1 #include <linux/kernel.h>
2 #include <linux/kthread.h>
3 #include <linux/spinlock.h>
5 #include <linux/interrupt.h>
6 #include <linux/linkage.h>
7 #include <linux/sched.h>
8 #include <linux/uaccess.h>
9 #include <asm/irq_vectors.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/kthread.h>
15 #include <asm/uaccess.h>
16 #include <linux/smp.h>
18 #include <palacios/vmm.h>
19 #include <palacios/vmm_host_events.h>
34 static struct v3_vm_info * irq_to_guest_map[256];
37 extern unsigned int cpu_khz;
39 extern int cpu_list[NR_CPUS];
40 extern int cpu_list_len;
44 * Prints a message to the console.
46 static void palacios_print(const char * fmt, ...) {
58 * Allocates a contiguous region of pages of the requested size.
59 * Returns the physical address of the first page in the region.
61 static void * palacios_allocate_pages(int num_pages, unsigned int alignment) {
62 void * pg_addr = NULL;
64 pg_addr = (void *)alloc_palacios_pgs(num_pages, alignment);
65 pg_allocs += num_pages;
72 * Frees a page previously allocated via palacios_allocate_page().
73 * Note that palacios_allocate_page() can allocate multiple pages with
74 * a single call while palacios_free_page() only frees a single page.
77 static void palacios_free_pages(void * page_paddr, int num_pages) {
78 pg_frees += num_pages;
79 free_palacios_pgs((uintptr_t)page_paddr, num_pages);
84 * Allocates 'size' bytes of kernel memory.
85 * Returns the kernel virtual address of the memory allocated.
88 palacios_alloc(unsigned int size) {
91 if (irqs_disabled()) {
92 addr = kmalloc(size, GFP_ATOMIC);
94 addr = kmalloc(size, GFP_KERNEL);
103 * Frees memory that was previously allocated by palacios_alloc().
116 * Converts a kernel virtual address to the corresponding physical address.
119 palacios_vaddr_to_paddr(
123 return (void*) __pa(vaddr);
128 * Converts a physical address to the corresponding kernel virtual address.
131 palacios_paddr_to_vaddr(
139 * Runs a function on the specified CPU.
142 // For now, do call only on local CPU
146 void (*fn)(void *arg),
152 // We set wait to 1, but I'm not sure this is necessary
153 smp_call_function_single(cpu_id, fn, arg, 1);
158 struct lnx_thread_arg {
159 int (*fn)(void * arg);
164 static int lnx_thread_target(void * arg) {
165 struct lnx_thread_arg * thread_info = (struct lnx_thread_arg *)arg;
168 printk("Daemonizing new Palacios thread (name=%s)\n", thread_info->name);
170 daemonize(thread_info->name);
171 allow_signal(SIGKILL);
175 ret = thread_info->fn(thread_info->arg);
181 printk("Palacios Thread (%s) EXITTING\n", thread_info->name);
185 return 0; // should not get here.
189 * Creates a kernel thread.
192 palacios_start_kernel_thread(
193 int (*fn) (void * arg),
195 char * thread_name) {
197 struct lnx_thread_arg * thread_info = kmalloc(sizeof(struct lnx_thread_arg), GFP_KERNEL);
199 thread_info->fn = fn;
200 thread_info->arg = arg;
201 thread_info->name = thread_name;
203 return kthread_run( lnx_thread_target, thread_info, thread_name );
208 * Starts a kernel thread on the specified CPU.
211 palacios_start_thread_on_cpu(int cpu_id,
212 int (*fn)(void * arg),
214 char * thread_name ) {
215 struct task_struct * thread = NULL;
216 struct lnx_thread_arg * thread_info = kmalloc(sizeof(struct lnx_thread_arg), GFP_KERNEL);
218 thread_info->fn = fn;
219 thread_info->arg = arg;
220 thread_info->name = thread_name;
223 thread = kthread_create( lnx_thread_target, thread_info, thread_name );
225 if (IS_ERR(thread)) {
226 printk("Palacios error creating thread: %s\n", thread_name);
230 if (set_cpus_allowed_ptr(thread, cpumask_of(cpu_id)) != 0) {
231 kthread_stop(thread);
235 wake_up_process(thread);
242 * Rebind a kernel thread to the specified CPU
243 * The thread will be running on target CPU on return
244 * non-zero return means failure
247 palacios_move_thread_to_cpu(int new_cpu_id,
249 struct task_struct * thread = (struct task_struct *)thread_ptr;
251 printk("Moving thread (%p) to cpu %d\n", thread, new_cpu_id);
253 if (thread == NULL) {
258 * Bind to the specified CPU. When this call returns,
259 * the thread should be running on the target CPU.
261 return set_cpus_allowed_ptr(thread, cpumask_of(new_cpu_id));
266 * Returns the CPU ID that the caller is running on.
269 palacios_get_cpu(void)
272 /* We want to call smp_processor_id()
273 * But this is not safe if kernel preemption is possible
274 * We need to ensure that the palacios threads are bound to a give cpu
277 unsigned int cpu_id = get_cpu();
283 * Interrupts the physical CPU corresponding to the specified logical guest cpu.
286 * This is dependent on the implementation of xcall_reschedule(). Currently
287 * xcall_reschedule does not explicitly call schedule() on the destination CPU,
288 * but instead relies on the return to user space to handle it. Because
289 * palacios is a kernel thread schedule will not be called, which is correct.
290 * If it ever changes to induce side effects, we'll need to figure something
294 #include <asm/apic.h>
297 palacios_interrupt_cpu(
298 struct v3_vm_info * vm,
304 smp_send_reschedule(cpu_id);
306 apic->send_IPI_mask(cpumask_of(cpu_id), vector);
312 * Dispatches an interrupt to Palacios for handling.
315 palacios_dispatch_interrupt( int vector, void * dev, struct pt_regs * regs ) {
316 struct v3_interrupt intr = {
318 .error = regs->orig_ax,
322 if (irq_to_guest_map[vector]) {
323 v3_deliver_irq(irq_to_guest_map[vector], &intr);
329 * Instructs the kernel to forward the specified IRQ to Palacios.
332 palacios_hook_interrupt(struct v3_vm_info * vm,
333 unsigned int vector ) {
334 printk("hooking vector %d\n", vector);
336 if (irq_to_guest_map[vector]) {
338 "%s: Interrupt vector %u is already hooked.\n",
344 "%s: Hooking interrupt vector %u to vm %p.\n",
345 __func__, vector, vm);
347 irq_to_guest_map[vector] = vm;
350 * NOTE: Normally PCI devices are supposed to be level sensitive,
351 * but we need them to be edge sensitive so that they are
352 * properly latched by Palacios. Leaving them as level
353 * sensitive would lead to an interrupt storm.
355 //ioapic_set_trigger_for_vector(vector, ioapic_edge_sensitive);
357 //set_idtvec_handler(vector, palacios_dispatch_interrupt);
359 panic("unexpected vector for hooking\n");
366 printk("hooking vector: %d\n", vector);
374 error = request_irq((vector - 32),
375 (void *)palacios_dispatch_interrupt,
377 "interrupt_for_palacios",
381 printk("error code for request_irq is %d\n", error);
382 panic("request vector %d failed",vector);
392 * Acknowledges an interrupt.
395 palacios_ack_interrupt(
400 printk("Pretending to ack interrupt, vector=%d\n",vector);
405 * Returns the CPU frequency in kilohertz.
408 palacios_get_cpu_khz(void)
410 printk("cpu_khz is %u\n",cpu_khz);
413 printk("faking cpu_khz to 1000000\n");
422 * Yield the CPU so other host OS tasks can run.
425 palacios_yield_cpu(void)
435 * Returns NULL on failure.
438 palacios_mutex_alloc(void)
440 spinlock_t *lock = kmalloc(sizeof(spinlock_t), GFP_KERNEL);
443 spin_lock_init(lock);
453 palacios_mutex_free(void * mutex) {
461 palacios_mutex_lock(void * mutex, int must_spin) {
462 spin_lock((spinlock_t *)mutex);
469 palacios_mutex_unlock(
473 spin_unlock((spinlock_t *)mutex);
477 * Structure used by the Palacios hypervisor to interface with the host kernel.
479 static struct v3_os_hooks palacios_os_hooks = {
480 .print = palacios_print,
481 .allocate_pages = palacios_allocate_pages,
482 .free_pages = palacios_free_pages,
483 .malloc = palacios_alloc,
484 .free = palacios_free,
485 .vaddr_to_paddr = palacios_vaddr_to_paddr,
486 .paddr_to_vaddr = palacios_paddr_to_vaddr,
487 .hook_interrupt = palacios_hook_interrupt,
488 .ack_irq = palacios_ack_interrupt,
489 .get_cpu_khz = palacios_get_cpu_khz,
490 .start_kernel_thread = palacios_start_kernel_thread,
491 .yield_cpu = palacios_yield_cpu,
492 .mutex_alloc = palacios_mutex_alloc,
493 .mutex_free = palacios_mutex_free,
494 .mutex_lock = palacios_mutex_lock,
495 .mutex_unlock = palacios_mutex_unlock,
496 .get_cpu = palacios_get_cpu,
497 .interrupt_cpu = palacios_interrupt_cpu,
498 .call_on_cpu = palacios_xcall,
499 .start_thread_on_cpu = palacios_start_thread_on_cpu,
500 .move_thread_to_cpu = palacios_move_thread_to_cpu,
506 int palacios_vmm_init( void )
508 int num_cpus = num_online_cpus();
509 char * cpu_mask = NULL;
511 if (cpu_list_len > 0) {
516 cpu_mask = kmalloc((num_cpus / 8) + 1, GFP_KERNEL);
517 memset(cpu_mask, 0, (num_cpus / 8) + 1);
519 for (i = 0; i < cpu_list_len; i++) {
520 if (cpu_list[i] >= num_cpus) {
521 printk("CPU (%d) exceeds number of available CPUs. Ignoring...\n", cpu_list[i]);
525 major = cpu_list[i] / 8;
526 minor = cpu_list[i] % 8;
528 *(cpu_mask + major) |= (0x1 << minor);
532 memset(irq_to_guest_map, 0, sizeof(struct v3_vm_info *) * 256);
534 printk("palacios_init starting - calling init_v3\n");
536 Init_V3(&palacios_os_hooks, cpu_mask, num_cpus);
543 int palacios_vmm_exit( void ) {