* (such as the example in Documentation/lguest/lguest.c) is called the
* Launcher.
*
- * Secondly, we only run specially modified Guests, not normal kernels. When
- * you set CONFIG_LGUEST to 'y' or 'm', this automatically sets
- * CONFIG_LGUEST_GUEST=y, which compiles this file into the kernel so it knows
- * how to be a Guest. This means that you can use the same kernel you boot
- * normally (ie. as a Host) as a Guest.
+ * Secondly, we only run specially modified Guests, not normal kernels: setting
+ * CONFIG_LGUEST_GUEST to "y" compiles this file into the kernel so it knows
+ * how to be a Guest at boot time. This means that you can use the same kernel
+ * you boot normally (ie. as a Host) as a Guest.
*
* These Guests know that they cannot do privileged operations, such as disable
* interrupts, and that they have to ask the Host to do such things explicitly.
* This file consists of all the replacements for such low-level native
* hardware operations: these special Guest versions call the Host.
*
- * So how does the kernel know it's a Guest? The Guest starts at a special
- * entry point marked with a magic string, which sets up a few things then
- * calls here. We replace the native functions various "paravirt" structures
- * with our Guest versions, then boot like normal. :*/
+ * So how does the kernel know it's a Guest? We'll see that later, but let's
+ * just say that we end up here where we replace the native functions various
+ * "paravirt" structures with our Guest versions, then boot like normal. :*/
/*
* Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
#include <linux/lguest_launcher.h>
#include <linux/virtio_console.h>
#include <linux/pm.h>
+#include <asm/apic.h>
+#include <asm/lguest.h>
#include <asm/paravirt.h>
#include <asm/param.h>
#include <asm/page.h>
#include <asm/mce.h>
#include <asm/io.h>
#include <asm/i387.h>
+#include <asm/stackprotector.h>
+#include <asm/reboot.h> /* for struct machine_ops */
/*G:010 Welcome to the Guest!
*
* behaving in simplified but equivalent ways. In particular, the Guest is the
* same kernel as the Host (or at least, built from the same source code). :*/
-/* Declarations for definitions in lguest_guest.S */
-extern char lguest_noirq_start[], lguest_noirq_end[];
-extern const char lgstart_cli[], lgend_cli[];
-extern const char lgstart_sti[], lgend_sti[];
-extern const char lgstart_popf[], lgend_popf[];
-extern const char lgstart_pushf[], lgend_pushf[];
-extern const char lgstart_iret[], lgend_iret[];
-extern void lguest_iret(void);
-
struct lguest_data lguest_data = {
.hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
.noirq_start = (u32)lguest_noirq_start,
.blocked_interrupts = { 1 }, /* Block timer interrupts */
.syscall_vec = SYSCALL_VECTOR,
};
-static cycle_t clock_base;
/*G:037 async_hcall() is pretty simple: I'm quite proud of it really. We have a
* ring buffer of stored hypercalls which the Host will run though next time we
- * do a normal hypercall. Each entry in the ring has 4 slots for the hypercall
+ * do a normal hypercall. Each entry in the ring has 5 slots for the hypercall
* arguments, and a "hcall_status" word which is 0 if the call is ready to go,
* and 255 once the Host has finished with it.
*
* effect of causing the Host to run all the stored calls in the ring buffer
* which empties it for next time! */
static void async_hcall(unsigned long call, unsigned long arg1,
- unsigned long arg2, unsigned long arg3)
+ unsigned long arg2, unsigned long arg3,
+ unsigned long arg4)
{
/* Note: This code assumes we're uniprocessor. */
static unsigned int next_call;
local_irq_save(flags);
if (lguest_data.hcall_status[next_call] != 0xFF) {
/* Table full, so do normal hcall which will flush table. */
- hcall(call, arg1, arg2, arg3);
+ kvm_hypercall4(call, arg1, arg2, arg3, arg4);
} else {
lguest_data.hcalls[next_call].arg0 = call;
lguest_data.hcalls[next_call].arg1 = arg1;
lguest_data.hcalls[next_call].arg2 = arg2;
lguest_data.hcalls[next_call].arg3 = arg3;
+ lguest_data.hcalls[next_call].arg4 = arg4;
/* Arguments must all be written before we mark it to go */
wmb();
lguest_data.hcall_status[next_call] = 0;
* lguest_leave_lazy_mode().
*
* So, when we're in lazy mode, we call async_hcall() to store the call for
- * future processing. */
-static void lazy_hcall(unsigned long call,
+ * future processing: */
+static void lazy_hcall1(unsigned long call,
+ unsigned long arg1)
+{
+ if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
+ kvm_hypercall1(call, arg1);
+ else
+ async_hcall(call, arg1, 0, 0, 0);
+}
+
+static void lazy_hcall2(unsigned long call,
+ unsigned long arg1,
+ unsigned long arg2)
+{
+ if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
+ kvm_hypercall2(call, arg1, arg2);
+ else
+ async_hcall(call, arg1, arg2, 0, 0);
+}
+
+static void lazy_hcall3(unsigned long call,
unsigned long arg1,
unsigned long arg2,
unsigned long arg3)
{
if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
- hcall(call, arg1, arg2, arg3);
+ kvm_hypercall3(call, arg1, arg2, arg3);
else
- async_hcall(call, arg1, arg2, arg3);
+ async_hcall(call, arg1, arg2, arg3, 0);
}
+#ifdef CONFIG_X86_PAE
+static void lazy_hcall4(unsigned long call,
+ unsigned long arg1,
+ unsigned long arg2,
+ unsigned long arg3,
+ unsigned long arg4)
+{
+ if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
+ kvm_hypercall4(call, arg1, arg2, arg3, arg4);
+ else
+ async_hcall(call, arg1, arg2, arg3, arg4);
+}
+#endif
+
/* When lazy mode is turned off reset the per-cpu lazy mode variable and then
- * issue a hypercall to flush any stored calls. */
-static void lguest_leave_lazy_mode(void)
+ * issue the do-nothing hypercall to flush any stored calls. */
+static void lguest_leave_lazy_mmu_mode(void)
+{
+ kvm_hypercall0(LHCALL_FLUSH_ASYNC);
+ paravirt_leave_lazy_mmu();
+}
+
+static void lguest_end_context_switch(struct task_struct *next)
{
- paravirt_leave_lazy(paravirt_get_lazy_mode());
- hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0);
+ kvm_hypercall0(LHCALL_FLUSH_ASYNC);
+ paravirt_end_context_switch(next);
}
-/*G:033
+/*G:032
* After that diversion we return to our first native-instruction
* replacements: four functions for interrupt control.
*
*
* So instead we keep an "irq_enabled" field inside our "struct lguest_data",
* which the Guest can update with a single instruction. The Host knows to
- * check there when it wants to deliver an interrupt.
+ * check there before it tries to deliver an interrupt.
*/
-/* save_flags() is expected to return the processor state (ie. "eflags"). The
- * eflags word contains all kind of stuff, but in practice Linux only cares
+/* save_flags() is expected to return the processor state (ie. "flags"). The
+ * flags word contains all kind of stuff, but in practice Linux only cares
* about the interrupt flag. Our "save_flags()" just returns that. */
static unsigned long save_fl(void)
{
return lguest_data.irq_enabled;
}
-/* restore_flags() just sets the flags back to the value given. */
-static void restore_fl(unsigned long flags)
-{
- lguest_data.irq_enabled = flags;
-}
-
/* Interrupts go off... */
static void irq_disable(void)
{
lguest_data.irq_enabled = 0;
}
-/* Interrupts go on... */
-static void irq_enable(void)
-{
- lguest_data.irq_enabled = X86_EFLAGS_IF;
-}
+/* Let's pause a moment. Remember how I said these are called so often?
+ * Jeremy Fitzhardinge optimized them so hard early in 2009 that he had to
+ * break some rules. In particular, these functions are assumed to save their
+ * own registers if they need to: normal C functions assume they can trash the
+ * eax register. To use normal C functions, we use
+ * PV_CALLEE_SAVE_REGS_THUNK(), which pushes %eax onto the stack, calls the
+ * C function, then restores it. */
+PV_CALLEE_SAVE_REGS_THUNK(save_fl);
+PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
/*:*/
+
+/* These are in i386_head.S */
+extern void lg_irq_enable(void);
+extern void lg_restore_fl(unsigned long flags);
+
/*M:003 Note that we don't check for outstanding interrupts when we re-enable
* them (or when we unmask an interrupt). This seems to work for the moment,
* since interrupts are rare and we'll just get the interrupt on the next timer
- * tick, but when we turn on CONFIG_NO_HZ, we should revisit this. One way
+ * tick, but now we can run with CONFIG_NO_HZ, we should revisit this. One way
* would be to put the "irq_enabled" field in a page by itself, and have the
* Host write-protect it when an interrupt comes in when irqs are disabled.
- * There will then be a page fault as soon as interrupts are re-enabled. :*/
+ * There will then be a page fault as soon as interrupts are re-enabled.
+ *
+ * A better method is to implement soft interrupt disable generally for x86:
+ * instead of disabling interrupts, we set a flag. If an interrupt does come
+ * in, we then disable them for real. This is uncommon, so we could simply use
+ * a hypercall for interrupt control and not worry about efficiency. :*/
/*G:034
* The Interrupt Descriptor Table (IDT).
* address of the handler, and... well, who cares? The Guest just asks the
* Host to make the change anyway, because the Host controls the real IDT.
*/
-static void lguest_write_idt_entry(struct desc_struct *dt,
- int entrynum, u32 low, u32 high)
+static void lguest_write_idt_entry(gate_desc *dt,
+ int entrynum, const gate_desc *g)
{
+ /* The gate_desc structure is 8 bytes long: we hand it to the Host in
+ * two 32-bit chunks. The whole 32-bit kernel used to hand descriptors
+ * around like this; typesafety wasn't a big concern in Linux's early
+ * years. */
+ u32 *desc = (u32 *)g;
/* Keep the local copy up to date. */
- write_dt_entry(dt, entrynum, low, high);
+ native_write_idt_entry(dt, entrynum, g);
/* Tell Host about this new entry. */
- hcall(LHCALL_LOAD_IDT_ENTRY, entrynum, low, high);
+ kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1]);
}
/* Changing to a different IDT is very rare: we keep the IDT up-to-date every
* time it is written, so we can simply loop through all entries and tell the
* Host about them. */
-static void lguest_load_idt(const struct Xgt_desc_struct *desc)
+static void lguest_load_idt(const struct desc_ptr *desc)
{
unsigned int i;
struct desc_struct *idt = (void *)desc->address;
for (i = 0; i < (desc->size+1)/8; i++)
- hcall(LHCALL_LOAD_IDT_ENTRY, i, idt[i].a, idt[i].b);
+ kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, i, idt[i].a, idt[i].b);
}
/*
* controls the entire thing and the Guest asks it to make changes using the
* LOAD_GDT hypercall.
*
- * This is the opposite of the IDT code where we have a LOAD_IDT_ENTRY
- * hypercall and use that repeatedly to load a new IDT. I don't think it
- * really matters, but wouldn't it be nice if they were the same?
+ * This is the exactly like the IDT code.
*/
-static void lguest_load_gdt(const struct Xgt_desc_struct *desc)
+static void lguest_load_gdt(const struct desc_ptr *desc)
{
- BUG_ON((desc->size+1)/8 != GDT_ENTRIES);
- hcall(LHCALL_LOAD_GDT, __pa(desc->address), GDT_ENTRIES, 0);
+ unsigned int i;
+ struct desc_struct *gdt = (void *)desc->address;
+
+ for (i = 0; i < (desc->size+1)/8; i++)
+ kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b);
}
/* For a single GDT entry which changes, we do the lazy thing: alter our GDT,
* then tell the Host to reload the entire thing. This operation is so rare
* that this naive implementation is reasonable. */
-static void lguest_write_gdt_entry(struct desc_struct *dt,
- int entrynum, u32 low, u32 high)
+static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
+ const void *desc, int type)
{
- write_dt_entry(dt, entrynum, low, high);
- hcall(LHCALL_LOAD_GDT, __pa(dt), GDT_ENTRIES, 0);
+ native_write_gdt_entry(dt, entrynum, desc, type);
+ /* Tell Host about this new entry. */
+ kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, entrynum,
+ dt[entrynum].a, dt[entrynum].b);
}
/* OK, I lied. There are three "thread local storage" GDT entries which change
/* There's one problem which normal hardware doesn't have: the Host
* can't handle us removing entries we're currently using. So we clear
* the GS register here: if it's needed it'll be reloaded anyway. */
- loadsegment(gs, 0);
- lazy_hcall(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu, 0);
+ lazy_load_gs(0);
+ lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu);
}
/*G:038 That's enough excitement for now, back to ploughing through each of
/* The "cpuid" instruction is a way of querying both the CPU identity
* (manufacturer, model, etc) and its features. It was introduced before the
- * Pentium in 1993 and keeps getting extended by both Intel and AMD. As you
- * might imagine, after a decade and a half this treatment, it is now a giant
- * ball of hair. Its entry in the current Intel manual runs to 28 pages.
+ * Pentium in 1993 and keeps getting extended by both Intel, AMD and others.
+ * As you might imagine, after a decade and a half this treatment, it is now a
+ * giant ball of hair. Its entry in the current Intel manual runs to 28 pages.
*
* This instruction even it has its own Wikipedia entry. The Wikipedia entry
* has been translated into 4 languages. I am not making this up!
* anyone (including userspace) can just use the raw "cpuid" instruction and
* the Host won't even notice since it isn't privileged. So we try not to get
* too worked up about it. */
-static void lguest_cpuid(unsigned int *eax, unsigned int *ebx,
- unsigned int *ecx, unsigned int *edx)
+static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
+ unsigned int *cx, unsigned int *dx)
{
- int function = *eax;
+ int function = *ax;
- native_cpuid(eax, ebx, ecx, edx);
+ native_cpuid(ax, bx, cx, dx);
switch (function) {
+ case 0: /* ID and highest CPUID. Futureproof a little by sticking to
+ * older ones. */
+ if (*ax > 5)
+ *ax = 5;
+ break;
case 1: /* Basic feature request. */
/* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */
- *ecx &= 0x00002201;
- /* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, FPU. */
- *edx &= 0x07808101;
+ *cx &= 0x00002201;
+ /* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU, PAE. */
+ *dx &= 0x07808151;
/* The Host can do a nice optimization if it knows that the
* kernel mappings (addresses above 0xC0000000 or whatever
* PAGE_OFFSET is set to) haven't changed. But Linux calls
* flush_tlb_user() for both user and kernel mappings unless
* the Page Global Enable (PGE) feature bit is set. */
- *edx |= 0x00002000;
+ *dx |= 0x00002000;
+ /* We also lie, and say we're family id 5. 6 or greater
+ * leads to a rdmsr in early_init_intel which we can't handle.
+ * Family ID is returned as bits 8-12 in ax. */
+ *ax &= 0xFFFFF0FF;
+ *ax |= 0x00000500;
break;
case 0x80000000:
/* Futureproof this a little: if they ask how much extended
* processor information there is, limit it to known fields. */
- if (*eax > 0x80000008)
- *eax = 0x80000008;
+ if (*ax > 0x80000008)
+ *ax = 0x80000008;
+ break;
+ case 0x80000001:
+ /* Here we should fix nx cap depending on host. */
+ /* For this version of PAE, we just clear NX bit. */
+ *dx &= ~(1 << 20);
break;
}
}
* lazily after a task switch, and Linux uses that gratefully, but wouldn't a
* name like "FPUTRAP bit" be a little less cryptic?
*
- * We store cr0 (and cr3) locally, because the Host never changes it. The
- * Guest sometimes wants to read it and we'd prefer not to bother the Host
- * unnecessarily. */
-static unsigned long current_cr0, current_cr3;
+ * We store cr0 locally because the Host never changes it. The Guest sometimes
+ * wants to read it and we'd prefer not to bother the Host unnecessarily. */
+static unsigned long current_cr0;
static void lguest_write_cr0(unsigned long val)
{
- lazy_hcall(LHCALL_TS, val & X86_CR0_TS, 0, 0);
+ lazy_hcall1(LHCALL_TS, val & X86_CR0_TS);
current_cr0 = val;
}
* the vowels have been optimized out. */
static void lguest_clts(void)
{
- lazy_hcall(LHCALL_TS, 0, 0, 0);
+ lazy_hcall1(LHCALL_TS, 0);
current_cr0 &= ~X86_CR0_TS;
}
return lguest_data.cr2;
}
+/* See lguest_set_pte() below. */
+static bool cr3_changed = false;
+
/* cr3 is the current toplevel pagetable page: the principle is the same as
- * cr0. Keep a local copy, and tell the Host when it changes. */
+ * cr0. Keep a local copy, and tell the Host when it changes. The only
+ * difference is that our local copy is in lguest_data because the Host needs
+ * to set it upon our initial hypercall. */
static void lguest_write_cr3(unsigned long cr3)
{
- lazy_hcall(LHCALL_NEW_PGTABLE, cr3, 0, 0);
- current_cr3 = cr3;
+ lguest_data.pgdir = cr3;
+ lazy_hcall1(LHCALL_NEW_PGTABLE, cr3);
+ cr3_changed = true;
}
static unsigned long lguest_read_cr3(void)
{
- return current_cr3;
+ return lguest_data.pgdir;
}
/* cr4 is used to enable and disable PGE, but we don't care. */
* into a process' address space. We set the entry then tell the Host the
* toplevel and address this corresponds to. The Guest uses one pagetable per
* process, so we need to tell the Host which one we're changing (mm->pgd). */
+static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep)
+{
+#ifdef CONFIG_X86_PAE
+ lazy_hcall4(LHCALL_SET_PTE, __pa(mm->pgd), addr,
+ ptep->pte_low, ptep->pte_high);
+#else
+ lazy_hcall3(LHCALL_SET_PTE, __pa(mm->pgd), addr, ptep->pte_low);
+#endif
+}
+
static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
pte_t *ptep, pte_t pteval)
{
- *ptep = pteval;
- lazy_hcall(LHCALL_SET_PTE, __pa(mm->pgd), addr, pteval.pte_low);
+ native_set_pte(ptep, pteval);
+ lguest_pte_update(mm, addr, ptep);
+}
+
+/* The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
+ * to set a middle-level entry when PAE is activated.
+ * Again, we set the entry then tell the Host which page we changed,
+ * and the index of the entry we changed. */
+#ifdef CONFIG_X86_PAE
+static void lguest_set_pud(pud_t *pudp, pud_t pudval)
+{
+ native_set_pud(pudp, pudval);
+
+ /* 32 bytes aligned pdpt address and the index. */
+ lazy_hcall2(LHCALL_SET_PGD, __pa(pudp) & 0xFFFFFFE0,
+ (__pa(pudp) & 0x1F) / sizeof(pud_t));
+}
+
+static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
+{
+ native_set_pmd(pmdp, pmdval);
+ lazy_hcall2(LHCALL_SET_PMD, __pa(pmdp) & PAGE_MASK,
+ (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
}
+#else
-/* The Guest calls this to set a top-level entry. Again, we set the entry then
- * tell the Host which top-level page we changed, and the index of the entry we
- * changed. */
+/* The Guest calls lguest_set_pmd to set a top-level entry when PAE is not
+ * activated. */
static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
{
- *pmdp = pmdval;
- lazy_hcall(LHCALL_SET_PMD, __pa(pmdp)&PAGE_MASK,
- (__pa(pmdp)&(PAGE_SIZE-1))/4, 0);
+ native_set_pmd(pmdp, pmdval);
+ lazy_hcall2(LHCALL_SET_PGD, __pa(pmdp) & PAGE_MASK,
+ (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
}
+#endif
/* There are a couple of legacy places where the kernel sets a PTE, but we
* don't know the top level any more. This is useless for us, since we don't
* to forget all of them. Fortunately, this is very rare.
*
* ... except in early boot when the kernel sets up the initial pagetables,
- * which makes booting astonishingly slow. So we don't even tell the Host
- * anything changed until we've done the first page table switch. */
+ * which makes booting astonishingly slow: 1.83 seconds! So we don't even tell
+ * the Host anything changed until we've done the first page table switch,
+ * which brings boot back to 0.25 seconds. */
static void lguest_set_pte(pte_t *ptep, pte_t pteval)
{
- *ptep = pteval;
- /* Don't bother with hypercall before initial setup. */
- if (current_cr3)
- lazy_hcall(LHCALL_FLUSH_TLB, 1, 0, 0);
+ native_set_pte(ptep, pteval);
+ if (cr3_changed)
+ lazy_hcall1(LHCALL_FLUSH_TLB, 1);
+}
+
+#ifdef CONFIG_X86_PAE
+static void lguest_set_pte_atomic(pte_t *ptep, pte_t pte)
+{
+ native_set_pte_atomic(ptep, pte);
+ if (cr3_changed)
+ lazy_hcall1(LHCALL_FLUSH_TLB, 1);
+}
+
+void lguest_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
+{
+ native_pte_clear(mm, addr, ptep);
+ lguest_pte_update(mm, addr, ptep);
+}
+
+void lguest_pmd_clear(pmd_t *pmdp)
+{
+ lguest_set_pmd(pmdp, __pmd(0));
}
+#endif
/* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
* native page table operations. On native hardware you can set a new page
static void lguest_flush_tlb_single(unsigned long addr)
{
/* Simply set it to zero: if it was not, it will fault back in. */
- lazy_hcall(LHCALL_SET_PTE, current_cr3, addr, 0);
+ lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0);
}
/* This is what happens after the Guest has removed a large number of entries.
* have changed, ie. virtual addresses below PAGE_OFFSET. */
static void lguest_flush_tlb_user(void)
{
- lazy_hcall(LHCALL_FLUSH_TLB, 0, 0, 0);
+ lazy_hcall1(LHCALL_FLUSH_TLB, 0);
}
/* This is called when the kernel page tables have changed. That's not very
* slow), so it's worth separating this from the user flushing above. */
static void lguest_flush_tlb_kernel(void)
{
- lazy_hcall(LHCALL_FLUSH_TLB, 1, 0, 0);
+ lazy_hcall1(LHCALL_FLUSH_TLB, 1);
}
/*
{
unsigned int i;
- for (i = 0; i < LGUEST_IRQS; i++) {
- int vector = FIRST_EXTERNAL_VECTOR + i;
- if (vector != SYSCALL_VECTOR) {
- set_intr_gate(vector, interrupt[i]);
- set_irq_chip_and_handler(i, &lguest_irq_controller,
- handle_level_irq);
- }
+ for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
+ /* Some systems map "vectors" to interrupts weirdly. Lguest has
+ * a straightforward 1 to 1 mapping, so force that here. */
+ __get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR;
+ if (i != SYSCALL_VECTOR)
+ set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]);
}
/* This call is required to set up for 4k stacks, where we have
* separate stacks for hard and soft interrupts. */
irq_ctx_init(smp_processor_id());
}
+void lguest_setup_irq(unsigned int irq)
+{
+ irq_to_desc_alloc_node(irq, 0);
+ set_irq_chip_and_handler_name(irq, &lguest_irq_controller,
+ handle_level_irq, "level");
+}
+
/*
* Time.
*
return lguest_data.time.tv_sec;
}
-static cycle_t lguest_clock_read(void)
+/* The TSC is an Intel thing called the Time Stamp Counter. The Host tells us
+ * what speed it runs at, or 0 if it's unusable as a reliable clock source.
+ * This matches what we want here: if we return 0 from this function, the x86
+ * TSC clock will give up and not register itself. */
+static unsigned long lguest_tsc_khz(void)
{
- unsigned long sec, nsec;
+ return lguest_data.tsc_khz;
+}
- /* If the Host tells the TSC speed, we can trust that. */
- if (lguest_data.tsc_khz)
- return native_read_tsc();
+/* If we can't use the TSC, the kernel falls back to our lower-priority
+ * "lguest_clock", where we read the time value given to us by the Host. */
+static cycle_t lguest_clock_read(struct clocksource *cs)
+{
+ unsigned long sec, nsec;
- /* If we can't use the TSC, we read the time value written by the Host.
- * Since it's in two parts (seconds and nanoseconds), we risk reading
- * it just as it's changing from 99 & 0.999999999 to 100 and 0, and
- * getting 99 and 0. As Linux tends to come apart under the stress of
- * time travel, we must be careful: */
+ /* Since the time is in two parts (seconds and nanoseconds), we risk
+ * reading it just as it's changing from 99 & 0.999999999 to 100 and 0,
+ * and getting 99 and 0. As Linux tends to come apart under the stress
+ * of time travel, we must be careful: */
do {
/* First we read the seconds part. */
sec = lguest_data.time.tv_sec;
/* Now if the seconds part has changed, try again. */
} while (unlikely(lguest_data.time.tv_sec != sec));
- /* Our non-TSC clock is in real nanoseconds. */
+ /* Our lguest clock is in real nanoseconds. */
return sec*1000000000ULL + nsec;
}
-/* This is what we tell the kernel is our clocksource. */
+/* This is the fallback clocksource: lower priority than the TSC clocksource. */
static struct clocksource lguest_clock = {
.name = "lguest",
- .rating = 400,
+ .rating = 200,
.read = lguest_clock_read,
.mask = CLOCKSOURCE_MASK(64),
.mult = 1 << 22,
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
};
-/* The "scheduler clock" is just our real clock, adjusted to start at zero */
-static unsigned long long lguest_sched_clock(void)
-{
- return cyc2ns(&lguest_clock, lguest_clock_read() - clock_base);
-}
-
/* We also need a "struct clock_event_device": Linux asks us to set it to go
* off some time in the future. Actually, James Morris figured all this out, I
* just applied the patch. */
static int lguest_clockevent_set_next_event(unsigned long delta,
struct clock_event_device *evt)
{
+ /* FIXME: I don't think this can ever happen, but James tells me he had
+ * to put this code in. Maybe we should remove it now. Anyone? */
if (delta < LG_CLOCK_MIN_DELTA) {
if (printk_ratelimit())
printk(KERN_DEBUG "%s: small delta %lu ns\n",
- __FUNCTION__, delta);
+ __func__, delta);
return -ETIME;
}
- hcall(LHCALL_SET_CLOCKEVENT, delta, 0, 0);
+
+ /* Please wake us this far in the future. */
+ kvm_hypercall1(LHCALL_SET_CLOCKEVENT, delta);
return 0;
}
case CLOCK_EVT_MODE_UNUSED:
case CLOCK_EVT_MODE_SHUTDOWN:
/* A 0 argument shuts the clock down. */
- hcall(LHCALL_SET_CLOCKEVENT, 0, 0, 0);
+ kvm_hypercall0(LHCALL_SET_CLOCKEVENT);
break;
case CLOCK_EVT_MODE_ONESHOT:
/* This is what we expect. */
/* Set up the timer interrupt (0) to go to our simple timer routine */
set_irq_handler(0, lguest_time_irq);
- /* Our clock structure looks like arch/x86/kernel/tsc_32.c if we can
- * use the TSC, otherwise it's a dumb nanosecond-resolution clock.
- * Either way, the "rating" is set so high that it's always chosen over
- * any other clocksource. */
- if (lguest_data.tsc_khz)
- lguest_clock.mult = clocksource_khz2mult(lguest_data.tsc_khz,
- lguest_clock.shift);
- clock_base = lguest_clock_read();
clocksource_register(&lguest_clock);
- /* Now we've set up our clock, we can use it as the scheduler clock */
- pv_time_ops.sched_clock = lguest_sched_clock;
-
/* We can't set cpumask in the initializer: damn C limitations! Set it
* here and register our timer device. */
- lguest_clockevent.cpumask = cpumask_of_cpu(0);
+ lguest_clockevent.cpumask = cpumask_of(0);
clockevents_register_device(&lguest_clockevent);
/* Finally, we unblock the timer interrupt. */
* segment), the privilege level (we're privilege level 1, the Host is 0 and
* will not tolerate us trying to use that), the stack pointer, and the number
* of pages in the stack. */
-static void lguest_load_esp0(struct tss_struct *tss,
- struct thread_struct *thread)
+static void lguest_load_sp0(struct tss_struct *tss,
+ struct thread_struct *thread)
{
- lazy_hcall(LHCALL_SET_STACK, __KERNEL_DS|0x1, thread->esp0,
- THREAD_SIZE/PAGE_SIZE);
+ lazy_hcall3(LHCALL_SET_STACK, __KERNEL_DS | 0x1, thread->sp0,
+ THREAD_SIZE / PAGE_SIZE);
}
/* Let's just say, I wouldn't do debugging under a Guest. */
* code qualifies for Advanced. It will also never interrupt anything. It
* does, however, allow us to get through the Linux boot code. */
#ifdef CONFIG_X86_LOCAL_APIC
-static void lguest_apic_write(unsigned long reg, u32 v)
+static void lguest_apic_write(u32 reg, u32 v)
+{
+}
+
+static u32 lguest_apic_read(u32 reg)
+{
+ return 0;
+}
+
+static u64 lguest_apic_icr_read(void)
+{
+ return 0;
+}
+
+static void lguest_apic_icr_write(u32 low, u32 id)
{
+ /* Warn to see if there's any stray references */
+ WARN_ON(1);
}
-static u32 lguest_apic_read(unsigned long reg)
+static void lguest_apic_wait_icr_idle(void)
+{
+ return;
+}
+
+static u32 lguest_apic_safe_wait_icr_idle(void)
{
return 0;
}
+
+static void set_lguest_basic_apic_ops(void)
+{
+ apic->read = lguest_apic_read;
+ apic->write = lguest_apic_write;
+ apic->icr_read = lguest_apic_icr_read;
+ apic->icr_write = lguest_apic_icr_write;
+ apic->wait_icr_idle = lguest_apic_wait_icr_idle;
+ apic->safe_wait_icr_idle = lguest_apic_safe_wait_icr_idle;
+};
#endif
/* STOP! Until an interrupt comes in. */
static void lguest_safe_halt(void)
{
- hcall(LHCALL_HALT, 0, 0, 0);
+ kvm_hypercall0(LHCALL_HALT);
}
-/* Perhaps CRASH isn't the best name for this hypercall, but we use it to get a
- * message out when we're crashing as well as elegant termination like powering
- * off.
+/* The SHUTDOWN hypercall takes a string to describe what's happening, and
+ * an argument which says whether this to restart (reboot) the Guest or not.
*
* Note that the Host always prefers that the Guest speak in physical addresses
* rather than virtual addresses, so we use __pa() here. */
static void lguest_power_off(void)
{
- hcall(LHCALL_CRASH, __pa("Power down"), 0, 0);
+ kvm_hypercall2(LHCALL_SHUTDOWN, __pa("Power down"),
+ LGUEST_SHUTDOWN_POWEROFF);
}
/*
*/
static int lguest_panic(struct notifier_block *nb, unsigned long l, void *p)
{
- hcall(LHCALL_CRASH, __pa(p), 0, 0);
+ kvm_hypercall2(LHCALL_SHUTDOWN, __pa(p), LGUEST_SHUTDOWN_POWEROFF);
/* The hcall won't return, but to keep gcc happy, we're "done". */
return NOTIFY_DONE;
}
/* Setting up memory is fairly easy. */
static __init char *lguest_memory_setup(void)
{
- /* We do this here and not earlier because lockcheck barfs if we do it
- * before start_kernel() */
+ /* We do this here and not earlier because lockcheck used to barf if we
+ * did it before start_kernel(). I think we fixed that, so it'd be
+ * nice to move it back to lguest_init. Patch welcome... */
atomic_notifier_chain_register(&panic_notifier_list, &paniced);
/* The Linux bootloader header contains an "e820" memory map: the
* Launcher populated the first entry with our memory limit. */
- add_memory_region(boot_params.e820_map[0].addr,
+ e820_add_region(boot_params.e820_map[0].addr,
boot_params.e820_map[0].size,
boot_params.e820_map[0].type);
len = sizeof(scratch) - 1;
scratch[len] = '\0';
memcpy(scratch, buf, len);
- hcall(LHCALL_NOTIFY, __pa(scratch), 0, 0);
+ kvm_hypercall1(LHCALL_NOTIFY, __pa(scratch));
/* This routine returns the number of bytes actually written. */
return len;
}
+/* Rebooting also tells the Host we're finished, but the RESTART flag tells the
+ * Launcher to reboot us. */
+static void lguest_restart(char *reason)
+{
+ kvm_hypercall2(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART);
+}
+
/*G:050
* Patching (Powerfully Placating Performance Pedants)
*
- * We have already seen that pv_ops structures let us replace simple
- * native instructions with calls to the appropriate back end all throughout
- * the kernel. This allows the same kernel to run as a Guest and as a native
+ * We have already seen that pv_ops structures let us replace simple native
+ * instructions with calls to the appropriate back end all throughout the
+ * kernel. This allows the same kernel to run as a Guest and as a native
* kernel, but it's slow because of all the indirect branches.
*
* Remember that David Wheeler quote about "Any problem in computer science can
*
* Our current solution is to allow the paravirt back end to optionally patch
* over the indirect calls to replace them with something more efficient. We
- * patch the four most commonly called functions: disable interrupts, enable
- * interrupts, restore interrupts and save interrupts. We usually have 6 or 10
- * bytes to patch into: the Guest versions of these operations are small enough
- * that we can fit comfortably.
+ * patch two of the simplest of the most commonly called functions: disable
+ * interrupts and save interrupts. We usually have 6 or 10 bytes to patch
+ * into: the Guest versions of these operations are small enough that we can
+ * fit comfortably.
*
* First we need assembly templates of each of the patchable Guest operations,
- * and these are in lguest_asm.S. */
+ * and these are in i386_head.S. */
/*G:060 We construct a table from the assembler templates: */
static const struct lguest_insns
const char *start, *end;
} lguest_insns[] = {
[PARAVIRT_PATCH(pv_irq_ops.irq_disable)] = { lgstart_cli, lgend_cli },
- [PARAVIRT_PATCH(pv_irq_ops.irq_enable)] = { lgstart_sti, lgend_sti },
- [PARAVIRT_PATCH(pv_irq_ops.restore_fl)] = { lgstart_popf, lgend_popf },
[PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
};
return insn_len;
}
-/*G:030 Once we get to lguest_init(), we know we're a Guest. The pv_ops
- * structures in the kernel provide points for (almost) every routine we have
- * to override to avoid privileged instructions. */
+/*G:029 Once we get to lguest_init(), we know we're a Guest. The various
+ * pv_ops structures in the kernel provide points for (almost) every routine we
+ * have to override to avoid privileged instructions. */
__init void lguest_init(void)
{
/* We're under lguest, paravirt is enabled, and we're running at
pv_info.name = "lguest";
pv_info.paravirt_enabled = 1;
pv_info.kernel_rpl = 1;
+ pv_info.shared_kernel_pmd = 1;
/* We set up all the lguest overrides for sensitive operations. These
* are detailed with the operations themselves. */
/* interrupt-related operations */
pv_irq_ops.init_IRQ = lguest_init_IRQ;
- pv_irq_ops.save_fl = save_fl;
- pv_irq_ops.restore_fl = restore_fl;
- pv_irq_ops.irq_disable = irq_disable;
- pv_irq_ops.irq_enable = irq_enable;
+ pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl);
+ pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl);
+ pv_irq_ops.irq_disable = PV_CALLEE_SAVE(irq_disable);
+ pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable);
pv_irq_ops.safe_halt = lguest_safe_halt;
/* init-time operations */
pv_cpu_ops.cpuid = lguest_cpuid;
pv_cpu_ops.load_idt = lguest_load_idt;
pv_cpu_ops.iret = lguest_iret;
- pv_cpu_ops.load_esp0 = lguest_load_esp0;
+ pv_cpu_ops.load_sp0 = lguest_load_sp0;
pv_cpu_ops.load_tr_desc = lguest_load_tr_desc;
pv_cpu_ops.set_ldt = lguest_set_ldt;
pv_cpu_ops.load_tls = lguest_load_tls;
pv_cpu_ops.write_gdt_entry = lguest_write_gdt_entry;
pv_cpu_ops.write_idt_entry = lguest_write_idt_entry;
pv_cpu_ops.wbinvd = lguest_wbinvd;
- pv_cpu_ops.lazy_mode.enter = paravirt_enter_lazy_cpu;
- pv_cpu_ops.lazy_mode.leave = lguest_leave_lazy_mode;
+ pv_cpu_ops.start_context_switch = paravirt_start_context_switch;
+ pv_cpu_ops.end_context_switch = lguest_end_context_switch;
/* pagetable management */
pv_mmu_ops.write_cr3 = lguest_write_cr3;
pv_mmu_ops.set_pte = lguest_set_pte;
pv_mmu_ops.set_pte_at = lguest_set_pte_at;
pv_mmu_ops.set_pmd = lguest_set_pmd;
+#ifdef CONFIG_X86_PAE
+ pv_mmu_ops.set_pte_atomic = lguest_set_pte_atomic;
+ pv_mmu_ops.pte_clear = lguest_pte_clear;
+ pv_mmu_ops.pmd_clear = lguest_pmd_clear;
+ pv_mmu_ops.set_pud = lguest_set_pud;
+#endif
pv_mmu_ops.read_cr2 = lguest_read_cr2;
pv_mmu_ops.read_cr3 = lguest_read_cr3;
pv_mmu_ops.lazy_mode.enter = paravirt_enter_lazy_mmu;
- pv_mmu_ops.lazy_mode.leave = lguest_leave_lazy_mode;
+ pv_mmu_ops.lazy_mode.leave = lguest_leave_lazy_mmu_mode;
+ pv_mmu_ops.pte_update = lguest_pte_update;
+ pv_mmu_ops.pte_update_defer = lguest_pte_update;
#ifdef CONFIG_X86_LOCAL_APIC
/* apic read/write intercepts */
- pv_apic_ops.apic_write = lguest_apic_write;
- pv_apic_ops.apic_write_atomic = lguest_apic_write;
- pv_apic_ops.apic_read = lguest_apic_read;
+ set_lguest_basic_apic_ops();
#endif
/* time operations */
pv_time_ops.get_wallclock = lguest_get_wallclock;
pv_time_ops.time_init = lguest_time_init;
+ pv_time_ops.get_tsc_khz = lguest_tsc_khz;
/* Now is a good time to look at the implementations of these functions
* before returning to the rest of lguest_init(). */
* lguest_init() where the rest of the fairly chaotic boot setup
* occurs. */
- /* The native boot code sets up initial page tables immediately after
- * the kernel itself, and sets init_pg_tables_end so they're not
- * clobbered. The Launcher places our initial pagetables somewhere at
- * the top of our physical memory, so we don't need extra space: set
- * init_pg_tables_end to the end of the kernel. */
- init_pg_tables_end = __pa(pg0);
-
- /* Load the %fs segment register (the per-cpu segment register) with
- * the normal data segment to get through booting. */
- asm volatile ("mov %0, %%fs" : : "r" (__KERNEL_DS) : "memory");
-
- /* The Host uses the top of the Guest's virtual address space for the
- * Host<->Guest Switcher, and it tells us how big that is in
- * lguest_data.reserve_mem, set up on the LGUEST_INIT hypercall. */
+ /* The stack protector is a weird thing where gcc places a canary
+ * value on the stack and then checks it on return. This file is
+ * compiled with -fno-stack-protector it, so we got this far without
+ * problems. The value of the canary is kept at offset 20 from the
+ * %gs register, so we need to set that up before calling C functions
+ * in other files. */
+ setup_stack_canary_segment(0);
+ /* We could just call load_stack_canary_segment(), but we might as
+ * call switch_to_new_gdt() which loads the whole table and sets up
+ * the per-cpu segment descriptor register %fs as well. */
+ switch_to_new_gdt(0);
+
+ /* As described in head_32.S, we map the first 128M of memory. */
+ max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT;
+
+ /* The Host<->Guest Switcher lives at the top of our address space, and
+ * the Host told us how big it is when we made LGUEST_INIT hypercall:
+ * it put the answer in lguest_data.reserve_mem */
reserve_top_address(lguest_data.reserve_mem);
/* If we don't initialize the lock dependency checker now, it crashes
/* Math is always hard! */
new_cpu_data.hard_math = 1;
+ /* We don't have features. We have puppies! Puppies! */
#ifdef CONFIG_X86_MCE
mce_disabled = 1;
#endif
acpi_ht = 0;
#endif
- /* We set the perferred console to "hvc". This is the "hypervisor
+ /* We set the preferred console to "hvc". This is the "hypervisor
* virtual console" driver written by the PowerPC people, which we also
* adapted for lguest's use. */
add_preferred_console("hvc", 0, NULL);
virtio_cons_early_init(early_put_chars);
/* Last of all, we set the power management poweroff hook to point to
- * the Guest routine to power off. */
+ * the Guest routine to power off, and the reboot hook to our restart
+ * routine. */
pm_power_off = lguest_power_off;
+ machine_ops.restart = lguest_restart;
- /* Now we're set up, call start_kernel() in init/main.c and we proceed
+ /* Now we're set up, call i386_start_kernel() in head32.c and we proceed
* to boot as normal. It never returns. */
- start_kernel();
+ i386_start_kernel();
}
/*
* This marks the end of stage II of our journey, The Guest.
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff