svn commit: r190620 - in head/sys: amd64/acpica amd64/amd64
amd64/ia32 amd64/include amd64/linux32 conf
Konstantin Belousov
kib at FreeBSD.org
Wed Apr 1 06:09:27 PDT 2009
Author: kib
Date: Wed Apr 1 13:09:26 2009
New Revision: 190620
URL: http://svn.freebsd.org/changeset/base/190620
Log:
Save and restore segment registers on amd64 when entering and leaving
the kernel on amd64. Fill and read segment registers for mcontext and
signals. Handle traps caused by restoration of the
invalidated selectors.
Implement user-mode creation and manipulation of the process-specific
LDT descriptors for amd64, see sysarch(2).
Implement support for TSS i/o port access permission bitmap for amd64.
Context-switch LDT and TSS. Do not save and restore segment registers on
the context switch, that is handled by kernel enter/leave trampolines
now. Remove segment restore code from the signal trampolines for
freebsd/amd64, freebsd/ia32 and linux/i386 for the same reason.
Implement amd64-specific compat shims for sysarch.
Linuxolator (temporary ?) switched to use gsbase for thread_area pointer.
TODO:
Currently, gdb is not adapted to show segment registers from struct reg.
Also, no machine-depended ptrace command is added to set segment
registers for debugged process.
In collaboration with: pho
Discussed with: peter
Reviewed by: jhb
Linuxolator tested by: dchagin
Added:
head/sys/amd64/ia32/ia32_misc.c (contents, props changed)
Modified:
head/sys/amd64/acpica/acpi_switch.S
head/sys/amd64/amd64/apic_vector.S
head/sys/amd64/amd64/cpu_switch.S
head/sys/amd64/amd64/db_interface.c
head/sys/amd64/amd64/db_trace.c
head/sys/amd64/amd64/exception.S
head/sys/amd64/amd64/genassym.c
head/sys/amd64/amd64/machdep.c
head/sys/amd64/amd64/mp_machdep.c
head/sys/amd64/amd64/sys_machdep.c
head/sys/amd64/amd64/trap.c
head/sys/amd64/amd64/vm_machdep.c
head/sys/amd64/ia32/ia32_exception.S
head/sys/amd64/ia32/ia32_reg.c
head/sys/amd64/ia32/ia32_signal.c
head/sys/amd64/ia32/ia32_sigtramp.S
head/sys/amd64/include/asmacros.h
head/sys/amd64/include/frame.h
head/sys/amd64/include/md_var.h
head/sys/amd64/include/pcb.h
head/sys/amd64/include/pcpu.h
head/sys/amd64/include/proc.h
head/sys/amd64/include/segments.h
head/sys/amd64/include/sysarch.h
head/sys/amd64/linux32/linux32_locore.s
head/sys/amd64/linux32/linux32_machdep.c
head/sys/amd64/linux32/linux32_sysvec.c
head/sys/conf/files.amd64
Modified: head/sys/amd64/acpica/acpi_switch.S
==============================================================================
--- head/sys/amd64/acpica/acpi_switch.S Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/acpica/acpi_switch.S Wed Apr 1 13:09:26 2009 (r190620)
@@ -64,12 +64,17 @@ ENTRY(acpi_restorecpu)
/* Fetch PCB. */
movq WAKEUP_CTX(xpcb), %r11
- /* Restore segment registers. */
- mov WAKEUP_PCB(DS), %ds
- mov WAKEUP_PCB(ES), %es
- mov WAKEUP_XPCB(SS), %ss
- mov WAKEUP_PCB(FS), %fs
- mov WAKEUP_PCB(GS), %gs
+ /* Force kernel segment registers. */
+ movl $KDSEL, %eax
+ movw %ax, %ds
+ movl $KDSEL, %eax
+ movw %ax, %es
+ movl $KDSEL, %eax
+ movw %ax, %ss
+ movl $KUF32SEL, %eax
+ movw %ax, %fs
+ movl $KUG32SEL, %eax
+ movw %ax, %gs
movl $MSR_FSBASE, %ecx
movl WAKEUP_PCB(FSBASE), %eax
Modified: head/sys/amd64/amd64/apic_vector.S
==============================================================================
--- head/sys/amd64/amd64/apic_vector.S Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/apic_vector.S Wed Apr 1 13:09:26 2009 (r190620)
@@ -219,9 +219,7 @@ IDTVEC(cpustop)
movl $0, LA_EOI(%rax) /* End Of Interrupt to APIC */
call cpustop_handler
-
- POP_FRAME
- iretq
+ jmp doreti
/*
* Executed by a CPU when it receives an IPI_SUSPEND from another CPU.
@@ -251,6 +249,5 @@ IDTVEC(rendezvous)
call smp_rendezvous_action
movq lapic, %rax
movl $0, LA_EOI(%rax) /* End Of Interrupt to APIC */
- POP_FRAME /* Why not doreti? */
- iretq
+ jmp doreti
#endif /* SMP */
Modified: head/sys/amd64/amd64/cpu_switch.S
==============================================================================
--- head/sys/amd64/amd64/cpu_switch.S Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/cpu_switch.S Wed Apr 1 13:09:26 2009 (r190620)
@@ -75,8 +75,6 @@ ENTRY(cpu_throw)
1:
movq TD_PCB(%rdi),%r8 /* Old pcb */
movl PCPU(CPUID), %eax
- movq PCB_FSBASE(%r8),%r9
- movq PCB_GSBASE(%r8),%r10
/* release bit from old pm_active */
movq TD_PROC(%rdi), %rdx /* oldtd->td_proc */
movq P_VMSPACE(%rdx), %rdx /* proc->p_vmspace */
@@ -110,28 +108,6 @@ ENTRY(cpu_switch)
movq %rbx,PCB_RBX(%r8)
movq %rax,PCB_RIP(%r8)
- /*
- * Reread fs and gs bases. Explicit fs segment register load
- * by the usermode code may change actual fs base without
- * updating pcb_{fs,gs}base.
- *
- * %rdx still contains the mtx, save %rdx around rdmsr.
- */
- movq %rdx,%r11
- movl $MSR_FSBASE,%ecx
- rdmsr
- shlq $32,%rdx
- leaq (%rax,%rdx),%r9
- movl $MSR_KGSBASE,%ecx
- rdmsr
- shlq $32,%rdx
- leaq (%rax,%rdx),%r10
- movq %r11,%rdx
-
- testl $PCB_32BIT,PCB_FLAGS(%r8)
- jnz store_seg
-done_store_seg:
-
testl $PCB_DBREGS,PCB_FLAGS(%r8)
jnz store_dr /* static predict not taken */
done_store_dr:
@@ -192,36 +168,47 @@ sw1:
testl $TDP_KTHREAD,TD_PFLAGS(%rsi)
jnz do_kthread
- testl $PCB_32BIT,PCB_FLAGS(%r8)
- jnz load_seg
-done_load_seg:
-
- cmpq PCB_FSBASE(%r8),%r9
- jz 1f
- /* Restore userland %fs */
-restore_fsbase:
- movl $MSR_FSBASE,%ecx
+ /*
+ * Load ldt register
+ */
+ movq TD_PROC(%rsi),%rcx
+ cmpq $0, P_MD+MD_LDT(%rcx)
+ jne do_ldt
+ xorl %eax,%eax
+ld_ldt: lldt %ax
+
+ /* Restore fs base in GDT */
movl PCB_FSBASE(%r8),%eax
- movl PCB_FSBASE+4(%r8),%edx
- wrmsr
-1:
- cmpq PCB_GSBASE(%r8),%r10
- jz 2f
- /* Restore userland %gs */
- movl $MSR_KGSBASE,%ecx
+ movq PCPU(FS32P),%rdx
+ movw %ax,2(%rdx)
+ shrl $16,%eax
+ movb %al,4(%rdx)
+ shrl $8,%eax
+ movb %al,7(%rdx)
+
+ /* Restore gs base in GDT */
movl PCB_GSBASE(%r8),%eax
- movl PCB_GSBASE+4(%r8),%edx
- wrmsr
-2:
+ movq PCPU(GS32P),%rdx
+ movw %ax,2(%rdx)
+ shrl $16,%eax
+ movb %al,4(%rdx)
+ shrl $8,%eax
+ movb %al,7(%rdx)
-do_tss:
+do_kthread:
+ /* Do we need to reload tss ? */
+ movq PCPU(TSSP),%rax
+ movq PCB_TSSP(%r8),%rdx
+ testq %rdx,%rdx
+ cmovzq PCPU(COMMONTSSP),%rdx
+ cmpq %rax,%rdx
+ jne do_tss
+done_tss:
+ movq %r8,PCPU(RSP0)
+ movq %r8,PCPU(CURPCB)
/* Update the TSS_RSP0 pointer for the next interrupt */
- movq PCPU(TSSP), %rax
- movq %r8, PCPU(RSP0)
- movq %r8, PCPU(CURPCB)
- addq $COMMON_TSS_RSP0, %rax
- movq %rsi, PCPU(CURTHREAD) /* into next thread */
- movq %r8, (%rax)
+ movq %r8,COMMON_TSS_RSP0(%rdx)
+ movq %rsi,PCPU(CURTHREAD) /* into next thread */
/* Test if debug registers should be restored. */
testl $PCB_DBREGS,PCB_FLAGS(%r8)
@@ -250,45 +237,6 @@ done_load_dr:
* We use jumps rather than call in order to avoid the stack.
*/
-do_kthread:
- /*
- * Copy old fs/gsbase to new kthread pcb for future switches
- * This maintains curpcb->pcb_[fg]sbase as caches of the MSR
- */
- movq %r9,PCB_FSBASE(%r8)
- movq %r10,PCB_GSBASE(%r8)
- jmp do_tss
-
-store_seg:
- mov %gs,PCB_GS(%r8)
- testl $PCB_GS32BIT,PCB_FLAGS(%r8)
- jnz 2f
-1: mov %ds,PCB_DS(%r8)
- mov %es,PCB_ES(%r8)
- mov %fs,PCB_FS(%r8)
- jmp done_store_seg
-2: movq PCPU(GS32P),%rax
- movq (%rax),%rax
- movq %rax,PCB_GS32SD(%r8)
- jmp 1b
-
-load_seg:
- testl $PCB_GS32BIT,PCB_FLAGS(%r8)
- jnz 2f
-1: movl $MSR_GSBASE,%ecx
- rdmsr
- mov PCB_GS(%r8),%gs
- wrmsr
- mov PCB_DS(%r8),%ds
- mov PCB_ES(%r8),%es
- mov PCB_FS(%r8),%fs
- jmp restore_fsbase
- /* Restore userland %gs while preserving kernel gsbase */
-2: movq PCPU(GS32P),%rax
- movq PCB_GS32SD(%r8),%rcx
- movq %rcx,(%rax)
- jmp 1b
-
store_dr:
movq %dr7,%rax /* yes, do the save */
movq %dr0,%r15
@@ -325,6 +273,29 @@ load_dr:
movq %r11,%dr6
movq %rax,%dr7
jmp done_load_dr
+
+do_tss: movq %rdx,PCPU(TSSP)
+ movq %rdx,%rcx
+ movq PCPU(TSS),%rax
+ movw %rcx,2(%rax)
+ shrq $16,%rcx
+ movb %cl,4(%rax)
+ shrq $8,%rcx
+ movb %cl,7(%rax)
+ shrq $8,%rcx
+ movl %ecx,8(%rax)
+ movb $0x89,5(%rax) /* unset busy */
+ movl $TSSSEL,%eax
+ ltr %ax
+ jmp done_tss
+
+do_ldt: movq PCPU(LDT),%rax
+ movq P_MD+MD_LDT_SD(%rcx),%rdx
+ movq %rdx,(%rax)
+ movq P_MD+MD_LDT_SD+8(%rcx),%rdx
+ movq %rdx,8(%rax)
+ movl $LDTSEL,%eax
+ jmp ld_ldt
END(cpu_switch)
/*
@@ -398,12 +369,6 @@ ENTRY(savectx2)
movq (%rsp),%rax
movq %rax,PCB_RIP(%r8)
- mov %ds,PCB_DS(%r8)
- mov %es,PCB_ES(%r8)
- mov %ss,XPCB_SS(%r8)
- mov %fs,PCB_FS(%r8)
- mov %gs,PCB_GS(%r8)
-
movq %rbx,PCB_RBX(%r8)
movq %rsp,PCB_RSP(%r8)
movq %rbp,PCB_RBP(%r8)
Modified: head/sys/amd64/amd64/db_interface.c
==============================================================================
--- head/sys/amd64/amd64/db_interface.c Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/db_interface.c Wed Apr 1 13:09:26 2009 (r190620)
@@ -139,7 +139,11 @@ void
db_show_mdpcpu(struct pcpu *pc)
{
-#if 0
- db_printf("currentldt = 0x%x\n", pc->pc_currentldt);
-#endif
+ db_printf("curpmap = %p\n", pc->pc_curpmap);
+ db_printf("tssp = %p\n", pc->pc_tssp);
+ db_printf("commontssp = %p\n", pc->pc_commontssp);
+ db_printf("rsp0 = 0x%lx\n", pc->pc_rsp0);
+ db_printf("gs32p = %p\n", pc->pc_gs32p);
+ db_printf("ldt = %p\n", pc->pc_ldt);
+ db_printf("tss = %p\n", pc->pc_tss);
}
Modified: head/sys/amd64/amd64/db_trace.c
==============================================================================
--- head/sys/amd64/amd64/db_trace.c Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/db_trace.c Wed Apr 1 13:09:26 2009 (r190620)
@@ -69,12 +69,10 @@ static db_varfcn_t db_ss;
#define DB_OFFSET(x) (db_expr_t *)offsetof(struct trapframe, x)
struct db_variable db_regs[] = {
{ "cs", DB_OFFSET(tf_cs), db_frame },
-#if 0
{ "ds", DB_OFFSET(tf_ds), db_frame },
{ "es", DB_OFFSET(tf_es), db_frame },
{ "fs", DB_OFFSET(tf_fs), db_frame },
{ "gs", DB_OFFSET(tf_gs), db_frame },
-#endif
{ "ss", NULL, db_ss },
{ "rax", DB_OFFSET(tf_rax), db_frame },
{ "rcx", DB_OFFSET(tf_rcx), db_frame },
@@ -94,7 +92,7 @@ struct db_variable db_regs[] = {
{ "r15", DB_OFFSET(tf_r15), db_frame },
{ "rip", DB_OFFSET(tf_rip), db_frame },
{ "rflags", DB_OFFSET(tf_rflags), db_frame },
-#define DB_N_SHOW_REGS 20 /* Don't show registers after here. */
+#define DB_N_SHOW_REGS 24 /* Don't show registers after here. */
{ "dr0", NULL, db_dr0 },
{ "dr1", NULL, db_dr1 },
{ "dr2", NULL, db_dr2 },
@@ -357,7 +355,7 @@ db_nextframe(struct amd64_frame **fp, db
rbp = tf->tf_rbp;
switch (frame_type) {
case TRAP:
- db_printf("--- trap %#lr", tf->tf_trapno);
+ db_printf("--- trap %#r", tf->tf_trapno);
break;
case SYSCALL:
db_printf("--- syscall");
Modified: head/sys/amd64/amd64/exception.S
==============================================================================
--- head/sys/amd64/amd64/exception.S Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/exception.S Wed Apr 1 13:09:26 2009 (r190620)
@@ -42,6 +42,7 @@
#include <machine/asmacros.h>
#include <machine/psl.h>
#include <machine/trap.h>
+#include <machine/specialreg.h>
#include "assym.s"
@@ -99,7 +100,7 @@ MCOUNT_LABEL(btrap)
/* Traps that we leave interrupts disabled for.. */
#define TRAP_NOEN(a) \
subq $TF_RIP,%rsp; \
- movq $(a),TF_TRAPNO(%rsp) ; \
+ movl $(a),TF_TRAPNO(%rsp) ; \
movq $0,TF_ADDR(%rsp) ; \
movq $0,TF_ERR(%rsp) ; \
jmp alltraps_noen
@@ -111,7 +112,7 @@ IDTVEC(bpt)
/* Regular traps; The cpu does not supply tf_err for these. */
#define TRAP(a) \
subq $TF_RIP,%rsp; \
- movq $(a),TF_TRAPNO(%rsp) ; \
+ movl $(a),TF_TRAPNO(%rsp) ; \
movq $0,TF_ADDR(%rsp) ; \
movq $0,TF_ERR(%rsp) ; \
jmp alltraps
@@ -139,7 +140,7 @@ IDTVEC(xmm)
/* This group of traps have tf_err already pushed by the cpu */
#define TRAP_ERR(a) \
subq $TF_ERR,%rsp; \
- movq $(a),TF_TRAPNO(%rsp) ; \
+ movl $(a),TF_TRAPNO(%rsp) ; \
movq $0,TF_ADDR(%rsp) ; \
jmp alltraps
IDTVEC(tss)
@@ -164,6 +165,10 @@ alltraps:
testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
jz alltraps_testi /* already running with kernel GS.base */
swapgs
+ movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
alltraps_testi:
testl $PSL_I,TF_RFLAGS(%rsp)
jz alltraps_pushregs
@@ -185,6 +190,7 @@ alltraps_pushregs_no_rdi:
movq %r13,TF_R13(%rsp)
movq %r14,TF_R14(%rsp)
movq %r15,TF_R15(%rsp)
+ movl $TF_HASSEGS,TF_FLAGS(%rsp)
FAKE_MCOUNT(TF_RIP(%rsp))
#ifdef KDTRACE_HOOKS
/*
@@ -193,7 +199,7 @@ alltraps_pushregs_no_rdi:
* interrupt. For all other trap types, just handle them in
* the usual way.
*/
- cmpq $T_BPTFLT,TF_TRAPNO(%rsp)
+ cmpl $T_BPTFLT,TF_TRAPNO(%rsp)
jne calltrap
/* Check if there is no DTrace hook registered. */
@@ -228,13 +234,17 @@ calltrap:
.type alltraps_noen, at function
alltraps_noen:
testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
- jz alltraps_pushregs /* already running with kernel GS.base */
+ jz 1f /* already running with kernel GS.base */
swapgs
+1: movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
jmp alltraps_pushregs
IDTVEC(dblfault)
subq $TF_ERR,%rsp
- movq $T_DOUBLEFLT,TF_TRAPNO(%rsp)
+ movl $T_DOUBLEFLT,TF_TRAPNO(%rsp)
movq $0,TF_ADDR(%rsp)
movq $0,TF_ERR(%rsp)
movq %rdi,TF_RDI(%rsp)
@@ -252,6 +262,11 @@ IDTVEC(dblfault)
movq %r13,TF_R13(%rsp)
movq %r14,TF_R14(%rsp)
movq %r15,TF_R15(%rsp)
+ movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
+ movl $TF_HASSEGS,TF_FLAGS(%rsp)
testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
jz 1f /* already running with kernel GS.base */
swapgs
@@ -262,7 +277,7 @@ IDTVEC(dblfault)
IDTVEC(page)
subq $TF_ERR,%rsp
- movq $T_PAGEFLT,TF_TRAPNO(%rsp)
+ movl $T_PAGEFLT,TF_TRAPNO(%rsp)
testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
jz 1f /* already running with kernel GS.base */
swapgs
@@ -270,6 +285,10 @@ IDTVEC(page)
movq %rdi,TF_RDI(%rsp) /* free up a GP register */
movq %cr2,%rdi /* preserve %cr2 before .. */
movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */
+ movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
testl $PSL_I,TF_RFLAGS(%rsp)
jz alltraps_pushregs_no_rdi
sti
@@ -283,17 +302,19 @@ IDTVEC(page)
*/
IDTVEC(prot)
subq $TF_ERR,%rsp
- movq $T_PROTFLT,TF_TRAPNO(%rsp)
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
movq $0,TF_ADDR(%rsp)
movq %rdi,TF_RDI(%rsp) /* free up a GP register */
leaq doreti_iret(%rip),%rdi
cmpq %rdi,TF_RIP(%rsp)
- je 2f /* kernel but with user gsbase!! */
+ je 1f /* kernel but with user gsbase!! */
testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
- jz 1f /* already running with kernel GS.base */
-2:
- swapgs
-1:
+ jz 2f /* already running with kernel GS.base */
+1: swapgs
+2: movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
testl $PSL_I,TF_RFLAGS(%rsp)
jz alltraps_pushregs_no_rdi
sti
@@ -316,6 +337,10 @@ IDTVEC(fast_syscall)
movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */
movq PCPU(SCRATCH_RSP),%r11 /* %r11 already saved */
movq %r11,TF_RSP(%rsp) /* user stack pointer */
+ movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
sti
movq $KUDSEL,TF_SS(%rsp)
movq $KUCSEL,TF_CS(%rsp)
@@ -333,40 +358,11 @@ IDTVEC(fast_syscall)
movq %r13,TF_R13(%rsp) /* C preserved */
movq %r14,TF_R14(%rsp) /* C preserved */
movq %r15,TF_R15(%rsp) /* C preserved */
+ movl $TF_HASSEGS,TF_FLAGS(%rsp)
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rdi
call syscall
movq PCPU(CURPCB),%rax
- testq $PCB_FULLCTX,PCB_FLAGS(%rax)
- jne 3f
-1: /* Check for and handle AST's on return to userland */
- cli
- movq PCPU(CURTHREAD),%rax
- testl $TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
- je 2f
- sti
- movq %rsp, %rdi
- call ast
- jmp 1b
-2: /* restore preserved registers */
- MEXITCOUNT
- movq TF_RDI(%rsp),%rdi /* bonus; preserve arg 1 */
- movq TF_RSI(%rsp),%rsi /* bonus: preserve arg 2 */
- movq TF_RDX(%rsp),%rdx /* return value 2 */
- movq TF_RAX(%rsp),%rax /* return value 1 */
- movq TF_RBX(%rsp),%rbx /* C preserved */
- movq TF_RBP(%rsp),%rbp /* C preserved */
- movq TF_R12(%rsp),%r12 /* C preserved */
- movq TF_R13(%rsp),%r13 /* C preserved */
- movq TF_R14(%rsp),%r14 /* C preserved */
- movq TF_R15(%rsp),%r15 /* C preserved */
- movq TF_RFLAGS(%rsp),%r11 /* original %rflags */
- movq TF_RIP(%rsp),%rcx /* original %rip */
- movq TF_RSP(%rsp),%r9 /* user stack pointer */
- movq %r9,%rsp /* original %rsp */
- swapgs
- sysretq
-3: /* Requested full context restore, use doreti for that */
andq $~PCB_FULLCTX,PCB_FLAGS(%rax)
MEXITCOUNT
jmp doreti
@@ -405,7 +401,7 @@ IDTVEC(fast_syscall32)
IDTVEC(nmi)
subq $TF_RIP,%rsp
- movq $(T_NMI),TF_TRAPNO(%rsp)
+ movl $(T_NMI),TF_TRAPNO(%rsp)
movq $0,TF_ADDR(%rsp)
movq $0,TF_ERR(%rsp)
movq %rdi,TF_RDI(%rsp)
@@ -423,6 +419,11 @@ IDTVEC(nmi)
movq %r13,TF_R13(%rsp)
movq %r14,TF_R14(%rsp)
movq %r15,TF_R15(%rsp)
+ movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
+ movl $TF_HASSEGS,TF_FLAGS(%rsp)
xorl %ebx,%ebx
testb $SEL_RPL_MASK,TF_CS(%rsp)
jnz nmi_fromuserspace
@@ -515,9 +516,7 @@ outofnmi:
nocallchain:
#endif
testl %ebx,%ebx
- jz nmi_kernelexit
- swapgs
- jmp nmi_restoreregs
+ jnz doreti_exit
nmi_kernelexit:
/*
* Put back the preserved MSR_GSBASE value.
@@ -633,7 +632,55 @@ doreti_ast:
*/
doreti_exit:
MEXITCOUNT
- movq TF_RDI(%rsp),%rdi
+ movq PCPU(CURTHREAD),%r8
+ movq TD_PCB(%r8),%r8
+
+ /*
+ * Do not reload segment registers for kernel.
+ * Since we do not reload segments registers with sane
+ * values on kernel entry, descriptors referenced by
+ * segments registers may be not valid. This is fatal
+ * for the usermode, but is innocent for the kernel.
+ */
+ testb $SEL_RPL_MASK,TF_CS(%rsp)
+ jz ld_regs
+
+ testl $TF_HASSEGS,TF_FLAGS(%rsp)
+ je set_segs
+
+do_segs:
+ /* Restore %fs and fsbase */
+ movw TF_FS(%rsp),%ax
+ .globl ld_fs
+ld_fs: movw %ax,%fs
+ cmpw $KUF32SEL,%ax
+ jne 1f
+ movl $MSR_FSBASE,%ecx
+ movl PCB_FSBASE(%r8),%eax
+ movl PCB_FSBASE+4(%r8),%edx
+ wrmsr
+1:
+ /* Restore %gs and gsbase */
+ movw TF_GS(%rsp),%si
+ pushfq
+ cli
+ movl $MSR_GSBASE,%ecx
+ rdmsr
+ .globl ld_gs
+ld_gs: movw %si,%gs
+ wrmsr
+ popfq
+ cmpw $KUG32SEL,%si
+ jne 1f
+ movl $MSR_KGSBASE,%ecx
+ movl PCB_GSBASE(%r8),%eax
+ movl PCB_GSBASE+4(%r8),%edx
+ wrmsr
+1: .globl ld_es
+ld_es: movw TF_ES(%rsp),%es
+ .globl ld_ds
+ld_ds: movw TF_DS(%rsp),%ds
+ld_regs:movq TF_RDI(%rsp),%rdi
movq TF_RSI(%rsp),%rsi
movq TF_RDX(%rsp),%rdx
movq TF_RCX(%rsp),%rcx
@@ -657,6 +704,14 @@ doreti_exit:
doreti_iret:
iretq
+set_segs:
+ movw $KUDSEL,%ax
+ movw %ax,TF_DS(%rsp)
+ movw %ax,TF_ES(%rsp)
+ movw $KUF32SEL,TF_FS(%rsp)
+ movw $KUG32SEL,TF_GS(%rsp)
+ jmp do_segs
+
/*
* doreti_iret_fault. Alternative return code for
* the case where we get a fault in the doreti_exit code
@@ -671,7 +726,12 @@ doreti_iret_fault:
testl $PSL_I,TF_RFLAGS(%rsp)
jz 1f
sti
-1: movq %rdi,TF_RDI(%rsp)
+1: movw %fs,TF_FS(%rsp)
+ movw %gs,TF_GS(%rsp)
+ movw %es,TF_ES(%rsp)
+ movw %ds,TF_DS(%rsp)
+ movl $TF_HASSEGS,TF_FLAGS(%rsp)
+ movq %rdi,TF_RDI(%rsp)
movq %rsi,TF_RSI(%rsp)
movq %rdx,TF_RDX(%rsp)
movq %rcx,TF_RCX(%rsp)
@@ -686,11 +746,48 @@ doreti_iret_fault:
movq %r13,TF_R13(%rsp)
movq %r14,TF_R14(%rsp)
movq %r15,TF_R15(%rsp)
- movq $T_PROTFLT,TF_TRAPNO(%rsp)
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
movq $0,TF_ERR(%rsp) /* XXX should be the error code */
movq $0,TF_ADDR(%rsp)
FAKE_MCOUNT(TF_RIP(%rsp))
jmp calltrap
+
+ ALIGN_TEXT
+ .globl ds_load_fault
+ds_load_fault:
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
+ movzwl TF_DS(%rsp),%edx
+ movl %edx,TF_ERR(%rsp)
+ movw $KUDSEL,TF_DS(%rsp)
+ jmp calltrap
+
+ ALIGN_TEXT
+ .globl es_load_fault
+es_load_fault:
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
+ movzwl TF_ES(%rsp),%edx
+ movl %edx,TF_ERR(%rsp)
+ movw $KUDSEL,TF_ES(%rsp)
+ jmp calltrap
+
+ ALIGN_TEXT
+ .globl fs_load_fault
+fs_load_fault:
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
+ movzwl TF_FS(%rsp),%edx
+ movl %edx,TF_ERR(%rsp)
+ movw $KUF32SEL,TF_FS(%rsp)
+ jmp calltrap
+
+ ALIGN_TEXT
+ .globl gs_load_fault
+gs_load_fault:
+ popfq
+ movl $T_PROTFLT,TF_TRAPNO(%rsp)
+ movzwl TF_GS(%rsp),%edx
+ movl %edx,TF_ERR(%rsp)
+ movw $KUG32SEL,TF_GS(%rsp)
+ jmp calltrap
#ifdef HWPMC_HOOKS
ENTRY(end_exceptions)
#endif
Modified: head/sys/amd64/amd64/genassym.c
==============================================================================
--- head/sys/amd64/amd64/genassym.c Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/genassym.c Wed Apr 1 13:09:26 2009 (r190620)
@@ -79,6 +79,10 @@ ASSYM(P_VMSPACE, offsetof(struct proc, p
ASSYM(VM_PMAP, offsetof(struct vmspace, vm_pmap));
ASSYM(PM_ACTIVE, offsetof(struct pmap, pm_active));
+ASSYM(P_MD, offsetof(struct proc, p_md));
+ASSYM(MD_LDT, offsetof(struct mdproc, md_ldt));
+ASSYM(MD_LDT_SD, offsetof(struct mdproc, md_ldt_sd));
+
ASSYM(TD_LOCK, offsetof(struct thread, td_lock));
ASSYM(TD_FLAGS, offsetof(struct thread, td_flags));
ASSYM(TD_PCB, offsetof(struct thread, td_pcb));
@@ -132,16 +136,13 @@ ASSYM(PCB_RBX, offsetof(struct pcb, pcb_
ASSYM(PCB_RIP, offsetof(struct pcb, pcb_rip));
ASSYM(PCB_FSBASE, offsetof(struct pcb, pcb_fsbase));
ASSYM(PCB_GSBASE, offsetof(struct pcb, pcb_gsbase));
-ASSYM(PCB_DS, offsetof(struct pcb, pcb_ds));
-ASSYM(PCB_ES, offsetof(struct pcb, pcb_es));
-ASSYM(PCB_FS, offsetof(struct pcb, pcb_fs));
-ASSYM(PCB_GS, offsetof(struct pcb, pcb_gs));
ASSYM(PCB_DR0, offsetof(struct pcb, pcb_dr0));
ASSYM(PCB_DR1, offsetof(struct pcb, pcb_dr1));
ASSYM(PCB_DR2, offsetof(struct pcb, pcb_dr2));
ASSYM(PCB_DR3, offsetof(struct pcb, pcb_dr3));
ASSYM(PCB_DR6, offsetof(struct pcb, pcb_dr6));
ASSYM(PCB_DR7, offsetof(struct pcb, pcb_dr7));
+ASSYM(PCB_TSSP, offsetof(struct pcb, pcb_tssp));
ASSYM(PCB_DBREGS, PCB_DBREGS);
ASSYM(PCB_32BIT, PCB_32BIT);
ASSYM(PCB_GS32BIT, PCB_GS32BIT);
@@ -193,7 +194,13 @@ ASSYM(TF_CS, offsetof(struct trapframe,
ASSYM(TF_RFLAGS, offsetof(struct trapframe, tf_rflags));
ASSYM(TF_RSP, offsetof(struct trapframe, tf_rsp));
ASSYM(TF_SS, offsetof(struct trapframe, tf_ss));
+ASSYM(TF_DS, offsetof(struct trapframe, tf_ds));
+ASSYM(TF_ES, offsetof(struct trapframe, tf_es));
+ASSYM(TF_FS, offsetof(struct trapframe, tf_fs));
+ASSYM(TF_GS, offsetof(struct trapframe, tf_gs));
+ASSYM(TF_FLAGS, offsetof(struct trapframe, tf_flags));
ASSYM(TF_SIZE, sizeof(struct trapframe));
+ASSYM(TF_HASSEGS, TF_HASSEGS);
ASSYM(SIGF_HANDLER, offsetof(struct sigframe, sf_ahu.sf_handler));
ASSYM(SIGF_UC, offsetof(struct sigframe, sf_uc));
@@ -215,7 +222,11 @@ ASSYM(PC_SCRATCH_RSP, offsetof(struct pc
ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
ASSYM(PC_TSSP, offsetof(struct pcpu, pc_tssp));
ASSYM(PC_RSP0, offsetof(struct pcpu, pc_rsp0));
+ASSYM(PC_FS32P, offsetof(struct pcpu, pc_fs32p));
ASSYM(PC_GS32P, offsetof(struct pcpu, pc_gs32p));
+ASSYM(PC_LDT, offsetof(struct pcpu, pc_ldt));
+ASSYM(PC_COMMONTSSP, offsetof(struct pcpu, pc_commontssp));
+ASSYM(PC_TSS, offsetof(struct pcpu, pc_tss));
ASSYM(LA_VER, offsetof(struct LAPIC, version));
ASSYM(LA_TPR, offsetof(struct LAPIC, tpr));
@@ -230,6 +241,10 @@ ASSYM(KDSEL, GSEL(GDATA_SEL, SEL_KPL));
ASSYM(KUCSEL, GSEL(GUCODE_SEL, SEL_UPL));
ASSYM(KUDSEL, GSEL(GUDATA_SEL, SEL_UPL));
ASSYM(KUC32SEL, GSEL(GUCODE32_SEL, SEL_UPL));
+ASSYM(KUF32SEL, GSEL(GUFS32_SEL, SEL_UPL));
+ASSYM(KUG32SEL, GSEL(GUGS32_SEL, SEL_UPL));
+ASSYM(TSSSEL, GSEL(GPROC0_SEL, SEL_KPL));
+ASSYM(LDTSEL, GSEL(GUSERLDT_SEL, SEL_KPL));
ASSYM(SEL_RPL_MASK, SEL_RPL_MASK);
ASSYM(MSR_GSBASE, MSR_GSBASE);
Modified: head/sys/amd64/amd64/machdep.c
==============================================================================
--- head/sys/amd64/amd64/machdep.c Wed Apr 1 12:53:01 2009 (r190619)
+++ head/sys/amd64/amd64/machdep.c Wed Apr 1 13:09:26 2009 (r190620)
@@ -159,7 +159,7 @@ extern vm_offset_t ksym_start, ksym_end;
#define ICH_PMBASE 0x400
#define ICH_SMI_EN ICH_PMBASE + 0x30
-int _udatasel, _ucodesel, _ucode32sel;
+int _udatasel, _ucodesel, _ucode32sel, _ufssel, _ugssel;
int cold = 1;
@@ -192,6 +192,8 @@ struct mtx icu_lock;
struct mem_range_softc mem_range_softc;
+struct mtx dt_lock; /* lock for GDT and LDT */
+
static void
cpu_startup(dummy)
void *dummy;
@@ -278,7 +280,7 @@ cpu_startup(dummy)
* Send an interrupt to process.
*
* Stack is set up to allow sigcode stored
- * at top to call routine, followed by kcall
+ * at top to call routine, followed by call
* to sigreturn routine below. After sigreturn
* resets the signal mask, the stack, and the
* frame pointer, it returns to the user
@@ -316,6 +318,8 @@ sendsig(sig_t catcher, ksiginfo_t *ksi,
sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */
get_fpcontext(td, &sf.sf_uc.uc_mcontext);
fpstate_drop(td);
+ sf.sf_uc.uc_mcontext.mc_fsbase = td->td_pcb->pcb_fsbase;
+ sf.sf_uc.uc_mcontext.mc_gsbase = td->td_pcb->pcb_gsbase;
/* Allocate space for the signal handler context. */
if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
@@ -370,6 +374,11 @@ sendsig(sig_t catcher, ksiginfo_t *ksi,
regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
regs->tf_rflags &= ~(PSL_T | PSL_D);
regs->tf_cs = _ucodesel;
+ regs->tf_ds = _udatasel;
+ regs->tf_es = _udatasel;
+ regs->tf_fs = _ufssel;
+ regs->tf_gs = _ugssel;
+ regs->tf_flags = TF_HASSEGS;
PROC_LOCK(p);
mtx_lock(&psp->ps_mtx);
}
@@ -401,9 +410,16 @@ sigreturn(td, uap)
ksiginfo_t ksi;
error = copyin(uap->sigcntxp, &uc, sizeof(uc));
- if (error != 0)
+ if (error != 0) {
+ printf("sigreturn (pid %d): copyin failed\n", p->p_pid);
return (error);
+ }
ucp = &uc;
+ if ((ucp->uc_mcontext.mc_flags & ~_MC_FLAG_MASK) != 0) {
+ printf("sigreturn (pid %d): mc_flags %x\n", p->p_pid,
+ ucp->uc_mcontext.mc_flags);
+ return (EINVAL);
+ }
regs = td->td_frame;
rflags = ucp->uc_mcontext.mc_rflags;
/*
@@ -420,7 +436,8 @@ sigreturn(td, uap)
* one less debugger trap, so allowing it is fairly harmless.
*/
if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
- printf("sigreturn: rflags = 0x%lx\n", rflags);
+ printf("sigreturn (pid %d): rflags = 0x%lx\n", p->p_pid,
+ rflags);
return (EINVAL);
}
@@ -431,7 +448,7 @@ sigreturn(td, uap)
*/
cs = ucp->uc_mcontext.mc_cs;
if (!CS_SECURE(cs)) {
- printf("sigreturn: cs = 0x%x\n", cs);
+ printf("sigreturn (pid %d): cs = 0x%x\n", p->p_pid, cs);
ksiginfo_init_trap(&ksi);
ksi.ksi_signo = SIGBUS;
ksi.ksi_code = BUS_OBJERR;
@@ -442,9 +459,13 @@ sigreturn(td, uap)
}
ret = set_fpcontext(td, &ucp->uc_mcontext);
- if (ret != 0)
+ if (ret != 0) {
+ printf("sigreturn (pid %d): set_fpcontext\n", p->p_pid);
return (ret);
+ }
bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(*regs));
+ td->td_pcb->pcb_fsbase = ucp->uc_mcontext.mc_fsbase;
+ td->td_pcb->pcb_gsbase = ucp->uc_mcontext.mc_gsbase;
PROC_LOCK(p);
#if defined(COMPAT_43)
@@ -738,22 +759,16 @@ exec_setregs(td, entry, stack, ps_string
{
struct trapframe *regs = td->td_frame;
struct pcb *pcb = td->td_pcb;
+
+ mtx_lock(&dt_lock);
+ if (td->td_proc->p_md.md_ldt != NULL)
+ user_ldt_free(td);
+ else
+ mtx_unlock(&dt_lock);
- critical_enter();
- wrmsr(MSR_FSBASE, 0);
- wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */
pcb->pcb_fsbase = 0;
pcb->pcb_gsbase = 0;
- critical_exit();
pcb->pcb_flags &= ~(PCB_32BIT | PCB_GS32BIT);
- load_ds(_udatasel);
- load_es(_udatasel);
- load_fs(_udatasel);
- load_gs(_udatasel);
- pcb->pcb_ds = _udatasel;
- pcb->pcb_es = _udatasel;
- pcb->pcb_fs = _udatasel;
- pcb->pcb_gs = _udatasel;
pcb->pcb_initial_fpucw = __INITIAL_FPUCW__;
bzero((char *)regs, sizeof(struct trapframe));
@@ -763,6 +778,11 @@ exec_setregs(td, entry, stack, ps_string
regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T);
regs->tf_ss = _udatasel;
regs->tf_cs = _ucodesel;
+ regs->tf_ds = _udatasel;
+ regs->tf_es = _udatasel;
+ regs->tf_fs = _ufssel;
+ regs->tf_gs = _ugssel;
+ regs->tf_flags = TF_HASSEGS;
/*
* Reset the hardware debug registers if they were in use.
@@ -1380,12 +1400,12 @@ hammer_time(u_int64_t modulep, u_int64_t
/*
* make gdt memory segments
*/
- gdt_segs[GPROC0_SEL].ssd_base = (uintptr_t)&common_tss[0];
-
for (x = 0; x < NGDT; x++) {
- if (x != GPROC0_SEL && x != (GPROC0_SEL + 1))
+ if (x != GPROC0_SEL && x != (GPROC0_SEL + 1) &&
+ x != GUSERLDT_SEL && x != (GUSERLDT_SEL) + 1)
ssdtosd(&gdt_segs[x], &gdt[x]);
}
+ gdt_segs[GPROC0_SEL].ssd_base = (uintptr_t)&common_tss[0];
ssdtosyssd(&gdt_segs[GPROC0_SEL],
(struct system_segment_descriptor *)&gdt[GPROC0_SEL]);
@@ -1403,6 +1423,10 @@ hammer_time(u_int64_t modulep, u_int64_t
PCPU_SET(curthread, &thread0);
PCPU_SET(curpcb, thread0.td_pcb);
PCPU_SET(tssp, &common_tss[0]);
+ PCPU_SET(commontssp, &common_tss[0]);
+ PCPU_SET(tss, (struct system_segment_descriptor *)&gdt[GPROC0_SEL]);
+ PCPU_SET(ldt, (struct system_segment_descriptor *)&gdt[GUSERLDT_SEL]);
+ PCPU_SET(fs32p, &gdt[GUFS32_SEL]);
PCPU_SET(gs32p, &gdt[GUGS32_SEL]);
/*
@@ -1415,6 +1439,7 @@ hammer_time(u_int64_t modulep, u_int64_t
*/
mutex_init();
mtx_init(&icu_lock, "icu", NULL, MTX_SPIN | MTX_NOWITNESS);
+ mtx_init(&dt_lock, "descriptor tables", NULL, MTX_DEF);
/* exceptions */
for (x = 0; x < NIDT; x++)
@@ -1503,7 +1528,8 @@ hammer_time(u_int64_t modulep, u_int64_t
common_tss[0].tss_ist2 = (long) np;
/* Set the IO permission bitmap (empty due to tss seg limit) */
- common_tss[0].tss_iobase = sizeof(struct amd64tss);
+ common_tss[0].tss_iobase = sizeof(struct amd64tss) +
+ IOPAGES * PAGE_SIZE;
gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
ltr(gsel_tss);
@@ -1531,10 +1557,12 @@ hammer_time(u_int64_t modulep, u_int64_t
_ucodesel = GSEL(GUCODE_SEL, SEL_UPL);
_udatasel = GSEL(GUDATA_SEL, SEL_UPL);
_ucode32sel = GSEL(GUCODE32_SEL, SEL_UPL);
+ _ufssel = GSEL(GUFS32_SEL, SEL_UPL);
+ _ugssel = GSEL(GUGS32_SEL, SEL_UPL);
load_ds(_udatasel);
load_es(_udatasel);
- load_fs(_udatasel);
+ load_fs(_ufssel);
/* setup proc 0's pcb */
thread0.td_pcb->pcb_flags = 0;
@@ -1656,6 +1684,17 @@ fill_regs(struct thread *td, struct reg
regs->r_rflags = tp->tf_rflags;
regs->r_rsp = tp->tf_rsp;
regs->r_ss = tp->tf_ss;
+ if (tp->tf_flags & TF_HASSEGS) {
+ regs->r_ds = tp->tf_ds;
+ regs->r_es = tp->tf_es;
+ regs->r_fs = tp->tf_fs;
+ regs->r_gs = tp->tf_gs;
+ } else {
+ regs->r_ds = 0;
+ regs->r_es = 0;
+ regs->r_fs = 0;
+ regs->r_gs = 0;
+ }
return (0);
}
@@ -1689,6 +1728,13 @@ set_regs(struct thread *td, struct reg *
tp->tf_rflags = rflags;
tp->tf_rsp = regs->r_rsp;
tp->tf_ss = regs->r_ss;
+ if (0) { /* XXXKIB */
+ tp->tf_ds = regs->r_ds;
+ tp->tf_es = regs->r_es;
+ tp->tf_fs = regs->r_fs;
+ tp->tf_gs = regs->r_gs;
+ tp->tf_flags = TF_HASSEGS;
+ }
td->td_pcb->pcb_flags |= PCB_FULLCTX;
return (0);
}
@@ -1808,8 +1854,15 @@ get_mcontext(struct thread *td, mcontext
mcp->mc_cs = tp->tf_cs;
mcp->mc_rsp = tp->tf_rsp;
mcp->mc_ss = tp->tf_ss;
+ mcp->mc_ds = tp->tf_ds;
+ mcp->mc_es = tp->tf_es;
+ mcp->mc_fs = tp->tf_fs;
+ mcp->mc_gs = tp->tf_gs;
+ mcp->mc_flags = tp->tf_flags;
mcp->mc_len = sizeof(*mcp);
get_fpcontext(td, mcp);
+ mcp->mc_fsbase = td->td_pcb->pcb_fsbase;
+ mcp->mc_gsbase = td->td_pcb->pcb_gsbase;
return (0);
}
@@ -1827,7 +1880,8 @@ set_mcontext(struct thread *td, const mc
int ret;
tp = td->td_frame;
- if (mcp->mc_len != sizeof(*mcp))
+ if (mcp->mc_len != sizeof(*mcp) ||
+ (mcp->mc_flags & ~_MC_FLAG_MASK) != 0)
return (EINVAL);
rflags = (mcp->mc_rflags & PSL_USERCHANGE) |
(tp->tf_rflags & ~PSL_USERCHANGE);
@@ -1853,6 +1907,17 @@ set_mcontext(struct thread *td, const mc
tp->tf_rflags = rflags;
tp->tf_rsp = mcp->mc_rsp;
tp->tf_ss = mcp->mc_ss;
+ tp->tf_flags = mcp->mc_flags;
+ if (tp->tf_flags & TF_HASSEGS) {
+ tp->tf_ds = mcp->mc_ds;
+ tp->tf_es = mcp->mc_es;
*** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
More information about the svn-src-head
mailing list