(Bugzilla #1777) Import Mark Vojkovich's changes to NV driver since last

    sync with XFree86, including: Support for new NVIDIA chips. Egbert has
    a laptop that needs some DPMS ordering reversed. New chip support and
    workaround for Toshiba M30 laptop issue. PCI-Xpress chip support. DPMS
    support for DVI. Cursor changes for NV11 so that alpha blended cursors
    will work in conjunction with dithering (laptop panels). It previously
    did not. HW support for alpha blended stretch blits on NV5 and higher,
    though nothing is using them at this point. Add support for VBlank
    syncing the Xv Blit adaptor on hardware that supports it (NV11 and
    higher). Make console restoration a little more robust. We were making
    an assumption that failed on a newer laptop. Workaround for some NV31
    DVI artifacts.
Also add nVidia license statement to nv man page to comply with requirement
    that it appear in "user documentation"

18 files changed, 726 insertions, 238 deletions

diff --git a/src/nv_const.h b/src/nv_const.h
index f4ab437..bd01635 100644
--- a/src/nv_const.h
+++ b/src/nv_const.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_const.h,v 1.7 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_const.h,v 1.6 2001/12/07 00:09:55 mvojkovi Exp $ */
 
 #ifndef __NV_CONST_H__
 #define __NV_CONST_H__
diff --git a/src/nv_cursor.c b/src/nv_cursor.c
index 85ac42a..f22e3d7 100644
--- a/src/nv_cursor.c
+++ b/src/nv_cursor.c
@@ -37,7 +37,7 @@
 |*                                                                           *|
  \***************************************************************************/
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_cursor.c,v 1.12 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_cursor.c,v 1.13 2004/03/13 22:07:05 mvojkovi Exp $ */
 
 #include "nv_include.h"
 
@@ -239,28 +239,37 @@ NVLoadCursorARGB(ScrnInfoPtr pScrn, CursorPtr pCurs)
     NVPtr pNv = NVPTR(pScrn);
     CARD32 *image = pCurs->bits->argb;
     CARD32 *dst = (CARD32*)pNv->CURSOR;
+    CARD32 alpha, tmp;
     int x, y, w, h;
 
     w = pCurs->bits->width;
     h = pCurs->bits->height;
 
-#if X_BYTE_ORDER == X_BIG_ENDIAN
-    if((pNv->Chipset & 0x0ff0) == 0x0110) {
-       CARD32 tmp;
-
+    if((pNv->Chipset & 0x0ff0) == 0x0110) {  /* premultiply */
        for(y = 0; y < h; y++) {
           for(x = 0; x < w; x++) {
-              tmp = *image++;
-              *dst++ = BYTE_SWAP_32(tmp);
-          }
-          for(; x < 64; x++)
-              *dst++ = 0;
-       }
-    } else 
+             alpha = *image >> 24;
+             if(alpha == 0xff)
+                tmp = *image;
+             else {
+                tmp = (alpha << 24) |
+                         (((*image & 0xff) * alpha) / 255) |
+                        ((((*image & 0xff00) * alpha) / 255) & 0xff00) |
+                       ((((*image & 0xff0000) * alpha) / 255) & 0xff0000); 
+             }
+             image++;
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+             *dst++ = BYTE_SWAP_32(tmp);
+#else
+             *dst++ = tmp;
 #endif
-    {
+         }
+         for(; x < 64; x++)
+             *dst++ = 0;
+      }
+    } else {
        for(y = 0; y < h; y++) {
-          for(x = 0; x < w; x++) 
+          for(x = 0; x < w; x++)
               *dst++ = *image++;
           for(; x < 64; x++)
               *dst++ = 0;
@@ -299,9 +308,7 @@ NVCursorInit(ScreenPtr pScreen)
     infoPtr->UseHWCursor = NVUseHWCursor;
 
 #ifdef ARGB_CURSOR
-    if(pNv->alphaCursor &&
-       (((pNv->Chipset & 0x0ff0) != 0x0110) || !pNv->FPDither))
-    {
+    if(pNv->alphaCursor) {
        infoPtr->UseHWCursorARGB = NVUseHWCursorARGB;
        infoPtr->LoadCursorARGB = NVLoadCursorARGB;
     }
diff --git a/src/nv_dac.c b/src/nv_dac.c
index c1e093e..769179b 100644
--- a/src/nv_dac.c
+++ b/src/nv_dac.c
@@ -37,7 +37,7 @@
 |*                                                                           *|
  \***************************************************************************/
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dac.c,v 1.37 2003/09/08 20:00:27 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dac.c,v 1.41 2004/10/05 21:23:03 mvojkovi Exp $ */
 
 #include "nv_include.h"
 
@@ -58,7 +58,6 @@ NVDACInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
     int vertBlankStart  =  mode->CrtcVDisplay      - 1;
     int vertBlankEnd    =  mode->CrtcVTotal        - 1;
    
-
     NVPtr pNv = NVPTR(pScrn);
     NVRegPtr nvReg = &pNv->ModeReg;
     NVFBLayout *pLayout = &pNv->CurrentLayout;
@@ -215,26 +214,29 @@ NVDACInit(ScrnInfoPtr pScrn, DisplayModePtr mode)
     if(mode->Flags & V_DBLSCAN)
        nvReg->cursorConfig |= (1 << 4);
     if(pNv->alphaCursor) {
-        nvReg->cursorConfig |= 0x04011000;
+        if((pNv->Chipset & 0x0ff0) != 0x0110) 
+           nvReg->cursorConfig |= 0x04011000;
+        else
+           nvReg->cursorConfig |= 0x14011000;
         nvReg->general |= (1 << 29);
+    } else
+       nvReg->cursorConfig |= 0x02000000;
 
+    if(pNv->twoHeads) {
         if((pNv->Chipset & 0x0ff0) == 0x0110) {
-            nvReg->dither = pNv->PRAMDAC[0x0528/4] & ~0x00010000;
-            if(pNv->FPDither)
-               nvReg->dither |= 0x00010000;
-            else
-               nvReg->cursorConfig |= (1 << 28);
-        } else 
-        if((pNv->Chipset & 0x0ff0) >= 0x0170) {
+           nvReg->dither = pNv->PRAMDAC[0x0528/4] & ~0x00010000;
+           if(pNv->FPDither)
+              nvReg->dither |= 0x00010000;
+        } else {
            nvReg->dither = pNv->PRAMDAC[0x083C/4] & ~1;
-           nvReg->cursorConfig |= (1 << 28);
            if(pNv->FPDither)
               nvReg->dither |= 1;
-        } else {
-           nvReg->cursorConfig |= (1 << 28);
-        }
-    } else
-       nvReg->cursorConfig |= 0x02000000;
+        } 
+    }
+
+    nvReg->timingH = 0;
+    nvReg->timingV = 0;
+    nvReg->displayV = mode->CrtcVDisplay;
 
     return (TRUE);
 }
@@ -247,8 +249,6 @@ NVDACRestore(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, NVRegPtr nvReg,
     int restore = VGA_SR_MODE;
 
     if(primary) restore |= VGA_SR_CMAP | VGA_SR_FONTS;
-    else if((pNv->Chipset & 0xffff) == 0x0018) 
-	restore |= VGA_SR_CMAP;
     NVLoadStateExt(pNv, nvReg);
 #if defined(__powerpc__)
     restore &= ~VGA_SR_FONTS;
@@ -271,8 +271,6 @@ NVDACSave(ScrnInfoPtr pScrn, vgaRegPtr vgaReg, NVRegPtr nvReg,
     saveFonts = FALSE;
 #endif
 
-    NVLockUnlock(pNv, 0);
-
     vgaHWSave(pScrn, vgaReg, VGA_SR_CMAP | VGA_SR_MODE | 
                              (saveFonts? VGA_SR_FONTS : 0));
     NVUnloadStateExt(pNv, nvReg);
diff --git a/src/nv_dga.c b/src/nv_dga.c
index d4bf3ff..baafbfe 100644
--- a/src/nv_dga.c
+++ b/src/nv_dga.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dga.c,v 1.12 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dga.c,v 1.11 2002/01/25 21:56:06 tsi Exp $ */
 
 #include "nv_local.h"
 #include "nv_include.h"
diff --git a/src/nv_dma.h b/src/nv_dma.h
index 15b222f..ca0bf1b 100644
--- a/src/nv_dma.h
+++ b/src/nv_dma.h
@@ -38,7 +38,7 @@
 |*                                                                           *|
  \***************************************************************************/
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dma.h,v 1.1 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_dma.h,v 1.4 2004/03/20 01:52:16 mvojkovi Exp $ */
 
 #define SURFACE_FORMAT                                              0x00000300
 #define SURFACE_FORMAT_DEPTH8                                       0x00000001
@@ -145,9 +145,15 @@
 #define STRETCH_BLIT_FORMAT_DEPTH8                                  0x00000004
 #define STRETCH_BLIT_FORMAT_DEPTH16                                 0x00000007
 #define STRETCH_BLIT_FORMAT_DEPTH24                                 0x00000004
+#define STRETCH_BLIT_FORMAT_A8R8G8B8                                0x00000003
 #define STRETCH_BLIT_FORMAT_X8R8G8B8                                0x00000004
 #define STRETCH_BLIT_FORMAT_YUYV                                    0x00000005
 #define STRETCH_BLIT_FORMAT_UYVY                                    0x00000006
+/* STRETCH_BLIT_OPERATION is only supported on TNT2 and newer */
+#define STRETCH_BLIT_OPERATION                                      0x0000E304
+#define STRETCH_BLIT_OPERATION_ROP                                  0x00000001
+#define STRETCH_BLIT_OPERATION_COPY                                 0x00000003
+#define STRETCH_BLIT_OPERATION_BLEND                                0x00000002
 #define STRETCH_BLIT_CLIP_POINT                                     0x0000E308
 #define STRETCH_BLIT_CLIP_POINT_X                                   15:0 
 #define STRETCH_BLIT_CLIP_POINT_Y                                   31:16
diff --git a/src/nv_driver.c b/src/nv_driver.c
index b55ece9..77c72a7 100644
--- a/src/nv_driver.c
+++ b/src/nv_driver.c
@@ -1,4 +1,4 @@
-/* $XdotOrg: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_driver.c,v 1.4 2004/07/31 01:21:19 ajax Exp $ */
+/* $XdotOrg: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_driver.c,v 1.8 2004/11/10 02:53:44 alanc Exp $ */
 /* $XConsortium: nv_driver.c /main/3 1996/10/28 05:13:37 kaleb $ */
 /*
  * Copyright 1996-1997  David J. McKay
@@ -25,7 +25,7 @@
 /* Hacked together from mga driver and 3.3.4 NVIDIA driver by Jarno Paananen
    <jpaana@s2.org> */
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_driver.c,v 1.122 2004/01/10 22:31:53 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_driver.c,v 1.128 2004/10/09 22:53:25 mvojkovi Exp $ */
 
 #include "nv_include.h"
 
@@ -174,6 +174,8 @@ static SymTabRec NVKnownChipsets[] =
   { 0x10DE0323, "GeForce FX 5200SE" },
   { 0x10DE0324, "GeForce FX Go5200" },
   { 0x10DE0325, "GeForce FX Go5250" },
+  { 0x10DE0326, "GeForce FX 5500" },
+  { 0x10DE0327, "GeForce FX 5100" },
   { 0x10DE0328, "GeForce FX Go5200 32M/64M" },
 #if defined(__powerpc__)
   { 0x10DE0329, "GeForce FX 5200 (Mac)" },
@@ -181,15 +183,16 @@ static SymTabRec NVKnownChipsets[] =
   { 0x10DE0329, "0x0329" },
 #endif
   { 0x10DE032A, "Quadro NVS 280 PCI" },
-  { 0x10DE032B, "Quadro FX 500" },
-  { 0x10DE032C, "GeForce FX Go5300" },
+  { 0x10DE032B, "Quadro FX 500/600 PCI" },
+  { 0x10DE032C, "GeForce FX Go53xx Series" },
   { 0x10DE032D, "GeForce FX Go5100" },
   { 0x10DE032F, "0x032F" },
   { 0x10DE0330, "GeForce FX 5900 Ultra" },
   { 0x10DE0331, "GeForce FX 5900" },
   { 0x10DE0332, "GeForce FX 5900XT" },
   { 0x10DE0333, "GeForce FX 5950 Ultra" },
-  { 0x10DE0334, "0x0334" },
+  { 0x10DE033F, "Quadro FX 700" },
+  { 0x10DE0334, "GeForce FX 5900ZT" },
   { 0x10DE0338, "Quadro FX 3000" },
   { 0x10DE0341, "GeForce FX 5700 Ultra" },
   { 0x10DE0342, "GeForce FX 5700" },
@@ -203,6 +206,41 @@ static SymTabRec NVKnownChipsets[] =
   { 0x10DE034C, "Quadro FX Go1000" },
   { 0x10DE034E, "Quadro FX 1100" },
   { 0x10DE034F, "0x034F" },
+  { 0x10DE0040, "GeForce 6800 Ultra" },
+  { 0x10DE0041, "GeForce 6800" },
+  { 0x10DE0042, "GeForce 6800 LE" },
+  { 0x10DE0043, "0x0043" },
+  { 0x10DE0045, "GeForce 6800 GT" },
+  { 0x10DE0049, "0x0049" },
+  { 0x10DE004E, "Quadro FX 4000" },
+  { 0x10DE00C0, "0x00C0" },
+  { 0x10DE00C1, "0x00C1" },
+  { 0x10DE00C2, "0x00C2" },
+  { 0x10DE00C8, "0x00C8" },
+  { 0x10DE00C9, "0x00C9" },
+  { 0x10DE00CC, "0x00CC" },
+  { 0x10DE00CE, "0x00CE" },
+  { 0x10DE0140, "GeForce 6600 GT" },
+  { 0x10DE0141, "GeForce 6600" },
+  { 0x10DE0142, "0x0142" },
+  { 0x10DE0143, "0x0143" },
+  { 0x10DE0144, "0x0144" },
+  { 0x10DE0145, "GeForce 6610 XL" },
+  { 0x10DE0146, "0x0146" },
+  { 0x10DE0147, "0x0147" },
+  { 0x10DE0148, "0x0148" },
+  { 0x10DE0149, "0x0149" },
+  { 0x10DE014B, "0x014B" },
+  { 0x10DE014C, "0x014C" },
+  { 0x10DE014D, "0x014D" },
+  { 0x10DE014E, "Quadro FX 540" },
+  { 0x10DE014F, "GeForce 6200" },
+  { 0x10DE0160, "0x0160" },
+  { 0x10DE0166, "0x0166" },
+  { 0x10DE0210, "0x0210" },
+  { 0x10DE0211, "0x0211" },
+  { 0x10DE021D, "0x021D" },
+  { 0x10DE021E, "0x021E" },
   {-1, NULL}
 };
 
@@ -350,7 +388,8 @@ typedef enum {
     OPTION_VIDEO_KEY,
     OPTION_FLAT_PANEL,
     OPTION_FP_DITHER,
-    OPTION_CRTC_NUMBER
+    OPTION_CRTC_NUMBER,
+    OPTION_FP_SCALE
 } NVOpts;
 
 
@@ -365,6 +404,7 @@ static const OptionInfoRec NVOptions[] = {
     { OPTION_FLAT_PANEL,	"FlatPanel",	OPTV_BOOLEAN,	{0}, FALSE },
     { OPTION_FP_DITHER, 	"FPDither",	OPTV_BOOLEAN,	{0}, FALSE },
     { OPTION_CRTC_NUMBER,	"CrtcNumber",	OPTV_INTEGER,	{0}, FALSE },
+    { OPTION_FP_SCALE,          "FPScale",      OPTV_BOOLEAN,   {0}, FALSE },
     { -1,                       NULL,           OPTV_NONE,      {0}, FALSE }
 };
 
@@ -494,6 +534,36 @@ NVGetScrnInfoRec(PciChipsets *chips, int chip)
 
 #define MAX_CHIPS MAXSCREENS
 
+
+static CARD32 
+NVGetPCIXpressChip (pciVideoPtr pVideo)
+{
+    volatile CARD32 *regs;
+    CARD32 pciid, pcicmd;
+    PCITAG Tag = ((pciConfigPtr)(pVideo->thisCard))->tag;
+
+    pcicmd = pciReadLong(Tag, PCI_CMD_STAT_REG);
+    pciWriteLong(Tag, PCI_CMD_STAT_REG, pcicmd | PCI_CMD_MEM_ENABLE);
+    
+    regs = xf86MapPciMem(-1, VIDMEM_MMIO, Tag, pVideo->memBase[0], 0x2000);
+
+    pciid = regs[0x1800/4];
+
+    xf86UnMapVidMem(-1, (pointer)regs, 0x2000);
+
+    pciWriteLong(Tag, PCI_CMD_STAT_REG, pcicmd);
+
+    if((pciid & 0x0000ffff) == 0x000010DE) 
+       pciid = 0x10DE0000 | (pciid >> 16);
+    else 
+    if((pciid & 0xffff0000) == 0xDE100000) /* wrong endian */
+       pciid = 0x10DE0000 | ((pciid << 8) & 0x0000ff00) |
+                            ((pciid >> 8) & 0x000000ff);
+
+    return pciid;
+}
+
+
 /* Mandatory */
 static Bool
 NVProbe(DriverPtr drv, int flags)
@@ -523,7 +593,11 @@ NVProbe(DriverPtr drv, int flags)
            ((*ppPci)->vendor == PCI_VENDOR_NVIDIA)) 
         {
             SymTabRec *nvchips = NVKnownChipsets;
-            int token = ((*ppPci)->vendor << 16) | (*ppPci)->chipType;
+            int pciid = ((*ppPci)->vendor << 16) | (*ppPci)->chipType;
+            int token = pciid;
+
+            if((token & 0xfff0) == 0x00F0) 
+                token = NVGetPCIXpressChip(*ppPci);
 
             while(nvchips->name) {
                if(token == nvchips->token)
@@ -532,10 +606,10 @@ NVProbe(DriverPtr drv, int flags)
             }
 
             if(nvchips->name) { /* found one */
-               NVChipsets[numUsed].token = nvchips->token;
+               NVChipsets[numUsed].token = pciid;
                NVChipsets[numUsed].name = nvchips->name;
-               NVPciChipsets[numUsed].numChipset = nvchips->token;
-               NVPciChipsets[numUsed].PCIid = nvchips->token;
+               NVPciChipsets[numUsed].numChipset = pciid; 
+               NVPciChipsets[numUsed].PCIid = pciid;
                NVPciChipsets[numUsed].resList = RES_SHARED_VGA;
                numUsed++;
             } else if ((*ppPci)->vendor == PCI_VENDOR_NVIDIA) {
@@ -551,10 +625,19 @@ NVProbe(DriverPtr drv, int flags)
                case 0x0320:
                case 0x0330:
                case 0x0340:
-                   NVChipsets[numUsed].token = token;
+               case 0x0040:
+               case 0x00C0:
+               case 0x0120:
+               case 0x0140:
+               case 0x0160:
+               case 0x0130:
+               case 0x01D0:
+               case 0x0090:
+               case 0x0210:
+                   NVChipsets[numUsed].token = pciid;
                    NVChipsets[numUsed].name = "Unknown NVIDIA chip";
-                   NVPciChipsets[numUsed].numChipset = token;
-                   NVPciChipsets[numUsed].PCIid = token;
+                   NVPciChipsets[numUsed].numChipset = pciid;
+                   NVPciChipsets[numUsed].PCIid = pciid;
                    NVPciChipsets[numUsed].resList = RES_SHARED_VGA;
                    numUsed++;
                    break;
@@ -768,6 +851,17 @@ NVFreeScreen(int scrnIndex, int flags)
 static ModeStatus
 NVValidMode(int scrnIndex, DisplayModePtr mode, Bool verbose, int flags)
 {
+    NVPtr pNv = NVPTR(xf86Screens[scrnIndex]);
+
+    if(pNv->fpWidth && pNv->fpHeight) {
+      if((pNv->fpWidth < mode->HDisplay) || (pNv->fpHeight < mode->VDisplay)) {
+         xf86DrvMsg(scrnIndex, X_INFO, "Mode \"%s\" is larger than "
+                    "BIOS programmed panel size of %d x %d.  Removing.\n",
+                     mode->name, pNv->fpWidth, pNv->fpHeight);
+         return (MODE_BAD);
+      }
+    }
+
     return (MODE_OK);
 }
 
@@ -887,11 +981,16 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
     } else {
 	from = X_PROBED;
 	pNv->Chipset = (pNv->PciInfo->vendor << 16) | pNv->PciInfo->chipType;
+
+        if((pNv->Chipset & 0xfff0) == 0x00F0)
+            pNv->Chipset = NVGetPCIXpressChip(pNv->PciInfo);
+
 	pScrn->chipset = (char *)xf86TokenToString(NVKnownChipsets, 
                                                    pNv->Chipset);
         if(!pScrn->chipset)
           pScrn->chipset = "Unknown NVIDIA chipset";
     }
+
     if (pNv->pEnt->device->chipRev >= 0) {
 	pNv->ChipRev = pNv->pEnt->device->chipRev;
 	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "ChipRev override: %d\n",
@@ -1026,6 +1125,12 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
     }
     xf86DrvMsg(pScrn->scrnIndex, from, "Using %s cursor\n",
 		pNv->HWCursor ? "HW" : "SW");
+
+    pNv->FpScale = TRUE;
+    if (xf86GetOptValBool(pNv->Options, OPTION_FP_SCALE, &pNv->FpScale)) {
+        xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "Flat panel scaling %s\n",
+                   pNv->FpScale ? "on" : "off");
+    }
     if (xf86ReturnOptValBool(pNv->Options, OPTION_NOACCEL, FALSE)) {
 	pNv->NoAccel = TRUE;
 	xf86DrvMsg(pScrn->scrnIndex, X_CONFIG, "Acceleration disabled\n");
@@ -1178,12 +1283,46 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
 	return FALSE;
     }
 
-    pNv->alphaCursor = ((pNv->Chipset & 0x0ff0) >= 0x0110);
-
-    pNv->Architecture = (pNv->Chipset & 0x0f00) >> 4;
+    switch (pNv->Chipset & 0x0ff0) {
+    case 0x0100:   /* GeForce 256 */
+    case 0x0110:   /* GeForce2 MX */
+    case 0x0150:   /* GeForce2 */
+    case 0x0170:   /* GeForce4 MX */
+    case 0x0180:   /* GeForce4 MX (8x AGP) */
+    case 0x01A0:   /* nForce */
+    case 0x01F0:   /* nForce2 */
+         pNv->Architecture =  NV_ARCH_10;
+         break;
+    case 0x0200:   /* GeForce3 */
+    case 0x0250:   /* GeForce4 Ti */
+    case 0x0280:   /* GeForce4 Ti (8x AGP) */
+         pNv->Architecture =  NV_ARCH_20;
+         break;
+    case 0x0300:   /* GeForceFX 5800 */
+    case 0x0310:   /* GeForceFX 5600 */
+    case 0x0320:   /* GeForceFX 5200 */
+    case 0x0330:   /* GeForceFX 5900 */
+    case 0x0340:   /* GeForceFX 5700 */
+         pNv->Architecture =  NV_ARCH_30;
+         break;
+    case 0x0040:
+    case 0x00C0:
+    case 0x0120:
+    case 0x0130:
+    case 0x0140:
+    case 0x0160:
+    case 0x01D0:
+    case 0x0090:
+    case 0x0210:
+         pNv->Architecture =  NV_ARCH_40;
+         break;
+    default:
+         pNv->Architecture =  NV_ARCH_04;
+         break;
+    }
 
-    if(pNv->Architecture < NV_ARCH_10)
-        pNv->Architecture = NV_ARCH_04;
+    pNv->alphaCursor = (pNv->Architecture >= NV_ARCH_10) &&
+                       ((pNv->Chipset & 0x0ff0) != 0x0100);
 
     NVCommonSetup(pScrn);
 
@@ -1225,17 +1364,17 @@ NVPreInit(ScrnInfoPtr pScrn, int flags)
     clockRanges->minClock = pNv->MinVClockFreqKHz;
     clockRanges->maxClock = pNv->MaxVClockFreqKHz;
     clockRanges->clockIndex = -1;		/* programmable */
-    if(((pNv->Chipset & 0x0ff0) <= 0x0100) ||
-       ((pNv->Chipset & 0x0ff0) == 0x0150) ||
-       ((pNv->Chipset & 0x0ff0) >= 0x0300))
+    clockRanges->doubleScanAllowed = TRUE;
+    if((pNv->Architecture == NV_ARCH_20) ||
+         ((pNv->Architecture == NV_ARCH_10) && 
+           ((pNv->Chipset & 0x0ff0) != 0x0100) &&
+           ((pNv->Chipset & 0x0ff0) != 0x0150)))
     {
-       clockRanges->interlaceAllowed = TRUE;
-    } else {
-        /* No NV2x chips support interlaced modes and the only
-           NV1x chips that do are NV10 and NV15 */
+       /* HW is broken */
        clockRanges->interlaceAllowed = FALSE;
+    } else {
+       clockRanges->interlaceAllowed = TRUE;
     }
-    clockRanges->doubleScanAllowed = TRUE;
 
     if(pNv->FlatPanel == 1) {
        clockRanges->interlaceAllowed = FALSE;
@@ -1488,22 +1627,28 @@ NVRestore(ScrnInfoPtr pScrn)
     NVPtr pNv = NVPTR(pScrn);
     NVRegPtr nvReg = &pNv->SavedReg;
 
+    NVLockUnlock(pNv, 0);
+
     if(pNv->twoHeads) {
         VGA_WR08(pNv->PCIO, 0x03D4, 0x44);
-        VGA_WR08(pNv->PCIO, 0x03D5, nvReg->crtcOwner);
+        VGA_WR08(pNv->PCIO, 0x03D5, pNv->CRTCnumber * 0x3);
         NVLockUnlock(pNv, 0);
     }
 
-    NVLockUnlock(pNv, 0);
-
     /* Only restore text mode fonts/text for the primary card */
     vgaHWProtect(pScrn, TRUE);
     NVDACRestore(pScrn, vgaReg, nvReg, pNv->Primary);
+    if(pNv->twoHeads) {
+        VGA_WR08(pNv->PCIO, 0x03D4, 0x44);
+        VGA_WR08(pNv->PCIO, 0x03D5, nvReg->crtcOwner);
+    }
     vgaHWProtect(pScrn, FALSE);
 }
 
 static void NVBacklightEnable(NVPtr pNv,  Bool on)
 {
+    CARD32 fpcontrol = pNv->PRAMDAC[0x0848/4] & 0xCfffffCC;
+
     /* This is done differently on each laptop.  Here we
        define the ones we know for sure. */
 
@@ -1524,6 +1669,12 @@ static void NVBacklightEnable(NVPtr pNv,  Bool on)
       pNv->PCRTC0[0x081C/4] = tmp_pcrt;
     }
 #endif
+    
+    /* cut the TMDS output */
+    if(on) fpcontrol |= pNv->fpSyncs;
+    else fpcontrol |= 0x20000022;
+
+    pNv->PRAMDAC[0x0848/4] = fpcontrol;
 }
 
 static void
@@ -1533,6 +1684,8 @@ NVDPMSSetLCD(ScrnInfoPtr pScrn, int PowerManagementMode, int flags)
 
   if (!pScrn->vtSema) return;
 
+  vgaHWDPMSSet(pScrn, PowerManagementMode, flags);
+
   switch (PowerManagementMode) {
   case DPMSModeStandby:  /* HSync: Off, VSync: On */
   case DPMSModeSuspend:  /* HSync: On, VSync: Off */
@@ -1544,7 +1697,6 @@ NVDPMSSetLCD(ScrnInfoPtr pScrn, int PowerManagementMode, int flags)
   default:
     break;
   }
-  vgaHWDPMSSet(pScrn, PowerManagementMode, flags);
 }
 
 
@@ -1646,7 +1798,7 @@ NVScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
      * function.  If not, the visuals will need to be setup before calling
      * a fb ScreenInit() function and fixed up after.
      *
-     * For most PC hardware at depths >= 8, the defaults that cfb uses
+     * For most PC hardware at depths >= 8, the defaults that fb uses
      * are not appropriate.  In this driver, we fixup the visuals after.
      */
 
@@ -1830,6 +1982,13 @@ NVSave(ScrnInfoPtr pScrn)
     vgaHWPtr pVga = VGAHWPTR(pScrn);
     vgaRegPtr vgaReg = &pVga->SavedReg;
 
+    NVLockUnlock(pNv, 0);
+    if(pNv->twoHeads) {
+        VGA_WR08(pNv->PCIO, 0x03D4, 0x44);
+        VGA_WR08(pNv->PCIO, 0x03D5, pNv->CRTCnumber * 0x3);
+        NVLockUnlock(pNv, 0);
+    }
+
     NVDACSave(pScrn, vgaReg, nvReg, pNv->Primary);
 }
 
diff --git a/src/nv_hw.c b/src/nv_hw.c
index 67e3b5b..25b10d2 100644
--- a/src/nv_hw.c
+++ b/src/nv_hw.c
@@ -36,7 +36,7 @@
 |*     those rights set forth herein.                                        *|
 |*                                                                           *|
  \***************************************************************************/
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_hw.c,v 1.4 2003/11/03 05:11:25 tsi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_hw.c,v 1.11 2004/10/15 20:32:21 mvojkovi Exp $ */
 
 #include "nv_local.h"
 #include "compiler.h"
@@ -71,6 +71,12 @@ int NVShowHideCursor (
                                  (ShowHide & 0x01);
     VGA_WR08(pNv->PCIO, 0x3D4, 0x31);
     VGA_WR08(pNv->PCIO, 0x3D5, pNv->CurrentState->cursor1);
+
+    if(pNv->Architecture == NV_ARCH_40) {  /* HW bug */
+       volatile CARD32 curpos = pNv->PRAMDAC[0x0300/4];
+       pNv->PRAMDAC[0x0300/4] = curpos;
+    }
+
     return (current & 0x01);
 }
 
@@ -132,6 +138,26 @@ static void nvGetClocks(NVPtr pNv, unsigned int *MClk, unsigned int *NVClk)
 {
     unsigned int pll, N, M, MB, NB, P;
 
+    if(pNv->Architecture >= NV_ARCH_40) {
+       pll = pNv->PMC[0x4020/4];
+       P = (pll >> 16) & 0x03;
+       pll = pNv->PMC[0x4024/4];
+       M = pll & 0xFF;
+       N = (pll >> 8) & 0xFF;
+       MB = (pll >> 16) & 0xFF;
+       NB = (pll >> 24) & 0xFF;
+       *MClk = ((N * NB * pNv->CrystalFreqKHz) / (M * MB)) >> P;
+
+       pll = pNv->PMC[0x4000/4];
+       P = (pll >> 16) & 0x03;  
+       pll = pNv->PMC[0x4004/4];
+       M = pll & 0xFF;
+       N = (pll >> 8) & 0xFF;
+       MB = (pll >> 16) & 0xFF;
+       NB = (pll >> 24) & 0xFF;
+
+       *NVClk = ((N * NB * pNv->CrystalFreqKHz) / (M * MB)) >> P;
+    } else
     if(pNv->twoStagePLL) {
        pll = pNv->PRAMDAC0[0x0504/4];
        M = pll & 0xFF; 
@@ -202,6 +228,10 @@ static void nvGetClocks(NVPtr pNv, unsigned int *MClk, unsigned int *NVClk)
        P = (pll >> 16) & 0x0F;
        *NVClk = (N * pNv->CrystalFreqKHz / M) >> P;
     }
+
+#if 0
+    ErrorF("NVClock = %i MHz, MEMClock = %i MHz\n", *NVClk/1000, *MClk/1000);
+#endif
 }
 
 
@@ -632,12 +662,17 @@ static void nForceUpdateArbitrationSettings (
 {
     nv10_fifo_info fifo_data;
     nv10_sim_state sim_data;
-    unsigned int M, N, P, pll, MClk, NVClk;
-    unsigned int uMClkPostDiv, memctrl;
+    unsigned int M, N, P, pll, MClk, NVClk, memctrl;
 
-    uMClkPostDiv = (pciReadLong(pciTag(0, 0, 3), 0x6C) >> 8) & 0xf;
-    if(!uMClkPostDiv) uMClkPostDiv = 4; 
-    MClk = 400000 / uMClkPostDiv;
+    if((pNv->Chipset & 0x0FF0) == 0x01A0) {
+       unsigned int uMClkPostDiv;
+
+       uMClkPostDiv = (pciReadLong(pciTag(0, 0, 3), 0x6C) >> 8) & 0xf;
+       if(!uMClkPostDiv) uMClkPostDiv = 4; 
+       MClk = 400000 / uMClkPostDiv;
+    } else {
+       MClk = pciReadLong(pciTag(0, 0, 5), 0x4C) / 1000;
+    }
 
     pll = pNv->PRAMDAC0[0x0500/4];
     M = (pll >> 0)  & 0xFF; N = (pll >> 8)  & 0xFF; P = (pll >> 16) & 0x0F;
@@ -901,79 +936,167 @@ void NVLoadStateExt (
         pNv->PFB[0x02B4/4] = pNv->FbMapSize - 1;
     }
 
-    pNv->PRAMIN[0x0000] = 0x80000010;
-    pNv->PRAMIN[0x0001] = 0x80011201;  
-    pNv->PRAMIN[0x0002] = 0x80000011;
-    pNv->PRAMIN[0x0003] = 0x80011202; 
-    pNv->PRAMIN[0x0004] = 0x80000012;
-    pNv->PRAMIN[0x0005] = 0x80011203;
-    pNv->PRAMIN[0x0006] = 0x80000013;
-    pNv->PRAMIN[0x0007] = 0x80011204;
-    pNv->PRAMIN[0x0008] = 0x80000014;
-    pNv->PRAMIN[0x0009] = 0x80011205;
-    pNv->PRAMIN[0x000A] = 0x80000015;
-    pNv->PRAMIN[0x000B] = 0x80011206;
-    pNv->PRAMIN[0x000C] = 0x80000016;
-    pNv->PRAMIN[0x000D] = 0x80011207;
-    pNv->PRAMIN[0x000E] = 0x80000017;
-    pNv->PRAMIN[0x000F] = 0x80011208;
-    pNv->PRAMIN[0x0800] = 0x00003000;
-    pNv->PRAMIN[0x0801] = pNv->FbMapSize - 1;
-    pNv->PRAMIN[0x0802] = 0x00000002;
-    pNv->PRAMIN[0x0803] = 0x00000002;
-    if(pNv->Architecture >= NV_ARCH_10)
-       pNv->PRAMIN[0x0804] = 0x01008062;
-    else
-       pNv->PRAMIN[0x0804] = 0x01008042;
-    pNv->PRAMIN[0x0805] = 0x00000000;
-    pNv->PRAMIN[0x0806] = 0x12001200;
-    pNv->PRAMIN[0x0807] = 0x00000000;
-    pNv->PRAMIN[0x0808] = 0x01008043;
-    pNv->PRAMIN[0x0809] = 0x00000000;
-    pNv->PRAMIN[0x080A] = 0x00000000;
-    pNv->PRAMIN[0x080B] = 0x00000000;
-    pNv->PRAMIN[0x080C] = 0x01008044;
-    pNv->PRAMIN[0x080D] = 0x00000002;
-    pNv->PRAMIN[0x080E] = 0x00000000;
-    pNv->PRAMIN[0x080F] = 0x00000000;
-    pNv->PRAMIN[0x0810] = 0x01008019;
-    pNv->PRAMIN[0x0811] = 0x00000000;
-    pNv->PRAMIN[0x0812] = 0x00000000;
-    pNv->PRAMIN[0x0813] = 0x00000000;
-    pNv->PRAMIN[0x0814] = 0x0100A05C;
-    pNv->PRAMIN[0x0815] = 0x00000000;
-    pNv->PRAMIN[0x0816] = 0x00000000;
-    pNv->PRAMIN[0x0817] = 0x00000000;
-    pNv->PRAMIN[0x0818] = 0x0100805F;
-    pNv->PRAMIN[0x0819] = 0x00000000;
-    pNv->PRAMIN[0x081A] = 0x12001200;
-    pNv->PRAMIN[0x081B] = 0x00000000;
-    pNv->PRAMIN[0x081C] = 0x0100804A;
-    pNv->PRAMIN[0x081D] = 0x00000002;
-    pNv->PRAMIN[0x081E] = 0x00000000;
-    pNv->PRAMIN[0x081F] = 0x00000000;
-    pNv->PRAMIN[0x0820] = 0x01018077;
-    pNv->PRAMIN[0x0821] = 0x00000000;
-    pNv->PRAMIN[0x0822] = 0x01201200;
-    pNv->PRAMIN[0x0823] = 0x00000000;
-    pNv->PRAMIN[0x0824] = 0x00003002;
-    pNv->PRAMIN[0x0825] = 0x00007FFF;
-    pNv->PRAMIN[0x0826] = pNv->FbUsableSize | 0x00000002;
-    pNv->PRAMIN[0x0827] = 0x00000002;
+    if(pNv->Architecture >= NV_ARCH_40) {
+       pNv->PRAMIN[0x0000] = 0x80000010;
+       pNv->PRAMIN[0x0001] = 0x00101202;
+       pNv->PRAMIN[0x0002] = 0x80000011;
+       pNv->PRAMIN[0x0003] = 0x00101204;
+       pNv->PRAMIN[0x0004] = 0x80000012;
+       pNv->PRAMIN[0x0005] = 0x00101206;
+       pNv->PRAMIN[0x0006] = 0x80000013;
+       pNv->PRAMIN[0x0007] = 0x00101208;
+       pNv->PRAMIN[0x0008] = 0x80000014;
+       pNv->PRAMIN[0x0009] = 0x0010120A;
+       pNv->PRAMIN[0x000A] = 0x80000015;
+       pNv->PRAMIN[0x000B] = 0x0010120C;
+       pNv->PRAMIN[0x000C] = 0x80000016;
+       pNv->PRAMIN[0x000D] = 0x0010120E;
+       pNv->PRAMIN[0x000E] = 0x80000017;
+       pNv->PRAMIN[0x000F] = 0x00101210;
+       pNv->PRAMIN[0x0800] = 0x00003000;
+       pNv->PRAMIN[0x0801] = pNv->FbMapSize - 1;
+       pNv->PRAMIN[0x0802] = 0x00000002;
+       pNv->PRAMIN[0x0808] = 0x02080062;
+       pNv->PRAMIN[0x0809] = 0x00000000;
+       pNv->PRAMIN[0x080A] = 0x00001200;
+       pNv->PRAMIN[0x080B] = 0x00001200;
+       pNv->PRAMIN[0x080C] = 0x00000000;
+       pNv->PRAMIN[0x080D] = 0x00000000;
+       pNv->PRAMIN[0x0810] = 0x02080043;
+       pNv->PRAMIN[0x0811] = 0x00000000;
+       pNv->PRAMIN[0x0812] = 0x00000000;
+       pNv->PRAMIN[0x0813] = 0x00000000;
+       pNv->PRAMIN[0x0814] = 0x00000000;
+       pNv->PRAMIN[0x0815] = 0x00000000;
+       pNv->PRAMIN[0x0818] = 0x02080044;
+       pNv->PRAMIN[0x0819] = 0x02000000;
+       pNv->PRAMIN[0x081A] = 0x00000000;
+       pNv->PRAMIN[0x081B] = 0x00000000;
+       pNv->PRAMIN[0x081C] = 0x00000000;
+       pNv->PRAMIN[0x081D] = 0x00000000;
+       pNv->PRAMIN[0x0820] = 0x02080019;
+       pNv->PRAMIN[0x0821] = 0x00000000;
+       pNv->PRAMIN[0x0822] = 0x00000000;
+       pNv->PRAMIN[0x0823] = 0x00000000;
+       pNv->PRAMIN[0x0824] = 0x00000000;
+       pNv->PRAMIN[0x0825] = 0x00000000;
+       pNv->PRAMIN[0x0828] = 0x020A005C;
+       pNv->PRAMIN[0x0829] = 0x00000000;
+       pNv->PRAMIN[0x082A] = 0x00000000;
+       pNv->PRAMIN[0x082B] = 0x00000000;
+       pNv->PRAMIN[0x082C] = 0x00000000;
+       pNv->PRAMIN[0x082D] = 0x00000000;
+       pNv->PRAMIN[0x0830] = 0x0208009F;
+       pNv->PRAMIN[0x0831] = 0x00000000;
+       pNv->PRAMIN[0x0832] = 0x00001200;
+       pNv->PRAMIN[0x0833] = 0x00001200;
+       pNv->PRAMIN[0x0834] = 0x00000000;
+       pNv->PRAMIN[0x0835] = 0x00000000;
+       pNv->PRAMIN[0x0838] = 0x0208004A;
+       pNv->PRAMIN[0x0839] = 0x02000000;
+       pNv->PRAMIN[0x083A] = 0x00000000;
+       pNv->PRAMIN[0x083B] = 0x00000000;
+       pNv->PRAMIN[0x083C] = 0x00000000;
+       pNv->PRAMIN[0x083D] = 0x00000000;
+       pNv->PRAMIN[0x0840] = 0x02080077;
+       pNv->PRAMIN[0x0841] = 0x00000000;
+       pNv->PRAMIN[0x0842] = 0x00001200;
+       pNv->PRAMIN[0x0843] = 0x00001200;
+       pNv->PRAMIN[0x0844] = 0x00000000;
+       pNv->PRAMIN[0x0845] = 0x00000000;
+       pNv->PRAMIN[0x084C] = 0x00003002;
+       pNv->PRAMIN[0x084D] = 0x00007FFF;
+       pNv->PRAMIN[0x084E] = pNv->FbUsableSize | 0x00000002;
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
-    pNv->PRAMIN[0x0804] |= 0x00080000;
-    pNv->PRAMIN[0x0808] |= 0x00080000;
-    pNv->PRAMIN[0x080C] |= 0x00080000;
-    pNv->PRAMIN[0x0810] |= 0x00080000;
-    pNv->PRAMIN[0x0814] |= 0x00080000;
-    pNv->PRAMIN[0x0818] |= 0x00080000;
-    pNv->PRAMIN[0x081C] |= 0x00080000;
-    pNv->PRAMIN[0x0820] |= 0x00080000;
-
-    pNv->PRAMIN[0x080D] = 0x00000001;
-    pNv->PRAMIN[0x081D] = 0x00000001;
+       pNv->PRAMIN[0x080A] |= 0x01000000;
+       pNv->PRAMIN[0x0812] |= 0x01000000;
+       pNv->PRAMIN[0x081A] |= 0x01000000;
+       pNv->PRAMIN[0x0822] |= 0x01000000;
+       pNv->PRAMIN[0x082A] |= 0x01000000;
+       pNv->PRAMIN[0x0832] |= 0x01000000;
+       pNv->PRAMIN[0x083A] |= 0x01000000;
+       pNv->PRAMIN[0x0842] |= 0x01000000;  
+       pNv->PRAMIN[0x0819] = 0x01000000;
+       pNv->PRAMIN[0x0839] = 0x01000000;
 #endif
+    } else {
+       pNv->PRAMIN[0x0000] = 0x80000010;
+       pNv->PRAMIN[0x0001] = 0x80011201;  
+       pNv->PRAMIN[0x0002] = 0x80000011;
+       pNv->PRAMIN[0x0003] = 0x80011202; 
+       pNv->PRAMIN[0x0004] = 0x80000012;
+       pNv->PRAMIN[0x0005] = 0x80011203;
+       pNv->PRAMIN[0x0006] = 0x80000013;
+       pNv->PRAMIN[0x0007] = 0x80011204;
+       pNv->PRAMIN[0x0008] = 0x80000014;
+       pNv->PRAMIN[0x0009] = 0x80011205;
+       pNv->PRAMIN[0x000A] = 0x80000015;
+       pNv->PRAMIN[0x000B] = 0x80011206;
+       pNv->PRAMIN[0x000C] = 0x80000016;
+       pNv->PRAMIN[0x000D] = 0x80011207;
+       pNv->PRAMIN[0x000E] = 0x80000017;
+       pNv->PRAMIN[0x000F] = 0x80011208;
+       pNv->PRAMIN[0x0800] = 0x00003000;
+       pNv->PRAMIN[0x0801] = pNv->FbMapSize - 1;
+       pNv->PRAMIN[0x0802] = 0x00000002;
+       pNv->PRAMIN[0x0803] = 0x00000002;
+       if(pNv->Architecture >= NV_ARCH_10)
+          pNv->PRAMIN[0x0804] = 0x01008062;
+       else
+          pNv->PRAMIN[0x0804] = 0x01008042;
+       pNv->PRAMIN[0x0805] = 0x00000000;
+       pNv->PRAMIN[0x0806] = 0x12001200;
+       pNv->PRAMIN[0x0807] = 0x00000000;
+       pNv->PRAMIN[0x0808] = 0x01008043;
+       pNv->PRAMIN[0x0809] = 0x00000000;
+       pNv->PRAMIN[0x080A] = 0x00000000;
+       pNv->PRAMIN[0x080B] = 0x00000000;
+       pNv->PRAMIN[0x080C] = 0x01008044;
+       pNv->PRAMIN[0x080D] = 0x00000002;
+       pNv->PRAMIN[0x080E] = 0x00000000;
+       pNv->PRAMIN[0x080F] = 0x00000000;
+       pNv->PRAMIN[0x0810] = 0x01008019;
+       pNv->PRAMIN[0x0811] = 0x00000000;
+       pNv->PRAMIN[0x0812] = 0x00000000;
+       pNv->PRAMIN[0x0813] = 0x00000000;
+       pNv->PRAMIN[0x0814] = 0x0100A05C;
+       pNv->PRAMIN[0x0815] = 0x00000000;
+       pNv->PRAMIN[0x0816] = 0x00000000;
+       pNv->PRAMIN[0x0817] = 0x00000000;
+       if(pNv->WaitVSyncPossible)
+          pNv->PRAMIN[0x0818] = 0x0100809F;
+       else
+          pNv->PRAMIN[0x0818] = 0x0100805F;
+       pNv->PRAMIN[0x0819] = 0x00000000;
+       pNv->PRAMIN[0x081A] = 0x12001200;
+       pNv->PRAMIN[0x081B] = 0x00000000;
+       pNv->PRAMIN[0x081C] = 0x0100804A;
+       pNv->PRAMIN[0x081D] = 0x00000002;
+       pNv->PRAMIN[0x081E] = 0x00000000;
+       pNv->PRAMIN[0x081F] = 0x00000000;
+       pNv->PRAMIN[0x0820] = 0x01018077;
+       pNv->PRAMIN[0x0821] = 0x00000000;
+       pNv->PRAMIN[0x0822] = 0x12001200;
+       pNv->PRAMIN[0x0823] = 0x00000000;
+       pNv->PRAMIN[0x0824] = 0x00003002;
+       pNv->PRAMIN[0x0825] = 0x00007FFF;
+       pNv->PRAMIN[0x0826] = pNv->FbUsableSize | 0x00000002;
+       pNv->PRAMIN[0x0827] = 0x00000002;
+
+#if X_BYTE_ORDER == X_BIG_ENDIAN
+       pNv->PRAMIN[0x0804] |= 0x00080000;
+       pNv->PRAMIN[0x0808] |= 0x00080000;
+       pNv->PRAMIN[0x080C] |= 0x00080000;
+       pNv->PRAMIN[0x0810] |= 0x00080000;
+       pNv->PRAMIN[0x0814] |= 0x00080000;
+       pNv->PRAMIN[0x0818] |= 0x00080000;
+       pNv->PRAMIN[0x081C] |= 0x00080000;
+       pNv->PRAMIN[0x0820] |= 0x00080000;
+       pNv->PRAMIN[0x080D] = 0x00000001;
+       pNv->PRAMIN[0x081D] = 0x00000001;
+#endif
+    }
 
     if(pNv->Architecture < NV_ARCH_10) {
        if((pNv->Chipset & 0x0fff) == 0x0020) {
@@ -985,6 +1108,7 @@ void NVLoadStateExt (
        pNv->PGRAPH[0x0084/4] = 0x72111101;
        pNv->PGRAPH[0x0088/4] = 0x11D5F071;
        pNv->PGRAPH[0x008C/4] = 0x0004FF31;
+       pNv->PGRAPH[0x008C/4] = 0x4004FF31;
 
        pNv->PGRAPH[0x0140/4] = 0x00000000;
        pNv->PGRAPH[0x0100/4] = 0xFFFFFFFF;
@@ -993,6 +1117,7 @@ void NVLoadStateExt (
        pNv->PGRAPH[0x0720/4] = 0x00000001;
 
        pNv->PGRAPH[0x0810/4] = 0x00000000;
+       pNv->PGRAPH[0x0608/4] = 0xFFFFFFFF; 
     } else {
        pNv->PGRAPH[0x0080/4] = 0xFFFFFFFF;
        pNv->PGRAPH[0x0080/4] = 0x00000000;
@@ -1002,6 +1127,8 @@ void NVLoadStateExt (
        pNv->PGRAPH[0x0144/4] = 0x10010100;
        pNv->PGRAPH[0x0714/4] = 0xFFFFFFFF;
        pNv->PGRAPH[0x0720/4] = 0x00000001;
+       pNv->PGRAPH[0x0710/4] &= 0x0007ff00;
+       pNv->PGRAPH[0x0710/4] |= 0x00020100;
 
        if(pNv->Architecture == NV_ARCH_10) {
            pNv->PGRAPH[0x0084/4] = 0x00118700;
@@ -1017,8 +1144,54 @@ void NVLoadStateExt (
            pNv->PGRAPH[0x688/4] = pNv->FbMapSize - 1;
 
            pNv->PGRAPH[0x0810/4] = 0x00000000;
+           pNv->PGRAPH[0x0608/4] = 0xFFFFFFFF;
        } else {
-           if(pNv->Architecture >= NV_ARCH_30) {
+           if(pNv->Architecture >= NV_ARCH_40) {
+              pNv->PGRAPH[0x0084/4] = 0x401287c0;
+              pNv->PGRAPH[0x008C/4] = 0x60de8051;
+              pNv->PGRAPH[0x0090/4] = 0x00008000;
+              pNv->PGRAPH[0x0610/4] = 0x00be3c5f;
+
+              if((pNv->Chipset & 0xfff0) == 0x0040) {
+                 pNv->PGRAPH[0x09b0/4] = 0x83280fff;
+                 pNv->PGRAPH[0x09b4/4] = 0x000000a0;
+              } else {
+                 pNv->PGRAPH[0x0820/4] = 0x83280eff;
+                 pNv->PGRAPH[0x0824/4] = 0x000000a0;
+              }
+
+              switch(pNv->Chipset & 0xfff0) {
+              case 0x0040:
+                 pNv->PGRAPH[0x09b8/4] = 0x0078e366;
+                 pNv->PGRAPH[0x09bc/4] = 0x0000014c;
+                 break;
+              case 0x00C0:
+                 pNv->PGRAPH[0x0828/4] = 0x007596ff;
+                 pNv->PGRAPH[0x082C/4] = 0x00000108;
+                 break;
+              case 0x0160:
+                 pNv->PMC[0x1700/4] = pNv->PFB[0x020C/4];
+                 pNv->PMC[0x1704/4] = 0;
+                 pNv->PMC[0x1708/4] = 0;
+                 pNv->PMC[0x170C/4] = pNv->PFB[0x020C/4];
+                 pNv->PGRAPH[0x0860/4] = 0;
+                 pNv->PGRAPH[0x0864/4] = 0;
+                 pNv->PRAMDAC[0x0608/4] |= 0x00100000;
+                 break;
+              case 0x0140:
+                 pNv->PGRAPH[0x0828/4] = 0x0072cb77;
+                 pNv->PGRAPH[0x082C/4] = 0x00000108;
+                 break;
+              default:
+                 break;
+              };
+
+              pNv->PGRAPH[0x0b38/4] = 0x2ffff800;
+              pNv->PGRAPH[0x0b3c/4] = 0x00006000;
+              pNv->PGRAPH[0x032C/4] = 0x01000000; 
+              pNv->PGRAPH[0x0220/4] = 0x00001200;
+           } else
+           if(pNv->Architecture == NV_ARCH_30) {
               pNv->PGRAPH[0x0084/4] = 0x40108700;
               pNv->PGRAPH[0x0890/4] = 0x00140000;
               pNv->PGRAPH[0x008C/4] = 0xf00e0431;
@@ -1055,19 +1228,44 @@ void NVLoadStateExt (
            for(i = 0; i < 32; i++)
              pNv->PGRAPH[(0x0900/4) + i] = pNv->PFB[(0x0240/4) + i];
 
-           pNv->PGRAPH[0x09A4/4] = pNv->PFB[0x0200/4];
-           pNv->PGRAPH[0x09A8/4] = pNv->PFB[0x0204/4];
-           pNv->PGRAPH[0x0750/4] = 0x00EA0000;
-           pNv->PGRAPH[0x0754/4] = pNv->PFB[0x0200/4];
-           pNv->PGRAPH[0x0750/4] = 0x00EA0004;
-           pNv->PGRAPH[0x0754/4] = pNv->PFB[0x0204/4];
-
-           pNv->PGRAPH[0x0820/4] = 0;
-           pNv->PGRAPH[0x0824/4] = 0;
-           pNv->PGRAPH[0x0864/4] = pNv->FbMapSize - 1;
-           pNv->PGRAPH[0x0868/4] = pNv->FbMapSize - 1;
+           if(pNv->Architecture >= NV_ARCH_40) {
+              if((pNv->Chipset & 0xfff0) == 0x0040) {
+                 pNv->PGRAPH[0x09A4/4] = pNv->PFB[0x0200/4];
+                 pNv->PGRAPH[0x09A8/4] = pNv->PFB[0x0204/4];
+                 pNv->PGRAPH[0x69A4/4] = pNv->PFB[0x0200/4];
+                 pNv->PGRAPH[0x69A8/4] = pNv->PFB[0x0204/4];
+
+                 pNv->PGRAPH[0x0820/4] = 0;
+                 pNv->PGRAPH[0x0824/4] = 0;
+                 pNv->PGRAPH[0x0864/4] = pNv->FbMapSize - 1;
+                 pNv->PGRAPH[0x0868/4] = pNv->FbMapSize - 1;
+              } else {
+                 pNv->PGRAPH[0x09F0/4] = pNv->PFB[0x0200/4];
+                 pNv->PGRAPH[0x09F4/4] = pNv->PFB[0x0204/4];
+                 pNv->PGRAPH[0x69F0/4] = pNv->PFB[0x0200/4];
+                 pNv->PGRAPH[0x69F4/4] = pNv->PFB[0x0204/4];
+
+                 pNv->PGRAPH[0x0840/4] = 0;
+                 pNv->PGRAPH[0x0844/4] = 0;
+                 pNv->PGRAPH[0x08a0/4] = pNv->FbMapSize - 1;
+                 pNv->PGRAPH[0x08a4/4] = pNv->FbMapSize - 1;
+              }
+           } else {
+              pNv->PGRAPH[0x09A4/4] = pNv->PFB[0x0200/4];
+              pNv->PGRAPH[0x09A8/4] = pNv->PFB[0x0204/4];
+              pNv->PGRAPH[0x0750/4] = 0x00EA0000;
+              pNv->PGRAPH[0x0754/4] = pNv->PFB[0x0200/4];
+              pNv->PGRAPH[0x0750/4] = 0x00EA0004;
+              pNv->PGRAPH[0x0754/4] = pNv->PFB[0x0204/4];
+
+              pNv->PGRAPH[0x0820/4] = 0;
+              pNv->PGRAPH[0x0824/4] = 0;
+              pNv->PGRAPH[0x0864/4] = pNv->FbMapSize - 1;
+              pNv->PGRAPH[0x0868/4] = pNv->FbMapSize - 1;
+           }
 
            pNv->PGRAPH[0x0B20/4] = 0x00000000;
+           pNv->PGRAPH[0x0B04/4] = 0xFFFFFFFF;
        }
     }
     pNv->PGRAPH[0x053C/4] = 0;
@@ -1079,10 +1277,16 @@ void NVLoadStateExt (
     pNv->PFIFO[0x0141] = 0x00000001;
     pNv->PFIFO[0x0480] = 0x00000000;
     pNv->PFIFO[0x0494] = 0x00000000;
-    pNv->PFIFO[0x0481] = 0x00000100;
+    if(pNv->Architecture >= NV_ARCH_40)
+       pNv->PFIFO[0x0481] = 0x00010000;
+    else
+       pNv->PFIFO[0x0481] = 0x00000100;
     pNv->PFIFO[0x0490] = 0x00000000;
     pNv->PFIFO[0x0491] = 0x00000000;
-    pNv->PFIFO[0x048B] = 0x00001209;
+    if(pNv->Architecture >= NV_ARCH_40)
+       pNv->PFIFO[0x048B] = 0x00001213;
+    else
+       pNv->PFIFO[0x048B] = 0x00001209;
     pNv->PFIFO[0x0400] = 0x00000000;
     pNv->PFIFO[0x0414] = 0x00000000;
     pNv->PFIFO[0x0084] = 0x03000100;
@@ -1122,12 +1326,14 @@ void NVLoadStateExt (
         pNv->PMC[0x1588/4] = 0;
 
         pNv->PCRTC[0x0810/4] = state->cursorConfig;
+        pNv->PCRTC[0x0830/4] = state->displayV - 3;
+        pNv->PCRTC[0x0834/4] = state->displayV - 1;
     
         if(pNv->FlatPanel) {
            if((pNv->Chipset & 0x0ff0) == 0x0110) {
                pNv->PRAMDAC[0x0528/4] = state->dither;
            } else 
-           if((pNv->Chipset & 0x0ff0) >= 0x0170) {
+           if(pNv->twoHeads) {
                pNv->PRAMDAC[0x083C/4] = state->dither;
            }
     
@@ -1177,6 +1383,23 @@ void NVLoadStateExt (
        }
     } else {
        pNv->PRAMDAC[0x0848/4] = state->scale;
+
+       /* begin flat panel hacks */
+       /* This is unfortunate, but some chips need this register
+          tweaked or else you get artifacts where adjacent pixels are
+          swapped.  There are no hard rules for what to set here so all
+          we can do is experiment and apply hacks. */
+
+       if(((pNv->Chipset & 0xffff) == 0x0328) && (state->bpp == 32)) {
+          /* At least one NV34 laptop needs this workaround. */
+          pNv->PRAMDAC[0x0828/4] &= ~1;
+       }
+
+       if((pNv->Chipset & 0xfff0) == 0x0310) {
+          pNv->PRAMDAC[0x0828/4] |= 1;
+       }
+
+       /* end flat panel hacks */
     }
     pNv->PRAMDAC[0x0600/4] = state->general;
 
@@ -1240,7 +1463,7 @@ void NVUnloadStateExt
         if((pNv->Chipset & 0x0ff0) == 0x0110) {
            state->dither = pNv->PRAMDAC[0x0528/4];
         } else 
-        if((pNv->Chipset & 0x0ff0) >= 0x0170) {
+        if(pNv->twoHeads) {
             state->dither = pNv->PRAMDAC[0x083C/4];
         }
 
diff --git a/src/nv_include.h b/src/nv_include.h
index 2d4465b..acd3543 100644
--- a/src/nv_include.h
+++ b/src/nv_include.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_include.h,v 1.11 2001/12/07 00:09:56 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_include.h,v 1.9 2000/10/06 12:31:03 eich Exp $ */
 
 #ifndef __NV_INCLUDE_H__
 #define __NV_INCLUDE_H__
diff --git a/src/nv_local.h b/src/nv_local.h
index dc29a6f..ed58ef4 100644
--- a/src/nv_local.h
+++ b/src/nv_local.h
@@ -36,7 +36,7 @@
 |*     those rights set forth herein.                                        *|
 |*                                                                           *|
  \***************************************************************************/
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_local.h,v 1.8 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_local.h,v 1.8tsi Exp $ */
 
 #ifndef __NV_LOCAL_H__
 #define __NV_LOCAL_H__
@@ -60,12 +60,12 @@ typedef unsigned int   U032;
 /*
  * HW access macros.  These assume memory-mapped I/O, and not normal I/O space.
  */
-#define NV_WR08(p,i,d)  MMIO_OUT8((volatile pointer)(p), (i), (d))
-#define NV_RD08(p,i)    MMIO_IN8((volatile pointer)(p), (i))
-#define NV_WR16(p,i,d)  MMIO_OUT16((volatile pointer)(p), (i), (d))
-#define NV_RD16(p,i)    MMIO_IN16((volatile pointer)(p), (i))
-#define NV_WR32(p,i,d)  MMIO_OUT32((volatile pointer)(p), (i), (d))
-#define NV_RD32(p,i)    MMIO_IN32((volatile pointer)(p), (i))
+#define NV_WR08(p,i,d)  MMIO_OUT8((pointer)(p), (i), (d))
+#define NV_RD08(p,i)    MMIO_IN8((pointer)(p), (i))
+#define NV_WR16(p,i,d)  MMIO_OUT16((pointer)(p), (i), (d))
+#define NV_RD16(p,i)    MMIO_IN16((pointer)(p), (i))
+#define NV_WR32(p,i,d)  MMIO_OUT32((pointer)(p), (i), (d))
+#define NV_RD32(p,i)    MMIO_IN32((pointer)(p), (i))
 
 /* VGA I/O is now always done through MMIO */
 #define VGA_WR08(p,i,d) NV_WR08(p,i,d)
diff --git a/src/nv_proto.h b/src/nv_proto.h
index f526a71..5905309 100644
--- a/src/nv_proto.h
+++ b/src/nv_proto.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_proto.h,v 1.10 2003/07/31 20:24:29 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_proto.h,v 1.11 2004/03/20 01:52:16 mvojkovi Exp $ */
 
 #ifndef __NV_PROTO_H__
 #define __NV_PROTO_H__
@@ -37,6 +37,7 @@ void   NVSync(ScrnInfoPtr pScrn);
 void   NVResetGraphics(ScrnInfoPtr pScrn);
 void   NVDmaKickoff(NVPtr pNv);
 void   NVDmaWait(NVPtr pNv, int size);
+void   NVWaitVSync(NVPtr pNv);
 
 /* in nv_dga.c */
 Bool   NVDGAInit(ScreenPtr pScreen);
diff --git a/src/nv_setup.c b/src/nv_setup.c
index 3d58efa..cc157f6 100644
--- a/src/nv_setup.c
+++ b/src/nv_setup.c
@@ -37,7 +37,7 @@
 |*                                                                           *|
  \***************************************************************************/
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_setup.c,v 1.38 2003/09/08 20:00:27 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_setup.c,v 1.43 2004/08/26 22:38:47 mvojkovi Exp $ */
 
 #include "nv_include.h"
 
@@ -292,10 +292,10 @@ static void nv10GetConfig (NVPtr pNv)
     }
 #endif
 
-    if((pNv->Chipset && 0xffff) == 0x01a0) {
+    if(implementation == 0x01a0) {
         int amt = pciReadLong(pciTag(0, 0, 1), 0x7C);
         pNv->RamAmountKBytes = (((amt >> 6) & 31) + 1) * 1024;
-    } else if((pNv->Chipset & 0xffff) == 0x01f0) {
+    } else if(implementation == 0x01f0) {
         int amt = pciReadLong(pciTag(0, 0, 1), 0x84);
         pNv->RamAmountKBytes = (((amt >> 4) & 127) + 1) * 1024;
     } else {
@@ -304,10 +304,7 @@ static void nv10GetConfig (NVPtr pNv)
 
     pNv->CrystalFreqKHz = (pNv->PEXTDEV[0x0000/4] & (1 << 6)) ? 14318 : 13500;
     
-    if((implementation == 0x0170) ||
-       (implementation == 0x0180) ||
-       (implementation == 0x01F0) ||
-       (implementation >= 0x0250))
+    if(pNv->twoHeads && (implementation != 0x0110))
     {
        if(pNv->PEXTDEV[0x0000/4] & (1 << 22))
            pNv->CrystalFreqKHz = 27000;
@@ -381,15 +378,22 @@ NVCommonSetup(ScrnInfoPtr pScrn)
     pNv->PDIO0    = (U008*)pNv->REGS + 0x00681000;
     pNv->PVIO     = (U008*)pNv->REGS + 0x000C0000;
 
-    pNv->twoHeads =  (implementation >= 0x0110) &&
+    pNv->twoHeads =  (pNv->Architecture >= NV_ARCH_10) &&
+                     (implementation != 0x0100) &&
                      (implementation != 0x0150) &&
                      (implementation != 0x01A0) &&
                      (implementation != 0x0200);
 
-    pNv->fpScaler = (pNv->twoHeads && (implementation != 0x0110));
+    pNv->fpScaler = (pNv->FpScale && pNv->twoHeads && (implementation!=0x0110));
 
     pNv->twoStagePLL = (implementation == 0x0310) ||
-                       (implementation == 0x0340);
+                       (implementation == 0x0340) ||
+                       (pNv->Architecture >= NV_ARCH_40);
+
+    pNv->WaitVSyncPossible = (pNv->Architecture >= NV_ARCH_10) &&
+                             (implementation != 0x0100);
+
+    pNv->BlendingPossible = ((pNv->Chipset & 0xffff) != 0x0020);
 
     /* look for known laptop chips */
     switch(pNv->Chipset & 0xffff) {
@@ -403,6 +407,7 @@ NVCommonSetup(ScrnInfoPtr pScrn)
     case 0x017D:
     case 0x0186:
     case 0x0187:
+    case 0x0189:
     case 0x0286:
     case 0x028C:
     case 0x0316:
@@ -424,6 +429,15 @@ NVCommonSetup(ScrnInfoPtr pScrn)
     case 0x0349:
     case 0x034B:
     case 0x034C:
+    case 0x0160:
+    case 0x0166:
+    case 0x00C8:
+    case 0x00C9:
+    case 0x00CC:
+    case 0x0147:
+    case 0x0148:
+    case 0x0149:
+    case 0x014C:
         mobile = TRUE;
         break;
     default:
@@ -658,6 +672,7 @@ NVCommonSetup(ScrnInfoPtr pScrn)
     if(pNv->FlatPanel && !pNv->Television) {
        pNv->fpWidth = pNv->PRAMDAC[0x0820/4] + 1;
        pNv->fpHeight = pNv->PRAMDAC[0x0800/4] + 1;
+       pNv->fpSyncs = pNv->PRAMDAC[0x0848/4] & 0x30000033;
        xf86DrvMsg(pScrn->scrnIndex, X_PROBED, "Panel size is %i x %i\n",
                   pNv->fpWidth, pNv->fpHeight);
     }
@@ -665,7 +680,7 @@ NVCommonSetup(ScrnInfoPtr pScrn)
     if(monitorA)
       xf86SetDDCproperties(pScrn, monitorA);
 
-    if(!pNv->FlatPanel || (pScrn->depth != 24))
+    if(!pNv->FlatPanel || (pScrn->depth != 24) || !pNv->twoHeads)
         pNv->FPDither = FALSE;
 }
 
diff --git a/src/nv_shadow.c b/src/nv_shadow.c
index 5effc79..564220a 100644
--- a/src/nv_shadow.c
+++ b/src/nv_shadow.c
@@ -2,7 +2,7 @@
    Copyright (c) 1999,  The XFree86 Project Inc. 
    Written by Mark Vojkovich <markv@valinux.com>
 */
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_shadow.c,v 1.6 2001/01/22 21:32:36 dawes Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_shadow.c,v 1.5 2000/03/13 18:49:29 mvojkovi Exp $ */
 
 #include "nv_local.h"
 #include "nv_include.h"
diff --git a/src/nv_type.h b/src/nv_type.h
index 5730cb5..96ca4e4 100644
--- a/src/nv_type.h
+++ b/src/nv_type.h
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_type.h,v 1.44 2003/09/08 20:00:27 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_type.h,v 1.48 2004/08/26 22:38:47 mvojkovi Exp $ */
 
 #ifndef __NV_STRUCT_H__
 #define __NV_STRUCT_H__
@@ -13,6 +13,7 @@
 #define NV_ARCH_10  0x10
 #define NV_ARCH_20  0x20
 #define NV_ARCH_30  0x30
+#define NV_ARCH_40  0x40
 
 
 #define BITMASK(t,b) (((unsigned)(1U << (((t)-(b)+1)))-1)  << (b))
@@ -63,6 +64,9 @@ typedef struct _riva_hw_state
     U032 cursor0;
     U032 cursor1;
     U032 cursor2;
+    U032 timingH;
+    U032 timingV;
+    U032 displayV;
 } RIVA_HW_STATE, *NVRegPtr;
 
 
@@ -88,6 +92,7 @@ typedef struct {
     CARD32              ScratchBufferStart;
     Bool                NoAccel;
     Bool                HWCursor;
+    Bool                FpScale;
     Bool                ShadowFB;
     unsigned char *     ShadowPtr;
     int                 ShadowPitch;
@@ -151,6 +156,7 @@ typedef struct {
     Bool                fpScaler;
     int                 fpWidth;
     int                 fpHeight;
+    CARD32              fpSyncs;
 
     CARD32              dmaPut;
     CARD32              dmaCurrent;
@@ -159,6 +165,8 @@ typedef struct {
     CARD32              *dmaBase;
 
     CARD32              currentRop;
+    Bool                WaitVSyncPossible;
+    Bool                BlendingPossible;
 } NVRec, *NVPtr;
 
 #define NVPTR(p) ((NVPtr)((p)->driverPrivate))
diff --git a/src/nv_video.c b/src/nv_video.c
index f985461..7bfa94b 100644
--- a/src/nv_video.c
+++ b/src/nv_video.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_video.c,v 1.19 2003/09/01 20:54:26 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_video.c,v 1.23 2004/03/20 22:07:06 mvojkovi Exp $ */
 
 #include "xf86.h"
 #include "xf86_OSproc.h"
@@ -46,6 +46,7 @@ typedef struct _NVPortPrivRec {
    Bool		grabbedByV4L;
    Bool         iturbt_709;
    Bool         blitter;
+   Bool         SyncToVBlank;
    FBLinearPtr  linear;
    int pitch;
    int offset;
@@ -95,7 +96,7 @@ static void NVInitOffscreenImages (ScreenPtr pScreen);
 
 static Atom xvBrightness, xvContrast, xvColorKey, xvSaturation, 
             xvHue, xvAutopaintColorKey, xvSetDefaults, xvDoubleBuffer,
-            xvITURBT709;
+            xvITURBT709, xvSyncToVBlank;
 
 /* client libraries expect an encoding */
 static XF86VideoEncodingRec DummyEncoding =
@@ -114,9 +115,8 @@ XF86VideoFormatRec NVFormats[NUM_FORMATS_ALL] =
    {15, DirectColor}, {16, DirectColor}, {24, DirectColor}
 };
 
-#define NUM_ATTRIBUTES 9
-
-XF86AttributeRec NVAttributes[NUM_ATTRIBUTES] =
+#define NUM_OVERLAY_ATTRIBUTES 9
+XF86AttributeRec NVOverlayAttributes[NUM_OVERLAY_ATTRIBUTES] =
 {
    {XvSettable | XvGettable, 0, 1, "XV_DOUBLE_BUFFER"},
    {XvSettable | XvGettable, 0, (1 << 24) - 1, "XV_COLORKEY"},
@@ -129,6 +129,14 @@ XF86AttributeRec NVAttributes[NUM_ATTRIBUTES] =
    {XvSettable | XvGettable, 0, 1, "XV_ITURBT_709"}
 };
 
+#define NUM_BLIT_ATTRIBUTES 2
+XF86AttributeRec NVBlitAttributes[NUM_BLIT_ATTRIBUTES] =
+{
+   {XvSettable             , 0, 0, "XV_SET_DEFAULTS"},
+   {XvSettable | XvGettable, 0, 1, "XV_SYNC_TO_VBLANK"}
+};
+
+
 #define NUM_IMAGES_YUV 4
 #define NUM_IMAGES_ALL 5
 
@@ -285,7 +293,10 @@ void NVInitVideo (ScreenPtr pScreen)
     NVPtr         	pNv   = NVPTR(pScrn);
     int 		num_adaptors;
 
-    if((pScrn->bitsPerPixel != 8) && (pNv->Architecture >= NV_ARCH_10)) {
+    if((pScrn->bitsPerPixel != 8) && (pNv->Architecture >= NV_ARCH_10) &&
+         ((pNv->Architecture <= NV_ARCH_30) || 
+            ((pNv->Chipset & 0xfff0) == 0x0040)))
+    {
 	overlayAdaptor = NVSetupOverlayVideo(pScreen);
   
 	if(overlayAdaptor)
@@ -358,8 +369,13 @@ NVSetupBlitVideo (ScreenPtr pScreen)
     for(i = 0; i < NUM_BLIT_PORTS; i++)
        adapt->pPortPrivates[i].ptr = (pointer)(pPriv);
 
-    adapt->pAttributes          = NULL;
-    adapt->nAttributes          = 0;
+    if(pNv->WaitVSyncPossible) {
+       adapt->pAttributes          = NVBlitAttributes;
+       adapt->nAttributes          = NUM_BLIT_ATTRIBUTES;
+    } else {
+       adapt->pAttributes          = NULL;
+       adapt->nAttributes          = 0;
+    }
     adapt->pImages              = NVImages;
     adapt->nImages              = NUM_IMAGES_ALL;
     adapt->PutVideo             = NULL;
@@ -377,9 +393,12 @@ NVSetupBlitVideo (ScreenPtr pScreen)
     pPriv->grabbedByV4L         = FALSE;
     pPriv->blitter              = TRUE;
     pPriv->doubleBuffer         = FALSE;
+    pPriv->SyncToVBlank         = pNv->WaitVSyncPossible;
 
     pNv->blitAdaptor            = adapt;
 
+    xvSyncToVBlank              = MAKE_ATOM("XV_SYNC_TO_VBLANK");
+
     return adapt;
 }
 
@@ -409,8 +428,8 @@ NVSetupOverlayVideo (ScreenPtr pScreen)
     adapt->pPortPrivates        = (DevUnion*)(&adapt[1]);
     pPriv                       = (NVPortPrivPtr)(&adapt->pPortPrivates[1]);
     adapt->pPortPrivates[0].ptr = (pointer)(pPriv);
-    adapt->pAttributes          = NVAttributes;
-    adapt->nAttributes          = NUM_ATTRIBUTES;
+    adapt->pAttributes          = NVOverlayAttributes;
+    adapt->nAttributes          = NUM_OVERLAY_ATTRIBUTES;
     adapt->pImages              = NVImages;
     adapt->nImages              = NUM_IMAGES_YUV;
     adapt->PutVideo             = NULL;
@@ -432,7 +451,7 @@ NVSetupOverlayVideo (ScreenPtr pScreen)
     NVSetPortDefaults (pScrnInfo, pPriv);
     
     /* gotta uninit this someplace */
-    REGION_INIT(pScreen, &pPriv->clip, NullBox, 0); 
+    REGION_NULL(pScreen, &pPriv->clip);
     
     pNv->overlayAdaptor		= adapt;
     
@@ -571,8 +590,19 @@ NVPutBlitImage (
         NVDmaNext (pNv, SURFACE_FORMAT_DEPTH15);
     }
 
-    NVDmaStart(pNv, STRETCH_BLIT_FORMAT, 1);
-    NVDmaNext (pNv, format);
+    if(pPriv->SyncToVBlank) {
+       NVDmaKickoff(pNv);
+       NVWaitVSync(pNv);
+    }
+
+    if(pNv->BlendingPossible) {
+       NVDmaStart(pNv, STRETCH_BLIT_FORMAT, 2);
+       NVDmaNext (pNv, format);
+       NVDmaNext (pNv, STRETCH_BLIT_OPERATION_COPY);
+    } else {
+       NVDmaStart(pNv, STRETCH_BLIT_FORMAT, 1);
+       NVDmaNext (pNv, format);
+    }
 
     while(nbox--) {
        NVDmaStart(pNv, RECT_SOLID_COLOR, 1);
@@ -757,7 +787,20 @@ static int NVSetBlitPortAttribute
     pointer     data
 )
 {
-    return BadMatch;
+    NVPortPrivPtr pPriv = (NVPortPrivPtr)data;
+    NVPtr pNv = NVPTR(pScrnInfo);
+    
+    if ((attribute == xvSyncToVBlank) && pNv->WaitVSyncPossible) {
+        if ((value < 0) || (value > 1))
+            return BadValue;
+        pPriv->SyncToVBlank = value;
+    } else
+    if (attribute == xvSetDefaults) {
+        pPriv->SyncToVBlank = pNv->WaitVSyncPossible;
+    } else
+       return BadMatch;
+
+    return Success;
 }
 
 static int NVGetBlitPortAttribute
@@ -768,7 +811,14 @@ static int NVGetBlitPortAttribute
     pointer      data
 )
 {
-    return BadMatch;
+    NVPortPrivPtr pPriv = (NVPortPrivPtr)data;
+
+    if(attribute == xvSyncToVBlank)
+       *value = (pPriv->SyncToVBlank) ? 1 : 0;
+    else
+       return BadMatch;
+
+    return Success;
 }
 
 
@@ -958,7 +1008,7 @@ static int NVPutImage
     NVPtr pNv = NVPTR(pScrnInfo);
     INT32 xa, xb, ya, yb;
     unsigned char *dst_start;
-    int pitch, newSize, offset, s2offset, s3offset;
+    int newSize, offset, s2offset, s3offset;
     int srcPitch, srcPitch2, dstPitch;
     int top, left, right, bottom, npixels, nlines, bpp;
     Bool skip = FALSE;
@@ -1011,7 +1061,6 @@ static int NVPutImage
     }
 
     bpp = pScrnInfo->bitsPerPixel >> 3;
-    pitch = bpp * pScrnInfo->displayWidth;
 
     switch(id) {
     case FOURCC_YV12:
@@ -1435,8 +1484,8 @@ XF86OffscreenImageRec NVOffscreenImages[2] =
    NVGetSurfaceAttribute,
    NVSetSurfaceAttribute,
    2046, 2046,
-   NUM_ATTRIBUTES - 1,
-   &NVAttributes[1]
+   NUM_OVERLAY_ATTRIBUTES - 1,
+   &NVOverlayAttributes[1]
   },
  {
    &NVImages[2],
@@ -1448,8 +1497,8 @@ XF86OffscreenImageRec NVOffscreenImages[2] =
    NVGetSurfaceAttribute,
    NVSetSurfaceAttribute,
    2046, 2046,
-   NUM_ATTRIBUTES - 1,
-   &NVAttributes[1]
+   NUM_OVERLAY_ATTRIBUTES - 1,
+   &NVOverlayAttributes[1]
   },
 };
 
diff --git a/src/nv_xaa.c b/src/nv_xaa.c
index 2dd2bc1..0723788 100644
--- a/src/nv_xaa.c
+++ b/src/nv_xaa.c
@@ -37,7 +37,7 @@
 |*                                                                           *|
  \***************************************************************************/
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_xaa.c,v 1.32 2003/08/18 21:40:04 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/nv_xaa.c,v 1.35 2004/03/20 16:25:18 mvojkovi Exp $ */
 
 #include "nv_include.h"
 #include "xaalocal.h"
@@ -113,6 +113,13 @@ NVDmaKickoff(NVPtr pNv)
     }
 }
 
+
+/* There is a HW race condition with videoram command buffers.
+   You can't jump to the location of your put offset.  We write put
+   at the jump offset + SKIPS dwords with noop padding in between
+   to solve this problem */
+#define SKIPS  8
+
 void 
 NVDmaWait (
    NVPtr pNv,
@@ -129,21 +136,34 @@ NVDmaWait (
            pNv->dmaFree = pNv->dmaMax - pNv->dmaCurrent;
            if(pNv->dmaFree < size) {
                NVDmaNext(pNv, 0x20000000);
-               if(!dmaGet) {
-                   if(!pNv->dmaPut) /* corner case - idle */
-                      NVDmaKickoff(pNv);
+               if(dmaGet <= SKIPS) {
+                   if(pNv->dmaPut <= SKIPS) /* corner case - will be idle */
+                      WRITE_PUT(pNv, SKIPS + 1);
                    do { dmaGet = READ_GET(pNv); }
-                   while(!dmaGet);
+                   while(dmaGet <= SKIPS);
                }
-               WRITE_PUT(pNv, 0);
-               pNv->dmaCurrent = pNv->dmaPut = 0;
-               pNv->dmaFree = dmaGet - 1;
+               WRITE_PUT(pNv, SKIPS);
+               pNv->dmaCurrent = pNv->dmaPut = SKIPS;
+               pNv->dmaFree = dmaGet - (SKIPS + 1);
            }
        } else 
            pNv->dmaFree = dmaGet - pNv->dmaCurrent - 1;
     }
 }
 
+void
+NVWaitVSync(NVPtr pNv)
+{
+    NVDmaStart(pNv, 0x0000A12C, 1);
+    NVDmaNext (pNv, 0);
+    NVDmaStart(pNv, 0x0000A134, 1);
+    NVDmaNext (pNv, pNv->CRTCnumber);
+    NVDmaStart(pNv, 0x0000A100, 1);
+    NVDmaNext (pNv, 0);
+    NVDmaStart(pNv, 0x0000A130, 1);
+    NVDmaNext (pNv, 0);
+}
+
 /* 
   currentRop =  0-15  solid fill
                16-31  8x8 pattern fill
@@ -194,7 +214,7 @@ void NVResetGraphics(ScrnInfoPtr pScrn)
 {
     NVPtr pNv = NVPTR(pScrn);
     CARD32 surfaceFormat, patternFormat, rectFormat, lineFormat;
-    int pitch;
+    int pitch, i;
 
     if(pNv->NoAccel) return;
 
@@ -203,27 +223,30 @@ void NVResetGraphics(ScrnInfoPtr pScrn)
 
     pNv->dmaBase = (CARD32*)(&pNv->FbStart[pNv->FbUsableSize]);
 
-    pNv->dmaBase[0x0] = 0x00040000;
-    pNv->dmaBase[0x1] = 0x80000010;
-    pNv->dmaBase[0x2] = 0x00042000;
-    pNv->dmaBase[0x3] = 0x80000011;
-    pNv->dmaBase[0x4] = 0x00044000;
-    pNv->dmaBase[0x5] = 0x80000012;
-    pNv->dmaBase[0x6] = 0x00046000;
-    pNv->dmaBase[0x7] = 0x80000013;
-    pNv->dmaBase[0x8] = 0x00048000;
-    pNv->dmaBase[0x9] = 0x80000014;
-    pNv->dmaBase[0xA] = 0x0004A000;
-    pNv->dmaBase[0xB] = 0x80000015;
-    pNv->dmaBase[0xC] = 0x0004C000;
-    pNv->dmaBase[0xD] = 0x80000016;
-    pNv->dmaBase[0xE] = 0x0004E000;
-    pNv->dmaBase[0xF] = 0x80000017;
+    for(i = 0; i < SKIPS; i++)
+      pNv->dmaBase[i] = 0x00000000;
+
+    pNv->dmaBase[0x0 + SKIPS] = 0x00040000;
+    pNv->dmaBase[0x1 + SKIPS] = 0x80000010;
+    pNv->dmaBase[0x2 + SKIPS] = 0x00042000;
+    pNv->dmaBase[0x3 + SKIPS] = 0x80000011;
+    pNv->dmaBase[0x4 + SKIPS] = 0x00044000;
+    pNv->dmaBase[0x5 + SKIPS] = 0x80000012;
+    pNv->dmaBase[0x6 + SKIPS] = 0x00046000;
+    pNv->dmaBase[0x7 + SKIPS] = 0x80000013;
+    pNv->dmaBase[0x8 + SKIPS] = 0x00048000;
+    pNv->dmaBase[0x9 + SKIPS] = 0x80000014;
+    pNv->dmaBase[0xA + SKIPS] = 0x0004A000;
+    pNv->dmaBase[0xB + SKIPS] = 0x80000015;
+    pNv->dmaBase[0xC + SKIPS] = 0x0004C000;
+    pNv->dmaBase[0xD + SKIPS] = 0x80000016;
+    pNv->dmaBase[0xE + SKIPS] = 0x0004E000;
+    pNv->dmaBase[0xF + SKIPS] = 0x80000017;
 
     pNv->dmaPut = 0;
-    pNv->dmaCurrent = 16;
+    pNv->dmaCurrent = 16 + SKIPS;
     pNv->dmaMax = 8191;
-    pNv->dmaFree = 8175;
+    pNv->dmaFree = pNv->dmaMax - pNv->dmaCurrent;
 
     switch(pNv->CurrentLayout.depth) {
     case 24:
diff --git a/src/riva_hw.c b/src/riva_hw.c
index 9f50450..fbd9858 100644
--- a/src/riva_hw.c
+++ b/src/riva_hw.c
@@ -36,7 +36,7 @@
 |*     those rights set forth herein.                                        *|
 |*                                                                           *|
  \***************************************************************************/
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_hw.c,v 1.51 2003/07/31 20:24:30 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_hw.c,v 1.51tsi Exp $ */
 
 #include "riva_local.h"
 #include "compiler.h"
@@ -169,14 +169,13 @@ static int nv3_iterate(nv3_fifo_info *res_info, nv3_sim_state * state, nv3_arb_i
     int mburst_size = 32;
     int mmisses, gmisses, vmisses;
     int misses;
-    int vlwm, glwm, mlwm;
+    int vlwm, glwm;
     int last, next, cur;
     int max_gfsize ;
     long ns;
 
     vlwm = 0;
     glwm = 0;
-    mlwm = 0;
     vfsize = 0;
     gfsize = 0;
     cur = ainfo->cur;
diff --git a/src/riva_hw.h b/src/riva_hw.h
index ec2b90b..0680d7d 100644
--- a/src/riva_hw.h
+++ b/src/riva_hw.h
@@ -36,7 +36,7 @@
 |*     those rights set forth herein.                                        *|
 |*                                                                           *|
 \***************************************************************************/
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_hw.h,v 1.26 2003/07/31 20:24:31 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_hw.h,v 1.25 2003/05/04 01:20:52 mvojkovi Exp $ */
 #ifndef __RIVA_HW_H__
 #define __RIVA_HW_H__
 
diff --git a/src/riva_tbl.h b/src/riva_tbl.h
index bb2a329..4156c80 100644
--- a/src/riva_tbl.h
+++ b/src/riva_tbl.h
@@ -36,7 +36,7 @@
 |*     those rights set forth herein.                                        *|
 |*                                                                           *|
  \***************************************************************************/
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_tbl.h,v 1.10 2003/07/31 20:24:31 mvojkovi Exp $ */
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/nv/riva_tbl.h,v 1.9 2002/01/30 01:35:03 mvojkovi Exp $ */
 
 
 /*