Enhanced the repeat IO accelerations (enabled by --enable-repeat-speedups)

to request bulk IO operations to IO devices which are bulk IO aware.
Currently, I modified only harddrv.cc to be aware.  I added some
fields to the bx_devices_c class for the IO instructions to
place requests and receive responses from the IO device emulation.
Devices except the hard drive, don't monitor these fields so they
respond as normal.  The hard drive now monitors these fields for
bulk requests, and if enabled, it memcpy()'s data straight from
the disk buffer to memory.  This eliminates numerous inp/outp calling
sequences per disk sector.

I used the fields in bx_devices_c so that I would not have to
disrupt most IO device modules.  Enhancements can be made to
other devices if they use high-bandwidth IO via in/out instructions.
This commit is contained in:
Kevin Lawton 2002-09-09 16:56:56 +00:00
parent 6a9d2e662f
commit 414e97bc32
4 changed files with 128 additions and 42 deletions

View File

@ -1,5 +1,5 @@
/////////////////////////////////////////////////////////////////////////
// $Id: io.cc,v 1.9 2002-09-08 04:08:14 kevinlawton Exp $
// $Id: io.cc,v 1.10 2002-09-09 16:56:54 kevinlawton Exp $
/////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2001 MandrakeSoft S.A.
@ -36,6 +36,8 @@
void
BX_CPU_C::INSB_YbDX(BxInstruction_t *i)
{
@ -141,7 +143,7 @@ BX_CPU_C::INSW_YvDX(BxInstruction_t *i)
Bit32u laddrDst, paddrDst, wordsFitDst;
Bit8u *hostAddrDst;
bx_segment_reg_t *dstSegPtr;
unsigned pointerDelta;
int pointerDelta;
dstSegPtr = &BX_CPU_THIS_PTR sregs[BX_SREG_ES];
@ -172,7 +174,7 @@ BX_CPU_C::INSW_YvDX(BxInstruction_t *i)
if ( (paddrDst & 0xfff) > 0xffe )
goto noAcceleration;
wordsFitDst = (2 + (paddrDst & 0xfff)) >> 1;
pointerDelta = (unsigned) -2;
pointerDelta = -2;
}
else {
// Counting upward.
@ -224,23 +226,33 @@ BX_CPU_C::INSW_YvDX(BxInstruction_t *i)
for (j=0; j<wordCount; ) {
Bit16u temp16;
if (j) {
// Complete a tick for the last iteration, terminating
// with one less tick than iterations, since the main cpu
// loop also decrements one. Probably a little anal...
BX_TICK1();
bx_devices.bulkIOQuantumsTransferred = 0;
if ( GetEFlagsDFLogical()==0 ) { // Only do accel for DF=0
bx_devices.bulkIOHostAddr = (Bit32u) hostAddrDst;
bx_devices.bulkIOQuantumsRequested = (wordCount - j);
}
else
bx_devices.bulkIOQuantumsRequested = 0;
temp16 = BX_INP(DX, 2);
* (Bit16u *) hostAddrDst = temp16;
hostAddrDst += pointerDelta;
j++;
if ( bx_devices.bulkIOQuantumsTransferred ) {
hostAddrDst = (Bit8u*) bx_devices.bulkIOHostAddr;
j += bx_devices.bulkIOQuantumsTransferred;
}
else {
* (Bit16u *) hostAddrDst = temp16;
hostAddrDst += pointerDelta;
j++;
}
// Terminate early if there was an event.
if ( BX_CPU_THIS_PTR async_event )
break;
}
// Reset for next non-bulk IO.
bx_devices.bulkIOQuantumsRequested = 0;
wordCount = j;
// Decrement eCX. Note, the main loop will decrement 1 also, so
// decrement by one less than expected, like the case above.
BX_TICKN(j-1); // Main cpu loop also decrements one more.
if (i->as_32)
ECX -= (wordCount-1);
else
@ -478,23 +490,33 @@ BX_CPU_C::OUTSW_DXXv(BxInstruction_t *i)
for (j=0; j<wordCount; ) {
Bit16u temp16;
if (j) {
// Complete a tick for the last iteration, terminating
// with one less tick than iterations, since the main cpu
// loop also decrements one. Probably a little anal...
BX_TICK1();
bx_devices.bulkIOQuantumsTransferred = 0;
if ( GetEFlagsDFLogical()==0 ) { // Only do accel for DF=0
bx_devices.bulkIOHostAddr = (Bit32u) hostAddrSrc;
bx_devices.bulkIOQuantumsRequested = (wordCount - j);
}
else
bx_devices.bulkIOQuantumsRequested = 0;
temp16 = * (Bit16u *) hostAddrSrc;
hostAddrSrc += pointerDelta;
BX_OUTP(DX, temp16, 2);
j++;
if ( bx_devices.bulkIOQuantumsTransferred ) {
hostAddrSrc = (Bit8u*) bx_devices.bulkIOHostAddr;
j += bx_devices.bulkIOQuantumsTransferred;
}
else {
hostAddrSrc += pointerDelta;
j++;
}
// Terminate early if there was an event.
if ( BX_CPU_THIS_PTR async_event )
break;
}
// Reset for next non-bulk IO.
bx_devices.bulkIOQuantumsRequested = 0;
wordCount = j;
// Decrement eCX. Note, the main loop will decrement 1 also, so
// decrement by one less than expected, like the case above.
BX_TICKN(j-1); // Main cpu loop also decrements one more.
if (i->as_32)
ECX -= (wordCount-1);
else

View File

@ -1,5 +1,5 @@
/////////////////////////////////////////////////////////////////////////
// $Id: devices.cc,v 1.29 2002-08-27 19:54:46 bdenney Exp $
// $Id: devices.cc,v 1.30 2002-09-09 16:56:55 kevinlawton Exp $
/////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2002 MandrakeSoft S.A.
@ -84,7 +84,7 @@ bx_devices_c::init(BX_MEM_C *newmem)
{
unsigned i;
BX_DEBUG(("Init $Id: devices.cc,v 1.29 2002-08-27 19:54:46 bdenney Exp $"));
BX_DEBUG(("Init $Id: devices.cc,v 1.30 2002-09-09 16:56:55 kevinlawton Exp $"));
mem = newmem;
/* no read / write handlers defined */
@ -227,6 +227,11 @@ bx_devices_c::init(BX_MEM_C *newmem)
timer_handle = bx_pc_system.register_timer( this, timer_handler,
(unsigned) BX_IODEV_HANDLER_PERIOD, 1, 1);
// Clear fields for bulk IO acceleration transfers.
bulkIOHostAddr = 0;
bulkIOQuantumsRequested = 0;
bulkIOQuantumsTransferred = 0;
}

View File

@ -1,5 +1,5 @@
/////////////////////////////////////////////////////////////////////////
// $Id: harddrv.cc,v 1.69 2002-08-27 19:54:46 bdenney Exp $
// $Id: harddrv.cc,v 1.70 2002-09-09 16:56:56 kevinlawton Exp $
/////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2002 MandrakeSoft S.A.
@ -128,7 +128,7 @@ bx_hard_drive_c::~bx_hard_drive_c(void)
bx_hard_drive_c::init(bx_devices_c *d, bx_cmos_c *cmos)
{
BX_HD_THIS devices = d;
BX_DEBUG(("Init $Id: harddrv.cc,v 1.69 2002-08-27 19:54:46 bdenney Exp $"));
BX_DEBUG(("Init $Id: harddrv.cc,v 1.70 2002-09-09 16:56:56 kevinlawton Exp $"));
/* HARD DRIVE 0 */
@ -421,17 +421,41 @@ bx_hard_drive_c::read(Bit32u address, unsigned io_len)
if (BX_SELECTED_CONTROLLER.buffer_index >= 512)
BX_PANIC(("IO read(1f0): buffer_index >= 512"));
value32 = 0L;
switch(io_len){
case 4:
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+3] << 24);
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+2] << 16);
case 2:
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+1] << 8);
value32 |= BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index];
}
#if BX_SupportRepeatSpeedups
if (BX_HD_THIS devices->bulkIOQuantumsRequested) {
unsigned transferLen, quantumsMax;
BX_SELECTED_CONTROLLER.buffer_index += io_len;
quantumsMax =
(512 - BX_SELECTED_CONTROLLER.buffer_index) / io_len;
if ( quantumsMax == 0)
BX_PANIC(("IO read(1f0): not enough space for read"));
BX_HD_THIS devices->bulkIOQuantumsTransferred =
BX_HD_THIS devices->bulkIOQuantumsRequested;
if (quantumsMax < BX_HD_THIS devices->bulkIOQuantumsTransferred)
BX_HD_THIS devices->bulkIOQuantumsTransferred = quantumsMax;
transferLen = io_len * BX_HD_THIS devices->bulkIOQuantumsTransferred;
memcpy((Bit8u*) BX_HD_THIS devices->bulkIOHostAddr,
&BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index],
transferLen);
BX_HD_THIS devices->bulkIOHostAddr += transferLen;
BX_SELECTED_CONTROLLER.buffer_index += transferLen;
value32 = 0; // Value returned not important;
}
else
#endif
{
value32 = 0L;
switch(io_len){
case 4:
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+3] << 24);
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+2] << 16);
case 2:
value32 |= (BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+1] << 8);
value32 |= BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index];
}
BX_SELECTED_CONTROLLER.buffer_index += io_len;
}
// if buffer completely read
if (BX_SELECTED_CONTROLLER.buffer_index >= 512) {
@ -881,16 +905,40 @@ BX_DEBUG(("IO write to %04x = %02x", (unsigned) address, (unsigned) value));
if (BX_SELECTED_CONTROLLER.buffer_index >= 512)
BX_PANIC(("IO write(1f0): buffer_index >= 512"));
switch(io_len){
case 4:
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+3] = (Bit8u)(value >> 24);
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+2] = (Bit8u)(value >> 16);
case 2:
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+1] = (Bit8u)(value >> 8);
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index] = (Bit8u) value;
}
#if BX_SupportRepeatSpeedups
if (BX_HD_THIS devices->bulkIOQuantumsRequested) {
unsigned transferLen, quantumsMax;
BX_SELECTED_CONTROLLER.buffer_index += io_len;
quantumsMax =
(512 - BX_SELECTED_CONTROLLER.buffer_index) / io_len;
if ( quantumsMax == 0)
BX_PANIC(("IO write(1f0): not enough space for write"));
BX_HD_THIS devices->bulkIOQuantumsTransferred =
BX_HD_THIS devices->bulkIOQuantumsRequested;
if (quantumsMax < BX_HD_THIS devices->bulkIOQuantumsTransferred)
BX_HD_THIS devices->bulkIOQuantumsTransferred = quantumsMax;
transferLen = io_len * BX_HD_THIS devices->bulkIOQuantumsTransferred;
memcpy(
&BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index],
(Bit8u*) BX_HD_THIS devices->bulkIOHostAddr,
transferLen);
BX_HD_THIS devices->bulkIOHostAddr += transferLen;
BX_SELECTED_CONTROLLER.buffer_index += transferLen;
}
else
#endif
{
switch(io_len){
case 4:
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+3] = (Bit8u)(value >> 24);
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+2] = (Bit8u)(value >> 16);
case 2:
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index+1] = (Bit8u)(value >> 8);
BX_SELECTED_CONTROLLER.buffer[BX_SELECTED_CONTROLLER.buffer_index] = (Bit8u) value;
}
BX_SELECTED_CONTROLLER.buffer_index += io_len;
}
/* if buffer completely writtten */
if (BX_SELECTED_CONTROLLER.buffer_index >= 512) {

View File

@ -1,5 +1,5 @@
/////////////////////////////////////////////////////////////////////////
// $Id: iodev.h,v 1.16 2002-08-27 19:54:46 bdenney Exp $
// $Id: iodev.h,v 1.17 2002-09-09 16:56:56 kevinlawton Exp $
/////////////////////////////////////////////////////////////////////////
//
// Copyright (C) 2002 MandrakeSoft S.A.
@ -120,6 +120,17 @@ public:
bx_iodebug_c *iodebug;
#endif
// Some info to pass to devices which can handled bulk IO. This allows
// the interface to remain the same for IO devices which can't handle
// bulk IO. We should probably implement special INPBulk() and OUTBulk()
// functions which stick these values in the bx_devices_c class, and
// then call the normal functions rather than having gross globals
// variables.
Bit32u bulkIOHostAddr;
unsigned bulkIOQuantumsRequested;
unsigned bulkIOQuantumsTransferred;
private:
Bit8u read_handler_id[0x10000]; // 64K
struct {