mem.c

Simple Operating Systems （简称SOS）是一个可以运行在X86平台上（包括QEMU

/* Copyright (C) 2005 David Decotigny

   This program is free software; you can redistribute it and/or
   modify it under the terms of the GNU General Public License
   as published by the Free Software Foundation; either version 2
   of the License, or (at your option) any later version.
   
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
   USA. 
*/

#include <sos/assert.h>
#include <sos/kmalloc.h>
#include <sos/physmem.h>
#include <hwcore/paging.h>
#include <sos/kmem_slab.h>
#include <sos/list.h>
#include <hwcore/paging.h>
#include <drivers/devices.h>
#include <sos/kmem_vmm.h>
#include <sos/uaccess.h>
#include <sos/chardev.h>

#include "mem.h"


/**
 * A mapped mem/kmem resource
 */
struct kernel_remapped_resource
{
  int ref_cnt;
  struct sos_umem_vmm_mapped_resource mr;
};


/** Called after the virtual region has been inserted inside its
    address space */
static void resource_ref(struct sos_umem_vmm_vr * vr)
{
  /* Retrieve the mem/kmem structure associated with the mapped resource */
  struct kernel_remapped_resource * resource;
  resource
    = (struct kernel_remapped_resource*)
    sos_umem_vmm_get_mapped_resource_of_vr(vr)->custom_data;

  /* Increment ref counter */
  resource->ref_cnt ++;
}


/** Called when the virtual region is removed from its address
    space */
static void resource_unref(struct sos_umem_vmm_vr * vr)
{
  /* Retrieve the mem/kmem structure associated with the mapped resource */
  struct kernel_remapped_resource * resource;
  resource
    = (struct kernel_remapped_resource*)
    sos_umem_vmm_get_mapped_resource_of_vr(vr)->custom_data;
  
  /* Decrement ref coutner */
  SOS_ASSERT_FATAL(resource->ref_cnt > 0);
  resource->ref_cnt --;

  /* Free the resource if it becomes unused */
  if (resource->ref_cnt == 0)
    sos_kfree((sos_vaddr_t)resource);
}


/** MOST IMPORTANT callback ! Called when a thread page faults on the
    resource's mapping */
static sos_ret_t kmem_page_in(struct sos_umem_vmm_vr * vr,
			      sos_uaddr_t uaddr,
			      sos_bool_t write_access)
{
  sos_vaddr_t vaddr;
  sos_ret_t retval = SOS_OK;
  sos_paddr_t ppage_paddr;

  /* Compute address of kernel page */
  vaddr = uaddr - sos_umem_vmm_get_start_of_vr(vr)
    + sos_umem_vmm_get_offset_in_resource(vr);

  /* Don't allow demand paging of non kernel pages */
  if (vaddr >= SOS_PAGING_BASE_USER_ADDRESS)
    return -SOS_EFAULT;

  /* Lookup physical kernel page */
  ppage_paddr = sos_paging_get_paddr(SOS_PAGE_ALIGN_INF(vaddr));

  /* Cannot access unmapped kernel pages */
  if (! ppage_paddr)
    return -SOS_EFAULT;
  
  /* Remap it in user space */
  retval = sos_paging_map(ppage_paddr,
			  SOS_PAGE_ALIGN_INF(uaddr),
			  TRUE,
			  sos_umem_vmm_get_prot_of_vr(vr));

  return retval;
}


/** The callbacks for a mapped kmem resource */
static struct sos_umem_vmm_vr_ops kmem_ops = (struct sos_umem_vmm_vr_ops)
{
  .ref     = resource_ref,
  .unref   = resource_unref,
  .page_in = kmem_page_in,
};


/** The callback that gets called when the resource gets mapped */
static sos_ret_t kmem_mmap(struct sos_umem_vmm_vr *vr)
{
  return sos_umem_vmm_set_ops_of_vr(vr, &kmem_ops);
}


/** The function responsible for mapping the /dev/kmem resource in
    user space */
static
sos_ret_t sos_dev_kmem_map(struct sos_umem_vmm_as * dest_as,
			   sos_uaddr_t *uaddr,
			   sos_size_t size,
			   sos_vaddr_t offset,
			   sos_ui32_t access_rights,
			   sos_ui32_t flags)
{
  sos_ret_t retval;
  struct kernel_remapped_resource * kmem_resource;

  /* Allocate a new "descriptor" for the resource */
  kmem_resource
    = (struct kernel_remapped_resource*) sos_kmalloc(sizeof(*kmem_resource),
						     0);
  if (! kmem_resource)
    return -SOS_ENOMEM;

  memset(kmem_resource, 0x0, sizeof(*kmem_resource));
  kmem_resource->mr.allowed_access_rights 
    = SOS_VM_MAP_PROT_READ
    | SOS_VM_MAP_PROT_WRITE
    | SOS_VM_MAP_PROT_EXEC;
  kmem_resource->mr.custom_data    = kmem_resource;
  kmem_resource->mr.mmap           = kmem_mmap;

  /* Map it in user space */
  retval = sos_umem_vmm_map(dest_as, uaddr, size,
			    access_rights, flags,
			    & kmem_resource->mr, offset);
  if (SOS_OK != retval)
    {
      sos_kfree((sos_vaddr_t)kmem_resource);
      return retval;
    }

  return SOS_OK;
}


/** MOST IMPORTANT callback ! Called when a thread page faults on the
    resource's mapping */
static sos_ret_t physmem_page_in(struct sos_umem_vmm_vr * vr,
				 sos_uaddr_t uaddr,
				 sos_bool_t write_access)
{
  sos_ret_t retval = SOS_OK;
  sos_paddr_t ppage_paddr;

  /* Compute address of kernel page */
  ppage_paddr = uaddr - sos_umem_vmm_get_start_of_vr(vr)
    + sos_umem_vmm_get_offset_in_resource(vr);

  /* Remap page in user space */
  retval = sos_paging_map(SOS_PAGE_ALIGN_INF(ppage_paddr),
			  SOS_PAGE_ALIGN_INF(uaddr),
			  TRUE,
			  sos_umem_vmm_get_prot_of_vr(vr));
  return retval;
}


/** The callbacks for a mapped physmem resource */
static struct sos_umem_vmm_vr_ops physmem_ops = (struct sos_umem_vmm_vr_ops)
{
  .ref     = resource_ref,
  .unref   = resource_unref,
  .page_in = physmem_page_in,
};


/** The callback that gets called when the resource gets mapped */
static sos_ret_t physmem_mmap(struct sos_umem_vmm_vr *vr)
{
  return sos_umem_vmm_set_ops_of_vr(vr, &physmem_ops);
}


/** The function responsible for mapping the /dev/mem resource in
    user space */
static
sos_ret_t sos_dev_physmem_map(struct sos_umem_vmm_as * dest_as,
			      sos_uaddr_t *uaddr,
			      sos_size_t size,
			      sos_paddr_t offset,
			      sos_ui32_t access_rights,
			      sos_ui32_t flags)
{
  sos_ret_t retval;
  struct kernel_remapped_resource * physmem_resource;

  physmem_resource
    = (struct kernel_remapped_resource*) sos_kmalloc(sizeof(*physmem_resource),
						     0);
  if (! physmem_resource)
    return -SOS_ENOMEM;

  memset(physmem_resource, 0x0, sizeof(*physmem_resource));
  physmem_resource->mr.allowed_access_rights 
    = SOS_VM_MAP_PROT_READ
    | SOS_VM_MAP_PROT_WRITE
    | SOS_VM_MAP_PROT_EXEC;
  physmem_resource->mr.custom_data    = physmem_resource;
  physmem_resource->mr.mmap           = physmem_mmap;

  retval = sos_umem_vmm_map(dest_as, uaddr, size,
			    access_rights, flags,
			    & physmem_resource->mr, offset);
  if (SOS_OK != retval)
    {
      sos_kfree((sos_vaddr_t)physmem_resource);
      return retval;
    }

  return SOS_OK;
}


/*
 * /dev/mem and /dev/kmem character device operations
 *
 * the "custom_data" field of the FS node is used to store the total
 * number of pages available
 */
#define GET_DEV_SIZE(fsnode) \
  ((sos_size_t)(fsnode)->custom_data)

static sos_ret_t dev_mem_fs_open(struct sos_fs_node        * fsnode,
				 struct sos_fs_opened_file * of,
				 void * chardev_class_custom_data)
{
  /* Make sure the device is supported by this driver and compute its
     "size" (use the custom_data field to store it) */
  switch (fsnode->dev_id.device_instance)
    {
      /* For /dev/kmem, go to the end of the kernel mapping */
    case SOS_CHARDEV_KMEM_MINOR:
      fsnode->custom_data = (void*)SOS_PAGING_BASE_USER_ADDRESS;
      return SOS_OK;
      break;

      /* For /dev/mem, go to the end of physical memory */
    case SOS_CHARDEV_PHYSMEM_MINOR:
      {
	sos_size_t ram_pages = 0;
	sos_physmem_get_state(& ram_pages, NULL);
	fsnode->custom_data = (void*)(ram_pages << SOS_PAGE_SHIFT);
      }
      return SOS_OK;
      break;

    default:
      break;
    }

  return -SOS_ENODEV;
}


static sos_ret_t dev_mem_fs_seek(struct sos_fs_opened_file *this,
				 sos_lsoffset_t offset,
				 sos_seek_whence_t whence,
				 /* out */ sos_lsoffset_t * result_position)
{
  /* Make sure the device is supported by this driver */
  struct sos_fs_node * fsnode = sos_fs_nscache_get_fs_node(this->direntry);

  /* Artificiallly update the position in the "file" */
  sos_lsoffset_t ref_offs;
  sos_lsoffset_t dev_size = GET_DEV_SIZE(fsnode);

  *result_position = this->position;
  switch (whence)
    {
    case SOS_SEEK_SET:
      ref_offs = 0;
      break;

    case SOS_SEEK_CUR:
      ref_offs = this->position;
      break;

    case SOS_SEEK_END:
      ref_offs = dev_size;
      break;

    default:
      return -SOS_EINVAL;
    }

  /* Forbid accesses "before" the start of the device */
  if (offset < -ref_offs)
    return -SOS_EINVAL;

  /* Forbid accesses "after" the end of the device */
  else if (ref_offs + offset > dev_size)
    return -SOS_EINVAL;
  
  this->position = ref_offs + offset;
  *result_position = this->position;
  return SOS_OK;
}


typedef enum { DO_READ, DO_WRITE } dev_mem_access_type_t;
static sos_ret_t dev_mem_fs_access(struct sos_fs_opened_file *this,
				   sos_uaddr_t user_buf,
				   sos_size_t * /* in/out */len,
				   dev_mem_access_type_t access_type)
{
  struct sos_fs_node * fsnode = sos_fs_nscache_get_fs_node(this->direntry);
  sos_vaddr_t   physmem_transfer_kernel_page = 0; /* Used for /dev/mem only */
  sos_uoffset_t offs;
  sos_size_t    accesslen = 0;

  /* Readjust copy length to match the size of the device */
  if (this->position + *len >= GET_DEV_SIZE(fsnode))
    *len = GET_DEV_SIZE(fsnode) - this->position;

  /* Ignore zero-size requests */
  if (*len <= 0)
    return SOS_OK;

  /* For /dev/mem device, prepare a kernel page to copy the physical
     pages before transferring to user space */
  if (SOS_CHARDEV_PHYSMEM_MINOR == fsnode->dev_id.device_instance)
    {
      physmem_transfer_kernel_page = sos_kmem_vmm_alloc(1, 0);
      if (! physmem_transfer_kernel_page)
	return -SOS_ENOMEM;
    }

  /* Try to copy the data in page-size chunks */
  offs = this->position;
  while (offs < this->position + *len)
    {
      /* Retrieve page address of data in kernel memory */
      sos_uoffset_t page_boundary = SOS_PAGE_ALIGN_INF(offs);
      sos_vaddr_t   page_vaddr;
      sos_uoffset_t offset_in_page;
      sos_uoffset_t accesslen_in_page;
      sos_ret_t retval;

      /* For /dev/mem device, we need to map the page in kernel memory
	 before */
      if (SOS_CHARDEV_PHYSMEM_MINOR == fsnode->dev_id.device_instance)
	{
	  retval = sos_paging_map(page_boundary,
				  physmem_transfer_kernel_page,
				  FALSE,
				  (access_type==DO_WRITE)?
				    SOS_VM_MAP_PROT_WRITE
				    :SOS_VM_MAP_PROT_READ);
	  if (SOS_OK != retval)
	    break;

	  page_vaddr = physmem_transfer_kernel_page;
	}
      /* For /dev/kmem device, the page should already be in kernel space */
      else if (! sos_kmem_vmm_is_valid_vaddr(page_boundary))
	break; /* No: page is not mapped in kernel space ! */
      else
	page_vaddr = page_boundary; /* Yes, page is mapped */

      /* Now copy the data from kernel to user space */
      offset_in_page = offs - page_boundary;
      accesslen_in_page  = SOS_PAGE_SIZE - offset_in_page;
      if (accesslen + accesslen_in_page > *len)
	accesslen_in_page = *len - accesslen;

      if (access_type==DO_WRITE)
	retval = sos_memcpy_from_user(page_vaddr + offset_in_page,
				      user_buf + accesslen,
				      accesslen_in_page);
      else
	retval = sos_memcpy_to_user(user_buf + accesslen,
				    page_vaddr + offset_in_page,
				    accesslen_in_page);
      
      /* Now, for /dev/mem, unmap the page from kernel */
      if (SOS_CHARDEV_PHYSMEM_MINOR == fsnode->dev_id.device_instance)
	sos_paging_unmap(physmem_transfer_kernel_page);
      
      /* Go to next page if possible */
      if (retval < 0)
	break;

      accesslen += retval;
      /* If transfer was interrupted, stop here */
      if (retval < accesslen_in_page)
	break;

      /* Go on to next page */
      offs = page_boundary + SOS_PAGE_SIZE;
    }

  /* Release the temporary page for physical mem transfers */
  if (SOS_CHARDEV_PHYSMEM_MINOR == fsnode->dev_id.device_instance)
    sos_kmem_vmm_free(physmem_transfer_kernel_page);

  /* Update the position in the "file" */
  *len = accesslen;
  this->position += accesslen;
  return SOS_OK;
}


static sos_ret_t dev_mem_fs_read(struct sos_fs_opened_file *this,
				   sos_uaddr_t dest_buf,
				   sos_size_t * /* in/out */len)
{
  return dev_mem_fs_access(this, dest_buf, len, DO_READ);
}


static sos_ret_t dev_mem_fs_write(struct sos_fs_opened_file *this,
				  sos_uaddr_t src_buf,
				  sos_size_t * /* in/out */len)
{
  return dev_mem_fs_access(this, src_buf, len, DO_WRITE);
}


static sos_ret_t dev_mem_fs_mmap(struct sos_fs_opened_file *this,
				 sos_uaddr_t *uaddr, sos_size_t size,
				 sos_ui32_t access_rights,
				 sos_ui32_t flags,
				 sos_luoffset_t offset)
{
  struct sos_fs_node * fsnode = sos_fs_nscache_get_fs_node(this->direntry);
  if (SOS_CHARDEV_PHYSMEM_MINOR == fsnode->dev_id.device_instance)
    return sos_dev_physmem_map(sos_process_get_address_space(this->owner),
			       uaddr, size, offset, access_rights, flags);

  return sos_dev_kmem_map(sos_process_get_address_space(this->owner),
			  uaddr, size, offset, access_rights, flags);
}


static struct sos_chardev_ops dev_mem_fs_ops
  = (struct sos_chardev_ops) {
    .open  = dev_mem_fs_open,
    .close = NULL,
    .seek  = dev_mem_fs_seek,
    .read  = dev_mem_fs_read,
    .write = dev_mem_fs_write,
    .mmap  = dev_mem_fs_mmap,
    .fcntl = NULL,
    .ioctl = NULL
  };


sos_ret_t sos_dev_mem_chardev_setup()
{
  return sos_chardev_register_class(SOS_CHARDEV_MEM_MAJOR,
				    & dev_mem_fs_ops,
				    NULL);
}