#include <linux/lguest.h>
#include <linux/lguest_launcher.h>
#include <linux/virtio_console.h>
+#include <linux/pm.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/reboot.h> /* for struct machine_ops */
/*G:010 Welcome to the Guest!
*
};
static cycle_t clock_base;
-/*G:035 Notice the lazy_hcall() above, rather than hcall(). This is our first
- * real optimization trick!
- *
- * When lazy_mode is set, it means we're allowed to defer all hypercalls and do
- * them as a batch when lazy_mode is eventually turned off. Because hypercalls
- * are reasonably expensive, batching them up makes sense. For example, a
- * large mmap might update dozens of page table entries: that code calls
- * paravirt_enter_lazy_mmu(), does the dozen updates, then calls
- * lguest_leave_lazy_mode().
- *
- * So, when we're in lazy mode, we call async_hypercall() to store the call for
- * future processing. When lazy mode is turned off we issue a hypercall to
- * flush the stored calls.
- */
-static void lguest_leave_lazy_mode(void)
-{
- paravirt_leave_lazy(paravirt_get_lazy_mode());
- hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0);
-}
-
-static void lazy_hcall(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);
- else
- async_hcall(call, arg1, arg2, arg3);
-}
-
-/* async_hcall() is pretty simple: I'm quite proud of it really. We have a
+/*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
* arguments, and a "hcall_status" word which is 0 if the call is ready to go,
* full and we just make the hypercall directly. This has the nice side
* effect of causing the Host to run all the stored calls in the ring buffer
* which empties it for next time! */
-void async_hcall(unsigned long call,
- unsigned long arg1, unsigned long arg2, unsigned long arg3)
+static void async_hcall(unsigned long call, unsigned long arg1,
+ unsigned long arg2, unsigned long arg3)
{
/* Note: This code assumes we're uniprocessor. */
static unsigned int next_call;
}
local_irq_restore(flags);
}
-/*:*/
+
+/*G:035 Notice the lazy_hcall() above, rather than hcall(). This is our first
+ * real optimization trick!
+ *
+ * When lazy_mode is set, it means we're allowed to defer all hypercalls and do
+ * them as a batch when lazy_mode is eventually turned off. Because hypercalls
+ * are reasonably expensive, batching them up makes sense. For example, a
+ * large munmap might update dozens of page table entries: that code calls
+ * paravirt_enter_lazy_mmu(), does the dozen updates, then calls
+ * 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,
+ unsigned long arg1,
+ unsigned long arg2,
+ unsigned long arg3)
+{
+ if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
+ hcall(call, arg1, arg2, arg3);
+ else
+ async_hcall(call, arg1, arg2, arg3);
+}
+
+/* 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)
+{
+ paravirt_leave_lazy(paravirt_get_lazy_mode());
+ hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0);
+}
/*G:033
- * Here are our first native-instruction replacements: four functions for
- * interrupt control.
+ * After that diversion we return to our first native-instruction
+ * replacements: four functions for interrupt control.
*
* The simplest way of implementing these would be to have "turn interrupts
* off" and "turn interrupts on" hypercalls. Unfortunately, this is too slow:
* check there when it wants 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. */
+/* restore_flags() just sets the flags back to the value given. */
static void restore_fl(unsigned long flags)
{
lguest_data.irq_enabled = flags;
* 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)
{
+ 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);
+ hcall(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;
* 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?
*/
-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);
/* 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);
+ native_write_gdt_entry(dt, entrynum, desc, type);
hcall(LHCALL_LOAD_GDT, __pa(dt), GDT_ENTRIES, 0);
}
* 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 1: /* Basic feature request. */
/* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */
- *ecx &= 0x00002201;
+ *cx &= 0x00002201;
/* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, FPU. */
- *edx &= 0x07808101;
+ *dx &= 0x07808101;
/* 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;
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;
}
}
* it. The Host needs to know when the Guest wants to change them, so we have
* a whole series of functions like read_cr0() and write_cr0().
*
- * We start with CR0. CR0 allows you to turn on and off all kinds of basic
+ * We start with cr0. cr0 allows you to turn on and off all kinds of basic
* features, but Linux only really cares about one: the horrifically-named Task
* Switched (TS) bit at bit 3 (ie. 8)
*
current_cr0 &= ~X86_CR0_TS;
}
-/* CR2 is the virtual address of the last page fault, which the Guest only ever
+/* cr2 is the virtual address of the last page fault, which the Guest only ever
* reads. The Host kindly writes this into our "struct lguest_data", so we
* just read it out of there. */
static unsigned long lguest_read_cr2(void)
return lguest_data.cr2;
}
-/* CR3 is the current toplevel pagetable page: the principle is the same as
+/* 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. */
static void lguest_write_cr3(unsigned long cr3)
{
return current_cr3;
}
-/* CR4 is used to enable and disable PGE, but we don't care. */
+/* cr4 is used to enable and disable PGE, but we don't care. */
static unsigned long lguest_read_cr4(void)
{
return 0;
* maps virtual addresses to physical addresses using "page tables". We could
* use one huge index of 1 million entries: each address is 4 bytes, so that's
* 1024 pages just to hold the page tables. But since most virtual addresses
- * are unused, we use a two level index which saves space. The CR3 register
+ * are unused, we use a two level index which saves space. The cr3 register
* contains the physical address of the top level "page directory" page, which
* contains physical addresses of up to 1024 second-level pages. Each of these
* second level pages contains up to 1024 physical addresses of actual pages,
*
* Here's a diagram, where arrows indicate physical addresses:
*
- * CR3 ---> +---------+
+ * cr3 ---> +---------+
* | --------->+---------+
* | | | PADDR1 |
* Top-level | | PADDR2 |
*
* ... 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.
- */
+ * anything changed until we've done the first page table switch. */
static void lguest_set_pte(pte_t *ptep, pte_t pteval)
{
*ptep = pteval;
/* Set up the timer interrupt (0) to go to our simple timer routine */
set_irq_handler(0, lguest_time_irq);
- /* Our clock structure look like arch/i386/kernel/tsc.c if we can use
- * the TSC, otherwise it's a dumb nanosecond-resolution clock. Either
- * way, the "rating" is initialized so high that it's always chosen
- * over any other clocksource. */
+ /* 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);
* to work. They're pretty simple.
*/
-/* The Guest needs to tell the host what stack it expects traps to use. For
+/* The Guest needs to tell the Host what stack it expects traps to use. For
* native hardware, this is part of the Task State Segment mentioned above in
* lguest_load_tr_desc(), but to help hypervisors there's this special call.
*
* 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,
+static void lguest_load_sp0(struct tss_struct *tss,
struct thread_struct *thread)
{
- lazy_hcall(LHCALL_SET_STACK, __KERNEL_DS|0x1, thread->esp0,
+ lazy_hcall(LHCALL_SET_STACK, __KERNEL_DS|0x1, thread->sp0,
THREAD_SIZE/PAGE_SIZE);
}
* 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, unsigned long v)
+static void lguest_apic_write(unsigned long reg, u32 v)
{
}
-static unsigned long lguest_apic_read(unsigned long reg)
+static u32 lguest_apic_read(unsigned long reg)
{
return 0;
}
* rather than virtual addresses, so we use __pa() here. */
static void lguest_power_off(void)
{
- hcall(LHCALL_CRASH, __pa("Power down"), 0, 0);
+ hcall(LHCALL_SHUTDOWN, __pa("Power down"), LGUEST_SHUTDOWN_POWEROFF, 0);
}
/*
*/
static int lguest_panic(struct notifier_block *nb, unsigned long l, void *p)
{
- hcall(LHCALL_CRASH, __pa(p), 0, 0);
+ hcall(LHCALL_SHUTDOWN, __pa(p), LGUEST_SHUTDOWN_POWEROFF, 0);
/* The hcall won't return, but to keep gcc happy, we're "done". */
return NOTIFY_DONE;
}
return "LGUEST";
}
-/* Before virtqueues are set up, we use LHCALL_NOTIFY on normal memory to
- * produce console output. */
+/* We will eventually use the virtio console device to produce console output,
+ * but before that is set up we use LHCALL_NOTIFY on normal memory to produce
+ * console output. */
static __init int early_put_chars(u32 vtermno, const char *buf, int count)
{
char scratch[17];
unsigned int len = count;
+ /* We use a nul-terminated string, so we have to make a copy. Icky,
+ * huh? */
if (len > sizeof(scratch) - 1)
len = sizeof(scratch) - 1;
scratch[len] = '\0';
* 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 10
+ * 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.
*
return insn_len;
}
+static void lguest_restart(char *reason)
+{
+ hcall(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART, 0);
+}
+
/*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. */
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;
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 much it needs in
+ * Host<->Guest Switcher, and it tells us how big that is in
* lguest_data.reserve_mem, set up on the LGUEST_INIT hypercall. */
reserve_top_address(lguest_data.reserve_mem);
* the Guest routine to power off. */
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
* to boot as normal. It never returns. */
start_kernel();
/*
* This marks the end of stage II of our journey, The Guest.
*
- * It is now time for us to explore the nooks and crannies of the three Guest
- * devices and complete our understanding of the Guest in "make Drivers".
+ * It is now time for us to explore the layer of virtual drivers and complete
+ * our understanding of the Guest in "make Drivers".
*/