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>
27 // The following can be used to track heap bugs
28 // zero memory after allocation
29 #define ALLOC_ZERO_MEM 0
30 // pad allocations by this many bytes on both ends of block
40 static struct v3_vm_info * irq_to_guest_map[256];
43 extern unsigned int cpu_khz;
45 extern int cpu_list[NR_CPUS];
46 extern int cpu_list_len;
49 static char *print_buffer[NR_CPUS];
51 static void deinit_print_buffers(void)
55 for (i=0;i<NR_CPUS;i++) {
56 if (print_buffer[i]) {
57 palacios_free(print_buffer[i]);
63 static int init_print_buffers(void)
67 memset(print_buffer,0,sizeof(char*)*NR_CPUS);
69 #if !V3_PRINTK_OLD_STYLE_OUTPUT
71 for (i=0;i<NR_CPUS;i++) {
72 print_buffer[i] = palacios_alloc(V3_PRINTK_BUF_SIZE);
73 if (!print_buffer[i]) {
74 ERROR("Cannot allocate print buffer for cpu %d\n",i);
75 deinit_print_buffers();
78 memset(print_buffer[i],0,V3_PRINTK_BUF_SIZE);
88 * Prints a message to the console.
90 void palacios_print_scoped(void * vm, int vcore, const char *fmt, ...) {
92 #if V3_PRINTK_OLD_STYLE_OUTPUT
106 unsigned int cpu = palacios_get_cpu();
107 struct v3_guest *guest = (struct v3_guest *)vm;
109 buf = print_buffer[cpu];
112 printk(KERN_INFO "palacios (pcore %u): output skipped - no allocated buffer\n",cpu);
117 vsnprintf(buf,V3_PRINTK_BUF_SIZE, fmt, ap);
120 #if V3_PRINTK_CHECK_7BIT
124 for (i=0;i<strlen(buf);i++) {
131 printk(KERN_INFO "palacios (pcore %u): ALERT ALERT 8 BIT CHAR (c=%d) DETECTED\n", cpu,c);
138 printk(KERN_INFO "palacios (pcore %u vm %s vcore %u): %s",
144 printk(KERN_INFO "palacios (pcore %u vm %s): %s",
150 printk(KERN_INFO "palacios (pcore %u): %s",
163 * Allocates a contiguous region of pages of the requested size.
164 * Returns the physical address of the first page in the region.
166 void *palacios_allocate_pages(int num_pages, unsigned int alignment) {
167 void * pg_addr = NULL;
169 pg_addr = (void *)alloc_palacios_pgs(num_pages, alignment);
172 ERROR("ALERT ALERT Page allocation has FAILED Warning\n");
176 pg_allocs += num_pages;
183 * Frees a page previously allocated via palacios_allocate_page().
184 * Note that palacios_allocate_page() can allocate multiple pages with
185 * a single call while palacios_free_page() only frees a single page.
188 void palacios_free_pages(void * page_paddr, int num_pages) {
189 pg_frees += num_pages;
190 free_palacios_pgs((uintptr_t)page_paddr, num_pages);
195 palacios_alloc_extended(unsigned int size, unsigned int flags) {
198 addr = kmalloc(size+2*ALLOC_PAD, flags);
201 ERROR("ALERT ALERT kmalloc has FAILED FAILED FAILED\n");
208 memset(addr,0,size+2*ALLOC_PAD);
211 return addr+ALLOC_PAD;
216 * Allocates 'size' bytes of kernel memory.
217 * Returns the kernel virtual address of the memory allocated.
220 palacios_alloc(unsigned int size) {
222 // It is very important that this test remains since
223 // this function is used extensively throughout palacios and the linux
224 // module, both in places where interrupts are off and where they are on
225 // a GFP_KERNEL call, when done with interrupts off can lead to DEADLOCK
226 if (irqs_disabled()) {
227 return palacios_alloc_extended(size,GFP_ATOMIC);
229 return palacios_alloc_extended(size,GFP_KERNEL);
235 * Frees memory that was previously allocated by palacios_alloc().
243 kfree(addr-ALLOC_PAD);
248 * Converts a kernel virtual address to the corresponding physical address.
251 palacios_vaddr_to_paddr(
255 return (void*) __pa(vaddr);
260 * Converts a physical address to the corresponding kernel virtual address.
263 palacios_paddr_to_vaddr(
271 * Runs a function on the specified CPU.
276 void (*fn)(void *arg),
282 // We set wait to 1, but I'm not sure this is necessary
283 smp_call_function_single(cpu_id, fn, arg, 1);
289 #define MAX_THREAD_NAME 32
291 struct lnx_thread_arg {
292 int (*fn)(void * arg);
294 char name[MAX_THREAD_NAME];
297 static int lnx_thread_target(void * arg) {
298 struct lnx_thread_arg * thread_info = (struct lnx_thread_arg *)arg;
301 INFO("Daemonizing new Palacios thread (name=%s)\n", thread_info->name);
303 daemonize(thread_info->name);
304 allow_signal(SIGKILL);
308 ret = thread_info->fn(thread_info->arg);
311 INFO("Palacios Thread (%s) EXITING\n", thread_info->name);
313 palacios_free(thread_info);
318 return 0; // should not get here.
322 * Creates a kernel thread.
325 palacios_start_kernel_thread(
326 int (*fn) (void * arg),
328 char * thread_name) {
330 struct lnx_thread_arg * thread_info = palacios_alloc(sizeof(struct lnx_thread_arg));
333 ERROR("ALERT ALERT Unable to allocate thread\n");
337 thread_info->fn = fn;
338 thread_info->arg = arg;
339 strncpy(thread_info->name,thread_name,MAX_THREAD_NAME);
340 thread_info->name[MAX_THREAD_NAME-1] =0;
342 return kthread_run( lnx_thread_target, thread_info, thread_info->name );
347 * Starts a kernel thread on the specified CPU.
350 palacios_start_thread_on_cpu(int cpu_id,
351 int (*fn)(void * arg),
353 char * thread_name ) {
354 struct task_struct * thread = NULL;
355 struct lnx_thread_arg * thread_info = palacios_alloc(sizeof(struct lnx_thread_arg));
358 ERROR("ALERT ALERT Unable to allocate thread to start on cpu\n");
362 thread_info->fn = fn;
363 thread_info->arg = arg;
364 strncpy(thread_info->name,thread_name,MAX_THREAD_NAME);
365 thread_info->name[MAX_THREAD_NAME-1] =0;
367 thread = kthread_create( lnx_thread_target, thread_info, thread_info->name );
369 if (IS_ERR(thread)) {
370 WARNING("Palacios error creating thread: %s\n", thread_info->name);
371 palacios_free(thread_info);
375 if (set_cpus_allowed_ptr(thread, cpumask_of(cpu_id)) != 0) {
376 WARNING("Attempt to start thread on disallowed CPU\n");
377 kthread_stop(thread);
378 palacios_free(thread_info);
382 wake_up_process(thread);
389 * Rebind a kernel thread to the specified CPU
390 * The thread will be running on target CPU on return
391 * non-zero return means failure
394 palacios_move_thread_to_cpu(int new_cpu_id,
396 struct task_struct * thread = (struct task_struct *)thread_ptr;
398 INFO("Moving thread (%p) to cpu %d\n", thread, new_cpu_id);
400 if (thread == NULL) {
405 * Bind to the specified CPU. When this call returns,
406 * the thread should be running on the target CPU.
408 return set_cpus_allowed_ptr(thread, cpumask_of(new_cpu_id));
413 * Returns the CPU ID that the caller is running on.
416 palacios_get_cpu(void)
419 /* We want to call smp_processor_id()
420 * But this is not safe if kernel preemption is possible
421 * We need to ensure that the palacios threads are bound to a give cpu
424 unsigned int cpu_id = get_cpu();
430 * Interrupts the physical CPU corresponding to the specified logical guest cpu.
433 * This is dependent on the implementation of xcall_reschedule(). Currently
434 * xcall_reschedule does not explicitly call schedule() on the destination CPU,
435 * but instead relies on the return to user space to handle it. Because
436 * palacios is a kernel thread schedule will not be called, which is correct.
437 * If it ever changes to induce side effects, we'll need to figure something
441 #include <asm/apic.h>
444 palacios_interrupt_cpu(
445 struct v3_vm_info * vm,
451 smp_send_reschedule(cpu_id);
453 apic->send_IPI_mask(cpumask_of(cpu_id), vector);
459 * Dispatches an interrupt to Palacios for handling.
462 palacios_dispatch_interrupt( int vector, void * dev, struct pt_regs * regs ) {
463 struct v3_interrupt intr = {
465 .error = regs->orig_ax,
469 if (irq_to_guest_map[vector]) {
470 v3_deliver_irq(irq_to_guest_map[vector], &intr);
476 * Instructs the kernel to forward the specified IRQ to Palacios.
479 palacios_hook_interrupt(struct v3_vm_info * vm,
480 unsigned int vector ) {
481 INFO("hooking vector %d\n", vector);
483 if (irq_to_guest_map[vector]) {
485 "%s: Interrupt vector %u is already hooked.\n",
491 "%s: Hooking interrupt vector %u to vm %p.\n",
492 __func__, vector, vm);
494 irq_to_guest_map[vector] = vm;
497 * NOTE: Normally PCI devices are supposed to be level sensitive,
498 * but we need them to be edge sensitive so that they are
499 * properly latched by Palacios. Leaving them as level
500 * sensitive would lead to an interrupt storm.
502 //ioapic_set_trigger_for_vector(vector, ioapic_edge_sensitive);
504 //set_idtvec_handler(vector, palacios_dispatch_interrupt);
506 ERROR("unexpected vector for hooking\n");
514 DEBUG("hooking vector: %d\n", vector);
522 error = request_irq((vector - 32),
523 (void *)palacios_dispatch_interrupt,
525 "interrupt_for_palacios",
529 ERROR("error code for request_irq is %d\n", error);
530 ERROR("request vector %d failed", vector);
541 * Acknowledges an interrupt.
544 palacios_ack_interrupt(
549 DEBUG("Pretending to ack interrupt, vector=%d\n", vector);
554 * Returns the CPU frequency in kilohertz.
557 palacios_get_cpu_khz(void)
559 INFO("cpu_khz is %u\n", cpu_khz);
562 INFO("faking cpu_khz to 1000000\n");
571 * Yield the CPU so other host OS tasks can run.
572 * This will return immediately if there is no other thread that is runnable
573 * And there is no real bound on how long it will yield
576 palacios_yield_cpu(void)
583 * Yield the CPU so other host OS tasks can run.
584 * Given now immediately if there is no other thread that is runnable
585 * And there is no real bound on how long it will yield
587 void palacios_sleep_cpu(unsigned int us)
590 set_current_state(TASK_INTERRUPTIBLE);
592 unsigned int uspj = 1000000U/HZ;
593 unsigned int jiffies = us/uspj + ((us%uspj) !=0); // ceiling
594 schedule_timeout(jiffies);
601 void palacios_wakeup_cpu(void *thread)
603 wake_up_process(thread);
609 * Returns NULL on failure.
612 palacios_mutex_alloc(void)
614 spinlock_t *lock = palacios_alloc(sizeof(spinlock_t));
617 spin_lock_init(lock);
619 ERROR("ALERT ALERT Unable to allocate lock\n");
630 palacios_mutex_free(void * mutex) {
631 palacios_free(mutex);
638 palacios_mutex_lock(void * mutex, int must_spin) {
639 spin_lock((spinlock_t *)mutex);
644 * Locks a mutex, disabling interrupts on this core
647 palacios_mutex_lock_irqsave(void * mutex, int must_spin) {
651 spin_lock_irqsave((spinlock_t *)mutex,flags);
653 return (void *)flags;
661 palacios_mutex_unlock(
665 spin_unlock((spinlock_t *)mutex);
670 * Unlocks a mutex and restores previous interrupt state on this core
673 palacios_mutex_unlock_irqrestore(void *mutex, void *flags)
675 // This is correct, flags is opaque
676 spin_unlock_irqrestore((spinlock_t *)mutex,(unsigned long)flags);
680 * Structure used by the Palacios hypervisor to interface with the host kernel.
682 static struct v3_os_hooks palacios_os_hooks = {
683 .print = palacios_print_scoped,
684 .allocate_pages = palacios_allocate_pages,
685 .free_pages = palacios_free_pages,
686 .malloc = palacios_alloc,
687 .free = palacios_free,
688 .vaddr_to_paddr = palacios_vaddr_to_paddr,
689 .paddr_to_vaddr = palacios_paddr_to_vaddr,
690 .hook_interrupt = palacios_hook_interrupt,
691 .ack_irq = palacios_ack_interrupt,
692 .get_cpu_khz = palacios_get_cpu_khz,
693 .start_kernel_thread = palacios_start_kernel_thread,
694 .yield_cpu = palacios_yield_cpu,
695 .sleep_cpu = palacios_sleep_cpu,
696 .wakeup_cpu = palacios_wakeup_cpu,
697 .mutex_alloc = palacios_mutex_alloc,
698 .mutex_free = palacios_mutex_free,
699 .mutex_lock = palacios_mutex_lock,
700 .mutex_unlock = palacios_mutex_unlock,
701 .mutex_lock_irqsave = palacios_mutex_lock_irqsave,
702 .mutex_unlock_irqrestore= palacios_mutex_unlock_irqrestore,
703 .get_cpu = palacios_get_cpu,
704 .interrupt_cpu = palacios_interrupt_cpu,
705 .call_on_cpu = palacios_xcall,
706 .start_thread_on_cpu = palacios_start_thread_on_cpu,
707 .move_thread_to_cpu = palacios_move_thread_to_cpu,
713 int palacios_vmm_init( char *options )
715 int num_cpus = num_online_cpus();
716 char * cpu_mask = NULL;
718 if (cpu_list_len > 0) {
723 cpu_mask = palacios_alloc((num_cpus / 8) + 1);
726 ERROR("Cannot allocate cpu mask\n");
730 memset(cpu_mask, 0, (num_cpus / 8) + 1);
732 for (i = 0; i < cpu_list_len; i++) {
733 if (cpu_list[i] >= num_cpus) {
734 WARNING("CPU (%d) exceeds number of available CPUs. Ignoring...\n", cpu_list[i]);
738 major = cpu_list[i] / 8;
739 minor = cpu_list[i] % 8;
741 *(cpu_mask + major) |= (0x1 << minor);
745 memset(irq_to_guest_map, 0, sizeof(struct v3_vm_info *) * 256);
747 if (init_print_buffers()) {
748 ERROR("Cannot initialize print buffers\n");
749 palacios_free(cpu_mask);
753 INFO("palacios_init starting - calling init_v3\n");
755 Init_V3(&palacios_os_hooks, cpu_mask, num_cpus, options);
762 int palacios_vmm_exit( void ) {
766 INFO("palacios shutdown complete\n");
768 deinit_print_buffers();