summaryrefslogtreecommitdiff
path: root/sys/vm/vm_map.c
diff options
context:
space:
mode:
Diffstat (limited to 'sys/vm/vm_map.c')
-rw-r--r--sys/vm/vm_map.c327
1 files changed, 327 insertions, 0 deletions
diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index 7b0656b..26effdb 100644
--- a/sys/vm/vm_map.c
+++ b/sys/vm/vm_map.c
@@ -30,9 +30,70 @@
#include <sys/param.h>
#include <sys/types.h>
#include <sys/errno.h>
+#include <sys/proc.h>
+#include <sys/systm.h>
+#include <sys/syscall.h>
+#include <sys/syslog.h>
+#include <sys/mman.h>
+#include <sys/filedesc.h>
+#include <vm/dynalloc.h>
+#include <vm/vm_pager.h>
+#include <vm/vm_device.h>
#include <vm/pmap.h>
#include <vm/map.h>
#include <vm/vm.h>
+#include <assert.h>
+
+#define pr_trace(fmt, ...) kprintf("vm_map: " fmt, ##__VA_ARGS__)
+#define pr_error(...) pr_trace(__VA_ARGS__)
+
+RBT_GENERATE(lgdr_entries, mmap_entry, hd, mmap_entrycmp);
+
+static inline void
+mmap_dbg(void *addr, size_t len, int prot, int flags, int fildes, off_t off)
+{
+ pr_trace("addr=%p, len=%d, prot=%x\nflags=%x, fildes=%d, off=%d\n",
+ addr, len, prot, flags, fildes, off);
+}
+
+/*
+ * Add a memory mapping to the mmap ledger.
+ *
+ * @td: Process to add mapping to.
+ * @ep: Memory map entry to add.
+ * @len: Length of memory mapping in bytes.
+ */
+static inline int
+mmap_add(struct proc *td, struct mmap_entry *ep)
+{
+ struct mmap_entry *tmp;
+ struct mmap_lgdr *lp = td->mlgdr;
+
+ if (ep->size == 0) {
+ return -EINVAL;
+ }
+
+ tmp = RBT_INSERT(lgdr_entries, &lp->hd, ep);
+ __assert(tmp == NULL);
+ lp->nbytes += ep->size;
+ return 0;
+}
+
+/*
+ * Remove memory mapping from mmap ledger
+ *
+ * @td: Process to remove mapping from.
+ * @ep: Memory map entry to remove.
+ */
+static inline void
+mmap_remove(struct proc *td, struct mmap_entry *ep)
+{
+ struct mmap_lgdr *lp = td->mlgdr;
+
+ RBT_REMOVE(lgdr_entries, &lp->hd, ep);
+ lp->nbytes -= ep->size;
+ dynfree(ep);
+}
/*
* Create/destroy virtual memory mappings in a specific
@@ -85,6 +146,263 @@ vm_map_modify(struct vas vas, vaddr_t va, paddr_t pa, vm_prot_t prot, bool unmap
}
/*
+ * Create a physical to virtual memory mapping.
+ *
+ * @addr: Virtual address to map (NULL to be any).
+ * @len: The amount of bytes to map (must be page aligned)
+ * @prot: Protection flags (PROT_*)
+ * @fildes: File descriptor.
+ * @off: Offset.
+ *
+ * TODO: Fields to use: `fildes' and `off'
+ * XXX: Must be called after pid 1 is up and running to avoid
+ * crashes.
+ */
+void *
+mmap(void *addr, size_t len, int prot, int flags, int fildes, off_t off)
+{
+ struct vm_object *map_obj = NULL;
+ struct cdevsw *cdevp;
+ struct vm_page *pg;
+ struct mmap_entry *ep;
+ struct vnode *vp;
+ struct filedesc *fdp;
+ struct proc *td;
+ struct vas vas;
+ int error, npgs;
+ paddr_t pa;
+ vaddr_t va;
+ size_t misalign;
+ off_t page_off;
+
+ misalign = len & (DEFAULT_PAGESIZE - 1);
+ len = ALIGN_UP(len + misalign, DEFAULT_PAGESIZE);
+ npgs = len / DEFAULT_PAGESIZE;
+ vas = pmap_read_vas();
+
+ /* Validate flags */
+ if (ISSET(flags, MAP_FIXED)) {
+ pr_error("mmap: fixed mappings not yet supported\n");
+ mmap_dbg(addr, len, prot, flags, fildes, off);
+ return NULL;
+ }
+
+ /*
+ * Attempt to open the file if mapping
+ * is shared.
+ */
+ if (ISSET(flags, MAP_SHARED)) {
+ fdp = fd_get(fildes);
+ if (fdp == NULL) {
+ pr_error("mmap: no such fd (fd=%d)\n", fildes);
+ return NULL;
+ }
+
+ vp = fdp->vp;
+ if (vp->type != VCHR) {
+ /* TODO */
+ pr_error("mmap: only device files supported\n");
+ return NULL;
+ }
+
+ map_obj = dv_attach(vp->major, vp->dev, prot);
+ if (map_obj == NULL) {
+ kprintf("mmap: dv_attach() failure\n");
+ return NULL;
+ }
+
+ cdevp = map_obj->data;
+ if ((pa = cdevp->mmap(vp->dev, len, off, 0)) == 0) {
+ kprintf("mmap: dev mmap() gave 0\n");
+ return NULL;
+ }
+
+ /*
+ * If the address passed is NULL, just identity
+ * map everything.
+ *
+ * XXX: This is why the bounds check done in the
+ * cdev mmap() *must* be correct.
+ *
+ * TODO: Use copy-on-write for this instead. Since mapping
+ * certain devices may required a lot of memory to
+ * be referenced anyways, we could use a buffered
+ * copy-on-write technique where only a window of
+ * pages can be mapped on-demand and other pages
+ * freed when that window is exceeded.
+ */
+ if (addr == NULL) {
+ addr = (void *)pa;
+ }
+
+ va = ALIGN_DOWN((vaddr_t)addr, DEFAULT_PAGESIZE);
+ error = vm_map(vas, va, pa, prot, len);
+ if (error != 0) {
+ kprintf("mmap: map failed (error=%d)\n", error);
+ return NULL;
+ }
+
+ goto done;
+ }
+
+ /* Only allocate new obj if needed */
+ if (map_obj == NULL) {
+ map_obj = dynalloc(sizeof(*map_obj));
+ if (map_obj == NULL) {
+ kprintf("mmap: failed to allocate map object\n");
+ return NULL;
+ }
+ error = vm_obj_init(map_obj, &vm_anonops, 1);
+ if (error < 0) {
+ kprintf("mmap: vm_obj_init() returned %d\n", error);
+ kprintf("mmap: failed to init object\n");
+ return NULL;
+ }
+ }
+
+ /* XXX: Assuming private */
+ va = ALIGN_DOWN((vaddr_t)addr, DEFAULT_PAGESIZE);
+
+ for (int i = 0; i < npgs; ++i) {
+ pg = vm_pagealloc(map_obj, PALLOC_ZERO);
+ page_off = i * DEFAULT_PAGESIZE;
+
+ if (pg == NULL) {
+ /* TODO */
+ pr_error("mmap: failed to allocate page %d\n");
+ return NULL;
+ }
+
+ /* TODO: copy-on-write */
+ if (addr == NULL) {
+ va = pg->phys_addr;
+ addr = (void *)va;
+ }
+
+ pa = pg->phys_addr;
+ error = vm_map(vas, va + page_off, pa, prot, len);
+ if (error < 0) {
+ pr_error("mmap: failed to map page (retval=%x)\n", error);
+ return NULL;
+ }
+ }
+
+done:
+ /* Add entry to ledger */
+ td = this_td();
+ ep = dynalloc(sizeof(*ep));
+ if (ep == NULL) {
+ pr_error("mmap: failed to allocate mmap ledger entry\n");
+ return NULL;
+ }
+
+ ep->va_start = va;
+ ep->obj = map_obj;
+ ep->size = len;
+ mmap_add(td, ep);
+ return addr;
+}
+
+/*
+ * Remove mappings for entire pages that
+ * belong to the current process.
+ *
+ * XXX: POSIX munmap(3) requires `addr' to be page-aligned
+ * and will return -EINVAL if otherwise. However, with
+ * OUSI munmap(3), `addr' is rounded down to the nearest
+ * multiple of the machine page size.
+ */
+int
+munmap(void *addr, size_t len)
+{
+ int pgno;
+ vaddr_t va;
+ struct proc *td;
+ struct mmap_lgdr *lp;
+ struct mmap_entry find, *res;
+ struct vas vas;
+
+ if (addr == NULL || len == 0) {
+ return -EINVAL;
+ }
+
+ /* Apply machine specific addr/len adjustments */
+ va = ALIGN_DOWN((vaddr_t)addr, DEFAULT_PAGESIZE);
+ len = ALIGN_UP(len, DEFAULT_PAGESIZE);
+ pgno = va >> 12;
+
+ td = this_td();
+ __assert(td != NULL && "no pid 1");
+ vas = pmap_read_vas();
+
+ /*
+ * Try to get the mmap ledger, should not run into
+ * any issues as long as the PCB isn't borked. However,
+ * if it somehow is, just segfault ourselves.
+ */
+ if ((lp = td->mlgdr) == NULL) {
+ __sigraise(SIGSEGV);
+ return -EFAULT; /* Unreachable */
+ }
+
+ /* Lookup entry in ledger with virtual address */
+ find.va_start = va;
+ res = RBT_FIND(lgdr_entries, &lp->hd, &find);
+ if (res == NULL) {
+ pr_error("munmap: page %d not in ledger\n", pgno);
+ return -EINVAL;
+ }
+
+ vm_unmap(vas, va, len);
+ mmap_remove(td, res);
+ return 0;
+}
+
+/*
+ * mmap() syscall
+ *
+ * arg0 -> addr
+ * arg1 -> len
+ * arg2 -> prot
+ * arg3 -> flags
+ * arg4 -> fildes
+ * arg5 -> off
+ */
+scret_t
+sys_mmap(struct syscall_args *scargs)
+{
+ void *addr;
+ size_t len;
+ int prot, flags;
+ int fildes, off;
+
+ addr = (void *)scargs->arg0;
+ len = scargs->arg1;
+ prot = scargs->arg2 | PROT_USER;
+ flags = scargs->arg3;
+ fildes = scargs->arg4;
+ off = scargs->arg5;
+ return (scret_t)mmap(addr, len, prot, flags, fildes, off);
+}
+
+/*
+ * munmap() syscall
+ *
+ * arg0 -> addr
+ * arg1 -> len
+ */
+scret_t
+sys_munmap(struct syscall_args *scargs)
+{
+ void *addr;
+ size_t len;
+
+ addr = (void *)scargs->arg0;
+ len = scargs->arg1;
+ return (scret_t)munmap(addr, len);
+}
+
+/*
* Create a virtual memory mapping in a specific
* address space.
*
@@ -130,3 +448,12 @@ vm_unmap(struct vas vas, vaddr_t va, size_t count)
{
return vm_map_modify(vas, va, 0, 0, true, count);
}
+
+/*
+ * Helper for tree(3) and the mmap ledger.
+ */
+int
+mmap_entrycmp(const struct mmap_entry *a, const struct mmap_entry *b)
+{
+ return (a->va_start < b->va_start) ? -1 : a->va_start > b->va_start;
+}