mango/include/socks/vm.h

#ifndef SOCKS_VM_H_
#define SOCKS_VM_H_

#include <stddef.h>
#include <socks/types.h>
#include <socks/status.h>
#include <socks/queue.h>
#include <socks/locks.h>
#include <socks/machine/vm.h>

/* maximum number of NUMA nodes */
#define VM_MAX_NODES 64
/* maximum number of memory zones per node */
#define VM_MAX_ZONES (VM_ZONE_MAX + 1)
/* maximum number of supported page orders */
#define VM_MAX_PAGE_ORDERS (VM_PAGE_MAX_ORDER + 1)

#define VM_CHECK_ALIGN(p, mask) ((((p) & (mask)) == (p)) ? 1 : 0)

#define VM_CACHE_INITIALISED(c) ((c)->c_obj_count != 0)
#define VM_PAGE_IS_FREE(pg) (((pg)->p_flags & (VM_PAGE_RESERVED | VM_PAGE_ALLOC)) == 0)

#define vm_page_foreach(pg, i) \
	for (vm_page_t *i = (pg); i; i = vm_page_get_next_tail(i))

typedef phys_addr_t vm_alignment_t;
typedef unsigned int vm_node_id_t;

typedef struct vm_object {
	unsigned int reserved;
} vm_object_t;

typedef enum vm_prot {
	VM_PROT_READ  = 0x01u,
	VM_PROT_WRITE = 0x02u,
	VM_PROT_EXEC  = 0x04u,
	VM_PROT_USER  = 0x08u,
	VM_PROT_SVR   = 0x10u,
} vm_prot_t;

typedef enum vm_flags {
	VM_GET_DMA = 0x01u,
} vm_flags_t;

typedef enum vm_zone_id {
	/* NOTE that these are used as indices into the node_zones array in vm/zone.c
	   they need to be continuous, and must start at 0! */
	VM_ZONE_DMA     = 0u,
	VM_ZONE_NORMAL  = 1u,
	VM_ZONE_HIGHMEM = 2u,
	VM_ZONE_MIN     = VM_ZONE_DMA,
	VM_ZONE_MAX     = VM_ZONE_HIGHMEM,
} vm_zone_id_t;

typedef enum vm_page_order {
	VM_PAGE_4K = 0u,
	VM_PAGE_8K,
	VM_PAGE_16K,
	VM_PAGE_32K,
	VM_PAGE_64K,
	VM_PAGE_128K,
	VM_PAGE_256K,
	VM_PAGE_512K,
	VM_PAGE_1M,
	VM_PAGE_2M,
	VM_PAGE_4M,
	VM_PAGE_8M,
	VM_PAGE_16M,
	VM_PAGE_32M,
	VM_PAGE_64M,
	VM_PAGE_128M,
#if 0
	/* vm_page_t only has 4 bits to store the page order with.
	   the maximum order that can be stored in 4 bits is 15 (VM_PAGE_128M)
	   to use any of the page orders listed here, this field
	   will have to be expanded. */
	VM_PAGE_256M,
	VM_PAGE_512M,
	VM_PAGE_1G,
#endif
} vm_page_order_t;

typedef enum vm_page_flags {
	/* page is reserved (probably by a call to memblock_reserve()) and cannot be
	   returned by any allocation function */
	VM_PAGE_RESERVED   = 0x01u,
	/* page has been allocated by a zone's buddy allocator, and is in-use */
	VM_PAGE_ALLOC      = 0x02u,
	/* page is the first page of a huge-page */
	VM_PAGE_HEAD       = 0x04u,
	/* page is part of a huge-page */
	VM_PAGE_HUGE       = 0x08u,
} vm_page_flags_t;

typedef enum vm_memory_region_status {
	VM_REGION_FREE		= 0x01u,
	VM_REGION_RESERVED	= 0x02u,
} vm_memory_region_status_t;

typedef enum vm_cache_flags {
	VM_CACHE_OFFSLAB = 0x01u,
	VM_CACHE_DMA     = 0x02u
} vm_cache_flags_t;

typedef struct vm_zone_descriptor {
	vm_zone_id_t zd_id;
	vm_node_id_t zd_node;
	const char zd_name[32];
	phys_addr_t zd_base;
	phys_addr_t zd_limit;
} vm_zone_descriptor_t;

typedef struct vm_zone {
	vm_zone_descriptor_t z_info;
	spin_lock_t z_lock;

	queue_t z_free_pages[VM_MAX_PAGE_ORDERS];
	unsigned long z_size;
} vm_zone_t;

typedef struct vm_pg_data {
	vm_zone_t pg_zones[VM_MAX_ZONES];
} vm_pg_data_t;

typedef struct vm_region {
	vm_memory_region_status_t r_status;
	phys_addr_t r_base;
	phys_addr_t r_limit;
} vm_region_t;

typedef struct vm_cache {
	const char *c_name;
	vm_cache_flags_t c_flags;
	queue_entry_t c_list;

	queue_t c_slabs_full;
	queue_t c_slabs_partial;
	queue_t c_slabs_empty;

	spin_lock_t c_lock;

	/* number of objects that can be stored in a single slab */
	unsigned int c_obj_count;
	/* the size of object kept in the cache */
	unsigned int c_obj_size;
	/* combined size of vm_slab_t and the freelist */
	unsigned int c_hdr_size;
	/* power of 2 alignment for objects returned from the cache */
	unsigned int c_align;
	/* offset from one object to the next in a slab.
	   this may be different from c_obj_size depending
	   on the alignment settings for this cache. */
	unsigned int c_stride;
	/* size of page used for slabs */
	unsigned int c_page_order;
} vm_cache_t;

typedef struct vm_slab {
	vm_cache_t *s_cache;
	/* queue entry for vm_cache_t.c_slabs_* */
	queue_entry_t s_list;
	/* pointer to the first object slot. */
	void *s_objects;
	/* the number of objects allocated on the slab. */
	unsigned int s_obj_allocated;
	/* the index of the next free object.
	   if s_free is equal to FREELIST_END (defined in vm/cache.c)
	   there are no free slots left in the slab. */
	unsigned int s_free;
	/* list of free object slots.
	   when allocating:
	     - s_free should be set to the value of s_freelist[s_free]
	   when freeing:
	     - s_free should be set to the index of the object being freed.
	     - s_freelist[s_free] should be set to the previous value of s_free.
	   */
	unsigned int s_freelist[];
} vm_slab_t;

typedef struct vm_page {
	/* order of the page block that this page belongs too */
	uint16_t p_order : 4;
	/* the id of the NUMA node that this page belongs to */
	uint16_t p_node : 6;
	/* the id of the memory zone that this page belongs to */
	uint16_t p_zone : 3;
	/* some unused bits */
	uint16_t p_reserved : 3;

	/* vm_page_flags_t bitfields. */
	uint32_t p_flags;

	/* multi-purpose list.
	   the owner of the page can decide what to do with this.
	   some examples:
	     - the buddy allocator uses this to maintain its per-zone free-page lists.
        */
	queue_entry_t p_list;

	/* owner-specific data */
	union {
		struct {
			vm_slab_t *p_slab;
		};
	};

} __attribute__((aligned(2 * sizeof(unsigned long)))) vm_page_t;

extern kern_status_t vm_bootstrap(const vm_zone_descriptor_t *zones, size_t nr_zones);

extern vm_pg_data_t *vm_pg_data_get(vm_node_id_t node);

extern phys_addr_t vm_virt_to_phys(void *p);
extern void *vm_phys_to_virt(phys_addr_t p);

extern void vm_page_init_array();
extern vm_page_t *vm_page_get(phys_addr_t addr);
extern phys_addr_t vm_page_get_paddr(vm_page_t *pg);
extern vm_zone_t *vm_page_get_zone(vm_page_t *pg);
extern void *vm_page_get_vaddr(vm_page_t *pg);
extern size_t vm_page_get_pfn(vm_page_t *pg);
extern size_t vm_page_order_to_bytes(vm_page_order_t order);
extern size_t vm_page_order_to_pages(vm_page_order_t order);
extern vm_alignment_t vm_page_order_to_alignment(vm_page_order_t order);
extern vm_page_t *vm_page_alloc(vm_page_order_t order, vm_flags_t flags);
extern void vm_page_free(vm_page_t *pg);

extern int vm_page_split(vm_page_t *pg, vm_page_t **a, vm_page_t **b);
extern vm_page_t *vm_page_merge(vm_page_t *a, vm_page_t *b);
extern vm_page_t *vm_page_get_buddy(vm_page_t *pg);
extern vm_page_t *vm_page_get_next_tail(vm_page_t *pg);

extern size_t vm_bytes_to_pages(size_t bytes);

extern void vm_zone_init(vm_zone_t *z, const vm_zone_descriptor_t *zone_info);
extern vm_page_t *vm_zone_alloc_page(vm_zone_t *z, vm_page_order_t order, vm_flags_t flags);
extern void vm_zone_free_page(vm_zone_t *z, vm_page_t *pg);

extern vm_cache_t *vm_cache_create(const char *name, size_t objsz, vm_cache_flags_t flags);
extern void vm_cache_init(vm_cache_t *cache);
extern void vm_cache_destroy(vm_cache_t *cache);
extern void *vm_cache_alloc(vm_cache_t *cache, vm_flags_t flags);
extern void vm_cache_free(vm_cache_t *cache, void *p);

extern void kmalloc_init(void);
extern void *kmalloc(size_t count, vm_flags_t flags);
extern void *kzalloc(size_t count, vm_flags_t flags);
extern void kfree(void *p);

#endif