lapic_eoi)		/* lapic EOI */
+	addl	$24,%esp
+	ret
 END(interrupt)
diff --git a/i386/i386at/ioapic.c b/i386/i386at/ioapic.c
index 3373aa71..c5eb3536 100644
--- a/i386/i386at/ioapic.c
+++ b/i386/i386at/ioapic.c
@@ -234,12 +234,6 @@ ioapic_toggle(int pin, int mask)
 }
 
 void
-lapic_eoi(void)
-{
-    lapic->eoi.r = 0;
-}
-
-void
 ioapic_irq_eoi(int pin)
 {
     int apic = 0;