trial implementation of private file mapping

for review only.  will soon be reverted.

kernel/include/memobj.h

@@ -0,0 +1,22 @@
+#ifndef HEADER_MEMOBJ_H
+#define HEADER_MEMOBJ_H
+
+#include <ihk/types.h>
+#include <ihk/atomic.h>
+#include <ihk/lock.h>
+#include <list.h>
+
+struct memobj {
+	struct list_head	list;
+	ihk_atomic_t		ref;
+	uintptr_t		handle;
+	struct list_head	page_list;
+	ihk_spinlock_t		page_list_lock;
+};
+
+int memobj_create(int fd, int flags, int prot, struct memobj **objp, int *maxprotp);
+void memobj_ref(struct memobj *obj);
+void memobj_release(struct memobj *obj);
+int memobj_get_page(struct memobj *obj, off_t off, size_t pgsize, uintptr_t *physp);
+
+#endif /* HEADER_MEMOBJ_H */

kernel/include/page.h

@@ -2,16 +2,25 @@
 #define __HEADER_PAGE_H
 
 struct page {
-	struct list_head list;
-	uint64_t flags;
-	int64_t count;
+	struct list_head	list;
+	uint8_t			mode;
+	uint8_t			padding[3];
+	int32_t			count;
+	off_t			offset;
 };
 
-/* flags */
-#define	PAGE_IN_LIST	0x0001UL
+/* mode */
+enum page_mode {
+	PM_NONE =		0x00,
+	PM_PENDING_FREE =	0x01,
+	PM_PAGEIO =		0x02,
+	PM_MAPPED =		0x03,
+	PM_ANON_COW =		0x04,
+};
 
 struct page *phys_to_page(uintptr_t phys);
 uintptr_t page_to_phys(struct page *page);
+int page_unmap(struct page *page);
 
 void *allocate_pages(int npages, enum ihk_mc_ap_flag flag);
 void free_pages(void *va, int npages);

kernel/include/pager.h

@@ -0,0 +1,26 @@
+#ifndef HEADER_PAGER_H
+#define HEADER_PAGER_H
+
+#include <ihk/types.h>
+
+enum pager_op {
+	PAGER_REQ_CREATE =	0x0001,
+	PAGER_REQ_RELEASE =	0x0002,
+	PAGER_REQ_READ =	0x0003,
+};
+
+/*
+ * int pager_req_create(int fd, int flags, int prot, uintptr_t result_rpa);
+ */
+struct pager_create_result {
+	uintptr_t	handle;
+	int		maxprot;
+};
+
+/*
+ * int pager_req_release(uintptr_t handle);
+ */
+/*
+ * int pager_req_read(uintptr_t handle, off_t off, size_t size, uintptr_t buf_rpa);
+ */
+#endif /* HEADER_PAGER_H */

kernel/include/process.h

@@ -7,6 +7,7 @@
 #include <ihk/atomic.h>
 #include <list.h>
 #include <signal.h>
+#include <memobj.h>
 
 #define VR_NONE            0x0
 #define VR_STACK           0x1
@@ -14,13 +15,21 @@
 #define VR_IO_NOCACHE      0x100
 #define VR_REMOTE          0x200
 #define VR_DEMAND_PAGING   0x1000
+#define	VR_PRIVATE         0x2000
 #define	VR_PROT_NONE       0x00000000
 #define	VR_PROT_READ       0x00010000
 #define	VR_PROT_WRITE      0x00020000
 #define	VR_PROT_EXEC       0x00040000
 #define	VR_PROT_MASK       0x00070000
+#define	VR_MAXPROT_NONE    0x00000000
+#define	VR_MAXPROT_READ    0x00100000
+#define	VR_MAXPROT_WRITE   0x00200000
+#define	VR_MAXPROT_EXEC    0x00400000
+#define	VR_MAXPROT_MASK    0x00700000
 
 #define	PROT_TO_VR_FLAG(prot)	(((unsigned long)(prot) << 16) & VR_PROT_MASK)
+#define	VRFLAG_PROT_TO_MAXPROT(vrflag)	(((vrflag) & VR_PROT_MASK) << 4)
+#define	VRFLAG_MAXPROT_TO_PROT(vrflag)	(((vrflag) & VR_MAXPROT_MASK) >> 4)
 
 #define PS_RUNNING           0x1
 #define PS_INTERRUPTIBLE     0x2
@@ -41,6 +50,8 @@ struct vm_range {
 	struct list_head list;
 	unsigned long start, end;
 	unsigned long flag;
+	struct memobj *memobj;
+	off_t objoff;
 };
 
 struct vm_regions {
@@ -106,7 +117,6 @@ struct process_vm {
     // is protected by its own lock (see ihk/manycore/generic/page_alloc.c)
 };
 
-
 struct process *create_process(unsigned long user_pc);
 struct process *clone_process(struct process *org,
                               unsigned long pc, unsigned long sp);
@@ -114,10 +124,12 @@ void destroy_process(struct process *proc);
 void hold_process(struct process *proc);
 void free_process(struct process *proc);
 void free_process_memory(struct process *proc);
+void flush_process_memory(struct process *proc);
 
 int add_process_memory_range(struct process *process,
                              unsigned long start, unsigned long end,
-                             unsigned long phys, unsigned long flag);
+                             unsigned long phys, unsigned long flag,
+			     struct memobj *memobj, off_t objoff);
 int remove_process_memory_range(
 		struct process *process, unsigned long start, unsigned long end);
 int split_process_memory_range(struct process *process,
@@ -133,6 +145,7 @@ struct vm_range *next_process_memory_range(
 		struct process_vm *vm, struct vm_range *range);
 struct vm_range *previous_process_memory_range(
 		struct process_vm *vm, struct vm_range *range);
+int page_fault_process_memory_range(struct process *proc, struct vm_range *range, uintptr_t fault_addr, uint64_t reason);
 int remove_process_region(struct process *proc,
                           unsigned long start, unsigned long end);
 struct program_load_desc;

kernel/include/syscall.h

@@ -126,6 +126,8 @@ struct syscall_request {
 struct syscall_response {
 	unsigned long status;
 	long ret;
+	unsigned long fault_address;
+	unsigned long fault_reason;
 };
 
 struct syscall_post {
@@ -190,6 +192,7 @@ struct syscall_params {
 
 extern int do_syscall(struct syscall_request *req, ihk_mc_user_context_t *ctx);
 extern int obtain_clone_cpuid();
+extern long syscall_generic_forwarding(int n, ihk_mc_user_context_t *ctx);
 
 #define DECLARATOR(number,name)		__NR_##name = number,
 #define	SYSCALL_HANDLED(number,name)	DECLARATOR(number,name)