diff options
Diffstat (limited to 'kernel/drivers/lguest/x86')
| -rw-r--r-- | kernel/drivers/lguest/x86/core.c | 737 | ||||
| -rw-r--r-- | kernel/drivers/lguest/x86/switcher_32.S | 388 |
2 files changed, 1125 insertions, 0 deletions
diff --git a/kernel/drivers/lguest/x86/core.c b/kernel/drivers/lguest/x86/core.c new file mode 100644 index 000000000..30f2aef69 --- /dev/null +++ b/kernel/drivers/lguest/x86/core.c @@ -0,0 +1,737 @@ +/* + * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation. + * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI. + * + * 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. + */ +/*P:450 + * This file contains the x86-specific lguest code. It used to be all + * mixed in with drivers/lguest/core.c but several foolhardy code slashers + * wrestled most of the dependencies out to here in preparation for porting + * lguest to other architectures (see what I mean by foolhardy?). + * + * This also contains a couple of non-obvious setup and teardown pieces which + * were implemented after days of debugging pain. +:*/ +#include <linux/kernel.h> +#include <linux/start_kernel.h> +#include <linux/string.h> +#include <linux/console.h> +#include <linux/screen_info.h> +#include <linux/irq.h> +#include <linux/interrupt.h> +#include <linux/clocksource.h> +#include <linux/clockchips.h> +#include <linux/cpu.h> +#include <linux/lguest.h> +#include <linux/lguest_launcher.h> +#include <asm/paravirt.h> +#include <asm/param.h> +#include <asm/page.h> +#include <asm/pgtable.h> +#include <asm/desc.h> +#include <asm/setup.h> +#include <asm/lguest.h> +#include <asm/uaccess.h> +#include <asm/i387.h> +#include <asm/tlbflush.h> +#include "../lg.h" + +static int cpu_had_pge; + +static struct { + unsigned long offset; + unsigned short segment; +} lguest_entry; + +/* Offset from where switcher.S was compiled to where we've copied it */ +static unsigned long switcher_offset(void) +{ + return switcher_addr - (unsigned long)start_switcher_text; +} + +/* This cpu's struct lguest_pages (after the Switcher text page) */ +static struct lguest_pages *lguest_pages(unsigned int cpu) +{ + return &(((struct lguest_pages *)(switcher_addr + PAGE_SIZE))[cpu]); +} + +static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu); + +/*S:010 + * We approach the Switcher. + * + * Remember that each CPU has two pages which are visible to the Guest when it + * runs on that CPU. This has to contain the state for that Guest: we copy the + * state in just before we run the Guest. + * + * Each Guest has "changed" flags which indicate what has changed in the Guest + * since it last ran. We saw this set in interrupts_and_traps.c and + * segments.c. + */ +static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages) +{ + /* + * Copying all this data can be quite expensive. We usually run the + * same Guest we ran last time (and that Guest hasn't run anywhere else + * meanwhile). If that's not the case, we pretend everything in the + * Guest has changed. + */ + if (__this_cpu_read(lg_last_cpu) != cpu || cpu->last_pages != pages) { + __this_cpu_write(lg_last_cpu, cpu); + cpu->last_pages = pages; + cpu->changed = CHANGED_ALL; + } + + /* + * These copies are pretty cheap, so we do them unconditionally: */ + /* Save the current Host top-level page directory. + */ + pages->state.host_cr3 = __pa(current->mm->pgd); + /* + * Set up the Guest's page tables to see this CPU's pages (and no + * other CPU's pages). + */ + map_switcher_in_guest(cpu, pages); + /* + * Set up the two "TSS" members which tell the CPU what stack to use + * for traps which do directly into the Guest (ie. traps at privilege + * level 1). + */ + pages->state.guest_tss.sp1 = cpu->esp1; + pages->state.guest_tss.ss1 = cpu->ss1; + + /* Copy direct-to-Guest trap entries. */ + if (cpu->changed & CHANGED_IDT) + copy_traps(cpu, pages->state.guest_idt, default_idt_entries); + + /* Copy all GDT entries which the Guest can change. */ + if (cpu->changed & CHANGED_GDT) + copy_gdt(cpu, pages->state.guest_gdt); + /* If only the TLS entries have changed, copy them. */ + else if (cpu->changed & CHANGED_GDT_TLS) + copy_gdt_tls(cpu, pages->state.guest_gdt); + + /* Mark the Guest as unchanged for next time. */ + cpu->changed = 0; +} + +/* Finally: the code to actually call into the Switcher to run the Guest. */ +static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages) +{ + /* This is a dummy value we need for GCC's sake. */ + unsigned int clobber; + + /* + * Copy the guest-specific information into this CPU's "struct + * lguest_pages". + */ + copy_in_guest_info(cpu, pages); + + /* + * Set the trap number to 256 (impossible value). If we fault while + * switching to the Guest (bad segment registers or bug), this will + * cause us to abort the Guest. + */ + cpu->regs->trapnum = 256; + + /* + * Now: we push the "eflags" register on the stack, then do an "lcall". + * This is how we change from using the kernel code segment to using + * the dedicated lguest code segment, as well as jumping into the + * Switcher. + * + * The lcall also pushes the old code segment (KERNEL_CS) onto the + * stack, then the address of this call. This stack layout happens to + * exactly match the stack layout created by an interrupt... + */ + asm volatile("pushf; lcall *%4" + /* + * This is how we tell GCC that %eax ("a") and %ebx ("b") + * are changed by this routine. The "=" means output. + */ + : "=a"(clobber), "=b"(clobber) + /* + * %eax contains the pages pointer. ("0" refers to the + * 0-th argument above, ie "a"). %ebx contains the + * physical address of the Guest's top-level page + * directory. + */ + : "0"(pages), + "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)), + "m"(lguest_entry) + /* + * We tell gcc that all these registers could change, + * which means we don't have to save and restore them in + * the Switcher. + */ + : "memory", "%edx", "%ecx", "%edi", "%esi"); +} +/*:*/ + +unsigned long *lguest_arch_regptr(struct lg_cpu *cpu, size_t reg_off, bool any) +{ + switch (reg_off) { + case offsetof(struct pt_regs, bx): + return &cpu->regs->ebx; + case offsetof(struct pt_regs, cx): + return &cpu->regs->ecx; + case offsetof(struct pt_regs, dx): + return &cpu->regs->edx; + case offsetof(struct pt_regs, si): + return &cpu->regs->esi; + case offsetof(struct pt_regs, di): + return &cpu->regs->edi; + case offsetof(struct pt_regs, bp): + return &cpu->regs->ebp; + case offsetof(struct pt_regs, ax): + return &cpu->regs->eax; + case offsetof(struct pt_regs, ip): + return &cpu->regs->eip; + case offsetof(struct pt_regs, sp): + return &cpu->regs->esp; + } + + /* Launcher can read these, but we don't allow any setting. */ + if (any) { + switch (reg_off) { + case offsetof(struct pt_regs, ds): + return &cpu->regs->ds; + case offsetof(struct pt_regs, es): + return &cpu->regs->es; + case offsetof(struct pt_regs, fs): + return &cpu->regs->fs; + case offsetof(struct pt_regs, gs): + return &cpu->regs->gs; + case offsetof(struct pt_regs, cs): + return &cpu->regs->cs; + case offsetof(struct pt_regs, flags): + return &cpu->regs->eflags; + case offsetof(struct pt_regs, ss): + return &cpu->regs->ss; + } + } + + return NULL; +} + +/*M:002 + * There are hooks in the scheduler which we can register to tell when we + * get kicked off the CPU (preempt_notifier_register()). This would allow us + * to lazily disable SYSENTER which would regain some performance, and should + * also simplify copy_in_guest_info(). Note that we'd still need to restore + * things when we exit to Launcher userspace, but that's fairly easy. + * + * We could also try using these hooks for PGE, but that might be too expensive. + * + * The hooks were designed for KVM, but we can also put them to good use. +:*/ + +/*H:040 + * This is the i386-specific code to setup and run the Guest. Interrupts + * are disabled: we own the CPU. + */ +void lguest_arch_run_guest(struct lg_cpu *cpu) +{ + /* + * Remember the awfully-named TS bit? If the Guest has asked to set it + * we set it now, so we can trap and pass that trap to the Guest if it + * uses the FPU. + */ + if (cpu->ts && user_has_fpu()) + stts(); + + /* + * SYSENTER is an optimized way of doing system calls. We can't allow + * it because it always jumps to privilege level 0. A normal Guest + * won't try it because we don't advertise it in CPUID, but a malicious + * Guest (or malicious Guest userspace program) could, so we tell the + * CPU to disable it before running the Guest. + */ + if (boot_cpu_has(X86_FEATURE_SEP)) + wrmsr(MSR_IA32_SYSENTER_CS, 0, 0); + + /* + * Now we actually run the Guest. It will return when something + * interesting happens, and we can examine its registers to see what it + * was doing. + */ + run_guest_once(cpu, lguest_pages(raw_smp_processor_id())); + + /* + * Note that the "regs" structure contains two extra entries which are + * not really registers: a trap number which says what interrupt or + * trap made the switcher code come back, and an error code which some + * traps set. + */ + + /* Restore SYSENTER if it's supposed to be on. */ + if (boot_cpu_has(X86_FEATURE_SEP)) + wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); + + /* Clear the host TS bit if it was set above. */ + if (cpu->ts && user_has_fpu()) + clts(); + + /* + * If the Guest page faulted, then the cr2 register will tell us the + * bad virtual address. We have to grab this now, because once we + * re-enable interrupts an interrupt could fault and thus overwrite + * cr2, or we could even move off to a different CPU. + */ + if (cpu->regs->trapnum == 14) + cpu->arch.last_pagefault = read_cr2(); + /* + * Similarly, if we took a trap because the Guest used the FPU, + * we have to restore the FPU it expects to see. + * math_state_restore() may sleep and we may even move off to + * a different CPU. So all the critical stuff should be done + * before this. + */ + else if (cpu->regs->trapnum == 7 && !user_has_fpu()) + math_state_restore(); +} + +/*H:130 + * Now we've examined the hypercall code; our Guest can make requests. + * Our Guest is usually so well behaved; it never tries to do things it isn't + * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual + * infrastructure isn't quite complete, because it doesn't contain replacements + * for the Intel I/O instructions. As a result, the Guest sometimes fumbles + * across one during the boot process as it probes for various things which are + * usually attached to a PC. + * + * When the Guest uses one of these instructions, we get a trap (General + * Protection Fault) and come here. We queue this to be sent out to the + * Launcher to handle. + */ + +/* + * The eip contains the *virtual* address of the Guest's instruction: + * we copy the instruction here so the Launcher doesn't have to walk + * the page tables to decode it. We handle the case (eg. in a kernel + * module) where the instruction is over two pages, and the pages are + * virtually but not physically contiguous. + * + * The longest possible x86 instruction is 15 bytes, but we don't handle + * anything that strange. + */ +static void copy_from_guest(struct lg_cpu *cpu, + void *dst, unsigned long vaddr, size_t len) +{ + size_t to_page_end = PAGE_SIZE - (vaddr % PAGE_SIZE); + unsigned long paddr; + + BUG_ON(len > PAGE_SIZE); + + /* If it goes over a page, copy in two parts. */ + if (len > to_page_end) { + /* But make sure the next page is mapped! */ + if (__guest_pa(cpu, vaddr + to_page_end, &paddr)) + copy_from_guest(cpu, dst + to_page_end, + vaddr + to_page_end, + len - to_page_end); + else + /* Otherwise fill with zeroes. */ + memset(dst + to_page_end, 0, len - to_page_end); + len = to_page_end; + } + + /* This will kill the guest if it isn't mapped, but that + * shouldn't happen. */ + __lgread(cpu, dst, guest_pa(cpu, vaddr), len); +} + + +static void setup_emulate_insn(struct lg_cpu *cpu) +{ + cpu->pending.trap = 13; + copy_from_guest(cpu, cpu->pending.insn, cpu->regs->eip, + sizeof(cpu->pending.insn)); +} + +static void setup_iomem_insn(struct lg_cpu *cpu, unsigned long iomem_addr) +{ + cpu->pending.trap = 14; + cpu->pending.addr = iomem_addr; + copy_from_guest(cpu, cpu->pending.insn, cpu->regs->eip, + sizeof(cpu->pending.insn)); +} + +/*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */ +void lguest_arch_handle_trap(struct lg_cpu *cpu) +{ + unsigned long iomem_addr; + + switch (cpu->regs->trapnum) { + case 13: /* We've intercepted a General Protection Fault. */ + /* Hand to Launcher to emulate those pesky IN and OUT insns */ + if (cpu->regs->errcode == 0) { + setup_emulate_insn(cpu); + return; + } + break; + case 14: /* We've intercepted a Page Fault. */ + /* + * The Guest accessed a virtual address that wasn't mapped. + * This happens a lot: we don't actually set up most of the page + * tables for the Guest at all when we start: as it runs it asks + * for more and more, and we set them up as required. In this + * case, we don't even tell the Guest that the fault happened. + * + * The errcode tells whether this was a read or a write, and + * whether kernel or userspace code. + */ + if (demand_page(cpu, cpu->arch.last_pagefault, + cpu->regs->errcode, &iomem_addr)) + return; + + /* Was this an access to memory mapped IO? */ + if (iomem_addr) { + /* Tell Launcher, let it handle it. */ + setup_iomem_insn(cpu, iomem_addr); + return; + } + + /* + * OK, it's really not there (or not OK): the Guest needs to + * know. We write out the cr2 value so it knows where the + * fault occurred. + * + * Note that if the Guest were really messed up, this could + * happen before it's done the LHCALL_LGUEST_INIT hypercall, so + * lg->lguest_data could be NULL + */ + if (cpu->lg->lguest_data && + put_user(cpu->arch.last_pagefault, + &cpu->lg->lguest_data->cr2)) + kill_guest(cpu, "Writing cr2"); + break; + case 7: /* We've intercepted a Device Not Available fault. */ + /* + * If the Guest doesn't want to know, we already restored the + * Floating Point Unit, so we just continue without telling it. + */ + if (!cpu->ts) + return; + break; + case 32 ... 255: + /* + * These values mean a real interrupt occurred, in which case + * the Host handler has already been run. We just do a + * friendly check if another process should now be run, then + * return to run the Guest again. + */ + cond_resched(); + return; + case LGUEST_TRAP_ENTRY: + /* + * Our 'struct hcall_args' maps directly over our regs: we set + * up the pointer now to indicate a hypercall is pending. + */ + cpu->hcall = (struct hcall_args *)cpu->regs; + return; + } + + /* We didn't handle the trap, so it needs to go to the Guest. */ + if (!deliver_trap(cpu, cpu->regs->trapnum)) + /* + * If the Guest doesn't have a handler (either it hasn't + * registered any yet, or it's one of the faults we don't let + * it handle), it dies with this cryptic error message. + */ + kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)", + cpu->regs->trapnum, cpu->regs->eip, + cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault + : cpu->regs->errcode); +} + +/* + * Now we can look at each of the routines this calls, in increasing order of + * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(), + * deliver_trap() and demand_page(). After all those, we'll be ready to + * examine the Switcher, and our philosophical understanding of the Host/Guest + * duality will be complete. +:*/ +static void adjust_pge(void *on) +{ + if (on) + cr4_set_bits(X86_CR4_PGE); + else + cr4_clear_bits(X86_CR4_PGE); +} + +/*H:020 + * Now the Switcher is mapped and every thing else is ready, we need to do + * some more i386-specific initialization. + */ +void __init lguest_arch_host_init(void) +{ + int i; + + /* + * Most of the x86/switcher_32.S doesn't care that it's been moved; on + * Intel, jumps are relative, and it doesn't access any references to + * external code or data. + * + * The only exception is the interrupt handlers in switcher.S: their + * addresses are placed in a table (default_idt_entries), so we need to + * update the table with the new addresses. switcher_offset() is a + * convenience function which returns the distance between the + * compiled-in switcher code and the high-mapped copy we just made. + */ + for (i = 0; i < IDT_ENTRIES; i++) + default_idt_entries[i] += switcher_offset(); + + /* + * Set up the Switcher's per-cpu areas. + * + * Each CPU gets two pages of its own within the high-mapped region + * (aka. "struct lguest_pages"). Much of this can be initialized now, + * but some depends on what Guest we are running (which is set up in + * copy_in_guest_info()). + */ + for_each_possible_cpu(i) { + /* lguest_pages() returns this CPU's two pages. */ + struct lguest_pages *pages = lguest_pages(i); + /* This is a convenience pointer to make the code neater. */ + struct lguest_ro_state *state = &pages->state; + + /* + * The Global Descriptor Table: the Host has a different one + * for each CPU. We keep a descriptor for the GDT which says + * where it is and how big it is (the size is actually the last + * byte, not the size, hence the "-1"). + */ + state->host_gdt_desc.size = GDT_SIZE-1; + state->host_gdt_desc.address = (long)get_cpu_gdt_table(i); + + /* + * All CPUs on the Host use the same Interrupt Descriptor + * Table, so we just use store_idt(), which gets this CPU's IDT + * descriptor. + */ + store_idt(&state->host_idt_desc); + + /* + * The descriptors for the Guest's GDT and IDT can be filled + * out now, too. We copy the GDT & IDT into ->guest_gdt and + * ->guest_idt before actually running the Guest. + */ + state->guest_idt_desc.size = sizeof(state->guest_idt)-1; + state->guest_idt_desc.address = (long)&state->guest_idt; + state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1; + state->guest_gdt_desc.address = (long)&state->guest_gdt; + + /* + * We know where we want the stack to be when the Guest enters + * the Switcher: in pages->regs. The stack grows upwards, so + * we start it at the end of that structure. + */ + state->guest_tss.sp0 = (long)(&pages->regs + 1); + /* + * And this is the GDT entry to use for the stack: we keep a + * couple of special LGUEST entries. + */ + state->guest_tss.ss0 = LGUEST_DS; + + /* + * x86 can have a finegrained bitmap which indicates what I/O + * ports the process can use. We set it to the end of our + * structure, meaning "none". + */ + state->guest_tss.io_bitmap_base = sizeof(state->guest_tss); + + /* + * Some GDT entries are the same across all Guests, so we can + * set them up now. + */ + setup_default_gdt_entries(state); + /* Most IDT entries are the same for all Guests, too.*/ + setup_default_idt_entries(state, default_idt_entries); + + /* + * The Host needs to be able to use the LGUEST segments on this + * CPU, too, so put them in the Host GDT. + */ + get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; + get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; + } + + /* + * In the Switcher, we want the %cs segment register to use the + * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so + * it will be undisturbed when we switch. To change %cs and jump we + * need this structure to feed to Intel's "lcall" instruction. + */ + lguest_entry.offset = (long)switch_to_guest + switcher_offset(); + lguest_entry.segment = LGUEST_CS; + + /* + * Finally, we need to turn off "Page Global Enable". PGE is an + * optimization where page table entries are specially marked to show + * they never change. The Host kernel marks all the kernel pages this + * way because it's always present, even when userspace is running. + * + * Lguest breaks this: unbeknownst to the rest of the Host kernel, we + * switch to the Guest kernel. If you don't disable this on all CPUs, + * you'll get really weird bugs that you'll chase for two days. + * + * I used to turn PGE off every time we switched to the Guest and back + * on when we return, but that slowed the Switcher down noticibly. + */ + + /* + * We don't need the complexity of CPUs coming and going while we're + * doing this. + */ + get_online_cpus(); + if (cpu_has_pge) { /* We have a broader idea of "global". */ + /* Remember that this was originally set (for cleanup). */ + cpu_had_pge = 1; + /* + * adjust_pge is a helper function which sets or unsets the PGE + * bit on its CPU, depending on the argument (0 == unset). + */ + on_each_cpu(adjust_pge, (void *)0, 1); + /* Turn off the feature in the global feature set. */ + clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); + } + put_online_cpus(); +} +/*:*/ + +void __exit lguest_arch_host_fini(void) +{ + /* If we had PGE before we started, turn it back on now. */ + get_online_cpus(); + if (cpu_had_pge) { + set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); + /* adjust_pge's argument "1" means set PGE. */ + on_each_cpu(adjust_pge, (void *)1, 1); + } + put_online_cpus(); +} + + +/*H:122 The i386-specific hypercalls simply farm out to the right functions. */ +int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args) +{ + switch (args->arg0) { + case LHCALL_LOAD_GDT_ENTRY: + load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3); + break; + case LHCALL_LOAD_IDT_ENTRY: + load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3); + break; + case LHCALL_LOAD_TLS: + guest_load_tls(cpu, args->arg1); + break; + default: + /* Bad Guest. Bad! */ + return -EIO; + } + return 0; +} + +/*H:126 i386-specific hypercall initialization: */ +int lguest_arch_init_hypercalls(struct lg_cpu *cpu) +{ + u32 tsc_speed; + + /* + * The pointer to the Guest's "struct lguest_data" is the only argument. + * We check that address now. + */ + if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1, + sizeof(*cpu->lg->lguest_data))) + return -EFAULT; + + /* + * Having checked it, we simply set lg->lguest_data to point straight + * into the Launcher's memory at the right place and then use + * copy_to_user/from_user from now on, instead of lgread/write. I put + * this in to show that I'm not immune to writing stupid + * optimizations. + */ + cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1; + + /* + * We insist that the Time Stamp Counter exist and doesn't change with + * cpu frequency. Some devious chip manufacturers decided that TSC + * changes could be handled in software. I decided that time going + * backwards might be good for benchmarks, but it's bad for users. + * + * We also insist that the TSC be stable: the kernel detects unreliable + * TSCs for its own purposes, and we use that here. + */ + if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) + tsc_speed = tsc_khz; + else + tsc_speed = 0; + if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz)) + return -EFAULT; + + /* The interrupt code might not like the system call vector. */ + if (!check_syscall_vector(cpu->lg)) + kill_guest(cpu, "bad syscall vector"); + + return 0; +} +/*:*/ + +/*L:030 + * Most of the Guest's registers are left alone: we used get_zeroed_page() to + * allocate the structure, so they will be 0. + */ +void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start) +{ + struct lguest_regs *regs = cpu->regs; + + /* + * There are four "segment" registers which the Guest needs to boot: + * The "code segment" register (cs) refers to the kernel code segment + * __KERNEL_CS, and the "data", "extra" and "stack" segment registers + * refer to the kernel data segment __KERNEL_DS. + * + * The privilege level is packed into the lower bits. The Guest runs + * at privilege level 1 (GUEST_PL). + */ + regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL; + regs->cs = __KERNEL_CS|GUEST_PL; + + /* + * The "eflags" register contains miscellaneous flags. Bit 1 (0x002) + * is supposed to always be "1". Bit 9 (0x200) controls whether + * interrupts are enabled. We always leave interrupts enabled while + * running the Guest. + */ + regs->eflags = X86_EFLAGS_IF | X86_EFLAGS_FIXED; + + /* + * The "Extended Instruction Pointer" register says where the Guest is + * running. + */ + regs->eip = start; + + /* + * %esi points to our boot information, at physical address 0, so don't + * touch it. + */ + + /* There are a couple of GDT entries the Guest expects at boot. */ + setup_guest_gdt(cpu); +} diff --git a/kernel/drivers/lguest/x86/switcher_32.S b/kernel/drivers/lguest/x86/switcher_32.S new file mode 100644 index 000000000..40634b0db --- /dev/null +++ b/kernel/drivers/lguest/x86/switcher_32.S @@ -0,0 +1,388 @@ +/*P:900 + * This is the Switcher: code which sits at 0xFFC00000 (or 0xFFE00000) astride + * both the Host and Guest to do the low-level Guest<->Host switch. It is as + * simple as it can be made, but it's naturally very specific to x86. + * + * You have now completed Preparation. If this has whet your appetite; if you + * are feeling invigorated and refreshed then the next, more challenging stage + * can be found in "make Guest". + :*/ + +/*M:012 + * Lguest is meant to be simple: my rule of thumb is that 1% more LOC must + * gain at least 1% more performance. Since neither LOC nor performance can be + * measured beforehand, it generally means implementing a feature then deciding + * if it's worth it. And once it's implemented, who can say no? + * + * This is why I haven't implemented this idea myself. I want to, but I + * haven't. You could, though. + * + * The main place where lguest performance sucks is Guest page faulting. When + * a Guest userspace process hits an unmapped page we switch back to the Host, + * walk the page tables, find it's not mapped, switch back to the Guest page + * fault handler, which calls a hypercall to set the page table entry, then + * finally returns to userspace. That's two round-trips. + * + * If we had a small walker in the Switcher, we could quickly check the Guest + * page table and if the page isn't mapped, immediately reflect the fault back + * into the Guest. This means the Switcher would have to know the top of the + * Guest page table and the page fault handler address. + * + * For simplicity, the Guest should only handle the case where the privilege + * level of the fault is 3 and probably only not present or write faults. It + * should also detect recursive faults, and hand the original fault to the + * Host (which is actually really easy). + * + * Two questions remain. Would the performance gain outweigh the complexity? + * And who would write the verse documenting it? +:*/ + +/*M:011 + * Lguest64 handles NMI. This gave me NMI envy (until I looked at their + * code). It's worth doing though, since it would let us use oprofile in the + * Host when a Guest is running. +:*/ + +/*S:100 + * Welcome to the Switcher itself! + * + * This file contains the low-level code which changes the CPU to run the Guest + * code, and returns to the Host when something happens. Understand this, and + * you understand the heart of our journey. + * + * Because this is in assembler rather than C, our tale switches from prose to + * verse. First I tried limericks: + * + * There once was an eax reg, + * To which our pointer was fed, + * It needed an add, + * Which asm-offsets.h had + * But this limerick is hurting my head. + * + * Next I tried haikus, but fitting the required reference to the seasons in + * every stanza was quickly becoming tiresome: + * + * The %eax reg + * Holds "struct lguest_pages" now: + * Cherry blossoms fall. + * + * Then I started with Heroic Verse, but the rhyming requirement leeched away + * the content density and led to some uniquely awful oblique rhymes: + * + * These constants are coming from struct offsets + * For use within the asm switcher text. + * + * Finally, I settled for something between heroic hexameter, and normal prose + * with inappropriate linebreaks. Anyway, it aint no Shakespeare. + */ + +// Not all kernel headers work from assembler +// But these ones are needed: the ENTRY() define +// And constants extracted from struct offsets +// To avoid magic numbers and breakage: +// Should they change the compiler can't save us +// Down here in the depths of assembler code. +#include <linux/linkage.h> +#include <asm/asm-offsets.h> +#include <asm/page.h> +#include <asm/segment.h> +#include <asm/lguest.h> + +// We mark the start of the code to copy +// It's placed in .text tho it's never run here +// You'll see the trick macro at the end +// Which interleaves data and text to effect. +.text +ENTRY(start_switcher_text) + +// When we reach switch_to_guest we have just left +// The safe and comforting shores of C code +// %eax has the "struct lguest_pages" to use +// Where we save state and still see it from the Guest +// And %ebx holds the Guest shadow pagetable: +// Once set we have truly left Host behind. +ENTRY(switch_to_guest) + // We told gcc all its regs could fade, + // Clobbered by our journey into the Guest + // We could have saved them, if we tried + // But time is our master and cycles count. + + // Segment registers must be saved for the Host + // We push them on the Host stack for later + pushl %es + pushl %ds + pushl %gs + pushl %fs + // But the compiler is fickle, and heeds + // No warning of %ebp clobbers + // When frame pointers are used. That register + // Must be saved and restored or chaos strikes. + pushl %ebp + // The Host's stack is done, now save it away + // In our "struct lguest_pages" at offset + // Distilled into asm-offsets.h + movl %esp, LGUEST_PAGES_host_sp(%eax) + + // All saved and there's now five steps before us: + // Stack, GDT, IDT, TSS + // Then last of all the page tables are flipped. + + // Yet beware that our stack pointer must be + // Always valid lest an NMI hits + // %edx does the duty here as we juggle + // %eax is lguest_pages: our stack lies within. + movl %eax, %edx + addl $LGUEST_PAGES_regs, %edx + movl %edx, %esp + + // The Guest's GDT we so carefully + // Placed in the "struct lguest_pages" before + lgdt LGUEST_PAGES_guest_gdt_desc(%eax) + + // The Guest's IDT we did partially + // Copy to "struct lguest_pages" as well. + lidt LGUEST_PAGES_guest_idt_desc(%eax) + + // The TSS entry which controls traps + // Must be loaded up with "ltr" now: + // The GDT entry that TSS uses + // Changes type when we load it: damn Intel! + // For after we switch over our page tables + // That entry will be read-only: we'd crash. + movl $(GDT_ENTRY_TSS*8), %edx + ltr %dx + + // Look back now, before we take this last step! + // The Host's TSS entry was also marked used; + // Let's clear it again for our return. + // The GDT descriptor of the Host + // Points to the table after two "size" bytes + movl (LGUEST_PAGES_host_gdt_desc+2)(%eax), %edx + // Clear "used" from type field (byte 5, bit 2) + andb $0xFD, (GDT_ENTRY_TSS*8 + 5)(%edx) + + // Once our page table's switched, the Guest is live! + // The Host fades as we run this final step. + // Our "struct lguest_pages" is now read-only. + movl %ebx, %cr3 + + // The page table change did one tricky thing: + // The Guest's register page has been mapped + // Writable under our %esp (stack) -- + // We can simply pop off all Guest regs. + popl %eax + popl %ebx + popl %ecx + popl %edx + popl %esi + popl %edi + popl %ebp + popl %gs + popl %fs + popl %ds + popl %es + + // Near the base of the stack lurk two strange fields + // Which we fill as we exit the Guest + // These are the trap number and its error + // We can simply step past them on our way. + addl $8, %esp + + // The last five stack slots hold return address + // And everything needed to switch privilege + // From Switcher's level 0 to Guest's 1, + // And the stack where the Guest had last left it. + // Interrupts are turned back on: we are Guest. + iret + +// We tread two paths to switch back to the Host +// Yet both must save Guest state and restore Host +// So we put the routine in a macro. +#define SWITCH_TO_HOST \ + /* We save the Guest state: all registers first \ + * Laid out just as "struct lguest_regs" defines */ \ + pushl %es; \ + pushl %ds; \ + pushl %fs; \ + pushl %gs; \ + pushl %ebp; \ + pushl %edi; \ + pushl %esi; \ + pushl %edx; \ + pushl %ecx; \ + pushl %ebx; \ + pushl %eax; \ + /* Our stack and our code are using segments \ + * Set in the TSS and IDT \ + * Yet if we were to touch data we'd use \ + * Whatever data segment the Guest had. \ + * Load the lguest ds segment for now. */ \ + movl $(LGUEST_DS), %eax; \ + movl %eax, %ds; \ + /* So where are we? Which CPU, which struct? \ + * The stack is our clue: our TSS starts \ + * It at the end of "struct lguest_pages". \ + * Or we may have stumbled while restoring \ + * Our Guest segment regs while in switch_to_guest, \ + * The fault pushed atop that part-unwound stack. \ + * If we round the stack down to the page start \ + * We're at the start of "struct lguest_pages". */ \ + movl %esp, %eax; \ + andl $(~(1 << PAGE_SHIFT - 1)), %eax; \ + /* Save our trap number: the switch will obscure it \ + * (In the Host the Guest regs are not mapped here) \ + * %ebx holds it safe for deliver_to_host */ \ + movl LGUEST_PAGES_regs_trapnum(%eax), %ebx; \ + /* The Host GDT, IDT and stack! \ + * All these lie safely hidden from the Guest: \ + * We must return to the Host page tables \ + * (Hence that was saved in struct lguest_pages) */ \ + movl LGUEST_PAGES_host_cr3(%eax), %edx; \ + movl %edx, %cr3; \ + /* As before, when we looked back at the Host \ + * As we left and marked TSS unused \ + * So must we now for the Guest left behind. */ \ + andb $0xFD, (LGUEST_PAGES_guest_gdt+GDT_ENTRY_TSS*8+5)(%eax); \ + /* Switch to Host's GDT, IDT. */ \ + lgdt LGUEST_PAGES_host_gdt_desc(%eax); \ + lidt LGUEST_PAGES_host_idt_desc(%eax); \ + /* Restore the Host's stack where its saved regs lie */ \ + movl LGUEST_PAGES_host_sp(%eax), %esp; \ + /* Last the TSS: our Host is returned */ \ + movl $(GDT_ENTRY_TSS*8), %edx; \ + ltr %dx; \ + /* Restore now the regs saved right at the first. */ \ + popl %ebp; \ + popl %fs; \ + popl %gs; \ + popl %ds; \ + popl %es + +// The first path is trod when the Guest has trapped: +// (Which trap it was has been pushed on the stack). +// We need only switch back, and the Host will decode +// Why we came home, and what needs to be done. +return_to_host: + SWITCH_TO_HOST + iret + +// We are lead to the second path like so: +// An interrupt, with some cause external +// Has ajerked us rudely from the Guest's code +// Again we must return home to the Host +deliver_to_host: + SWITCH_TO_HOST + // But now we must go home via that place + // Where that interrupt was supposed to go + // Had we not been ensconced, running the Guest. + // Here we see the trickness of run_guest_once(): + // The Host stack is formed like an interrupt + // With EIP, CS and EFLAGS layered. + // Interrupt handlers end with "iret" + // And that will take us home at long long last. + + // But first we must find the handler to call! + // The IDT descriptor for the Host + // Has two bytes for size, and four for address: + // %edx will hold it for us for now. + movl (LGUEST_PAGES_host_idt_desc+2)(%eax), %edx + // We now know the table address we need, + // And saved the trap's number inside %ebx. + // Yet the pointer to the handler is smeared + // Across the bits of the table entry. + // What oracle can tell us how to extract + // From such a convoluted encoding? + // I consulted gcc, and it gave + // These instructions, which I gladly credit: + leal (%edx,%ebx,8), %eax + movzwl (%eax),%edx + movl 4(%eax), %eax + xorw %ax, %ax + orl %eax, %edx + // Now the address of the handler's in %edx + // We call it now: its "iret" drops us home. + jmp *%edx + +// Every interrupt can come to us here +// But we must truly tell each apart. +// They number two hundred and fifty six +// And each must land in a different spot, +// Push its number on stack, and join the stream. + +// And worse, a mere six of the traps stand apart +// And push on their stack an addition: +// An error number, thirty two bits long +// So we punish the other two fifty +// And make them push a zero so they match. + +// Yet two fifty six entries is long +// And all will look most the same as the last +// So we create a macro which can make +// As many entries as we need to fill. + +// Note the change to .data then .text: +// We plant the address of each entry +// Into a (data) table for the Host +// To know where each Guest interrupt should go. +.macro IRQ_STUB N TARGET + .data; .long 1f; .text; 1: + // Trap eight, ten through fourteen and seventeen + // Supply an error number. Else zero. + .if (\N <> 8) && (\N < 10 || \N > 14) && (\N <> 17) + pushl $0 + .endif + pushl $\N + jmp \TARGET + ALIGN +.endm + +// This macro creates numerous entries +// Using GAS macros which out-power C's. +.macro IRQ_STUBS FIRST LAST TARGET + irq=\FIRST + .rept \LAST-\FIRST+1 + IRQ_STUB irq \TARGET + irq=irq+1 + .endr +.endm + +// Here's the marker for our pointer table +// Laid in the data section just before +// Each macro places the address of code +// Forming an array: each one points to text +// Which handles interrupt in its turn. +.data +.global default_idt_entries +default_idt_entries: +.text + // The first two traps go straight back to the Host + IRQ_STUBS 0 1 return_to_host + // We'll say nothing, yet, about NMI + IRQ_STUB 2 handle_nmi + // Other traps also return to the Host + IRQ_STUBS 3 31 return_to_host + // All interrupts go via their handlers + IRQ_STUBS 32 127 deliver_to_host + // 'Cept system calls coming from userspace + // Are to go to the Guest, never the Host. + IRQ_STUB 128 return_to_host + IRQ_STUBS 129 255 deliver_to_host + +// The NMI, what a fabulous beast +// Which swoops in and stops us no matter that +// We're suspended between heaven and hell, +// (Or more likely between the Host and Guest) +// When in it comes! We are dazed and confused +// So we do the simplest thing which one can. +// Though we've pushed the trap number and zero +// We discard them, return, and hope we live. +handle_nmi: + addl $8, %esp + iret + +// We are done; all that's left is Mastery +// And "make Mastery" is a journey long +// Designed to make your fingers itch to code. + +// Here ends the text, the file and poem. +ENTRY(end_switcher_text) |