Add x86 hardware debugging register support

This adds using the x86 hardware debugging registers, either from the kernel
through db_set_hw_watchpoint, or from userland through i386_DEBUG_STATE.
While the kernel is using the registers, the userland values are ignored.

* i386/i386/db_interface.c (kernel_dr, ids): New variables.
(db_load_context, db_get_debug_state, db_set_debug_state, db_dr,
db_set_hw_watchpoint): New functions.
(kdb_trap): Use get_dr* instead of dr_addr[].
* i386/i386/db_interface.h (db_user_to_kernel_address, db_set_hw_watchpoint)
(db_dr, db_get_debug_state, db_set_debug_state, db_load_context): Add functions
prototypes.
(dr0, dr1, dr2, dr3): Remove functions prototypes.
* i386/i386/locore.S (dr6, dr0, dr1, dr2, dr3): Remove functions.
(dr_msk, dr_addr): Remove variables.
* i386/include/mach/i386/thread_status.h (i386_DEBUG_STATE): Add macro.
(i386_debug_state): Add structure.
(i386_DEBUG_STATE_COUNT): Add macro.
* i386/i386/thread.h: Include <mach/machine/thread_status.h>.
(i386_machine_state): Add `struct i386_debug_state ids' field.
* i386/i386/pcb.c: Include <i386/db_interface.h>.
(switch_ktss): Call db_load_context.
(thread_setstatus, thread_getstatus): Add I386_DEBUG_STATE case.
* i386/i386/proc_reg.h (get_dr0, set_dr0, get_dr1, set_dr1, get_dr2,
set_dr2, get_dr3, set_dr3, get_dr6, set_dr6, get_dr7, set_dr7): Add macros.

7 files changed, 336 insertions, 134 deletions

diff --git a/i386/i386/db_interface.c b/i386/i386/db_interface.c
index 66cc8b5..c6d7fb4 100644
--- a/i386/i386/db_interface.c
+++ b/i386/i386/db_interface.c
@@ -27,8 +27,6 @@
  * Interface to new debugger.
  */
 
-#if MACH_KDB
-
 #include <sys/reboot.h>
 #include <vm/pmap.h>
 
@@ -58,12 +56,149 @@
 #include <machine/db_interface.h>
 #include <machine/machspl.h>
 
+#if MACH_KDB
+/* Whether the kernel uses any debugging register.  */
+static int kernel_dr;
+#endif
+
+void db_load_context(pcb_t pcb)
+{
+#if MACH_KDB
+	int s = splhigh();
+
+	if (kernel_dr) {
+		splx(s);
+		return;
+	}
+#endif
+
+	/* Else set user debug registers */
+	set_dr0(pcb->ims.ids.dr[0]);
+	set_dr1(pcb->ims.ids.dr[1]);
+	set_dr2(pcb->ims.ids.dr[2]);
+	set_dr3(pcb->ims.ids.dr[3]);
+	set_dr7(pcb->ims.ids.dr[7]);
+#if MACH_KDB
+	splx(s);
+#endif
+}
+
+void db_get_debug_state(
+	pcb_t pcb,
+	struct i386_debug_state *state)
+{
+	*state = pcb->ims.ids;
+}
+
+kern_return_t db_set_debug_state(
+	pcb_t pcb,
+	const struct i386_debug_state *state)
+{
+	int i;
+
+	for (i = 0; i <= 3; i++)
+		if (state->dr[i] < VM_MIN_ADDRESS
+	 	 || state->dr[i] >= VM_MAX_ADDRESS)
+			return KERN_INVALID_ARGUMENT;
+
+	pcb->ims.ids = *state;
+
+	if (pcb == current_thread()->pcb)
+		db_load_context(pcb);
+
+	return KERN_SUCCESS;
+}
+
+#if MACH_KDB
+
 struct	 i386_saved_state *i386_last_saved_statep;
 struct	 i386_saved_state i386_nested_saved_state;
 unsigned i386_last_kdb_sp;
 
 extern	thread_t db_default_thread;
 
+static struct i386_debug_state ids;
+
+void db_dr (
+	int		num,
+	vm_offset_t	linear_addr,
+	int		type,
+	int		len,
+	int		persistence)
+{
+	int s = splhigh();
+	unsigned long dr7;
+
+	if (!kernel_dr) {
+	    if (!linear_addr) {
+		splx(s);
+		return;
+	    }
+	    kernel_dr = 1;
+	    /* Clear user debugging registers */
+	    set_dr7(0);
+	    set_dr0(0);
+	    set_dr1(0);
+	    set_dr2(0);
+	    set_dr3(0);
+	}
+
+	ids.dr[num] = linear_addr;
+	switch (num) {
+	    case 0: set_dr0(linear_addr); break;
+	    case 1: set_dr1(linear_addr); break;
+	    case 2: set_dr2(linear_addr); break;
+	    case 3: set_dr3(linear_addr); break;
+	}
+
+	/* Replace type/len/persistence for DRnum in dr7 */
+	dr7 = get_dr7 ();
+	dr7 &= ~(0xfUL << (4*num+16)) & ~(0x3UL << (2*num));
+	dr7 |= (((len << 2) | type) << (4*num+16)) | (persistence << (2*num));
+	set_dr7 (dr7);
+
+	if (kernel_dr) {
+	    if (!ids.dr[0] && !ids.dr[1] && !ids.dr[2] && !ids.dr[3]) {
+		/* Not used any more, switch back to user debugging registers */
+		kernel_dr = 0;
+		db_load_context(current_thread()->pcb);
+	    }
+	}
+	splx(s);
+}
+
+void db_set_hw_watchpoint(
+	int		num,
+	task_t		task,
+	db_addr_t	addr,
+	vm_size_t	size)
+{
+	unsigned int kern_addr;
+
+	if (size != 1 && size != 2 && size != 4)
+	    return;
+
+	if (addr & (size-1))
+	    /* Unaligned */
+	    return;
+
+	if (!addr) {
+	    db_dr (num, 0, 0, 0, 0);
+	    db_printf("Hardware watchpoint %d deleted\n", num);
+	}
+
+	if (task) {
+	    if (db_user_to_kernel_address(task, addr, &kern_addr, 1) < 0)
+		return;
+	    addr = kern_addr;
+	}
+	addr = kvtolin(addr);
+
+	db_dr (num, addr, I386_DB_TYPE_W, size-1, I386_DB_LOCAL|I386_DB_GLOBAL);
+
+	db_printf("Hardware watchpoint %d set for %x\n", num, addr);
+}
+
 /*
  * Print trap reason.
  */
@@ -96,15 +231,14 @@ kdb_trap(
 	switch (type) {
 	    case T_DEBUG:	/* single_step */
 	    {
-	    	extern int dr_addr[];
 		int addr;
-	    	int status = dr6();
+	    	int status = get_dr6();
 
 		if (status & 0xf) {	/* hmm hdw break */
-			addr =	status & 0x8 ? dr_addr[3] :
-				status & 0x4 ? dr_addr[2] :
-				status & 0x2 ? dr_addr[1] :
-					       dr_addr[0];
+			addr =	status & 0x8 ? get_dr3() :
+				status & 0x4 ? get_dr2() :
+				status & 0x2 ? get_dr1() :
+					       get_dr0();
 			regs->efl |= EFL_RF;
 			db_single_step_cmd(addr, 0, 1, "p");
 		}
diff --git a/i386/i386/db_interface.h b/i386/i386/db_interface.h
index 10a02e2..a8dfdce 100644
--- a/i386/i386/db_interface.h
+++ b/i386/i386/db_interface.h
@@ -53,6 +53,12 @@ extern boolean_t db_phys_eq (
 	task_t		task2,
 	vm_offset_t	addr2);
 
+extern int db_user_to_kernel_address(
+	task_t		task,
+	vm_offset_t	addr,
+	unsigned int	*kaddr,
+	int		flag);
+
 extern void db_task_name (task_t task);
 
 #define I386_DB_TYPE_X 0
@@ -67,9 +73,26 @@ extern void db_task_name (task_t task);
 #define I386_DB_LOCAL 1
 #define I386_DB_GLOBAL 2
 
-extern unsigned long dr0 (vm_offset_t linear_addr, int type, int len, int persistence);
-extern unsigned long dr1 (vm_offset_t linear_addr, int type, int len, int persistence);
-extern unsigned long dr2 (vm_offset_t linear_addr, int type, int len, int persistence);
-extern unsigned long dr3 (vm_offset_t linear_addr, int type, int len, int persistence);
+extern void db_set_hw_watchpoint(
+	int		num,
+	task_t		task,
+	vm_offset_t	addr,
+	vm_size_t	size);
+
+extern void db_dr (
+	int		num,
+	vm_offset_t	linear_addr,
+	int		type,
+	int		len,
+	int		persistence);
+
+extern void db_get_debug_state(
+	pcb_t pcb,
+	struct i386_debug_state *state);
+extern kern_return_t db_set_debug_state(
+	pcb_t pcb,
+	const struct i386_debug_state *state);
+
+extern void db_load_context(pcb_t pcb);
 
 #endif /* _I386_DB_INTERFACE_H_ */
diff --git a/i386/i386/locore.S b/i386/i386/locore.S
index 9aa8485..e1befa7 100644
--- a/i386/i386/locore.S
+++ b/i386/i386/locore.S
@@ -1411,128 +1411,6 @@ _inst_fetch_fault:
 
 
 
-ENTRY(dr6)
-#ifdef MACH_RING1
-	pushl	%ebx
-	movl	$6, %ebx
-	call	__hyp_get_debugreg
-	popl	%ebx
-#else /* MACH_RING1 */
-	movl	%db6, %eax
-#endif /* MACH_RING1 */
-	ret
-
-/*	dr<i>(address, type, len, persistence)
- */
-ENTRY(dr0)
-	movl	S_ARG0, %eax
-	movl	%eax,EXT(dr_addr)
-#ifdef MACH_RING1
-	pushl	%ebx
-	movl	$0,%ebx
-	movl	%eax,%ecx
-	call	__hyp_set_debugreg
-#else /* MACH_RING1 */
-	movl	%eax, %db0
-#endif /* MACH_RING1 */
-	movl	$0, %ecx
-	jmp	0f
-ENTRY(dr1)
-	movl	S_ARG0, %eax
-	movl	%eax,EXT(dr_addr)+1*4
-#ifdef MACH_RING1
-	pushl	%ebx
-	movl	$1,%ebx
-	movl	%eax,%ecx
-	call	__hyp_set_debugreg
-#else /* MACH_RING1 */
-	movl	%eax, %db1
-#endif /* MACH_RING1 */
-	movl	$2, %ecx
-	jmp	0f
-ENTRY(dr2)
-	movl	S_ARG0, %eax
-	movl	%eax,EXT(dr_addr)+2*4
-#ifdef MACH_RING1
-	pushl	%ebx
-	movl	$2,%ebx
-	movl	%eax,%ecx
-	call	__hyp_set_debugreg
-#else /* MACH_RING1 */
-	movl	%eax, %db2
-#endif /* MACH_RING1 */
-	movl	$4, %ecx
-	jmp	0f
-
-ENTRY(dr3)
-	movl	S_ARG0, %eax
-	movl	%eax,EXT(dr_addr)+3*4
-#ifdef MACH_RING1
-	pushl	%ebx
-	movl	$3,%ebx
-	movl	%eax,%ecx
-	call	__hyp_set_debugreg
-#else /* MACH_RING1 */
-	movl	%eax, %db3
-#endif /* MACH_RING1 */
-	movl	$6, %ecx
-
-0:
-	pushl	%ebp
-	movl	%esp, %ebp
-
-#ifdef MACH_RING1
-	movl	$7,%ebx
-	call	__hyp_get_debugreg
-	movl	%eax, %edx
-#else /* MACH_RING1 */
-	movl	%db7, %edx
-#endif /* MACH_RING1 */
-	movl	%edx,EXT(dr_addr)+4*4
-	andl	dr_msk(,%ecx,2),%edx	/* clear out new entry */
-	movl	%edx,EXT(dr_addr)+5*4
-	movzbl	B_ARG3, %eax
-	andb	$3, %al
-	shll	%cl, %eax
-	orl	%eax, %edx
-
-	movzbl	B_ARG1, %eax
-	andb	$3, %al
-	addb	%cl, %cl
-	addb	$0x10, %cl
-	shll	%cl, %eax
-	orl	%eax, %edx
-
-	movzbl	B_ARG2, %eax
-	andb	$3, %al
-	addb	$0x2, %cl
-	shll	%cl, %eax
-	orl	%eax, %edx
-
-#ifdef MACH_RING1
-	movl	$7,%ebx
-	movl	%edx, %ecx
-	call	__hyp_set_debugreg
-	popl	%ebx
-#else /* MACH_RING1 */
-	movl	%edx, %db7
-#endif /* MACH_RING1 */
-	movl	%edx,EXT(dr_addr)+7*4
-	movl	%edx, %eax
-	leave
-	ret
-
-	.data
-dr_msk:
-	.long	~0x000f0003
-	.long	~0x00f0000c
-	.long	~0x0f000030
-	.long	~0xf00000c0
-ENTRY(dr_addr)
-	.long	0,0,0,0
-	.long	0,0,0,0
-	.text
-
 /*
  * cpu_shutdown()
  * Force reboot
diff --git a/i386/i386/pcb.c b/i386/i386/pcb.c
index 7d632a6..97ec00e 100644
--- a/i386/i386/pcb.c
+++ b/i386/i386/pcb.c
@@ -46,6 +46,7 @@
 #include <i386/proc_reg.h>
 #include <i386/seg.h>
 #include <i386/user_ldt.h>
+#include <i386/db_interface.h>
 #include <i386/fpu.h>
 #include "eflags.h"
 #include "gdt.h"
@@ -215,6 +216,8 @@ void switch_ktss(pcb)
         pcb->ims.user_gdt, sizeof pcb->ims.user_gdt);
 #endif /* MACH_PV_DESCRIPTORS */
 
+	db_load_context(pcb);
+
 	/*
 	 * Load the floating-point context, if necessary.
 	 */
@@ -621,6 +624,21 @@ kern_return_t thread_setstatus(thread, flavor, tstate, count)
 		break;
 	    }
 
+	    case i386_DEBUG_STATE:
+	    {
+		register struct i386_debug_state *state;
+		kern_return_t ret;
+
+		if (count < i386_DEBUG_STATE_COUNT)
+		    return KERN_INVALID_ARGUMENT;
+
+		state = (struct i386_debug_state *) tstate;
+		ret = db_set_debug_state(thread->pcb, state);
+		if (ret)
+			return ret;
+		break;
+	    }
+
 	    default:
 		return(KERN_INVALID_ARGUMENT);
 	}
@@ -760,6 +778,20 @@ kern_return_t thread_getstatus(thread, flavor, tstate, count)
 		break;
 	    }
 
+	    case i386_DEBUG_STATE:
+	    {
+		register struct i386_debug_state *state;
+
+		if (*count < i386_DEBUG_STATE_COUNT)
+		    return KERN_INVALID_ARGUMENT;
+
+		state = (struct i386_debug_state *) tstate;
+		db_get_debug_state(thread->pcb, state);
+
+		*count = i386_DEBUG_STATE_COUNT;
+		break;
+	    }
+
 	    default:
 		return(KERN_INVALID_ARGUMENT);
 	}
diff --git a/i386/i386/proc_reg.h b/i386/i386/proc_reg.h
index f1b2c89..8083536 100644
--- a/i386/i386/proc_reg.h
+++ b/i386/i386/proc_reg.h
@@ -234,6 +234,132 @@ extern unsigned long cr3;
 	asm("jmp 0f\n" \
             "0:\n")
 
+#ifdef MACH_RING1
+#define get_dr0() hyp_get_debugreg(0)
+#else
+#define	get_dr0() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr0, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr0(value) hyp_set_debugreg(0, value)
+#else
+#define	set_dr0(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr0" : : "r" (_temp__)); \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define get_dr1() hyp_get_debugreg(1)
+#else
+#define	get_dr1() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr1, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr1(value) hyp_set_debugreg(1, value)
+#else
+#define	set_dr1(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr1" : : "r" (_temp__)); \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define get_dr2() hyp_get_debugreg(2)
+#else
+#define	get_dr2() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr2, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr2(value) hyp_set_debugreg(2, value)
+#else
+#define	set_dr2(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr2" : : "r" (_temp__)); \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define get_dr3() hyp_get_debugreg(3)
+#else
+#define	get_dr3() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr3, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr3(value) hyp_set_debugreg(3, value)
+#else
+#define	set_dr3(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr3" : : "r" (_temp__)); \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define get_dr6() hyp_get_debugreg(6)
+#else
+#define	get_dr6() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr6, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr6(value) hyp_set_debugreg(6, value)
+#else
+#define	set_dr6(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr6" : : "r" (_temp__)); \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define get_dr7() hyp_get_debugreg(7)
+#else
+#define	get_dr7() \
+    ({ \
+	register unsigned long _temp__; \
+	asm volatile("movl %%dr7, %0" : "=r" (_temp__)); \
+	_temp__; \
+    })
+#endif
+
+#ifdef MACH_RING1
+#define set_dr7(value) hyp_set_debugreg(7, value)
+#else
+#define	set_dr7(value) \
+    ({ \
+	register unsigned long _temp__ = (value); \
+	asm volatile("movl %0,%%dr7" : : "r" (_temp__)); \
+    })
+#endif
+
 #endif	/* __GNUC__ */
 #endif	/* __ASSEMBLER__ */
 
diff --git a/i386/i386/thread.h b/i386/i386/thread.h
index eddd25c..450ec55 100644
--- a/i386/i386/thread.h
+++ b/i386/i386/thread.h
@@ -36,6 +36,7 @@
 #include <mach/boolean.h>
 #include <mach/machine/vm_types.h>
 #include <mach/machine/fp_reg.h>
+#include <mach/machine/thread_status.h>
 
 #include <kern/lock.h>
 
@@ -168,6 +169,7 @@ struct i386_machine_state {
 	struct i386_fpsave_state *ifps;
 	struct v86_assist_state	v86s;
 	struct real_descriptor user_gdt[USER_GDT_SLOTS];
+	struct i386_debug_state ids;
 };
 
 typedef struct pcb {
diff --git a/i386/include/mach/i386/thread_status.h b/i386/include/mach/i386/thread_status.h
index 5f20355..ba1e3de 100644
--- a/i386/include/mach/i386/thread_status.h
+++ b/i386/include/mach/i386/thread_status.h
@@ -56,6 +56,7 @@
 #define i386_ISA_PORT_MAP_STATE	3
 #define	i386_V86_ASSIST_STATE	4
 #define	i386_REGS_SEGS_STATE	5
+#define	i386_DEBUG_STATE	6
 
 /*
  * This structure is used for both
@@ -144,4 +145,10 @@ struct v86_interrupt_table {
 #define	i386_V86_ASSIST_STATE_COUNT \
 	    (sizeof(struct i386_v86_assist_state)/sizeof(unsigned int))
 
+struct i386_debug_state {
+	unsigned int	dr[8];
+};
+#define	i386_DEBUG_STATE_COUNT \
+	    (sizeof(struct i386_debug_state)/sizeof(unsigned int))
+
 #endif	/* _MACH_I386_THREAD_STATUS_H_ */