kernel: vm_map: Add mmap() and munmap()

Signed-off-by: Ian Moffett <ian@osmora.org>

7 files changed, 268 insertions, 0 deletions

diff --git a/lib/libc/include/sys/syscall.h b/lib/libc/include/sys/syscall.h
index 55d7bbf..e0900c7 100644
--- a/lib/libc/include/sys/syscall.h
+++ b/lib/libc/include/sys/syscall.h
@@ -40,6 +40,8 @@
 #define SYS_close 4
 #define SYS_read  5
 #define SYS_lseek 6
+#define SYS_mmap  7
+#define SYS_munmap 8
 
 #if !defined(__ASSEMBLER__)
 __attribute__((__always_inline__))
diff --git a/sys/include/sys/proc.h b/sys/include/sys/proc.h
index 09aebf5..f74d1da 100644
--- a/sys/include/sys/proc.h
+++ b/sys/include/sys/proc.h
@@ -38,6 +38,7 @@
 #include <machine/frame.h>
 #include <machine/pcb.h>
 #include <vm/vm.h>
+#include <vm/map.h>
 
 #define PROC_MAX_FDS 256
 #define PROC_MAX_ADDR_RANGE 4
@@ -62,6 +63,8 @@ struct proc {
     uint8_t is_user;
     uint32_t signal;
     struct filedesc *fds[PROC_MAX_FDS];
+    struct spinlock mapspace_lock;
+    struct vm_mapspace mapspace;
     TAILQ_ENTRY(proc) link;
 };
 
diff --git a/sys/include/sys/syscall.h b/sys/include/sys/syscall.h
index 8ebc2da..59b6036 100644
--- a/sys/include/sys/syscall.h
+++ b/sys/include/sys/syscall.h
@@ -43,6 +43,8 @@ enum {
     SYS_close,
     SYS_read,
     SYS_lseek,
+    SYS_mmap,
+    SYS_munmap,
     __MAX_SYSCALLS
 };
 
diff --git a/sys/include/vm/map.h b/sys/include/vm/map.h
index 078a5e8..f482788 100644
--- a/sys/include/vm/map.h
+++ b/sys/include/vm/map.h
@@ -32,11 +32,49 @@
 
 #include <sys/types.h>
 #include <sys/cdefs.h>
+#include <sys/queue.h>
+#include <sys/syscall.h>
+#include <sys/spinlock.h>
 #include <vm/pmap.h>
+#include <vm/vm.h>
+
+#define MAP_SHARED      0x0001
+#define MAP_PRIVATE     0x0002
+#define MAP_ANONYMOUS   0x0010
+#define MAP_FAILED      ((void *)-1)
+
+/* Memory map table entry count */
+#define MTAB_ENTRIES 32
+
+struct vm_mapping {
+    TAILQ_ENTRY(vm_mapping) link;
+    struct vm_range range;
+    paddr_t physmem_base;
+
+    /* Private */
+    size_t vhash;           /* Virtual address hash */
+};
+
+typedef TAILQ_HEAD(, vm_mapping) vm_mapq_t;
+
+struct vm_mapspace {
+    vm_mapq_t mtab[MTAB_ENTRIES];   /* Map table */
+    size_t map_count;
+};
+
 
 int vm_map_create(struct vas vas, vaddr_t va, paddr_t pa, vm_prot_t prot,
                   size_t bytes);
 
 int vm_map_destroy(struct vas vas, vaddr_t va, size_t bytes);
 
+uint64_t sys_mmap(struct syscall_args *args);
+uint64_t sys_munmap(struct syscall_args *args);
+
+/* Mapespace operations */
+void vm_mapspace_insert(struct vm_mapspace *ms, struct vm_mapping *mapping);
+void vm_mapspace_remove(struct vm_mapspace *ms, struct vm_mapping *mapping);
+struct vm_mapping *vm_mapping_fetch(struct vm_mapspace *ms, vaddr_t va);
+void vm_free_mapq(vm_mapq_t *mapq);
+
 #endif  /* !_VM_MMAP_H_ */
diff --git a/sys/kern/kern_sched.c b/sys/kern/kern_sched.c
index 877c55d..214032a 100644
--- a/sys/kern/kern_sched.c
+++ b/sys/kern/kern_sched.c
@@ -265,6 +265,10 @@ sched_create_td(uintptr_t rip, char *argvp[], char *envp[], struct auxval auxv,
     td->tf = tf;
     td->addrsp = vas;
     td->is_user = is_user;
+    for (size_t i = 0; i < MTAB_ENTRIES; ++i) {
+        /* Init the memory mapping table */
+        TAILQ_INIT(&td->mapspace.mtab[i]);
+    }
     if (prog_range != NULL) {
         memcpy(exec_range, prog_range, sizeof(struct vm_range));
     }
@@ -288,6 +292,7 @@ static void
 sched_destroy_td(struct proc *td)
 {
     const struct vm_range *stack_range = &td->addr_range[ADDR_RANGE_STACK];
+    vm_mapq_t *mapq;
 
     processor_free_pcb(td);
 
@@ -308,6 +313,11 @@ sched_destroy_td(struct proc *td)
         fd_close_fdnum(td, i);
     }
 
+    for (size_t i = 0; i < MTAB_ENTRIES; ++i) {
+        mapq = &td->mapspace.mtab[i];
+        vm_free_mapq(mapq);
+    }
+
     pmap_free_vas(vm_get_ctx(), td->addrsp);
     dynfree(td);
 }
diff --git a/sys/kern/kern_syscall.c b/sys/kern/kern_syscall.c
index a630195..b4a86f3 100644
--- a/sys/kern/kern_syscall.c
+++ b/sys/kern/kern_syscall.c
@@ -32,6 +32,7 @@
 #include <sys/cdefs.h>
 #include <sys/types.h>
 #include <sys/filedesc.h>
+#include <vm/map.h>
 
 __noreturn static uint64_t
 sys_exit(struct syscall_args *args)
@@ -47,4 +48,6 @@ uint64_t(*g_syscall_table[__MAX_SYSCALLS])(struct syscall_args *args) = {
     sys_close,
     sys_read,
     sys_lseek,
+    sys_mmap,
+    sys_munmap
 };
diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index 7d79108..266d696 100644
--- a/sys/vm/vm_map.c
+++ b/sys/vm/vm_map.c
@@ -30,11 +30,208 @@
 #include <vm/map.h>
 #include <vm/vm.h>
 #include <vm/pmap.h>
+#include <vm/physseg.h>
+#include <vm/dynalloc.h>
 #include <sys/types.h>
 #include <sys/cdefs.h>
 #include <sys/panic.h>
+#include <sys/sched.h>
 #include <lib/assert.h>
 
+#define ALLOC_MAPPING() dynalloc(sizeof(struct vm_mapping))
+
+static size_t
+vm_hash_vaddr(vaddr_t va) {
+    va = (va ^ (va >> 30)) * (size_t)0xBF58476D1CE4E5B9;
+    va = (va ^ (va >> 27)) * (size_t)0x94D049BB133111EB;
+    va = va ^ (va >> 31);
+    return va;
+}
+
+/*
+ * Destroy a map queue.
+ */
+void
+vm_free_mapq(vm_mapq_t *mapq)
+{
+    struct vm_mapping *map;
+    size_t map_pages, granule;
+
+    granule = vm_get_page_size();
+    TAILQ_FOREACH(map, mapq, link) {
+        map_pages = (map->range.end - map->range.start) / granule;
+        vm_free_pageframe(map->range.start, map_pages);
+    }
+    dynfree(map);
+}
+
+/*
+ * Remove a mapping from a mapspace.
+ *
+ * @ms: Mapspace.
+ * @mapping: Mapping to remove.
+ */
+void
+vm_mapspace_remove(struct vm_mapspace *ms, struct vm_mapping *mapping)
+{
+    size_t vhash;
+    vm_mapq_t *mapq;
+
+    if (ms == NULL)
+        return;
+
+    vhash = vm_hash_vaddr(mapping->range.start);
+    mapq = &ms->mtab[vhash % MTAB_ENTRIES];
+    TAILQ_REMOVE(mapq, mapping, link);
+}
+
+/*
+ * Fetch a mapping from a mapspace.
+ *
+ * @ms: Mapspace.
+ * @va: Virtual address.
+ */
+struct vm_mapping *
+vm_mapping_fetch(struct vm_mapspace *ms, vaddr_t va)
+{
+    size_t vhash;
+    const vm_mapq_t *mapq;
+    struct vm_mapping *map;
+
+    if (ms == NULL)
+        return NULL;
+
+    vhash = vm_hash_vaddr(va);
+    mapq = &ms->mtab[vhash % MTAB_ENTRIES];
+
+    TAILQ_FOREACH(map, mapq, link) {
+        if (map->vhash == vhash) {
+            return map;
+        }
+    }
+
+    return NULL;
+}
+
+/*
+ * Insert a mapping into a mapspace.
+ *
+ * @ms: Target mapspace.
+ * @mapping: Mapping to insert.
+ */
+void
+vm_mapspace_insert(struct vm_mapspace *ms, struct vm_mapping *mapping)
+{
+    size_t vhash;
+    vm_mapq_t *q;
+
+    if (mapping == NULL || ms == NULL)
+        return;
+
+    vhash = vm_hash_vaddr(mapping->range.start);
+    mapping->vhash = vhash;
+
+    q = &ms->mtab[vhash % MTAB_ENTRIES];
+    TAILQ_INSERT_HEAD(q, mapping, link);
+}
+
+static int
+munmap(void *addr, size_t len)
+{
+    struct proc *td = this_td();
+    struct vm_mapping *mapping;
+
+    struct vm_mapspace *ms;
+    size_t map_len, granule;
+    vaddr_t map_start, map_end;
+
+    spinlock_acquire(&td->mapspace_lock);
+    ms = &td->mapspace;
+
+    granule = vm_get_page_size();
+    mapping = vm_mapping_fetch(ms, (vaddr_t)addr);
+    if (mapping == NULL) {
+        return -1;
+    }
+
+    map_start = mapping->range.start;
+    map_end = mapping->range.end;
+    map_len = map_end - map_start;
+
+    /* Release the mapping */
+    vm_map_destroy(td->addrsp, map_start, map_len);
+    vm_free_pageframe(mapping->range.start, map_len / granule);
+
+    /* Destroy the mapping descriptor */
+    vm_mapspace_remove(ms, mapping);
+    dynfree(mapping);
+    spinlock_release(&td->mapspace_lock);
+    return 0;
+}
+
+static void *
+mmap(void *addr, size_t len, int prot, int flags, int fildes, off_t off)
+{
+    const int PROT_MASK = PROT_WRITE | PROT_EXEC;
+    const size_t GRANULE = vm_get_page_size();
+    uintptr_t map_end, map_start;
+
+    struct proc *td = this_td();
+    struct vm_mapping *mapping = ALLOC_MAPPING();
+
+    size_t misalign = ((vaddr_t)addr) & (GRANULE - 1);
+    int status;
+    paddr_t physmem;
+
+    if ((prot & ~PROT_MASK) != 0)
+        /* Invalid prot */
+        return MAP_FAILED;
+
+    /* Allocate the physical memory */
+    physmem = vm_alloc_pageframe(len / GRANULE);
+    if (physmem == 0)
+        return MAP_FAILED;
+
+    /*
+     * Handle address being NULL.
+     *
+     * FIXME: XXX: We currently identity map physmem which
+     *             is probably not ideal.
+     */
+    if (addr == NULL) {
+        addr = (void *)physmem;
+    }
+
+    /* Handle an anonymous map request */
+    if (__TEST(flags, MAP_ANONYMOUS)) {
+        /*
+         * XXX: There is no need to worry about alignment yet
+         *      as vm_map_create() handles that internally.
+         */
+        prot |= PROT_USER;
+        status = vm_map_create(td->addrsp, (vaddr_t)addr, physmem, prot, len);
+        if (status != 0) {
+            vm_free_pageframe(physmem, len / GRANULE);
+            return MAP_FAILED;
+        }
+    } else {
+        return MAP_FAILED;
+    }
+
+    map_start = __ALIGN_DOWN((vaddr_t)addr, GRANULE);
+    map_end = map_start + __ALIGN_UP(len + misalign, GRANULE);
+
+    mapping->range.start = map_start;
+    mapping->range.end = map_end;
+    mapping->physmem_base = physmem;
+
+    /* Add to mapspace */
+    spinlock_acquire(&td->mapspace_lock);
+    vm_mapspace_insert(&td->mapspace, mapping);
+    spinlock_release(&td->mapspace_lock);
+    return (void *)addr;
+}
+
 /*
  * Internal routine for cleaning up.
  *
@@ -141,3 +338,16 @@ vm_map_destroy(struct vas vas, vaddr_t va, size_t bytes)
 
     return 0;
 }
+
+uint64_t
+sys_mmap(struct syscall_args *args)
+{
+    return (uintptr_t)mmap((void *)args->arg0, args->arg1, args->arg2,
+                           args->arg3, args->arg4, args->arg5);
+}
+
+uint64_t
+sys_munmap(struct syscall_args *args)
+{
+    return munmap((void *)args->arg0, args->arg1);
+}