2001-10-03 17:10:38 +04:00
|
|
|
/////////////////////////////////////////////////////////////////////////
|
2008-04-23 02:05:38 +04:00
|
|
|
// $Id: paging.cc,v 1.123 2008-04-22 22:05:38 sshwarts Exp $
|
2001-10-03 17:10:38 +04:00
|
|
|
/////////////////////////////////////////////////////////////////////////
|
|
|
|
//
|
2001-04-10 06:20:02 +04:00
|
|
|
// Copyright (C) 2001 MandrakeSoft S.A.
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
|
|
|
// MandrakeSoft S.A.
|
|
|
|
// 43, rue d'Aboukir
|
|
|
|
// 75002 Paris - France
|
|
|
|
// http://www.linux-mandrake.com/
|
|
|
|
// http://www.mandrakesoft.com/
|
|
|
|
//
|
|
|
|
// This library is free software; you can redistribute it and/or
|
|
|
|
// modify it under the terms of the GNU Lesser General Public
|
|
|
|
// License as published by the Free Software Foundation; either
|
|
|
|
// version 2 of the License, or (at your option) any later version.
|
|
|
|
//
|
|
|
|
// This library 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
|
|
|
|
// Lesser General Public License for more details.
|
|
|
|
//
|
|
|
|
// You should have received a copy of the GNU Lesser General Public
|
|
|
|
// License along with this library; if not, write to the Free Software
|
|
|
|
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
2007-11-18 02:28:33 +03:00
|
|
|
/////////////////////////////////////////////////////////////////////////
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2001-05-24 22:46:34 +04:00
|
|
|
#define NEED_CPU_REG_SHORTCUTS 1
|
2001-04-10 05:04:59 +04:00
|
|
|
#include "bochs.h"
|
2006-03-07 01:03:16 +03:00
|
|
|
#include "cpu.h"
|
merge in BRANCH-io-cleanup.
To see the commit logs for this use either cvsweb or
cvs update -r BRANCH-io-cleanup and then 'cvs log' the various files.
In general this provides a generic interface for logging.
logfunctions:: is a class that is inherited by some classes, and also
. allocated as a standalone global called 'genlog'. All logging uses
. one of the ::info(), ::error(), ::ldebug(), ::panic() methods of this
. class through 'BX_INFO(), BX_ERROR(), BX_DEBUG(), BX_PANIC()' macros
. respectively.
.
. An example usage:
. BX_INFO(("Hello, World!\n"));
iofunctions:: is a class that is allocated once by default, and assigned
as the iofunction of each logfunctions instance. It is this class that
maintains the file descriptor and other output related code, at this
point using vfprintf(). At some future point, someone may choose to
write a gui 'console' for bochs to which messages would be redirected
simply by assigning a different iofunction class to the various logfunctions
objects.
More cleanup is coming, but this works for now. If you want to see alot
of debugging output, in main.cc, change onoff[LOGLEV_DEBUG]=0 to =1.
Comments, bugs, flames, to me: todd@fries.net
2001-05-15 18:49:57 +04:00
|
|
|
#define LOG_THIS BX_CPU_THIS_PTR
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
// X86 Registers Which Affect Paging:
|
|
|
|
// ==================================
|
|
|
|
//
|
|
|
|
// CR0:
|
|
|
|
// bit 31: PG, Paging (386+)
|
|
|
|
// bit 16: WP, Write Protect (486+)
|
|
|
|
// 0: allow supervisor level writes into user level RO pages
|
|
|
|
// 1: inhibit supervisor level writes into user level RO pages
|
|
|
|
//
|
|
|
|
// CR3:
|
|
|
|
// bit 31..12: PDBR, Page Directory Base Register (386+)
|
|
|
|
// bit 4: PCD, Page level Cache Disable (486+)
|
|
|
|
// Controls caching of current page directory. Affects only the processor's
|
|
|
|
// internal caches (L1 and L2).
|
|
|
|
// This flag ignored if paging disabled (PG=0) or cache disabled (CD=1).
|
|
|
|
// Values:
|
|
|
|
// 0: Page Directory can be cached
|
|
|
|
// 1: Page Directory not cached
|
|
|
|
// bit 3: PWT, Page level Writes Transparent (486+)
|
|
|
|
// Controls write-through or write-back caching policy of current page
|
|
|
|
// directory. Affects only the processor's internal caches (L1 and L2).
|
|
|
|
// This flag ignored if paging disabled (PG=0) or cache disabled (CD=1).
|
|
|
|
// Values:
|
|
|
|
// 0: write-back caching enabled
|
|
|
|
// 1: write-through caching enabled
|
|
|
|
//
|
|
|
|
// CR4:
|
|
|
|
// bit 4: PSE, Page Size Extension (Pentium+)
|
|
|
|
// 0: 4KByte pages (typical)
|
|
|
|
// 1: 4MByte or 2MByte pages
|
|
|
|
// bit 5: PAE, Physical Address Extension (Pentium Pro+)
|
|
|
|
// 0: 32bit physical addresses
|
|
|
|
// 1: 36bit physical addresses
|
|
|
|
// bit 7: PGE, Page Global Enable (Pentium Pro+)
|
|
|
|
// The global page feature allows frequently used or shared pages
|
|
|
|
// to be marked as global (PDE or PTE bit 8). Global pages are
|
|
|
|
// not flushed from TLB on a task switch or write to CR3.
|
|
|
|
// Values:
|
|
|
|
// 0: disables global page feature
|
|
|
|
// 1: enables global page feature
|
|
|
|
//
|
2007-09-20 21:33:35 +04:00
|
|
|
// page size extention and physical address size extention matrix (legacy mode)
|
|
|
|
// ==============================================================================
|
|
|
|
// CR0.PG CR4.PAE CR4.PSE PDPE.PS PDE.PS | page size physical address size
|
|
|
|
// ==============================================================================
|
|
|
|
// 0 X X R X | -- paging disabled
|
|
|
|
// 1 0 0 R X | 4K 32bits
|
|
|
|
// 1 0 1 R 0 | 4K 32bits
|
|
|
|
// 1 0 1 R 1 | 4M 32bits
|
|
|
|
// 1 1 X R 0 | 4K 36bits
|
|
|
|
// 1 1 X R 1 | 2M 36bits
|
|
|
|
|
|
|
|
// page size extention and physical address size extention matrix (long mode)
|
|
|
|
// ==============================================================================
|
|
|
|
// CR0.PG CR4.PAE CR4.PSE PDPE.PS PDE.PS | page size physical address size
|
|
|
|
// ==============================================================================
|
|
|
|
// 1 1 X 0 0 | 4K 52bits
|
|
|
|
// 1 1 X 0 1 | 2M 52bits
|
|
|
|
// 1 1 X 1 - | 1G 52bits
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
|
|
|
|
// Page Directory/Table Entry format when P=0:
|
|
|
|
// ===========================================
|
|
|
|
//
|
|
|
|
// 31.. 1: available
|
|
|
|
// 0: P=0
|
|
|
|
|
|
|
|
// Page Directory Entry format when P=1 (4-Kbyte Page Table):
|
|
|
|
// ==========================================================
|
|
|
|
//
|
|
|
|
// 31..12: page table base address
|
|
|
|
// 11.. 9: available
|
|
|
|
// 8: G (Pentium Pro+), 0=reserved otherwise
|
|
|
|
// 7: PS (Pentium+), 0=reserved otherwise
|
|
|
|
// 6: 0=reserved
|
|
|
|
// 5: A (386+)
|
|
|
|
// 4: PCD (486+), 0=reserved otherwise
|
|
|
|
// 3: PWT (486+), 0=reserved otherwise
|
|
|
|
// 2: U/S (386+)
|
|
|
|
// 1: R/W (386+)
|
|
|
|
// 0: P=1 (386+)
|
|
|
|
|
|
|
|
// Page Table Entry format when P=1 (4-Kbyte Page):
|
|
|
|
// ================================================
|
|
|
|
//
|
2004-12-14 01:26:36 +03:00
|
|
|
// 63..63: NX |
|
2004-10-21 22:20:40 +04:00
|
|
|
// 62..52: available | Long mode
|
2008-02-03 00:46:54 +03:00
|
|
|
// 51..32: page base address |
|
2001-04-10 05:04:59 +04:00
|
|
|
// 31..12: page base address
|
|
|
|
// 11.. 9: available
|
|
|
|
// 8: G (Pentium Pro+), 0=reserved otherwise
|
2004-10-21 22:20:40 +04:00
|
|
|
// 7: PAT
|
2001-04-10 05:04:59 +04:00
|
|
|
// 6: D (386+)
|
|
|
|
// 5: A (386+)
|
|
|
|
// 4: PCD (486+), 0=reserved otherwise
|
|
|
|
// 3: PWT (486+), 0=reserved otherwise
|
|
|
|
// 2: U/S (386+)
|
|
|
|
// 1: R/W (386+)
|
|
|
|
// 0: P=1 (386+)
|
|
|
|
|
|
|
|
// Page Directory/Table Entry Fields Defined:
|
|
|
|
// ==========================================
|
2004-10-21 22:20:40 +04:00
|
|
|
// NX: No Execute
|
|
|
|
// This bit controls the ability to execute code from all physical
|
|
|
|
// pages mapped by the table entry.
|
|
|
|
// 0: Code can be executed from the mapped physical pages
|
|
|
|
// 1: Code cannot be executed
|
|
|
|
// The NX bit can only be set when the no-execute page-protection
|
2008-02-03 00:46:54 +03:00
|
|
|
// feature is enabled by setting EFER.NXE=1, If EFER.NXE=0, the
|
|
|
|
// NX bit is treated as reserved. In this case, #PF occurs if the
|
2004-10-21 22:20:40 +04:00
|
|
|
// NX bit is not cleared to zero.
|
|
|
|
//
|
2001-04-10 05:04:59 +04:00
|
|
|
// G: Global flag
|
|
|
|
// Indiciates a global page when set. When a page is marked
|
|
|
|
// global and the PGE flag in CR4 is set, the page table or
|
|
|
|
// directory entry for the page is not invalidated in the TLB
|
|
|
|
// when CR3 is loaded or a task switch occurs. Only software
|
|
|
|
// clears and sets this flag. For page directory entries that
|
|
|
|
// point to page tables, this flag is ignored and the global
|
|
|
|
// characteristics of a page are set in the page table entries.
|
|
|
|
//
|
|
|
|
// PS: Page Size flag
|
|
|
|
// Only used in page directory entries. When PS=0, the page
|
|
|
|
// size is 4KBytes and the page directory entry points to a
|
|
|
|
// page table. When PS=1, the page size is 4MBytes for
|
|
|
|
// normal 32-bit addressing and 2MBytes if extended physical
|
2004-10-21 22:20:40 +04:00
|
|
|
// addressing.
|
|
|
|
//
|
|
|
|
// PAT: Page-Attribute Table
|
|
|
|
// This bit is only present in the lowest level of the page
|
2008-02-03 00:46:54 +03:00
|
|
|
// translation hierarchy. The PAT bit is the high-order bit
|
|
|
|
// of a 3-bit index into the PAT register. The other two
|
|
|
|
// bits involved in forming the index are the PCD and PWT
|
2004-10-21 22:20:40 +04:00
|
|
|
// bits.
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
|
|
|
// D: Dirty bit:
|
|
|
|
// Processor sets the Dirty bit in the 2nd-level page table before a
|
|
|
|
// write operation to an address mapped by that page table entry.
|
|
|
|
// Dirty bit in directory entries is undefined.
|
|
|
|
//
|
|
|
|
// A: Accessed bit:
|
|
|
|
// Processor sets the Accessed bits in both levels of page tables before
|
|
|
|
// a read/write operation to a page.
|
|
|
|
//
|
|
|
|
// PCD: Page level Cache Disable
|
|
|
|
// Controls caching of individual pages or page tables.
|
|
|
|
// This allows a per-page based mechanism to disable caching, for
|
|
|
|
// those pages which contained memory mapped IO, or otherwise
|
|
|
|
// should not be cached. Processor ignores this flag if paging
|
|
|
|
// is not used (CR0.PG=0) or the cache disable bit is set (CR0.CD=1).
|
|
|
|
// Values:
|
|
|
|
// 0: page or page table can be cached
|
|
|
|
// 1: page or page table is not cached (prevented)
|
|
|
|
//
|
|
|
|
// PWT: Page level Write Through
|
|
|
|
// Controls the write-through or write-back caching policy of individual
|
|
|
|
// pages or page tables. Processor ignores this flag if paging
|
|
|
|
// is not used (CR0.PG=0) or the cache disable bit is set (CR0.CD=1).
|
|
|
|
// Values:
|
|
|
|
// 0: write-back caching
|
|
|
|
// 1: write-through caching
|
|
|
|
//
|
|
|
|
// U/S: User/Supervisor level
|
|
|
|
// 0: Supervisor level - for the OS, drivers, etc.
|
|
|
|
// 1: User level - application code and data
|
|
|
|
//
|
|
|
|
// R/W: Read/Write access
|
|
|
|
// 0: read-only access
|
|
|
|
// 1: read/write access
|
|
|
|
//
|
|
|
|
// P: Present
|
|
|
|
// 0: Not present
|
|
|
|
// 1: Present
|
|
|
|
// ==========================================
|
|
|
|
|
|
|
|
|
|
|
|
// Combined page directory/page table protection:
|
|
|
|
// ==============================================
|
|
|
|
// There is one column for the combined effect on a 386
|
|
|
|
// and one column for the combined effect on a 486+ CPU.
|
|
|
|
//
|
|
|
|
// +----------------+-----------------+----------------+----------------+
|
|
|
|
// | Page Directory| Page Table | Combined 386 | Combined 486+ |
|
|
|
|
// |Privilege Type | Privilege Type | Privilege Type| Privilege Type|
|
|
|
|
// |----------------+-----------------+----------------+----------------|
|
|
|
|
// |User R | User R | User R | User R |
|
|
|
|
// |User R | User RW | User R | User R |
|
|
|
|
// |User RW | User R | User R | User R |
|
|
|
|
// |User RW | User RW | User RW | User RW |
|
|
|
|
// |User R | Supervisor R | User R | Supervisor RW |
|
|
|
|
// |User R | Supervisor RW | User R | Supervisor RW |
|
|
|
|
// |User RW | Supervisor R | User R | Supervisor RW |
|
|
|
|
// |User RW | Supervisor RW | User RW | Supervisor RW |
|
|
|
|
// |Supervisor R | User R | User R | Supervisor RW |
|
|
|
|
// |Supervisor R | User RW | User R | Supervisor RW |
|
|
|
|
// |Supervisor RW | User R | User R | Supervisor RW |
|
|
|
|
// |Supervisor RW | User RW | User RW | Supervisor RW |
|
|
|
|
// |Supervisor R | Supervisor R | Supervisor RW | Supervisor RW |
|
|
|
|
// |Supervisor R | Supervisor RW | Supervisor RW | Supervisor RW |
|
|
|
|
// |Supervisor RW | Supervisor R | Supervisor RW | Supervisor RW |
|
|
|
|
// |Supervisor RW | Supervisor RW | Supervisor RW | Supervisor RW |
|
|
|
|
// +----------------+-----------------+----------------+----------------+
|
|
|
|
|
|
|
|
// Page Fault Error Code Format:
|
|
|
|
// =============================
|
|
|
|
//
|
|
|
|
// bits 31..4: Reserved
|
|
|
|
// bit 3: RSVD (Pentium Pro+)
|
|
|
|
// 0: fault caused by reserved bits set to 1 in a page directory
|
|
|
|
// when the PSE or PAE flags in CR4 are set to 1
|
|
|
|
// 1: fault was not caused by reserved bit violation
|
|
|
|
// bit 2: U/S (386+)
|
|
|
|
// 0: fault originated when in supervior mode
|
|
|
|
// 1: fault originated when in user mode
|
|
|
|
// bit 1: R/W (386+)
|
|
|
|
// 0: access causing the fault was a read
|
|
|
|
// 1: access causing the fault was a write
|
|
|
|
// bit 0: P (386+)
|
|
|
|
// 0: fault caused by a nonpresent page
|
|
|
|
// 1: fault caused by a page level protection violation
|
|
|
|
|
|
|
|
// Some paging related notes:
|
|
|
|
// ==========================
|
|
|
|
//
|
|
|
|
// - When the processor is running in supervisor level, all pages are both
|
|
|
|
// readable and writable (write-protect ignored). When running at user
|
|
|
|
// level, only pages which belong to the user level are accessible;
|
|
|
|
// read/write & read-only are readable, read/write are writable.
|
|
|
|
//
|
|
|
|
// - If the Present bit is 0 in either level of page table, an
|
|
|
|
// access which uses these entries will generate a page fault.
|
|
|
|
//
|
|
|
|
// - (A)ccess bit is used to report read or write access to a page
|
|
|
|
// or 2nd level page table.
|
|
|
|
//
|
|
|
|
// - (D)irty bit is used to report write access to a page.
|
|
|
|
//
|
|
|
|
// - Processor running at CPL=0,1,2 maps to U/S=0
|
|
|
|
// Processor running at CPL=3 maps to U/S=1
|
|
|
|
//
|
|
|
|
// - Pentium+ processors have separate TLB's for data and instruction caches
|
|
|
|
// - Pentium Pro+ processors maintain separate 4K and 4M TLBs.
|
2007-08-30 20:48:10 +04:00
|
|
|
|
2001-04-10 05:04:59 +04:00
|
|
|
#define BX_INVALID_TLB_ENTRY 0xffffffff
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2001-04-10 05:04:59 +04:00
|
|
|
#if BX_CPU_LEVEL >= 4
|
|
|
|
# define BX_PRIV_CHECK_SIZE 32
|
|
|
|
#else
|
|
|
|
# define BX_PRIV_CHECK_SIZE 16
|
|
|
|
#endif
|
|
|
|
|
2006-09-20 21:02:20 +04:00
|
|
|
static unsigned priv_check[BX_PRIV_CHECK_SIZE];
|
|
|
|
|
2001-04-10 05:04:59 +04:00
|
|
|
// The 'priv_check' array is used to decide if the current access
|
|
|
|
// has the proper paging permissions. An index is formed, based
|
|
|
|
// on parameters such as the access type and level, the write protect
|
|
|
|
// flag and values cached in the TLB. The format of the index into this
|
|
|
|
// array is:
|
|
|
|
//
|
|
|
|
// |4 |3 |2 |1 |0 |
|
|
|
|
// |wp|us|us|rw|rw|
|
|
|
|
// | | | | |
|
|
|
|
// | | | | +---> r/w of current access
|
|
|
|
// | | +--+------> u/s,r/w combined of page dir & table (cached)
|
|
|
|
// | +------------> u/s of current access
|
2007-07-09 19:16:14 +04:00
|
|
|
// +---------------> Current CR0.WP value
|
2002-09-05 07:09:59 +04:00
|
|
|
|
|
|
|
|
|
|
|
// Each entry in the TLB cache has 3 entries:
|
2007-12-17 00:03:46 +03:00
|
|
|
//
|
2002-09-05 07:09:59 +04:00
|
|
|
// lpf: Linear Page Frame (page aligned linear address of page)
|
2007-12-17 00:03:46 +03:00
|
|
|
// bits 32..12 Linear page frame.
|
|
|
|
// bits 11...0 Invalidate index.
|
|
|
|
//
|
2002-09-05 07:09:59 +04:00
|
|
|
// ppf: Physical Page Frame (page aligned phy address of page)
|
2005-06-15 00:55:57 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// hostPageAddr:
|
|
|
|
// Host Page Frame address used for direct access to
|
|
|
|
// the mem.vector[] space allocated for the guest physical
|
|
|
|
// memory. If this is zero, it means that a pointer
|
|
|
|
// to the host space could not be generated, likely because
|
|
|
|
// that page of memory is not standard memory (it might
|
|
|
|
// be memory mapped IO, ROM, etc).
|
2005-06-15 00:55:57 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// accessBits:
|
|
|
|
//
|
|
|
|
// bit 31: Page is a global page.
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// The following bits are used for a very efficient permissions
|
2002-09-05 07:09:59 +04:00
|
|
|
// check. The goal is to be able, using only the current privilege
|
|
|
|
// level and access type, to determine if the page tables allow the
|
|
|
|
// access to occur or at least should rewalk the page tables. On
|
|
|
|
// the first read access, permissions are set to only read, so a
|
|
|
|
// rewalk is necessary when a subsequent write fails the tests.
|
|
|
|
// This allows for the dirty bit to be set properly, but for the
|
|
|
|
// test to be efficient. Note that the CR0.WP flag is not present.
|
|
|
|
// The values in the following flags is based on the current CR0.WP
|
|
|
|
// value, necessitating a TLB flush when CR0.WP changes.
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
2002-09-05 07:09:59 +04:00
|
|
|
// The test is:
|
2008-02-15 22:03:54 +03:00
|
|
|
// OK = 0x01 << ( (W<<2) | CPL ) [W:1=write, 0=read]
|
2008-02-03 00:46:54 +03:00
|
|
|
//
|
2005-06-15 00:55:57 +04:00
|
|
|
// Thus for reads, it is:
|
2008-02-15 22:03:54 +03:00
|
|
|
// OK = 0x01 << ( CPL )
|
2005-06-15 00:55:57 +04:00
|
|
|
// And for writes:
|
2008-02-15 22:03:54 +03:00
|
|
|
// OK = 0x10 << ( CPL )
|
2007-12-17 00:03:46 +03:00
|
|
|
//
|
|
|
|
// bit 15: a Write from CPL=3 is OK
|
|
|
|
// bit 14: a Write from CPL=2 is OK
|
|
|
|
// bit 13: a Write from CPL=1 is OK
|
|
|
|
// bit 12: a Write from CPL=0 is OK
|
2005-06-15 00:55:57 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// bit 11: a Read from CPL=3 is OK
|
|
|
|
// bit 10: a Read from CPL=2 is OK
|
|
|
|
// bit 9: a Read from CPL=1 is OK
|
|
|
|
// bit 8: a Read from CPL=0 is OK
|
2005-06-15 00:55:57 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// And the lowest bits are as above, except that they also indicate
|
2008-02-03 00:46:54 +03:00
|
|
|
// that hostPageAddr is valid, so we do not separately need to test
|
2005-06-15 00:55:57 +04:00
|
|
|
// that pointer against NULL. These have smaller constants for us
|
|
|
|
// to be able to use smaller encodings in the trace generators. Note
|
2007-12-17 00:03:46 +03:00
|
|
|
// that whenever bit n (n=0..7) is set, then also n+8 is set.
|
2005-06-15 00:55:57 +04:00
|
|
|
// (The opposite is of course not true)
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
2007-12-17 00:03:46 +03:00
|
|
|
// bit 7: a Write from CPL=3 is OK, hostPageAddr is valid
|
|
|
|
// bit 6: a Write from CPL=2 is OK, hostPageAddr is valid
|
|
|
|
// bit 5: a Write from CPL=1 is OK, hostPageAddr is valid
|
|
|
|
// bit 4: a Write from CPL=0 is OK, hostPageAddr is valid
|
|
|
|
//
|
|
|
|
// bit 3: a Read from CPL=3 is OK, hostPageAddr is valid
|
|
|
|
// bit 2: a Read from CPL=2 is OK, hostPageAddr is valid
|
|
|
|
// bit 1: a Read from CPL=1 is OK, hostPageAddr is valid
|
|
|
|
// bit 0: a Read from CPL=0 is OK, hostPageAddr is valid
|
2005-06-15 00:55:57 +04:00
|
|
|
//
|
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
#define TLB_WriteUserOK 0x8000
|
|
|
|
#define TLB_WriteSysOK 0x7000
|
|
|
|
#define TLB_ReadUserOK 0x0800
|
|
|
|
#define TLB_ReadSysOK 0x0700
|
|
|
|
#define TLB_WriteUserPtrOK 0x0080
|
|
|
|
#define TLB_WriteSysPtrOK 0x0070
|
|
|
|
#define TLB_ReadUserPtrOK 0x0008
|
|
|
|
#define TLB_ReadSysPtrOK 0x0007
|
2002-09-05 07:09:59 +04:00
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
#define TLB_GlobalPage 0x80000000
|
2004-12-17 01:21:35 +03:00
|
|
|
|
2002-09-17 00:23:38 +04:00
|
|
|
// === TLB Instrumentation section ==============================
|
|
|
|
|
|
|
|
// Note: this is an approximation of what Peter Tattam had.
|
|
|
|
|
|
|
|
#define InstrumentTLB 0
|
|
|
|
|
|
|
|
#if InstrumentTLB
|
|
|
|
static unsigned tlbLookups=0;
|
|
|
|
static unsigned tlbMisses=0;
|
|
|
|
static unsigned tlbGlobalFlushes=0;
|
|
|
|
static unsigned tlbNonGlobalFlushes=0;
|
|
|
|
|
|
|
|
#define InstrTLB_StatsMask 0xfffff
|
|
|
|
|
|
|
|
#define InstrTLB_Stats() {\
|
|
|
|
if ((tlbLookups & InstrTLB_StatsMask) == 0) { \
|
|
|
|
BX_INFO(("TLB lookup:%8d miss:%8d %6.2f%% flush:%8d %6.2f%%", \
|
|
|
|
tlbLookups, \
|
|
|
|
tlbMisses, \
|
|
|
|
tlbMisses * 100.0 / tlbLookups, \
|
|
|
|
(tlbGlobalFlushes+tlbNonGlobalFlushes), \
|
|
|
|
(tlbGlobalFlushes+tlbNonGlobalFlushes) * 100.0 / tlbLookups \
|
|
|
|
)); \
|
|
|
|
tlbLookups = tlbMisses = tlbGlobalFlushes = tlbNonGlobalFlushes = 0; \
|
|
|
|
} \
|
|
|
|
}
|
|
|
|
#define InstrTLB_Increment(v) (v)++
|
|
|
|
|
|
|
|
#else
|
|
|
|
#define InstrTLB_Stats()
|
|
|
|
#define InstrTLB_Increment(v)
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// ==============================================================
|
|
|
|
|
|
|
|
|
2003-03-03 02:59:12 +03:00
|
|
|
void BX_CPP_AttrRegparmN(2)
|
2002-09-07 09:21:28 +04:00
|
|
|
BX_CPU_C::pagingCR0Changed(Bit32u oldCR0, Bit32u newCR0)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2002-09-07 09:21:28 +04:00
|
|
|
// Modification of PG,PE flushes TLB cache according to docs.
|
|
|
|
// Additionally, the TLB strategy is based on the current value of
|
|
|
|
// WP, so if that changes we must also flush the TLB.
|
2008-02-15 22:03:54 +03:00
|
|
|
if ((oldCR0 & 0x80010001) != (newCR0 & 0x80010001))
|
2002-09-07 09:21:28 +04:00
|
|
|
TLB_flush(1); // 1 = Flush Global entries also.
|
|
|
|
|
|
|
|
if (bx_dbg.paging)
|
2006-02-28 20:47:33 +03:00
|
|
|
BX_INFO(("pagingCR0Changed: (0x%x -> 0x%x)", oldCR0, newCR0));
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2003-03-03 02:59:12 +03:00
|
|
|
void BX_CPP_AttrRegparmN(2)
|
2002-09-07 09:21:28 +04:00
|
|
|
BX_CPU_C::pagingCR4Changed(Bit32u oldCR4, Bit32u newCR4)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2002-09-07 09:21:28 +04:00
|
|
|
// Modification of PGE,PAE,PSE flushes TLB cache according to docs.
|
2006-04-05 21:31:35 +04:00
|
|
|
if ((oldCR4 & 0x000000b0) != (newCR4 & 0x000000b0))
|
2002-09-07 09:21:28 +04:00
|
|
|
TLB_flush(1); // 1 = Flush Global entries also.
|
|
|
|
|
|
|
|
if (bx_dbg.paging)
|
2006-02-28 20:47:33 +03:00
|
|
|
BX_INFO(("pagingCR4Changed: (0x%x -> 0x%x)", oldCR4, newCR4));
|
2005-04-14 20:44:40 +04:00
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_PAE
|
2006-04-05 21:31:35 +04:00
|
|
|
if ((oldCR4 & 0x00000020) != (newCR4 & 0x00000020)) {
|
2006-10-04 23:08:40 +04:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PAE() && !long_mode())
|
2005-04-14 20:44:40 +04:00
|
|
|
BX_CPU_THIS_PTR cr3_masked = BX_CPU_THIS_PTR cr3 & 0xffffffe0;
|
|
|
|
else
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_CPU_THIS_PTR cr3_masked = BX_CPU_THIS_PTR cr3 & BX_CONST64(0x000ffffffffff000);
|
2005-04-14 20:44:40 +04:00
|
|
|
}
|
|
|
|
#endif
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2003-03-03 02:59:12 +03:00
|
|
|
void BX_CPP_AttrRegparmN(1)
|
2006-04-05 21:31:35 +04:00
|
|
|
BX_CPU_C::CR3_change(bx_phy_address value)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
|
|
|
if (bx_dbg.paging) {
|
2001-05-30 22:56:02 +04:00
|
|
|
BX_INFO(("CR3_change(): flush TLB cache"));
|
2002-09-17 00:23:38 +04:00
|
|
|
BX_INFO(("Page Directory Base %08x", (unsigned) value));
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
// flush TLB even if value does not change
|
2002-09-07 09:21:28 +04:00
|
|
|
TLB_flush(0); // 0 = Don't flush Global entries.
|
2002-09-17 00:23:38 +04:00
|
|
|
BX_CPU_THIS_PTR cr3 = value;
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_PAE
|
2006-10-04 23:08:40 +04:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PAE() && !long_mode())
|
2004-06-18 18:11:11 +04:00
|
|
|
BX_CPU_THIS_PTR cr3_masked = value & 0xffffffe0;
|
|
|
|
else
|
|
|
|
#endif
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_CPU_THIS_PTR cr3_masked = value & BX_CONST64(0x000ffffffffff000);
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2006-09-20 21:02:20 +04:00
|
|
|
// Called to initialize the TLB upon startup.
|
|
|
|
// Unconditional initialization of all TLB entries.
|
2006-04-29 21:21:49 +04:00
|
|
|
void BX_CPU_C::TLB_init(void)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2006-09-20 21:02:20 +04:00
|
|
|
unsigned i, wp, us_combined, rw_combined, us_current, rw_current;
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2004-10-21 22:20:40 +04:00
|
|
|
for (i=0; i<BX_TLB_SIZE; i++)
|
2001-04-10 05:04:59 +04:00
|
|
|
BX_CPU_THIS_PTR TLB.entry[i].lpf = BX_INVALID_TLB_ENTRY;
|
2006-09-20 21:02:20 +04:00
|
|
|
|
2001-04-10 05:04:59 +04:00
|
|
|
//
|
|
|
|
// Setup privilege check matrix.
|
|
|
|
//
|
|
|
|
for (i=0; i<BX_PRIV_CHECK_SIZE; i++) {
|
|
|
|
wp = (i & 0x10) >> 4;
|
|
|
|
us_current = (i & 0x08) >> 3;
|
|
|
|
us_combined = (i & 0x04) >> 2;
|
|
|
|
rw_combined = (i & 0x02) >> 1;
|
|
|
|
rw_current = (i & 0x01) >> 0;
|
|
|
|
if (wp) { // when write protect on
|
|
|
|
if (us_current > us_combined) // user access, supervisor page
|
|
|
|
priv_check[i] = 0;
|
|
|
|
else if (rw_current > rw_combined) // RW access, RO page
|
|
|
|
priv_check[i] = 0;
|
|
|
|
else
|
|
|
|
priv_check[i] = 1;
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
else { // when write protect off
|
|
|
|
if (us_current == 0) // Supervisor mode access, anything goes
|
|
|
|
priv_check[i] = 1;
|
|
|
|
else {
|
|
|
|
// user mode access
|
|
|
|
if (us_combined == 0) // user access, supervisor Page
|
|
|
|
priv_check[i] = 0;
|
|
|
|
else if (rw_current > rw_combined) // RW access, RO page
|
|
|
|
priv_check[i] = 0;
|
|
|
|
else
|
|
|
|
priv_check[i] = 1;
|
|
|
|
}
|
|
|
|
}
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2006-03-02 01:32:24 +03:00
|
|
|
void BX_CPU_C::TLB_flush(bx_bool invalidateGlobal)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2002-09-17 00:23:38 +04:00
|
|
|
#if InstrumentTLB
|
|
|
|
if (invalidateGlobal)
|
|
|
|
InstrTLB_Increment(tlbGlobalFlushes);
|
|
|
|
else
|
|
|
|
InstrTLB_Increment(tlbNonGlobalFlushes);
|
2002-09-06 23:21:55 +04:00
|
|
|
#endif
|
2002-09-06 18:58:56 +04:00
|
|
|
|
2002-09-17 00:23:38 +04:00
|
|
|
for (unsigned i=0; i<BX_TLB_SIZE; i++) {
|
|
|
|
// To be conscious of the native cache line usage, only
|
|
|
|
// write to (invalidate) entries which need it.
|
2005-06-15 00:55:57 +04:00
|
|
|
bx_TLB_entry *tlbEntry = &BX_CPU_THIS_PTR TLB.entry[i];
|
|
|
|
if (tlbEntry->lpf != BX_INVALID_TLB_ENTRY) {
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2006-02-28 20:47:33 +03:00
|
|
|
if (invalidateGlobal || !(tlbEntry->accessBits & TLB_GlobalPage))
|
2002-09-10 07:52:32 +04:00
|
|
|
#endif
|
2005-06-15 00:55:57 +04:00
|
|
|
tlbEntry->lpf = BX_INVALID_TLB_ENTRY;
|
2002-09-17 00:23:38 +04:00
|
|
|
}
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2006-03-02 01:32:24 +03:00
|
|
|
void BX_CPU_C::TLB_invlpg(bx_address laddr)
|
|
|
|
{
|
2007-12-27 02:07:44 +03:00
|
|
|
unsigned TLB_index = BX_TLB_INDEX_OF(laddr, 0);
|
2006-03-02 01:32:24 +03:00
|
|
|
BX_CPU_THIS_PTR TLB.entry[TLB_index].lpf = BX_INVALID_TLB_ENTRY;
|
|
|
|
}
|
|
|
|
|
2008-03-23 00:29:41 +03:00
|
|
|
void BX_CPP_AttrRegparmN(1) BX_CPU_C::INVLPG(bxInstruction_c* i)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
|
|
|
#if BX_CPU_LEVEL >= 4
|
|
|
|
invalidate_prefetch_q();
|
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!real_mode() && CPL!=0) {
|
|
|
|
BX_ERROR(("INVLPG: priveledge check failed, generate #GP(0)"));
|
2001-04-10 05:04:59 +04:00
|
|
|
exception(BX_GP_EXCEPTION, 0, 0);
|
2006-06-10 02:29:07 +04:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2008-01-10 22:37:56 +03:00
|
|
|
BX_CPU_CALL_METHODR(i->ResolveModrm, (i));
|
2008-04-07 22:39:17 +04:00
|
|
|
bx_address laddr = BX_CPU_THIS_PTR get_laddr(i->seg(), RMAddr(i));
|
2008-01-17 01:39:55 +03:00
|
|
|
BX_INSTR_TLB_CNTRL(BX_CPU_ID, BX_INSTR_INVLPG, laddr);
|
2006-03-02 01:32:24 +03:00
|
|
|
TLB_invlpg(laddr);
|
2001-04-10 05:04:59 +04:00
|
|
|
#else
|
|
|
|
// not supported on < 486
|
2005-08-05 16:47:33 +04:00
|
|
|
BX_INFO(("INVLPG: required i486, use --enable-cpu=4 option"));
|
2001-04-10 05:04:59 +04:00
|
|
|
UndefinedOpcode(i);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
2007-08-30 20:48:10 +04:00
|
|
|
// error checking order - page not present, reserved bits, protection
|
|
|
|
#define ERROR_NOT_PRESENT 0x00
|
|
|
|
#define ERROR_PROTECTION 0x01
|
|
|
|
#define ERROR_RESERVED 0x08
|
|
|
|
#define ERROR_CODE_ACCESS 0x10
|
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
void BX_CPU_C::page_fault(unsigned fault, bx_address laddr, unsigned user, unsigned rw, unsigned access_type)
|
2007-08-30 20:48:10 +04:00
|
|
|
{
|
|
|
|
unsigned error_code = fault;
|
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
error_code |= (user << 2) | (rw << 1);
|
2007-08-30 20:48:10 +04:00
|
|
|
#if BX_SUPPORT_X86_64
|
2008-04-01 00:56:27 +04:00
|
|
|
if (BX_CPU_THIS_PTR efer.get_NXE() && (access_type == CODE_ACCESS))
|
2007-08-30 20:48:10 +04:00
|
|
|
error_code |= ERROR_CODE_ACCESS; // I/D = 1
|
|
|
|
#endif
|
|
|
|
BX_CPU_THIS_PTR cr2 = laddr;
|
|
|
|
|
|
|
|
TLB_invlpg(laddr); // Invalidate TLB entry
|
|
|
|
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
BX_DEBUG(("page fault for address %08x%08x @ %08x%08x",
|
|
|
|
GET32H(laddr), GET32L(laddr), GET32H(RIP), GET32L(RIP)));
|
|
|
|
#else
|
|
|
|
BX_DEBUG(("page fault for address %08x @ %08x", laddr, EIP));
|
|
|
|
#endif
|
|
|
|
|
|
|
|
exception(BX_PF_EXCEPTION, error_code, 0);
|
|
|
|
}
|
|
|
|
|
2008-04-20 00:00:28 +04:00
|
|
|
#define PAGING_PML4_RESERVED_BITS BX_CONST64(0x00000080) /* bits 7 */
|
|
|
|
#define PAGING_PDPE_RESERVED_BITS BX_CONST64(0x00000080) /* bits 7 - we not support 1G paging */
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
#define PAGE_DIRECTORY_NX_BIT (BX_CONST64(0x8000000000000000))
|
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Translate a linear address to a physical address
|
2007-12-17 00:03:46 +03:00
|
|
|
bx_phy_address BX_CPU_C::translate_linear(bx_address laddr, unsigned curr_pl, unsigned rw, unsigned access_type)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2007-08-30 20:48:10 +04:00
|
|
|
Bit32u accessBits, combined_access = 0;
|
2002-09-04 12:59:13 +04:00
|
|
|
unsigned priv_index;
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2004-12-14 01:26:36 +03:00
|
|
|
// note - we assume physical memory < 4gig so for brevity & speed, we'll use
|
|
|
|
// 32 bit entries although cr3 is expanded to 64 bits.
|
2008-02-11 23:52:10 +03:00
|
|
|
bx_phy_address paddress, ppf, poffset = PAGE_OFFSET(laddr);
|
2004-12-17 01:21:35 +03:00
|
|
|
bx_bool isWrite = (rw >= BX_WRITE); // write or r-m-w
|
2007-12-17 00:03:46 +03:00
|
|
|
unsigned pl = (curr_pl == 3);
|
|
|
|
|
2007-11-20 20:15:33 +03:00
|
|
|
InstrTLB_Increment(tlbLookups);
|
|
|
|
InstrTLB_Stats();
|
|
|
|
|
2007-10-31 01:15:42 +03:00
|
|
|
bx_address lpf = LPFOf(laddr);
|
2007-12-27 02:07:44 +03:00
|
|
|
unsigned TLB_index = BX_TLB_INDEX_OF(lpf, 0);
|
2006-10-04 23:47:24 +04:00
|
|
|
bx_TLB_entry *tlbEntry = &BX_CPU_THIS_PTR TLB.entry[TLB_index];
|
2004-10-21 22:20:40 +04:00
|
|
|
|
2008-04-06 00:41:00 +04:00
|
|
|
// already looked up TLB for code access
|
|
|
|
if (access_type != CODE_ACCESS && tlbEntry->lpf == lpf)
|
2006-10-04 23:47:24 +04:00
|
|
|
{
|
|
|
|
paddress = tlbEntry->ppf | poffset;
|
|
|
|
accessBits = tlbEntry->accessBits;
|
2004-10-21 22:20:40 +04:00
|
|
|
|
2007-12-17 00:03:46 +03:00
|
|
|
if (accessBits & (0x0100 << ((isWrite<<2) | curr_pl)))
|
2006-10-04 23:47:24 +04:00
|
|
|
return(paddress);
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2006-10-04 23:47:24 +04:00
|
|
|
// The current access does not have permission according to the info
|
|
|
|
// in our TLB cache entry. Re-walk the page tables, in case there is
|
|
|
|
// updated information in the memory image, and let the long path code
|
|
|
|
// generate an exception if one is warranted.
|
|
|
|
}
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2008-04-20 00:00:28 +04:00
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
BX_DEBUG(("page walk for address 0x%08x:%08x", GET32H(laddr), GET32L(laddr)));
|
|
|
|
#else
|
|
|
|
BX_DEBUG(("page walk for address 0x%08x", laddr));
|
|
|
|
#endif
|
|
|
|
|
2006-10-04 23:47:24 +04:00
|
|
|
InstrTLB_Increment(tlbMisses);
|
|
|
|
|
|
|
|
#if BX_SUPPORT_PAE
|
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PAE())
|
|
|
|
{
|
2007-09-20 21:33:35 +04:00
|
|
|
bx_phy_address pdpe_addr;
|
|
|
|
Bit64u pdpe, pde, pte;
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
Bit64u pml4;
|
|
|
|
#endif
|
|
|
|
unsigned nx_fault = 0;
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2002-09-17 01:55:57 +04:00
|
|
|
#if BX_SUPPORT_X86_64
|
2008-04-20 00:00:28 +04:00
|
|
|
if (long_mode()) {
|
2002-09-17 00:23:38 +04:00
|
|
|
// Get PML4 entry
|
2007-12-23 20:21:28 +03:00
|
|
|
bx_phy_address pml4_addr = (bx_phy_address)(BX_CPU_THIS_PTR cr3_masked |
|
|
|
|
((laddr & BX_CONST64(0x0000ff8000000000)) >> 36));
|
2004-12-14 01:26:36 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pml4_addr, 8, &pml4);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pml4_addr, 8, BX_READ, (Bit8u*)(&pml4));
|
2005-01-20 22:37:43 +03:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pml4 & 0x01)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PML4: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
|
|
|
}
|
|
|
|
if (pml4 & PAGING_PML4_RESERVED_BITS) {
|
2008-04-20 00:00:28 +04:00
|
|
|
BX_DEBUG(("PML4: reserved bit is set PML4=%08x:%08x", GET32H(pml4), GET32L(pml4)));
|
2007-09-20 21:33:35 +04:00
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
if (pml4 & PAGE_DIRECTORY_NX_BIT) {
|
2008-04-01 00:56:27 +04:00
|
|
|
if (! BX_CPU_THIS_PTR efer.get_NXE()) {
|
2006-10-28 16:31:23 +04:00
|
|
|
BX_DEBUG(("PML4: NX bit set when EFER.NXE is disabled"));
|
2007-08-30 20:48:10 +04:00
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
|
|
|
if (access_type == CODE_ACCESS) {
|
|
|
|
BX_DEBUG(("PML4: non-executable page fault occured"));
|
2007-09-20 21:33:35 +04:00
|
|
|
nx_fault = 1;
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
2007-09-20 21:33:35 +04:00
|
|
|
if (pml4 & BX_CONST64(0x000fffff00000000)) {
|
2008-04-11 18:30:15 +04:00
|
|
|
BX_PANIC(("PML4 0x%08x%08x: Only 32 bit physical address space is emulated !", GET32H(pml4), GET32L(pml4)));
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
if (!(pml4 & 0x20)) {
|
|
|
|
pml4 |= 0x20;
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pml4_addr, 8, &pml4);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pml4_addr, 8, BX_WRITE, (Bit8u*)(&pml4));
|
2006-10-04 23:08:40 +04:00
|
|
|
}
|
|
|
|
|
2007-12-23 20:21:28 +03:00
|
|
|
pdpe_addr = (bx_phy_address)((pml4 & BX_CONST64(0x000ffffffffff000)) |
|
|
|
|
((laddr & BX_CONST64(0x0000007fc0000000)) >> 27));
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2002-09-17 01:55:57 +04:00
|
|
|
else
|
|
|
|
#endif
|
2004-10-30 01:15:48 +04:00
|
|
|
{
|
2007-12-23 20:21:28 +03:00
|
|
|
pdpe_addr = (bx_phy_address) (BX_CPU_THIS_PTR cr3_masked | ((laddr & 0xc0000000) >> 27));
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pdpe_addr, 8, &pdpe);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pdpe_addr, 8, BX_READ, (Bit8u*)(&pdpe));
|
2005-01-20 22:37:43 +03:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
if (!(pdpe & 0x01)) {
|
|
|
|
BX_DEBUG(("PAE PDPE: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
|
|
|
}
|
|
|
|
if (pdpe & PAGING_PDPE_RESERVED_BITS) {
|
2008-04-20 00:00:28 +04:00
|
|
|
BX_DEBUG(("PAE PDPE: reserved bit is set: PDPE=%08x:%08x", GET32H(pdpe), GET32L(pdpe)));
|
2007-09-20 21:33:35 +04:00
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
#if BX_SUPPORT_X86_64
|
2007-09-20 21:33:35 +04:00
|
|
|
if (pdpe & PAGE_DIRECTORY_NX_BIT) {
|
2008-04-01 00:56:27 +04:00
|
|
|
if (! BX_CPU_THIS_PTR efer.get_NXE()) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PDPE: NX bit set when EFER.NXE is disabled"));
|
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
|
|
|
}
|
|
|
|
if (access_type == CODE_ACCESS) {
|
|
|
|
BX_DEBUG(("PDPE: non-executable page fault occured"));
|
|
|
|
nx_fault = 1;
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
2007-09-20 21:33:35 +04:00
|
|
|
if (pdpe & BX_CONST64(0x000fffff00000000)) {
|
2008-04-11 18:30:15 +04:00
|
|
|
BX_PANIC(("PAE PDPE 0x%08x%08x: Only 32 bit physical address space is emulated !", GET32H(pdpe), GET32L(pdpe)));
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-12-23 20:21:28 +03:00
|
|
|
bx_phy_address pde_addr = (bx_phy_address)((pdpe & BX_CONST64(0x000ffffffffff000))
|
|
|
|
| ((laddr & 0x3fe00000) >> 18));
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pde_addr, 8, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 8, BX_READ, (Bit8u*)(&pde));
|
2005-01-20 22:37:43 +03:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pde & 0x01)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PAE PDE: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
if (pde & PAGE_DIRECTORY_NX_BIT) {
|
2008-04-01 00:56:27 +04:00
|
|
|
if (! BX_CPU_THIS_PTR efer.get_NXE()) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PDE: NX bit set when EFER.NXE is disabled"));
|
2007-08-30 20:48:10 +04:00
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
|
|
|
if (access_type == CODE_ACCESS) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PDE: non-executable page fault occured"));
|
|
|
|
nx_fault = 1;
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
|
|
|
#endif
|
2007-09-20 21:33:35 +04:00
|
|
|
if (pde & BX_CONST64(0x000fffff00000000)) {
|
2008-04-11 18:30:15 +04:00
|
|
|
BX_PANIC(("PAE PDE 0x%08x%08x: Only 32 bit physical address space is emulated !", GET32H(pde), GET32L(pde)));
|
2006-10-04 23:08:40 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
// Ignore CR4.PSE in PAE mode
|
2002-09-17 00:23:38 +04:00
|
|
|
if (pde & 0x80) {
|
|
|
|
// Combined access is just access from the pde (no pte involved).
|
2007-09-20 21:33:35 +04:00
|
|
|
combined_access = (pde) & 0x06; // U/S and R/W
|
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
if (long_mode()) {
|
|
|
|
combined_access &= (pml4 & pdpe) & 0x06;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2002-09-17 00:23:38 +04:00
|
|
|
// Make up the physical page frame address.
|
2007-12-23 20:21:28 +03:00
|
|
|
ppf = (bx_phy_address)((pde & BX_CONST64(0x000fffffffe00000)) | (laddr & 0x001ff000));
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2007-12-06 21:35:33 +03:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PGE())
|
|
|
|
combined_access |= (pde & TLB_GlobalPage); // G
|
2002-09-17 00:23:38 +04:00
|
|
|
#endif
|
|
|
|
|
|
|
|
priv_index =
|
|
|
|
#if BX_CPU_LEVEL >= 4
|
2007-07-09 19:16:14 +04:00
|
|
|
(BX_CPU_THIS_PTR cr0.get_WP() << 4) | // bit 4
|
2002-09-17 00:23:38 +04:00
|
|
|
#endif
|
2007-07-09 19:16:14 +04:00
|
|
|
(pl<<3) | // bit 3
|
|
|
|
(combined_access & 0x06) | // bit 2,1
|
2007-09-20 21:33:35 +04:00
|
|
|
(isWrite); // bit 0
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
if (!priv_check[priv_index] || nx_fault)
|
2007-08-30 20:48:10 +04:00
|
|
|
page_fault(ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Update PDPE A bit if needed.
|
|
|
|
if (!(pdpe & 0x20)) {
|
|
|
|
pdpe |= 0x20;
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pdpe_addr, 8, &pdpe);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pdpe_addr, 8, BX_WRITE, (Bit8u*)(&pdpe));
|
2007-09-20 21:33:35 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
// Update PDE A/D bits if needed.
|
|
|
|
if (((pde & 0x20)==0) || (isWrite && ((pde & 0x40)==0))) {
|
2002-09-17 00:23:38 +04:00
|
|
|
pde |= (0x20 | (isWrite<<6)); // Update A and possibly D bits
|
2006-10-04 23:08:40 +04:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pde_addr, 8, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 8, BX_WRITE, (Bit8u*)(&pde));
|
2002-09-17 00:23:38 +04:00
|
|
|
}
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2007-09-20 21:33:35 +04:00
|
|
|
else {
|
|
|
|
// 4k pages, Get page table entry
|
2008-02-03 00:46:54 +03:00
|
|
|
bx_phy_address pte_addr = (bx_phy_address)((pde & BX_CONST64(0x000ffffffffff000)) |
|
2007-12-23 20:21:28 +03:00
|
|
|
((laddr & 0x001ff000) >> 9));
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pte_addr, 8, &pte);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pte_addr, 8, BX_READ, (Bit8u*)(&pte));
|
2004-12-14 01:26:36 +03:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pte & 0x01)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PAE PTE: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
|
|
|
#if BX_SUPPORT_X86_64
|
2005-03-03 23:24:52 +03:00
|
|
|
if (pte & PAGE_DIRECTORY_NX_BIT) {
|
2008-04-01 00:56:27 +04:00
|
|
|
if (! BX_CPU_THIS_PTR efer.get_NXE()) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PTE: NX bit set when EFER.NXE is disabled"));
|
2007-08-30 20:48:10 +04:00
|
|
|
page_fault(ERROR_RESERVED | ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
|
|
|
if (access_type == CODE_ACCESS) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PTE: non-executable page fault occured"));
|
|
|
|
nx_fault = 1;
|
2006-10-28 16:31:23 +04:00
|
|
|
}
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2005-01-20 22:37:43 +03:00
|
|
|
#endif
|
2007-09-20 21:33:35 +04:00
|
|
|
if (pte & BX_CONST64(0x000fffff00000000)) {
|
2008-04-11 18:30:15 +04:00
|
|
|
BX_PANIC(("PAE PTE 0x%08x%08x: Only 32 bit physical address space is emulated !", GET32H(pte), GET32L(pte)));
|
2006-10-04 23:08:40 +04:00
|
|
|
}
|
|
|
|
|
2005-03-03 23:24:52 +03:00
|
|
|
combined_access = (pde & pte) & 0x06; // U/S and R/W
|
2007-09-20 21:33:35 +04:00
|
|
|
#if BX_SUPPORT_X86_64
|
|
|
|
if (long_mode()) {
|
|
|
|
combined_access &= (pml4 & pdpe) & 0x06;
|
|
|
|
}
|
|
|
|
#endif
|
2005-03-03 23:24:52 +03:00
|
|
|
|
2002-09-17 00:23:38 +04:00
|
|
|
// Make up the physical page frame address.
|
2007-12-23 20:21:28 +03:00
|
|
|
ppf = (bx_phy_address)(pte & BX_CONST64(0x000ffffffffff000));
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2007-12-06 21:35:33 +03:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PGE())
|
|
|
|
combined_access |= (pte & TLB_GlobalPage); // G
|
2002-09-17 00:23:38 +04:00
|
|
|
#endif
|
|
|
|
|
|
|
|
priv_index =
|
|
|
|
#if BX_CPU_LEVEL >= 4
|
2007-07-09 19:16:14 +04:00
|
|
|
(BX_CPU_THIS_PTR cr0.get_WP() << 4) | // bit 4
|
2002-09-17 00:23:38 +04:00
|
|
|
#endif
|
2007-07-09 19:16:14 +04:00
|
|
|
(pl<<3) | // bit 3
|
|
|
|
(combined_access & 0x06) | // bit 2,1
|
2007-09-20 21:33:35 +04:00
|
|
|
(isWrite); // bit 0
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2008-02-03 00:46:54 +03:00
|
|
|
if (!priv_check[priv_index] || nx_fault)
|
2007-08-30 20:48:10 +04:00
|
|
|
page_fault(ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Update PDPE A bit if needed.
|
|
|
|
if (!(pdpe & 0x20)) {
|
|
|
|
pdpe |= 0x20;
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pdpe_addr, 8, &pdpe);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pdpe_addr, 8, BX_WRITE, (Bit8u*)(&pdpe));
|
2007-09-20 21:33:35 +04:00
|
|
|
}
|
|
|
|
|
2002-09-17 00:23:38 +04:00
|
|
|
// Update PDE A bit if needed.
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pde & 0x20)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
pde |= 0x20;
|
2006-10-04 23:08:40 +04:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pde_addr, 8, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 8, BX_WRITE, (Bit8u*)(&pde));
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2002-09-17 00:23:38 +04:00
|
|
|
|
|
|
|
// Update PTE A/D bits if needed.
|
2007-09-20 21:33:35 +04:00
|
|
|
if (((pte & 0x20)==0) || (isWrite && ((pte & 0x40)==0))) {
|
2002-09-17 00:23:38 +04:00
|
|
|
pte |= (0x20 | (isWrite<<6)); // Update A and possibly D bits
|
2006-10-04 23:08:40 +04:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pte_addr, 8, &pte);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pte_addr, 8, BX_WRITE, (Bit8u*)(&pte));
|
2002-09-17 00:23:38 +04:00
|
|
|
}
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2004-10-21 22:20:40 +04:00
|
|
|
}
|
2004-12-14 01:26:36 +03:00
|
|
|
else
|
2005-11-27 00:36:51 +03:00
|
|
|
#endif // #if BX_SUPPORT_PAE
|
2004-12-14 01:26:36 +03:00
|
|
|
{
|
2007-09-11 00:47:08 +04:00
|
|
|
// CR4.PAE==0 (and EFER.LMA==0)
|
2006-10-04 23:08:40 +04:00
|
|
|
Bit32u pde, pte;
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
|
2007-12-23 20:21:28 +03:00
|
|
|
bx_phy_address pde_addr = (bx_phy_address) (BX_CPU_THIS_PTR cr3_masked |
|
|
|
|
((laddr & 0xffc00000) >> 20));
|
2004-12-14 01:26:36 +03:00
|
|
|
|
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pde_addr, 4, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 4, BX_READ, (Bit8u*)(&pde));
|
2004-12-14 01:26:36 +03:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pde & 0x01)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PDE: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
#if BX_SUPPORT_LARGE_PAGES
|
2004-12-14 01:26:36 +03:00
|
|
|
if ((pde & 0x80) && (BX_CPU_THIS_PTR cr4.get_PSE()))
|
|
|
|
{
|
2007-09-20 21:33:35 +04:00
|
|
|
// Note: when the PSE and PAE flags in CR4 are set, the
|
|
|
|
// processor generates a PF if the reserved bits are not zero.
|
2004-12-14 01:26:36 +03:00
|
|
|
|
2008-04-23 02:05:38 +04:00
|
|
|
if (pde & 0x0001e000) {
|
|
|
|
BX_PANIC(("PSE PDE 0x%08x: Only 32 bit physical address space is emulated !", pde));
|
|
|
|
}
|
|
|
|
|
2004-12-14 01:26:36 +03:00
|
|
|
// Combined access is just access from the pde (no pte involved).
|
2007-09-20 21:33:35 +04:00
|
|
|
combined_access = pde & 0x06; // U/S and R/W
|
2004-12-14 01:26:36 +03:00
|
|
|
// make up the physical frame number
|
2006-10-04 23:47:24 +04:00
|
|
|
ppf = (pde & 0xffc00000) | (laddr & 0x003ff000);
|
2002-09-17 00:23:38 +04:00
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2007-12-06 21:35:33 +03:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PGE())
|
|
|
|
combined_access |= pde & TLB_GlobalPage; // {G}
|
2002-09-10 07:52:32 +04:00
|
|
|
#endif
|
2002-06-19 19:49:07 +04:00
|
|
|
|
2004-12-14 01:26:36 +03:00
|
|
|
priv_index =
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#if BX_CPU_LEVEL >= 4
|
2007-07-09 19:16:14 +04:00
|
|
|
(BX_CPU_THIS_PTR cr0.get_WP() << 4) | // bit 4
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#endif
|
2007-07-09 19:16:14 +04:00
|
|
|
(pl<<3) | // bit 3
|
|
|
|
(combined_access & 0x06) | // bit 2,1
|
2007-09-20 21:33:35 +04:00
|
|
|
(isWrite); // bit 0
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
if (!priv_check[priv_index])
|
|
|
|
page_fault(ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2002-06-19 19:49:07 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Update PDE A/D bits if needed.
|
|
|
|
if (((pde & 0x20)==0) || (isWrite && ((pde & 0x40)==0))) {
|
2004-12-14 01:26:36 +03:00
|
|
|
pde |= (0x20 | (isWrite<<6)); // Update A and possibly D bits
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pde_addr, 4, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 4, BX_WRITE, (Bit8u*)(&pde));
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
}
|
2006-10-04 23:08:40 +04:00
|
|
|
else // else normal 4K page...
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#endif
|
2004-12-14 01:26:36 +03:00
|
|
|
{
|
|
|
|
// Get page table entry
|
2007-12-23 20:21:28 +03:00
|
|
|
bx_phy_address pte_addr = (bx_phy_address)((pde & 0xfffff000) | ((laddr & 0x003ff000) >> 10));
|
2004-12-14 01:26:36 +03:00
|
|
|
|
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pte_addr, 4, &pte);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pte_addr, 4, BX_READ, (Bit8u*)(&pte));
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2006-10-04 23:08:40 +04:00
|
|
|
if (!(pte & 0x01)) {
|
2007-09-20 21:33:35 +04:00
|
|
|
BX_DEBUG(("PTE: entry not present"));
|
|
|
|
page_fault(ERROR_NOT_PRESENT, laddr, pl, isWrite, access_type);
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
|
|
|
|
2006-10-04 23:47:24 +04:00
|
|
|
// 386 and 486+ have different behaviour for combining
|
2004-12-14 01:26:36 +03:00
|
|
|
// privilege from PDE and PTE.
|
2001-04-10 05:04:59 +04:00
|
|
|
#if BX_CPU_LEVEL == 3
|
2004-12-14 01:26:36 +03:00
|
|
|
combined_access = (pde | pte) & 0x04; // U/S
|
|
|
|
combined_access |= (pde & pte) & 0x02; // R/W
|
2001-04-10 05:04:59 +04:00
|
|
|
#else // 486+
|
2004-12-14 01:26:36 +03:00
|
|
|
combined_access = (pde & pte) & 0x06; // U/S and R/W
|
2007-09-20 21:33:35 +04:00
|
|
|
#endif
|
|
|
|
|
|
|
|
// Make up the physical page frame address.
|
|
|
|
ppf = pte & 0xfffff000;
|
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2008-04-22 00:17:45 +04:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PGE())
|
2005-06-15 00:55:57 +04:00
|
|
|
combined_access |= (pte & TLB_GlobalPage); // G
|
2001-04-10 05:04:59 +04:00
|
|
|
#endif
|
|
|
|
|
2004-12-14 01:26:36 +03:00
|
|
|
priv_index =
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#if BX_CPU_LEVEL >= 4
|
2007-07-09 19:16:14 +04:00
|
|
|
(BX_CPU_THIS_PTR cr0.get_WP() << 4) | // bit 4
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#endif
|
2007-07-09 19:16:14 +04:00
|
|
|
(pl<<3) | // bit 3
|
|
|
|
(combined_access & 0x06) | // bit 2,1
|
2007-09-20 21:33:35 +04:00
|
|
|
(isWrite); // bit 0
|
2002-06-19 19:49:07 +04:00
|
|
|
|
2007-08-30 20:48:10 +04:00
|
|
|
if (!priv_check[priv_index])
|
|
|
|
page_fault(ERROR_PROTECTION, laddr, pl, isWrite, access_type);
|
2002-06-19 19:49:07 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Update PDE A bit if needed.
|
2006-10-04 23:47:24 +04:00
|
|
|
if (!(pde & 0x20)) {
|
2004-12-14 01:26:36 +03:00
|
|
|
pde |= 0x20;
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pde_addr, 4, &pde);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pde_addr, 4, BX_WRITE, (Bit8u*)(&pde));
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
2002-06-19 19:49:07 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// Update PTE A/D bits if needed.
|
|
|
|
if (((pte & 0x20)==0) || (isWrite && ((pte & 0x40)==0))) {
|
2004-12-14 01:26:36 +03:00
|
|
|
pte |= (0x20 | (isWrite<<6)); // Update A and possibly D bits
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, pte_addr, 4, &pte);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_PHY_MEMORY_ACCESS(BX_CPU_ID, pte_addr, 4, BX_WRITE, (Bit8u*)(&pte));
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
}
|
2004-10-30 01:15:48 +04:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
// Calculate physical memory address and fill in TLB cache entry
|
|
|
|
paddress = ppf | poffset;
|
2007-09-20 21:33:35 +04:00
|
|
|
|
2008-04-22 00:17:45 +04:00
|
|
|
tlbEntry->lpf = lpf;
|
|
|
|
tlbEntry->ppf = ppf;
|
2002-09-04 12:59:13 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
// b3: Write User OK
|
|
|
|
// b2: Write Sys OK
|
|
|
|
// b1: Read User OK
|
|
|
|
// b0: Read Sys OK
|
2006-10-04 23:47:24 +04:00
|
|
|
if (combined_access & 4) { // User
|
2005-06-15 00:55:57 +04:00
|
|
|
// User priv; read from {user,sys} OK.
|
|
|
|
accessBits = (TLB_ReadUserOK | TLB_ReadSysOK);
|
2008-04-22 00:17:45 +04:00
|
|
|
if (isWrite) { // Current operation is a write (Dirty bit updated)
|
2005-03-03 23:24:52 +03:00
|
|
|
if (combined_access & 2) {
|
2005-06-15 00:55:57 +04:00
|
|
|
// R/W access from {user,sys} OK.
|
|
|
|
accessBits |= (TLB_WriteUserOK | TLB_WriteSysOK);
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
|
|
|
else {
|
2005-06-15 00:55:57 +04:00
|
|
|
accessBits |= TLB_WriteSysOK; // read only page, only {sys} write allowed
|
2005-03-03 23:24:52 +03:00
|
|
|
}
|
2002-09-04 12:59:13 +04:00
|
|
|
}
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
2002-09-04 12:59:13 +04:00
|
|
|
else { // System
|
2005-06-15 00:55:57 +04:00
|
|
|
accessBits = TLB_ReadSysOK; // System priv; read from {sys} OK.
|
2006-10-04 23:47:24 +04:00
|
|
|
if (isWrite) { // Current operation is a write (Dirty bit updated)
|
2005-06-15 00:55:57 +04:00
|
|
|
accessBits |= TLB_WriteSysOK; // write from {sys} OK.
|
2002-09-04 12:59:13 +04:00
|
|
|
}
|
2004-12-14 01:26:36 +03:00
|
|
|
}
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_GLOBAL_PAGES
|
2005-06-15 00:55:57 +04:00
|
|
|
accessBits |= combined_access & TLB_GlobalPage; // Global bit
|
2002-09-10 07:52:32 +04:00
|
|
|
#endif
|
2007-09-20 21:33:35 +04:00
|
|
|
|
Integrated patches for:
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
2002-09-02 00:12:09 +04:00
|
|
|
#if BX_SupportGuest2HostTLB
|
2004-12-14 01:26:36 +03:00
|
|
|
// Attempt to get a host pointer to this physical page. Put that
|
|
|
|
// pointer in the TLB cache. Note if the request is vetoed, NULL
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
// will be returned, and it's OK to OR zero in anyways.
|
2008-04-22 00:17:45 +04:00
|
|
|
tlbEntry->hostPageAddr = (bx_hostpageaddr_t) BX_CPU_THIS_PTR mem->getHostMemAddr(BX_CPU_THIS,
|
2006-03-26 23:39:37 +04:00
|
|
|
A20ADDR(ppf), rw, access_type);
|
2005-06-15 00:55:57 +04:00
|
|
|
|
2008-04-22 00:17:45 +04:00
|
|
|
if (tlbEntry->hostPageAddr) {
|
2005-06-15 00:55:57 +04:00
|
|
|
// All access allowed also via direct pointer
|
2008-02-03 00:46:54 +03:00
|
|
|
accessBits |= (accessBits & 0xff00) >> 8;
|
2005-06-15 00:55:57 +04:00
|
|
|
}
|
2004-10-21 22:20:40 +04:00
|
|
|
#endif
|
2008-04-22 00:17:45 +04:00
|
|
|
tlbEntry->accessBits = accessBits;
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-09-20 21:33:35 +04:00
|
|
|
return paddress;
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
|
2002-10-03 08:53:53 +04:00
|
|
|
#if BX_DEBUGGER || BX_DISASM || BX_INSTRUMENTATION || BX_GDBSTUB
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2006-06-17 16:09:55 +04:00
|
|
|
bx_bool BX_CPU_C::dbg_xlate_linear2phy(bx_address laddr, bx_phy_address *phy)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
2007-07-09 19:16:14 +04:00
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PG() == 0) {
|
2007-12-30 20:53:12 +03:00
|
|
|
*phy = (bx_phy_address) laddr;
|
2006-06-17 16:09:55 +04:00
|
|
|
return 1;
|
2004-12-17 01:21:35 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-10-09 00:45:30 +04:00
|
|
|
bx_phy_address paddress;
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
// see if page is in the TLB first
|
2008-02-11 23:52:10 +03:00
|
|
|
bx_address lpf = LPFOf(laddr);
|
2007-12-27 02:07:44 +03:00
|
|
|
unsigned TLB_index = BX_TLB_INDEX_OF(lpf, 0);
|
2005-06-15 00:55:57 +04:00
|
|
|
bx_TLB_entry *tlbEntry = &BX_CPU_THIS_PTR TLB.entry[TLB_index];
|
|
|
|
|
2007-11-11 23:44:07 +03:00
|
|
|
if (tlbEntry->lpf == lpf) {
|
2008-02-11 23:52:10 +03:00
|
|
|
paddress = tlbEntry->ppf | PAGE_OFFSET(laddr);
|
2001-04-10 05:04:59 +04:00
|
|
|
*phy = paddress;
|
2006-06-17 16:09:55 +04:00
|
|
|
return 1;
|
2004-10-21 22:20:40 +04:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-10-09 00:45:30 +04:00
|
|
|
bx_phy_address pt_address = BX_CPU_THIS_PTR cr3_masked;
|
|
|
|
bx_address offset_mask = 0xfff;
|
|
|
|
|
2005-11-27 00:36:51 +03:00
|
|
|
#if BX_SUPPORT_PAE
|
2005-02-16 21:58:48 +03:00
|
|
|
if (BX_CPU_THIS_PTR cr4.get_PAE()) {
|
|
|
|
int levels = 3;
|
2005-02-24 00:18:24 +03:00
|
|
|
#if BX_SUPPORT_X86_64
|
2006-10-04 23:08:40 +04:00
|
|
|
if (long_mode())
|
2005-02-16 21:58:48 +03:00
|
|
|
levels = 4;
|
|
|
|
#endif
|
|
|
|
for (int level = levels - 1; level >= 0; --level) {
|
|
|
|
Bit64u pte;
|
|
|
|
pt_address += 8 * ((laddr >> (12 + 9*level)) & 511);
|
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pt_address, 8, &pte);
|
|
|
|
if (!(pte & 1))
|
|
|
|
goto page_fault;
|
2007-12-30 20:53:12 +03:00
|
|
|
pt_address = (bx_phy_address)(pte & BX_CONST64(0x000ffffffffff000));
|
2005-02-16 21:58:48 +03:00
|
|
|
if (level == 1 && (pte & 0x80)) { // PSE page
|
|
|
|
offset_mask = 0x1fffff;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2007-12-23 20:21:28 +03:00
|
|
|
paddress = pt_address + (bx_phy_address)(laddr & offset_mask);
|
2008-02-03 00:46:54 +03:00
|
|
|
}
|
2005-11-17 20:52:00 +03:00
|
|
|
else // not PAE
|
|
|
|
#endif
|
|
|
|
{
|
2005-02-16 21:58:48 +03:00
|
|
|
for (int level = 1; level >= 0; --level) {
|
|
|
|
Bit32u pte;
|
|
|
|
pt_address += 4 * ((laddr >> (12 + 10*level)) & 1023);
|
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, pt_address, 4, &pte);
|
|
|
|
if (!(pte & 1))
|
|
|
|
goto page_fault;
|
|
|
|
pt_address = pte & 0xfffff000;
|
|
|
|
if (level == 1 && (pte & 0x80)) { // PSE page
|
|
|
|
offset_mask = 0x3fffff;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2007-12-23 20:21:28 +03:00
|
|
|
paddress = pt_address + (bx_phy_address)(laddr & offset_mask);
|
2005-01-20 22:37:43 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
*phy = paddress;
|
2006-06-17 16:09:55 +04:00
|
|
|
return 1;
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
page_fault:
|
|
|
|
*phy = 0;
|
2006-06-17 16:09:55 +04:00
|
|
|
return 0;
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
void BX_CPU_C::access_write_linear(bx_address laddr, unsigned len, unsigned curr_pl, void *data)
|
2001-04-10 05:04:59 +04:00
|
|
|
{
|
|
|
|
#if BX_X86_DEBUGGER
|
2008-03-29 21:18:08 +03:00
|
|
|
hwbreakpoint_match(laddr, len, BX_WRITE);
|
2001-04-10 05:04:59 +04:00
|
|
|
#endif
|
|
|
|
|
2007-12-27 02:07:44 +03:00
|
|
|
Bit32u pageOffset = PAGE_OFFSET(laddr);
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2007-07-09 19:16:14 +04:00
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PG()) {
|
2001-04-10 05:04:59 +04:00
|
|
|
/* check for reference across multiple pages */
|
2008-02-15 22:03:54 +03:00
|
|
|
if ((pageOffset + len) <= 4096) {
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
// Access within single page.
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 =
|
2008-03-29 21:18:08 +03:00
|
|
|
dtranslate_linear(laddr, curr_pl, BX_WRITE);
|
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 1;
|
|
|
|
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr, BX_CPU_THIS_PTR address_xlation.paddress1, len, BX_WRITE);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr, BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
len, curr_pl, BX_WRITE, (Bit8u*) data);
|
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1, len, data);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
// access across 2 pages
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 =
|
|
|
|
dtranslate_linear(laddr, curr_pl, BX_WRITE);
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1 = 4096 - pageOffset;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2 = len -
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 2;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2 =
|
|
|
|
dtranslate_linear(laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
curr_pl, BX_WRITE);
|
|
|
|
|
|
|
|
#ifdef BX_LITTLE_ENDIAN
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
|
|
|
BX_WRITE, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_WRITE, ((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress2,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2,
|
|
|
|
((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
2008-03-29 21:18:08 +03:00
|
|
|
#else // BX_BIG_ENDIAN
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
|
|
|
BX_WRITE, ((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_WRITE, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress2,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, data);
|
2008-03-29 21:18:08 +03:00
|
|
|
#endif
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
// Paging off.
|
|
|
|
if ((pageOffset + len) <= 4096) {
|
|
|
|
// Access within single page.
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 = (bx_phy_address) laddr;
|
2001-04-10 05:04:59 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 1;
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr, (bx_phy_address) laddr, len, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr, (bx_phy_address) laddr, len,
|
2008-04-19 18:13:43 +04:00
|
|
|
curr_pl, BX_WRITE, (Bit8u*) data);
|
2001-04-10 05:04:59 +04:00
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
#if BX_SupportGuest2HostTLB
|
|
|
|
unsigned tlbIndex = BX_TLB_INDEX_OF(laddr, 0);
|
|
|
|
bx_TLB_entry *tlbEntry = &BX_CPU_THIS_PTR TLB.entry[tlbIndex];
|
|
|
|
bx_address lpf = LPFOf(laddr);
|
|
|
|
|
|
|
|
if (tlbEntry->lpf == lpf) {
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, (bx_phy_address) laddr, len, data);
|
|
|
|
return;
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2008-03-29 21:18:08 +03:00
|
|
|
// We haven't seen this page, or it's been bumped before.
|
|
|
|
|
|
|
|
tlbEntry->lpf = lpf;
|
|
|
|
tlbEntry->ppf = (bx_phy_address) lpf;
|
|
|
|
// TLB.entry[tlbIndex].ppf field not used for PG==0.
|
|
|
|
// Request a direct write pointer so we can do either R or W.
|
|
|
|
tlbEntry->hostPageAddr = (bx_hostpageaddr_t)
|
|
|
|
BX_CPU_THIS_PTR mem->getHostMemAddr(BX_CPU_THIS, A20ADDR(lpf), BX_WRITE, DATA_ACCESS);
|
|
|
|
|
|
|
|
if (tlbEntry->hostPageAddr) {
|
|
|
|
// Got direct write pointer OK. Mark for any operation to succeed.
|
|
|
|
tlbEntry->accessBits = (TLB_ReadSysOK | TLB_ReadUserOK | TLB_WriteSysOK | TLB_WriteUserOK |
|
|
|
|
TLB_ReadSysPtrOK | TLB_ReadUserPtrOK | TLB_WriteSysPtrOK | TLB_WriteUserPtrOK);
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
2008-03-29 21:18:08 +03:00
|
|
|
else
|
|
|
|
tlbEntry->accessBits = 0;
|
|
|
|
#endif // BX_SupportGuest2HostTLB
|
|
|
|
|
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS, (bx_phy_address) laddr, len, data);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
// Access spans two pages.
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 = (bx_phy_address) laddr;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1 = 4096 - pageOffset;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2 = len -
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 2;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2 = (bx_phy_address) (laddr +
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1);
|
|
|
|
|
|
|
|
#ifdef BX_LITTLE_ENDIAN
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
2008-04-19 18:13:43 +04:00
|
|
|
BX_WRITE, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, data);
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, BX_WRITE);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_WRITE, ((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2,
|
|
|
|
((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
|
|
|
#else // BX_BIG_ENDIAN
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
|
|
|
BX_WRITE, ((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, BX_WRITE);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_WRITE, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->writePhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, data);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void BX_CPU_C::access_read_linear(bx_address laddr, unsigned len, unsigned curr_pl, unsigned xlate_rw, void *data)
|
|
|
|
{
|
2008-03-30 00:12:11 +03:00
|
|
|
BX_ASSERT(xlate_rw == BX_READ || xlate_rw == BX_RW);
|
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
#if BX_X86_DEBUGGER
|
|
|
|
hwbreakpoint_match(laddr, len, xlate_rw);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
Bit32u pageOffset = PAGE_OFFSET(laddr);
|
|
|
|
|
|
|
|
if (BX_CPU_THIS_PTR cr0.get_PG()) {
|
|
|
|
/* check for reference across multiple pages */
|
|
|
|
if ((pageOffset + len) <= 4096) {
|
|
|
|
// Access within single page.
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 =
|
|
|
|
dtranslate_linear(laddr, curr_pl, xlate_rw);
|
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 1;
|
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1, len, data);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1, len, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1, len, curr_pl,
|
2008-04-19 18:13:43 +04:00
|
|
|
BX_READ, (Bit8u*) data);
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
else {
|
2006-03-26 23:39:37 +04:00
|
|
|
// access across 2 pages
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 =
|
2007-12-17 00:03:46 +03:00
|
|
|
dtranslate_linear(laddr, curr_pl, xlate_rw);
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1 = 4096 - pageOffset;
|
2007-11-20 20:15:33 +03:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2 = len -
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1;
|
2001-04-10 05:04:59 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 2;
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2 =
|
|
|
|
dtranslate_linear(laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
2007-12-17 00:03:46 +03:00
|
|
|
curr_pl, xlate_rw);
|
2001-04-10 05:04:59 +04:00
|
|
|
|
|
|
|
#ifdef BX_LITTLE_ENDIAN
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, data);
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
2008-04-19 18:13:43 +04:00
|
|
|
BX_READ, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress2,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2,
|
|
|
|
((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_READ, ((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
2001-04-10 05:04:59 +04:00
|
|
|
#else // BX_BIG_ENDIAN
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress1,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
|
|
|
BX_READ, ((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, BX_CPU_THIS_PTR address_xlation.paddress2,
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, data);
|
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_READ, (Bit8u*) data);
|
2001-04-10 05:04:59 +04:00
|
|
|
#endif
|
|
|
|
}
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
else {
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
// Paging off.
|
2008-02-15 22:03:54 +03:00
|
|
|
if ((pageOffset + len) <= 4096) {
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
// Access within single page.
|
2007-12-23 20:21:28 +03:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 = (bx_phy_address) laddr;
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 1;
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr, (bx_phy_address) laddr, len, xlate_rw);
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
#if BX_SupportGuest2HostTLB
|
|
|
|
unsigned tlbIndex = BX_TLB_INDEX_OF(laddr, 0);
|
|
|
|
bx_TLB_entry *tlbEntry = &BX_CPU_THIS_PTR TLB.entry[tlbIndex];
|
|
|
|
bx_address lpf = LPFOf(laddr);
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
if (tlbEntry->lpf == lpf) {
|
2007-12-23 20:21:28 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, (bx_phy_address) laddr, len, data);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr, (bx_phy_address) laddr, len,
|
2008-04-19 18:13:43 +04:00
|
|
|
curr_pl, BX_READ, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
return;
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2008-03-29 21:18:08 +03:00
|
|
|
// We haven't seen this page, or it's been bumped before.
|
|
|
|
|
|
|
|
tlbEntry->lpf = lpf;
|
|
|
|
tlbEntry->ppf = (bx_phy_address) lpf;
|
|
|
|
// Request a direct write pointer so we can do either R or W.
|
|
|
|
tlbEntry->hostPageAddr = (bx_hostpageaddr_t)
|
|
|
|
BX_CPU_THIS_PTR mem->getHostMemAddr(BX_CPU_THIS, A20ADDR(lpf), BX_WRITE, DATA_ACCESS);
|
|
|
|
if (! tlbEntry->hostPageAddr) {
|
|
|
|
// Direct write vetoed. Try requesting only direct reads.
|
2005-06-15 00:55:57 +04:00
|
|
|
tlbEntry->hostPageAddr = (bx_hostpageaddr_t)
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->getHostMemAddr(BX_CPU_THIS, A20ADDR(lpf), BX_READ, DATA_ACCESS);
|
2005-06-15 00:55:57 +04:00
|
|
|
if (tlbEntry->hostPageAddr) {
|
2008-03-29 21:18:08 +03:00
|
|
|
// Got direct read pointer OK.
|
|
|
|
tlbEntry->accessBits =
|
|
|
|
(TLB_ReadSysOK | TLB_ReadUserOK | TLB_ReadSysPtrOK | TLB_ReadUserPtrOK);
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2002-09-10 04:01:01 +04:00
|
|
|
else
|
2005-06-15 00:55:57 +04:00
|
|
|
tlbEntry->accessBits = 0;
|
2008-03-29 21:18:08 +03:00
|
|
|
}
|
|
|
|
else {
|
|
|
|
// Got direct write pointer OK. Mark for any operation to succeed.
|
|
|
|
tlbEntry->accessBits = (TLB_ReadSysOK | TLB_ReadUserOK | TLB_WriteSysOK | TLB_WriteUserOK |
|
|
|
|
TLB_ReadSysPtrOK | TLB_ReadUserPtrOK | TLB_WriteSysPtrOK | TLB_WriteUserPtrOK);
|
|
|
|
}
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
#endif // BX_SupportGuest2HostTLB
|
|
|
|
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS, (bx_phy_address) laddr, len, data);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr, (bx_phy_address) laddr, len,
|
2008-04-19 18:13:43 +04:00
|
|
|
curr_pl, BX_READ, (Bit8u*) data);
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
else {
|
2006-03-26 23:39:37 +04:00
|
|
|
// Access spans two pages.
|
2007-12-23 20:21:28 +03:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1 = (bx_phy_address) laddr;
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1 = 4096 - pageOffset;
|
2007-11-20 20:15:33 +03:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len2 = len -
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
BX_CPU_THIS_PTR address_xlation.len1;
|
|
|
|
BX_CPU_THIS_PTR address_xlation.pages = 2;
|
2007-12-23 20:21:28 +03:00
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2 = (bx_phy_address) (laddr +
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1);
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
|
|
|
|
#ifdef BX_LITTLE_ENDIAN
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, data);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
2008-04-19 18:13:43 +04:00
|
|
|
BX_READ, (Bit8u*) data);
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2,
|
|
|
|
((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_READ, ((Bit8u*)data) + BX_CPU_THIS_PTR address_xlation.len1);
|
Now, when you compile with --enable-guest2host-tlb, non-paged
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
2002-09-05 06:31:24 +04:00
|
|
|
#else // BX_BIG_ENDIAN
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len1, curr_pl,
|
|
|
|
BX_READ, ((Bit8u*)data) + (len - BX_CPU_THIS_PTR address_xlation.len1));
|
2008-03-29 21:18:08 +03:00
|
|
|
BX_CPU_THIS_PTR mem->readPhysicalPage(BX_CPU_THIS,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, data);
|
2008-04-19 17:21:23 +04:00
|
|
|
BX_INSTR_LIN_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, xlate_rw);
|
|
|
|
BX_DBG_LIN_MEMORY_ACCESS(BX_CPU_ID, laddr + BX_CPU_THIS_PTR address_xlation.len1,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.paddress2,
|
|
|
|
BX_CPU_THIS_PTR address_xlation.len2, curr_pl,
|
|
|
|
BX_READ, (Bit8u*) data);
|
2005-02-16 21:58:48 +03:00
|
|
|
#endif
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|
2005-02-16 21:58:48 +03:00
|
|
|
}
|
2001-04-10 05:04:59 +04:00
|
|
|
}
|