support vdso which borrows clocksource from linux

kernel/ap.c

@@ -40,6 +40,7 @@ static void ap_wait(void)
 
 	kmalloc_init();
 	sched_init();
+	arch_start_pvclock();
 
 	if (find_command_line("hidos")) {
 		init_host_syscall_channel();

kernel/host.c

@@ -86,6 +86,7 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 	struct process_vm *vm = proc->vm;
 	struct address_space *as = vm->address_space;
 	long aout_base;
+	int error;
 	
 	n = p->num_sections;
 
@@ -324,6 +325,14 @@ int prepare_process_ranges_args_envs(struct thread *thread,
 
 	dkprintf("env OK\n");
 
+	if (pn->enable_vdso) {
+		error = arch_map_vdso(vm);
+		if (error) {
+			kprintf("ERROR: mapping vdso pages. %d\n", error);
+			goto err;
+		}
+	}
+
 	p->rprocess = (unsigned long)thread;
 	p->rpgtable = virt_to_phys(as->page_table);
 

kernel/include/auxvec.h

@@ -1,6 +1,8 @@
 #ifndef _LINUX_AUXVEC_H
 #define _LINUX_AUXVEC_H
 
+#include <arch/auxvec.h>
+
 /* Symbolic values for the entries in the auxiliary table
    put on the initial stack */
 #define AT_NULL   0	/* end of vector */

kernel/include/process.h

@@ -292,7 +292,7 @@ struct user
 	unsigned long int  u_debugreg [8];
 };
 
-#define	AUXV_LEN	16
+#define	AUXV_LEN	18
 
 struct vm_range {
 	struct list_head list;
@@ -557,6 +557,8 @@ struct process_vm {
 	struct process *proc;		/* process that reside on the same page */
 	void *opt;
 	void (*free_cb)(struct process_vm *, void *);
+	void *vdso_addr;
+	void *vvar_addr;
  	
 	ihk_spinlock_t page_table_lock;
 	ihk_spinlock_t memory_range_lock;

kernel/include/syscall.h

@@ -38,6 +38,7 @@
 #define SCD_MSG_SYSCALL_ONESIDE         0x4
 #define SCD_MSG_SEND_SIGNAL		0x8
 #define	SCD_MSG_CLEANUP_PROCESS		0x9
+#define SCD_MSG_GET_VDSO_INFO           0xa
 
 #define	SCD_MSG_PROCFS_CREATE		0x10
 #define	SCD_MSG_PROCFS_DELETE		0x11
@@ -176,6 +177,8 @@ struct program_load_desc {
 	int pgid;
 	int cred[8];
 	int reloc;
+	char enable_vdso;
+	char padding[7];
 	unsigned long entry;
 	unsigned long user_start;
 	unsigned long user_end;
@@ -361,5 +364,25 @@ intptr_t do_mmap(intptr_t addr0, size_t len0, int prot, int flags, int fd,
 		off_t off0);
 typedef int32_t key_t;
 int do_shmget(key_t key, size_t size, int shmflg);
+struct process_vm;
+int arch_map_vdso(struct process_vm *vm);	/* arch dependent */
+int arch_setup_vdso(void);
+
+#define VDSO_MAXPAGES 2
+struct vdso {
+	long busy;
+	int vdso_npages;
+	char vvar_is_global;
+	char hpet_is_global;
+	char pvti_is_global;
+	char padding;
+	long vdso_physlist[VDSO_MAXPAGES];
+	void *vvar_virt;
+	long vvar_phys;
+	void *hpet_virt;
+	long hpet_phys;
+	void *pvti_virt;
+	long pvti_phys;
+};
 
 #endif

kernel/init.c

@@ -343,6 +343,8 @@ static void post_init(void)
 		ihk_mc_spinlock_init(&syscall_lock);
 	}
 
+	arch_setup_vdso();
+	arch_start_pvclock();
 	ap_start();
 
 	sysfs_init();

kernel/process.c

@@ -498,23 +498,30 @@ static int copy_user_pte(void *arg0, page_table_t src_pt, pte_t *src_ptep, void
 		goto out;
 	}
 
-	dkprintf("copy_user_pte(): 0x%lx PTE found\n", pgaddr);
-	dkprintf("copy_user_pte(): page size: %d\n", pgsize);
-
-	npages = pgsize / PAGE_SIZE;
-	virt = ihk_mc_alloc_aligned_pages(npages, pgalign, IHK_MC_AP_NOWAIT);
-	if (!virt) {
-		kprintf("ERROR: copy_user_pte() allocating new page\n");
-		error = -ENOMEM;
-		goto out;
+	if (args->new_vrflag & VR_REMOTE) {
+		phys = src_phys;
+		attr = pte_get_attr(src_ptep, pgsize);
 	}
-	phys = virt_to_phys(virt);
-	dkprintf("copy_user_pte(): phys page allocated\n");
+	else {
+		dkprintf("copy_user_pte(): 0x%lx PTE found\n", pgaddr);
+		dkprintf("copy_user_pte(): page size: %d\n", pgsize);
 
-	memcpy(virt, src_kvirt, pgsize);
-	dkprintf("copy_user_pte(): memcpy OK\n");
+		npages = pgsize / PAGE_SIZE;
+		virt = ihk_mc_alloc_aligned_pages(npages, pgalign, IHK_MC_AP_NOWAIT);
+		if (!virt) {
+			kprintf("ERROR: copy_user_pte() allocating new page\n");
+			error = -ENOMEM;
+			goto out;
+		}
+		phys = virt_to_phys(virt);
+		dkprintf("copy_user_pte(): phys page allocated\n");
+
+		memcpy(virt, src_kvirt, pgsize);
+		dkprintf("copy_user_pte(): memcpy OK\n");
+
+		attr = arch_vrflag_to_ptattr(args->new_vrflag, PF_POPULATE, NULL);
+	}
 
-	attr = arch_vrflag_to_ptattr(args->new_vrflag, PF_POPULATE, NULL);
 	error = ihk_mc_pt_set_range(args->new_vm->address_space->page_table, args->new_vm, pgaddr, pgaddr+pgsize, phys, attr);
 	if (error) {
 		args->fault_addr = (intptr_t)pgaddr;
@@ -1815,6 +1822,11 @@ int init_process_stack(struct thread *thread, struct program_load_desc *pn,
 	p[s_ind--] = AT_CLKTCK;
 	p[s_ind--] = at_rand; /* AT_RANDOM */
 	p[s_ind--] = AT_RANDOM;
+#ifndef AT_SYSINFO_EHDR
+#define AT_SYSINFO_EHDR AT_IGNORE
+#endif
+	p[s_ind--] = (long)(thread->vm->vdso_addr);
+	p[s_ind--] = (thread->vm->vdso_addr)? AT_SYSINFO_EHDR: AT_IGNORE;
 
 	/* Save auxiliary vector for later use. */
 	memcpy(proc->saved_auxv, &p[s_ind + 1], sizeof(proc->saved_auxv));