i386 Vmi segment changes
RFC PATCH 11/24 i386 Vmi segment changes
Make the Linux kernel able to run at CPL 0, 1, or 2. There are a few limited places where CPL-0 is assumed, and they can be converted very efficiently into a supervisor CPL check instead of a CPL-0 check. This step prepares the kernel for running in direct execution under a hypervisor.
Note the user_mode_vm macro used in ptrace.h is very similar to the flag mixing of EFLAGS and CS used to test in one branch in entry.S.
To make the COMPARE_SEGMENT_REG macro work, it must contain both the push and the pop, requiring an additional load of EAX after the possible stack fixup. This is because FIXUP_ESPFIX_STACK destroys %EAX again with a call to C code.
In all, the overhead is couple of instructions, and no extra branches. Note that I prefer to use the testing of selectors in the form:
SELECTOR_CLEAR_RPL(sel) == __KERNEL_SEL
Instead of SELECTOR == (%seg), where %seg is the live register value. This is because moves from segment registers are more costly than a single ALU instruction, and both cost a temporary register. Still, pushing the raw %cs value in the NMI after sysenter debug trap fixup code instead of __KERNEL_CS | RPL is just as efficient, since fetching RPL would require a %cs load anyway.
The switch from __KERNEL_DS to __USER_DS at one point is for convenience, since they are both equivalent for %ds and %es segments; __KERNEL_DS, contrary to the name, is only useful for the %ss segment.
include/asm-i386/ptrace.h +3 -2
@@ -60,6 +60,7 @@ struct pt_regs {#ifdef __KERNEL__#include <asm/vm86.h> +#include <asm/segment.h> struct task_struct ;extern void send_sigtrap (struct task_struct *tsk,struct pt_regs *regs,int error_code);
@@ -73,11 +74,11 @@ extern void send_sigtrap(struct task_str*/ static inline int user_mode (struct pt_regs *regs) { -return (regs->xcs &3 ) !=0 ; +return (regs->xcs &SEGMENT_RPL_MASK ) ==3 ; }static inline int user_mode_vm (struct pt_regs *regs) { -return ((regs->xcs &3 ) | (regs->eflags &VM_MASK )) !=0 ; +return (((regs->xcs &SEGMENT_RPL_MASK ) | (regs->eflags &VM_MASK )) >=3 ); }#define instruction_pointer (regs) ((regs)->eip)#if defined(CONFIG_SMP ) && defined(CONFIG_FRAME_POINTER )
include/asm-i386/segment.h +5 -0
@@ -112,4 +112,9 @@*/ #define IDT_ENTRIES 256 +#define SEGMENT_RPL_MASK 0x03 +#define SEGMENT_TI_MASK 0x04 + +#include <mach_segment.h> +#endif
include/asm-i386/mach-vmi/mach_segment.h +56 -0
56 lines not on this page · raw message
include/asm-i386/mach-default/mach_segment.h +39 -0
@@ -0,0 +1,39 @@ +/* +* Copyright (C) 2005, VMware, Inc. +* +* All rights reserved. +* +* This program is free software; you can redistribute it and/or modify +* it under the terms of the GNU General Public License as published by +* the Free Software Foundation; either version 2 of the License, or +* (at your option) any later version. +* +* This program is distributed in the hope that it will be useful, but +* WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or +* NON INFRINGEMENT. See the GNU General Public License for more +* details. +* +* You should have received a copy of the GNU General Public License +* along with this program; if not, write to the Free Software +* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. +* +* Send feedback to [email protected] +* +*/ + + +#ifndef __MACH_SEGMENT_H +#define __MACH_SEGMENT_H + +#define get_kernel_rpl () 0 + +#define COMPARE_SEGMENT_STACK (segment, offset) \ + cmpw $segment, offset(%esp); + +#define COMPARE_SEGMENT_REG (segment, reg) \ + pushl %eax; \ + mov reg, %eax; \ + cmpw $segment,%ax; \ + popl %eax; +#endif
arch/i386/kernel/entry.S +12 -15
@@ -145,9 +145,11 @@ ret_from_exception: ret_from_intr: GET_THREAD_INFO(%ebp) movl EFLAGS(%esp), %eax # mix EFLAGS and CS + andl $VM_MASK, %eax movb CS(%esp), %al - testl $(VM_MASK | 3), %eax - jz resume_kernel + andb $SEGMENT_RPL_MASK, %al + cmpl $SEGMENT_RPL_MASK, %eax + jb resume_kernel # returning to kernel or vm86-space ENTRY(resume_userspace) CLI # make sure we don't miss an interrupt # setting need_resched or sigpending
@@ -382,17 +384,14 @@ syscall_badsys: /* put ESP to the proper location */ \ movl %eax, %esp; #define UNWIND_ESPFIX_STACK \ - pushl %eax; \ - movl %ss, %eax; \ - /* see if on 16bit stack */ \ - cmpw $__ESPFIX_SS, %ax; \ + COMPARE_SEGMENT_REG(__ESPFIX_SS, %ss) \ jne 28f; \ - movl $__KERNEL_DS, %edx; \ + movl $__USER_DS, %edx; \ movl %edx, %ds; \ movl %edx, %es; \ /* switch to 32bit stack */ \ FIXUP_ESPFIX_STACK \ -28: popl %eax; +28:; /* * Build the entry stubs and pointer table with
@@ -451,6 +450,7 @@ error_code: pushl %es UNWIND_ESPFIX_STACK popl %ecx + movl EAX(%esp), %eax movl ES(%esp), %edi # get the function address movl ORIG_EAX(%esp), %edx # get the error code movl %eax, ORIG_EAX(%esp)
@@ -501,12 +501,12 @@ device_not_available_emulate: * the instruction that would have done it for sysenter. */ #define FIX_STACK(offset, ok, label) \ - cmpw $__KERNEL_CS,4(%esp); \ + COMPARE_SEGMENT_STACK(__KERNEL_CS, 4) \ jne ok; \ label: \ movl TSS_sysenter_esp0+offset(%esp),%esp; \ pushfl; \ - pushl $__KERNEL_CS; \ + push %cs; \ pushl $sysenter_past_esp KPROBE_ENTRY(debug)
@@ -530,10 +530,7 @@ debug_stack_correct: * fault happened on the sysenter path. */ ENTRY(nmi) - pushl %eax - movl %ss, %eax - cmpw $__ESPFIX_SS, %ax - popl %eax + COMPARE_SEGMENT_REG(__ESPFIX_SS, %ss) je nmi_16bit_stack cmpl $sysenter_entry,(%esp) je nmi_stack_fixup
@@ -560,7 +557,7 @@ nmi_stack_fixup: FIX_STACK(12,nmi_stack_correct, 1) jmp nmi_stack_correct nmi_debug_stack_check: - cmpw $__KERNEL_CS,16(%esp) + COMPARE_SEGMENT_STACK(__KERNEL_CS, 16) jne nmi_stack_correct cmpl $debug,(%esp) jb nmi_stack_correct
arch/i386/kernel/process.c +1 -1
@@ -348,7 +348,7 @@ int kernel_thread(int (*fn)(void *), voi regs.xes = __USER_DS; regs.orig_eax =-1 ; regs.eip = (unsigned long ) kernel_thread_helper; - regs.xcs = __KERNEL_CS; + regs.xcs = __KERNEL_CS |get_kernel_rpl (); regs.eflags =X86_EFLAGS_IF |X86_EFLAGS_SF |X86_EFLAGS_PF |0x2 ;/* Ok, create the new process.. */
arch/i386/kernel/traps.c +2 -2
@@ -970,10 +970,10 @@ fastcall void setup_x86_bogus_stack(unsimemcpy ((void *)(stack_bot + iret_frame16_off), ®s->eip,20 );/* fill in the switch pointers */ switch16_ptr[0 ] = (regs->esp &0xffff0000 ) | iret_frame16_off; - switch16_ptr[1 ] = __ESPFIX_SS; + switch16_ptr[1 ] = __ESPFIX_SS |get_kernel_rpl (); switch32_ptr[0 ] = (unsigned long )stk +sizeof (struct pt_regs ) +8 -CPU_16BIT_STACK_SIZE ; - switch32_ptr[1 ] = __KERNEL_DS; + switch32_ptr[1 ] = __KERNEL_DS |get_kernel_rpl (); } fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)