add setting gs/fsbase

* i386/i386/i386asm.sym: add offsets for asm
* i386/i386/pcb.c: switch FSBASE/GSBASE on context switch and
  implement accessors in thread setstatus/getstatus
* i386/i386/thread.h: add new state to thread saved state
* kern/thread.c: add i386_FSGS_BASE_STATE handler
* x86_64/locore.S: fix fs/gs handling, skipping the base address and
  avoid resetting it by manually re-loading fs/gs
Message-Id: <20230419194703.410575-5-luca@orpolo.org>

5 files changed, 116 insertions, 21 deletions

diff --git a/i386/i386/i386asm.sym b/i386/i386/i386asm.sym
index 1b9b40bb..fd0be557 100644
--- a/i386/i386/i386asm.sym
+++ b/i386/i386/i386asm.sym
@@ -108,6 +108,8 @@ offset	i386_saved_state	r	r12
 offset	i386_saved_state	r	r13
 offset	i386_saved_state	r	r14
 offset	i386_saved_state	r	r15
+offset	i386_saved_state	r	fsbase
+offset	i386_saved_state	r	gsbase
 #endif
 
 offset	i386_interrupt_state	i	eip
diff --git a/i386/i386/pcb.c b/i386/i386/pcb.c
index 61125fe8..8a9e3bf4 100644
--- a/i386/i386/pcb.c
+++ b/i386/i386/pcb.c
@@ -51,6 +51,7 @@
 #include "eflags.h"
 #include "gdt.h"
 #include "ldt.h"
+#include "msr.h"
 #include "ktss.h"
 #include "pcb.h"
 
@@ -372,7 +373,10 @@ thread_t switch_context(
 	 *	Load the rest of the user state for the new thread
 	 */
 	switch_ktss(new->pcb);
-
+#if defined(__x86_64__) && !defined(USER32)
+        wrmsr(MSR_REG_FSBASE, new->pcb->iss.fsbase);
+        wrmsr(MSR_REG_GSBASE, new->pcb->iss.gsbase);
+#endif
 	return Switch_context(old, continuation, new);
 }
 
@@ -667,7 +671,23 @@ kern_return_t thread_setstatus(
 			return ret;
 		break;
 	    }
-
+#if defined(__x86_64__) && !defined(USER32)
+	    case i386_FSGS_BASE_STATE:
+            {
+                    struct i386_fsgs_base_state *state;
+                    if (count < i386_FSGS_BASE_STATE_COUNT)
+                            return KERN_INVALID_ARGUMENT;
+
+                    state = (struct i386_fsgs_base_state *) tstate;
+                    thread->pcb->iss.fsbase = state->fs_base;
+                    thread->pcb->iss.gsbase = state->gs_base;
+                    if (thread == current_thread()) {
+                            wrmsr(MSR_REG_FSBASE, state->fs_base);
+                            wrmsr(MSR_REG_GSBASE, state->gs_base);
+                    }
+                    break;
+            }
+#endif
 	    default:
 		return(KERN_INVALID_ARGUMENT);
 	}
@@ -843,7 +863,20 @@ kern_return_t thread_getstatus(
 		*count = i386_DEBUG_STATE_COUNT;
 		break;
 	    }
-
+#if defined(__x86_64__) && !defined(USER32)
+	    case i386_FSGS_BASE_STATE:
+            {
+                    struct i386_fsgs_base_state *state;
+                    if (*count < i386_FSGS_BASE_STATE_COUNT)
+                            return KERN_INVALID_ARGUMENT;
+
+                    state = (struct i386_fsgs_base_state *) tstate;
+                    state->fs_base = thread->pcb->iss.fsbase;
+                    state->fs_base = thread->pcb->iss.gsbase;
+                    *count = i386_FSGS_BASE_STATE_COUNT;
+                    break;
+            }
+#endif
 	    default:
 		return(KERN_INVALID_ARGUMENT);
 	}
diff --git a/i386/i386/thread.h b/i386/i386/thread.h
index 933b43d8..b5fc5ffb 100644
--- a/i386/i386/thread.h
+++ b/i386/i386/thread.h
@@ -51,6 +51,10 @@
  */
 
 struct i386_saved_state {
+#ifdef __x86_64__
+	unsigned long	fsbase;
+	unsigned long	gsbase;
+#endif
 	unsigned long	gs;
 	unsigned long	fs;
 	unsigned long	es;
diff --git a/kern/thread.c b/kern/thread.c
index 20c11024..a324eed7 100644
--- a/kern/thread.c
+++ b/kern/thread.c
@@ -1472,6 +1472,9 @@ kern_return_t thread_set_state(
 	if (flavor == i386_DEBUG_STATE && thread == current_thread())
 		/* This state can be set directly for the curren thread.  */
 		return thread_setstatus(thread, flavor, new_state, new_state_count);
+	if (flavor == i386_FSGS_BASE_STATE && thread == current_thread())
+		/* This state can be set directly for the curren thread.  */
+		return thread_setstatus(thread, flavor, new_state, new_state_count);
 #endif
 
 	if (thread == THREAD_NULL || thread == current_thread())
diff --git a/x86_64/locore.S b/x86_64/locore.S
index 0d7cdd0e..366ef292 100644
--- a/x86_64/locore.S
+++ b/x86_64/locore.S
@@ -34,6 +34,7 @@
 #include <i386/i386/trap.h>
 #include <i386/i386/seg.h>
 #include <i386/i386/ldt.h>
+#include <i386/i386/msr.h>
 #include <i386/i386/i386asm.h>
 #include <i386/i386/cpu_number.h>
 #include <i386/i386/xen.h>
@@ -41,6 +42,46 @@
 #define pusha pushq %rax ; pushq %rcx ; pushq %rdx ; pushq %rbx ; subq $8,%rsp ; pushq %rbp ; pushq %rsi ; pushq %rdi ; pushq %r8 ; pushq %r9 ; pushq %r10 ; pushq %r11 ; pushq %r12 ; pushq %r13 ; pushq %r14 ; pushq %r15
 #define popa popq %r15 ; popq %r14 ; popq %r13 ; popq %r12 ; popq %r11 ; popq %r10 ; popq %r9 ; popq %r8 ; popq %rdi ; popq %rsi ; popq %rbp ; addq $8,%rsp ; popq %rbx ; popq %rdx ; popq %rcx ; popq %rax
 
+#ifdef USER32
+#define PUSH_FSGS		\
+	pushq	%fs		;\
+	pushq	%gs		;\
+	subq	$16,%rsp
+#else
+#define PUSH_FSGS		\
+	subq	$32,%rsp
+#endif
+
+#ifdef USER32
+#define POP_FSGS		\
+	popq	%gs		;\
+	popq	%fs		;\
+	addq	$16,%rsp
+#else
+#define POP_FSGS		\
+	addq	$32,%rsp
+#endif
+
+#ifdef USER32
+#define PUSH_FSGS_ISR		\
+	pushq	%fs		;\
+	pushq	%gs
+#else
+#define PUSH_FSGS_ISR		\
+	subq	$16,%rsp
+#endif
+
+#ifdef USER32
+#define POP_FSGS_ISR		\
+	popq	%gs		;\
+	popq	%fs
+#else
+#define POP_FSGS_ISR		\
+	addq	$16,%rsp
+#endif
+
+
+
 /*
  * Fault recovery.
  */
@@ -405,6 +446,11 @@ push_fs:
 	pushq	%fs			/* restore fs, */
 push_gs:
 	pushq	%gs			/* restore gs. */
+#ifdef USER32
+push_gsbase:
+	pushq	$0
+	pushq	$0
+#endif
 push_segregs:
 	movq	%rax,R_TRAPNO(%rsp)	/* set trap number */
 	movq	%rdx,R_ERR(%rsp)	/* set error code */
@@ -470,8 +516,7 @@ trap_push_segs:
 	pushq	%rax
 	movq	%es,%rax			/* and the segment registers */
 	pushq	%rax
-	pushq	%fs
-	pushq	%gs
+	PUSH_FSGS
 
 	/* Note that we have to load the segment registers
 	   even if this is a trap from the kernel,
@@ -480,9 +525,10 @@ trap_push_segs:
 	mov	%ss,%ax			/* switch to kernel data segment */
 	mov	%ax,%ds			/* (same as kernel stack segment) */
 	mov	%ax,%es
+#ifdef USER32
 	mov	%ax,%fs
 	mov	%ax,%gs
-
+#endif
 trap_set_segs:
 	cld				/* clear direction flag */
 	testl	$(EFL_VM),R_EFLAGS(%rsp) /* in V86 mode? */
@@ -536,10 +582,17 @@ _return_to_user:
  */
 
 _return_from_kernel:
+	addq	$16,%rsp		/* skip FS/GS base */
+#ifndef USER32
+_kret_popl_gs:
+_kret_popl_fs:
+	addq	$16,%rsp		/* skip FS/GS selector */
+#else
 _kret_popl_gs:
 	popq	%gs			/* restore segment registers */
 _kret_popl_fs:
 	popq	%fs
+#endif
 _kret_popl_es:
 	popq	%rax
 	movq	%rax,%es
@@ -695,14 +748,15 @@ ENTRY(all_intrs)
 	pushq	%rdx
 	movq	%es,%rdx
 	pushq	%rdx
-	pushq	%fs
-	pushq	%gs
+	PUSH_FSGS_ISR
+
 	mov	%ss,%dx			/* switch to kernel segments */
 	mov	%dx,%ds
 	mov	%dx,%es
+#ifdef USER32
 	mov	%dx,%fs
 	mov	%dx,%gs
-
+#endif
 	CPU_NUMBER(%edx)
 
 	movq	CX(EXT(int_stack_top),%edx),%rcx
@@ -741,8 +795,7 @@ LEXT(return_to_iret)			/* ( label for kdb_kintr and hardclock) */
 	cmpq	$0,CX(EXT(need_ast),%edx)
 	jnz	ast_from_interrupt	/* take it if so */
 1:
-	pop	%gs			/* restore segment regs */
-	pop	%fs
+	POP_FSGS_ISR
 	pop	%rdx
 	mov	%rdx,%es
 	pop	%rdx
@@ -796,9 +849,7 @@ stack_overflowed:
  *	ss
  */
 ast_from_interrupt:
-	pop	%gs			/* restore all registers ... */
-	pop	%fs
-	pop	%rdx
+	POP_FSGS
 	mov	%rdx,%es
 	pop	%rdx
 	mov	%rdx,%ds
@@ -818,14 +869,15 @@ ast_from_interrupt:
 	push	%rdx
 	mov	%es,%rdx
 	push	%rdx
-	push	%fs
-	push	%gs
+	PUSH_FSGS
+
 	mov	%ss,%dx			/* switch to kernel segments */
 	mov	%dx,%ds
 	mov	%dx,%es
+#ifdef USER32
 	mov	%dx,%fs
 	mov	%dx,%gs
-
+#endif
 	CPU_NUMBER(%edx)
 	TIME_TRAP_UENTRY
 
@@ -947,14 +999,12 @@ kdb_from_iret_i:			/* on interrupt stack */
 	push	%rdx
 	mov	%es,%rdx
 	push	%rdx
-	push	%fs
-	push	%gs
+	PUSH_FSGS
 	movq	%rsp,%rdx		/* pass regs, */
 	movq	$0,%rsi			/* code, */
 	movq	$-1,%rdi		/* type to kdb */
 	call	EXT(kdb_trap)
-	pop	%gs			/* restore segment registers */
-	pop	%fs
+	POP_FSGS
 	pop	%rdx
 	mov	%rdx,%es
 	pop	%rdx
@@ -1039,6 +1089,7 @@ ttd_from_iret_i:			/* on interrupt stack */
 	push	%rdx
 	push	%fs
 	push	%gs
+	ud2	// TEST it
 	movq	%rsp,%rdx		/* pass regs, */
 	movq	$0,%rsi			/* code, */
 	movq	$-1,%rdi		/* type to kdb */
@@ -1087,6 +1138,8 @@ syscall_entry_2:
 	pushq	%rdx
 	pushq	%fs
 	pushq	%gs
+	pushq	$0	// gsbase
+	pushq	$0	// fsbase
 
 	mov	%ss,%dx			/* switch to kernel data segment */
 	mov	%dx,%ds