Merge tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-tracegrafted

Pull tracing fixes from Steven Rostedt:

 - Don't destroy user event fields when removal fails

   User event fields are destroyed before the event is removed from
   visibility. But that can fail leaving the still visible event with no
   fields. Move the destroying of the fields to after the event is
   successfully removed from visibility.

 - Initialize function graph state is fork before calling
   copy_exec_state()

   For non-CLONE_VM forks, copy_exec_state() allocates a new
   task_exec_state. If that allocation fails, ftrace_graph_exit_task()
   will free the tasks ret_stack pointer. Since that pointer is still
   using the parent's ret_stack, it mistakenly frees the parent's
   pointer too.

   Call ftrace_graph_init() on the task first which will NULL out the
   new tasks's ret_stack and if the copy fails, it will not free
   anything.

 - Remove FGRAPH_MAX_INDEX

   The macro FGRAPH_MAX_INDEX was added but never used. Remove it.

 - Save ent_size in function graph printing of nested functions

   The function graph tracer needs to look at the next event to see if
   the next event is the return of the current function entry. If it is,
   it prints a single line:

	ktime_get();

   Otherwise it prints it like a nested function:

	tick_nohz_irq_exit() {
	    ktime_get();
	    kcpustat_irq_exit();
	}

   In order to look at the next event, it must save the current event so
   that it has the information to print from it. It saves the event in
   the iterator descriptor called "ent". What it doesn't save is the
   ent_size of the event which is now used to know if the function graph
   arguments are to be printed. The peek doesn't save the size so the
   size used happens to be that of the size of the last event that was
   seen.

   Save the entry event size in the iterator descriptor so that the
   correct size is used.

 - Fix several errors with freeing data in the histogram code

   The histogram code had a lot of leaked or or incorrect accounting
   when failures happen. Correct them.

 - Fix histogram regression of .percent and .graph modifiers

   Up until 6.3 histogram values could have "percent" or "graph"
   modifiers that changed how they were printed. But a change that added
   restricting histograms values from being strings, stack traces and
   other modifiers inadvertently prevented them from using the percent
   and graph modifiers, which were legal use cases for values.

   Put back the percent and graph modifiers.

 - Fix various typos in the comments

 - Set the trace_clock before initializing a histogram with clock
   argument

   The histogram API allows the user to specific which trace clock to
   use via a "clock=" string. The histogram is set up first before the
   clock is checked. If the passed in clock is not valid, it exits
   without fully fixing up the histogram leaving it on the list and a
   use-after-free can trigger.

   Update the clock argument first and if it fails then exit gracefully
   before the histogram trigger is placed on any lists.

 - Restore :mod: trailer after parsing in ftrace_set_clr_event

   The function ftrace_set_clr_event() modifies the parse string and
   needs to put it back to what was passed in. It searches for ":mod:"
   via a strsep() but fails to put back the first ':' in the string.

   Add back the ':' in the passed in string.

 - Take trace_array reference when opening a tracer options file

   The options files are dynamically created and some tracers add their
   own options. When a tracer adds their own list of options, the
   trace_array holding them has an array to hold the list of options for
   each tracer. This array increases in size via a krealloc(), and the
   new entry gets a newly allocated array to hold the options of the new
   tracer being added.

   The element in each entry of the tracer's option array holds a
   pointer back to the trace_array, a pointer to the tracer it is
   associated to, a pointer to the flags of the option.

   The issue is that these arrays are freed when the trace_array is
   freed when its instance it represents is removed from the instances
   directory. There's a race that an open of one of these options files
   can happen when the instance is being removed.

   Add a new helper function to be called by the open function of the
   options file to iterate all existing trace_arrays under a lock and
   find the one that has the given option element in one of it's tracer
   arrays. If found, then update the associated trace_array's reference
   counter to keep it from being freed. If not found, have the open call
   return -ENODEV.

 - Disable interrupts when acquiring the lock in rb_wake_up_waiters()

   The function rb_wake_up_waiters() assumes it will be called in
   interrupt context and does not disable irqs when taking
   cpu_buffer->reader_lock, which can be called in hard interrupt
   context. The issue is in PREEMPT_RT, this function is called in
   thread context leaving this lock open to a deadlock.

   Take the lock with interrupts disabled.

 - Use rcu_assign_pointer() for tmp_ops filter hash

   The tmp_ops used in update_ftrace_direct_mod() assigns its
   filter_hash field directly, but that field is annotated as __rcu and
   sparse complains. Assign it with rcu_assign_pointer()

 - Fix use-after-free in enable_trigger_private_data_free()

   The trace_event_call is accessed through the event_trigger_data's
   trace_event_file pointer to put the trace_event_call on freeing. The
   issue is that the trace_event_file data may have been freed already
   causing a use-after-free. Add a field to the event_trigger_data that
   points directly to the trace_event_call so that it can decrement its
   reference directly without needing to go through the
   trace_event_file.

 - Fix accounting of buffer data remote headers

   trace_buffer_desc_size() and trace_remote_alloc_buffer() undercount
   the number of pages is needed for the asked for size as it doesn't
   take into account the meta data on each page. Add a helper function
   to do the calculation properly and use that in these functions.

 - Catch nr_page_va overflow in ring_buffer_desc sizing

   The number of pages per remote ring buffer is capped by
   ring_buffer_desc::nr_page_va (32 bits). A buffer_size large enough to
   overflow that field would silently allocate a descriptor smaller than
   what was asked for.

 - Do not resize the subbuf order if any per_cpu buffer is disabled

   The mmapping of ring buffers disables resizing the subbuffers, but it
   is done per-cpu whereas the subbuf size change is done for all the
   per_cpu buffers under the buffer->mutex. It could change the size of
   some while the mapping is happening on others. Have the resize of the
   subbuf order check all the per_cpu buffers under the lock to see if
   any of them is disabled before starting and causing an inconsistency
   between buffers that are being mapped.

* tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace: (25 commits)
  ring-buffer: Check resize_disabled before publishing the new subbuf order
  tracing/remotes: Catch nr_page_va overflow in ring_buffer_desc sizing
  tracing/remotes: Account for ring buffer page header in size calculation
  tracing: Don't dereference trace_event_file in deferred trigger free
  ftrace: Use rcu_assign_pointer() for tmp_ops filter hash
  ring-buffer: Acquire the lock with irqsave in rb_wake_up_waiters()
  tracing: Take trace_array reference when opening a tracer options file
  tracing: Fix ring_buffer_read_page_size() kernel-doc
  tracing: Restore :mod: trailer after parsing in ftrace_set_clr_event()
  tracing: Fix memory corruption from a "STACKTRACE" histogram key
  tracing: Fix memory corruption from the histogram stacktrace modifier
  tracing: Undo the registration when enabling the histogram trigger fails
  tracing: Take the reference before publishing the named histogram trigger
  tracing: Set the trace clock before registering the histogram trigger
  tracing: Fix typo "preceeded" in comment
  tracing: Fix typo "availabe" in comment
  tracing: Let histogram values keep the percent and graph modifiers
  tracing: Keep the entry count when the histogram stats allocation fails
  tracing: Free histogram the field rejected for a bad modifier
  tracing: Free histogram the var ref when its initialization fails
  ...

8 files changed, 2445 insertions, 0 deletions

diff --git a/drivers/video/fbdev/kyro/Makefile b/drivers/video/fbdev/kyro/Makefile
new file mode 100644
index 000000000..b05abe693
--- /dev/null
+++ b/drivers/video/fbdev/kyro/Makefile
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Makefile for the Kyro framebuffer driver
+#
+
+obj-$(CONFIG_FB_KYRO)	+= kyrofb.o
+
+kyrofb-objs	:= STG4000Ramdac.o STG4000VTG.o STG4000OverlayDevice.o \
+		   STG4000InitDevice.o fbdev.o
diff --git a/drivers/video/fbdev/kyro/STG4000InitDevice.c b/drivers/video/fbdev/kyro/STG4000InitDevice.c
new file mode 100644
index 000000000..79886a246
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000InitDevice.c
@@ -0,0 +1,323 @@
+/*
+ *  linux/drivers/video/kyro/STG4000InitDevice.c
+ *
+ *  Copyright (C) 2000 Imagination Technologies Ltd
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+#include <linux/pci.h>
+
+#include "STG4000Reg.h"
+#include "STG4000Interface.h"
+
+/* SDRAM fixed settings */
+#define SDRAM_CFG_0   0x49A1
+#define SDRAM_CFG_1   0xA732
+#define SDRAM_CFG_2   0x31
+#define SDRAM_ARB_CFG 0xA0
+#define SDRAM_REFRESH 0x20
+
+/* Reset values */
+#define PMX2_SOFTRESET_DAC_RST		0x0001
+#define PMX2_SOFTRESET_C1_RST		0x0004
+#define PMX2_SOFTRESET_C2_RST		0x0008
+#define PMX2_SOFTRESET_3D_RST		0x0010
+#define PMX2_SOFTRESET_VIDIN_RST	0x0020
+#define PMX2_SOFTRESET_TLB_RST		0x0040
+#define PMX2_SOFTRESET_SD_RST		0x0080
+#define PMX2_SOFTRESET_VGA_RST		0x0100
+#define PMX2_SOFTRESET_ROM_RST		0x0200	/* reserved bit, do not reset */
+#define PMX2_SOFTRESET_TA_RST		0x0400
+#define PMX2_SOFTRESET_REG_RST		0x4000
+#define PMX2_SOFTRESET_ALL		0x7fff
+
+/* Core clock freq */
+#define CORE_PLL_FREQ 1000000
+
+/* Reference Clock freq */
+#define REF_FREQ 14318
+
+/* PCI Registers */
+static u16 CorePllControl = 0x70;
+
+#define	PCI_CONFIG_SUBSYS_ID	0x2e
+
+/* Misc */
+#define CORE_PLL_MODE_REG_0_7      3
+#define CORE_PLL_MODE_REG_8_15     2
+#define CORE_PLL_MODE_CONFIG_REG   1
+#define DAC_PLL_CONFIG_REG         0
+
+#define STG_MAX_VCO 500000
+#define STG_MIN_VCO 100000
+
+/* PLL Clock */
+#define    STG4K3_PLL_SCALER      8	/* scale numbers by 2^8 for fixed point calc */
+#define    STG4K3_PLL_MIN_R       2	/* Minimum multiplier */
+#define    STG4K3_PLL_MAX_R       33	/* Max */
+#define    STG4K3_PLL_MIN_F       2	/* Minimum divisor */
+#define    STG4K3_PLL_MAX_F       513	/* Max */
+#define    STG4K3_PLL_MIN_OD      0	/* Min output divider (shift) */
+#define    STG4K3_PLL_MAX_OD      2	/* Max */
+#define    STG4K3_PLL_MIN_VCO_SC  (100000000 >> STG4K3_PLL_SCALER)	/* Min VCO rate */
+#define    STG4K3_PLL_MAX_VCO_SC  (500000000 >> STG4K3_PLL_SCALER)	/* Max VCO rate */
+#define    STG4K3_PLL_MINR_VCO_SC (100000000 >> STG4K3_PLL_SCALER)	/* Min VCO rate (restricted) */
+#define    STG4K3_PLL_MAXR_VCO_SC (500000000 >> STG4K3_PLL_SCALER)	/* Max VCO rate (restricted) */
+#define    STG4K3_PLL_MINR_VCO    100000000	/* Min VCO rate (restricted) */
+#define    STG4K3_PLL_MAX_VCO     500000000	/* Max VCO rate */
+#define    STG4K3_PLL_MAXR_VCO    500000000	/* Max VCO rate (restricted) */
+
+#define OS_DELAY(X) \
+{ \
+volatile u32 i,count=0; \
+    for(i=0;i<X;i++) count++; \
+}
+
+static u32 InitSDRAMRegisters(volatile STG4000REG __iomem *pSTGReg,
+			      u32 dwSubSysID, u32 dwRevID)
+{
+	static const u8 adwSDRAMArgCfg0[] = { 0xa0, 0x80, 0xa0, 0xa0, 0xa0 };
+	static const u16 adwSDRAMCfg1[] = { 0x8732, 0x8732, 0xa732, 0xa732, 0x8732 };
+	static const u16 adwSDRAMCfg2[] = { 0x87d2, 0x87d2, 0xa7d2, 0x87d2, 0xa7d2 };
+	static const u8 adwSDRAMRsh[] = { 36, 39, 40 };
+	static const u8 adwChipSpeed[] = { 110, 120, 125 };
+	u32 dwMemTypeIdx;
+	u32 dwChipSpeedIdx;
+
+	/* Get memory tpye and chip speed indexs from the SubSysDevID */
+	dwMemTypeIdx = (dwSubSysID & 0x70) >> 4;
+	dwChipSpeedIdx = (dwSubSysID & 0x180) >> 7;
+
+	if (dwMemTypeIdx > 4 || dwChipSpeedIdx > 2)
+		return 0;
+
+	/* Program SD-RAM interface */
+	STG_WRITE_REG(SDRAMArbiterConf, adwSDRAMArgCfg0[dwMemTypeIdx]);
+	if (dwRevID < 5) {
+		STG_WRITE_REG(SDRAMConf0, 0x49A1);
+		STG_WRITE_REG(SDRAMConf1, adwSDRAMCfg1[dwMemTypeIdx]);
+	} else {
+		STG_WRITE_REG(SDRAMConf0, 0x4DF1);
+		STG_WRITE_REG(SDRAMConf1, adwSDRAMCfg2[dwMemTypeIdx]);
+	}
+
+	STG_WRITE_REG(SDRAMConf2, 0x31);
+	STG_WRITE_REG(SDRAMRefresh, adwSDRAMRsh[dwChipSpeedIdx]);
+
+	return adwChipSpeed[dwChipSpeedIdx] * 10000;
+}
+
+u32 ProgramClock(u32 refClock,
+		   u32 coreClock,
+		   u32 * FOut, u32 * ROut, u32 * POut)
+{
+	u32 R = 0, F = 0, OD = 0, ODIndex = 0;
+	u32 ulBestR = 0, ulBestF = 0, ulBestOD = 0;
+	u32 ulBestClk = 0, ulBestScore = 0;
+	u32 ulScore, ulPhaseScore, ulVcoScore;
+	u32 ulTmp = 0, ulVCO;
+	u32 ulScaleClockReq, ulMinClock, ulMaxClock;
+	static const unsigned char ODValues[] = { 1, 2, 0 };
+
+	/* Translate clock in Hz */
+	coreClock *= 100;	/* in Hz */
+	refClock *= 1000;	/* in Hz */
+
+	/* Work out acceptable clock
+	 * The method calculates ~ +- 0.4% (1/256)
+	 */
+	ulMinClock = coreClock - (coreClock >> 8);
+	ulMaxClock = coreClock + (coreClock >> 8);
+
+	/* Scale clock required for use in calculations */
+	ulScaleClockReq = coreClock >> STG4K3_PLL_SCALER;
+
+	/* Iterate through post divider values */
+	for (ODIndex = 0; ODIndex < 3; ODIndex++) {
+		OD = ODValues[ODIndex];
+		R = STG4K3_PLL_MIN_R;
+
+		/* loop for pre-divider from min to max  */
+		while (R <= STG4K3_PLL_MAX_R) {
+			/* estimate required feedback multiplier */
+			ulTmp = R * (ulScaleClockReq << OD);
+
+			/* F = ClkRequired * R * (2^OD) / Fref */
+			F = (u32)(ulTmp / (refClock >> STG4K3_PLL_SCALER));
+
+			/* compensate for accuracy */
+			if (F > STG4K3_PLL_MIN_F)
+				F--;
+
+
+			/*
+			 * We should be close to our target frequency (if it's
+			 * achievable with current OD & R) let's iterate
+			 * through F for best fit
+			 */
+			while ((F >= STG4K3_PLL_MIN_F) &&
+			       (F <= STG4K3_PLL_MAX_F)) {
+				/* Calc VCO at full accuracy */
+				ulVCO = refClock / R;
+				ulVCO = F * ulVCO;
+
+				/*
+				 * Check it's within restricted VCO range
+				 * unless of course the desired frequency is
+				 * above the restricted range, then test
+				 * against VCO limit
+				 */
+				if ((ulVCO >= STG4K3_PLL_MINR_VCO) &&
+				    ((ulVCO <= STG4K3_PLL_MAXR_VCO) ||
+				     ((coreClock > STG4K3_PLL_MAXR_VCO)
+				      && (ulVCO <= STG4K3_PLL_MAX_VCO)))) {
+					ulTmp = (ulVCO >> OD);	/* Clock = VCO / (2^OD) */
+
+					/* Is this clock good enough? */
+					if ((ulTmp >= ulMinClock)
+					    && (ulTmp <= ulMaxClock)) {
+						ulPhaseScore = (((refClock / R) - (refClock / STG4K3_PLL_MAX_R))) / ((refClock - (refClock / STG4K3_PLL_MAX_R)) >> 10);
+
+						ulVcoScore = ((ulVCO - STG4K3_PLL_MINR_VCO)) / ((STG4K3_PLL_MAXR_VCO - STG4K3_PLL_MINR_VCO) >> 10);
+						ulScore = ulPhaseScore + ulVcoScore;
+
+						if (!ulBestScore) {
+							ulBestOD = OD;
+							ulBestF = F;
+							ulBestR = R;
+							ulBestClk = ulTmp;
+							ulBestScore =
+							    ulScore;
+						}
+						/* is this better, ( aim for highest Score) */
+			/*--------------------------------------------------------------------------
+                             Here we want to use a scoring system which will take account of both the
+                            value at the phase comparater and the VCO output
+                             to do this we will use a cumulative score between the two
+                          The way this ends up is that we choose the first value in the loop anyway
+                          but we shall keep this code in case new restrictions come into play
+                          --------------------------------------------------------------------------*/
+						if ((ulScore >= ulBestScore) && (OD > 0)) {
+							ulBestOD = OD;
+							ulBestF = F;
+							ulBestR = R;
+							ulBestClk = ulTmp;
+							ulBestScore =
+							    ulScore;
+						}
+					}
+				}
+				F++;
+			}
+			R++;
+		}
+	}
+
+	/*
+	   did we find anything?
+	   Then return RFOD
+	 */
+	if (ulBestScore) {
+		*ROut = ulBestR;
+		*FOut = ulBestF;
+
+		if ((ulBestOD == 2) || (ulBestOD == 3)) {
+			*POut = 3;
+		} else
+			*POut = ulBestOD;
+
+	}
+
+	return (ulBestClk);
+}
+
+int SetCoreClockPLL(volatile STG4000REG __iomem *pSTGReg, struct pci_dev *pDev)
+{
+	u32 F, R, P;
+	u16 core_pll = 0, sub;
+	u32 tmp;
+	u32 ulChipSpeed;
+
+	STG_WRITE_REG(IntMask, 0xFFFF);
+
+	/* Disable Primary Core Thread0 */
+	tmp = STG_READ_REG(Thread0Enable);
+	CLEAR_BIT(0);
+	STG_WRITE_REG(Thread0Enable, tmp);
+
+	/* Disable Primary Core Thread1 */
+	tmp = STG_READ_REG(Thread1Enable);
+	CLEAR_BIT(0);
+	STG_WRITE_REG(Thread1Enable, tmp);
+
+	STG_WRITE_REG(SoftwareReset,
+		      PMX2_SOFTRESET_REG_RST | PMX2_SOFTRESET_ROM_RST);
+	STG_WRITE_REG(SoftwareReset,
+		      PMX2_SOFTRESET_REG_RST | PMX2_SOFTRESET_TA_RST |
+		      PMX2_SOFTRESET_ROM_RST);
+
+	/* Need to play around to reset TA */
+	STG_WRITE_REG(TAConfiguration, 0);
+	STG_WRITE_REG(SoftwareReset,
+		      PMX2_SOFTRESET_REG_RST | PMX2_SOFTRESET_ROM_RST);
+	STG_WRITE_REG(SoftwareReset,
+		      PMX2_SOFTRESET_REG_RST | PMX2_SOFTRESET_TA_RST |
+		      PMX2_SOFTRESET_ROM_RST);
+
+	pci_read_config_word(pDev, PCI_CONFIG_SUBSYS_ID, &sub);
+
+	ulChipSpeed = InitSDRAMRegisters(pSTGReg, (u32)sub,
+		                         (u32)pDev->revision);
+
+	if (ulChipSpeed == 0)
+		return -EINVAL;
+
+	ProgramClock(REF_FREQ, CORE_PLL_FREQ, &F, &R, &P);
+
+	core_pll |= ((P) | ((F - 2) << 2) | ((R - 2) << 11));
+
+	/* Set Core PLL Control to Core PLL Mode  */
+
+	/* Send bits 0:7 of the Core PLL Mode register */
+	tmp = ((CORE_PLL_MODE_REG_0_7 << 8) | (core_pll & 0x00FF));
+	pci_write_config_word(pDev, CorePllControl, tmp);
+	/* Without some delay between the PCI config writes the clock does
+	   not reliably set when the code is compiled -O3
+	 */
+	OS_DELAY(1000000);
+
+	tmp |= SET_BIT(14);
+	pci_write_config_word(pDev, CorePllControl, tmp);
+	OS_DELAY(1000000);
+
+	/* Send bits 8:15 of the Core PLL Mode register */
+	tmp =
+	    ((CORE_PLL_MODE_REG_8_15 << 8) | ((core_pll & 0xFF00) >> 8));
+	pci_write_config_word(pDev, CorePllControl, tmp);
+	OS_DELAY(1000000);
+
+	tmp |= SET_BIT(14);
+	pci_write_config_word(pDev, CorePllControl, tmp);
+	OS_DELAY(1000000);
+
+	STG_WRITE_REG(SoftwareReset, PMX2_SOFTRESET_ALL);
+
+#if 0
+	/* Enable Primary Core Thread0 */
+	tmp = ((STG_READ_REG(Thread0Enable)) | SET_BIT(0));
+	STG_WRITE_REG(Thread0Enable, tmp);
+
+	/* Enable Primary Core Thread1 */
+	tmp = ((STG_READ_REG(Thread1Enable)) | SET_BIT(0));
+	STG_WRITE_REG(Thread1Enable, tmp);
+#endif
+
+	return 0;
+}
diff --git a/drivers/video/fbdev/kyro/STG4000Interface.h b/drivers/video/fbdev/kyro/STG4000Interface.h
new file mode 100644
index 000000000..b7c83d5df
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000Interface.h
@@ -0,0 +1,61 @@
+/*
+ *  linux/drivers/video/kyro/STG4000Interface.h
+ *
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#ifndef _STG4000INTERFACE_H
+#define _STG4000INTERFACE_H
+
+#include <linux/pci.h>
+#include <video/kyro.h>
+
+/*
+ * Ramdac Setup
+ */
+extern int InitialiseRamdac(volatile STG4000REG __iomem *pSTGReg, u32 displayDepth,
+			    u32 displayWidth, u32 displayHeight,
+			    s32 HSyncPolarity, s32 VSyncPolarity,
+			    u32 *pixelClock);
+
+extern void DisableRamdacOutput(volatile STG4000REG __iomem *pSTGReg);
+extern void EnableRamdacOutput(volatile STG4000REG __iomem *pSTGReg);
+
+/*
+ * Timing generator setup
+ */
+extern void DisableVGA(volatile STG4000REG __iomem *pSTGReg);
+extern void StopVTG(volatile STG4000REG __iomem *pSTGReg);
+extern void StartVTG(volatile STG4000REG __iomem *pSTGReg);
+extern void SetupVTG(volatile STG4000REG __iomem *pSTGReg,
+		     const struct kyrofb_info * pTiming);
+
+extern u32 ProgramClock(u32 refClock, u32 coreClock, u32 *FOut, u32 *ROut, u32 *POut);
+extern int SetCoreClockPLL(volatile STG4000REG __iomem *pSTGReg, struct pci_dev *pDev);
+
+/*
+ * Overlay setup
+ */
+extern void ResetOverlayRegisters(volatile STG4000REG __iomem *pSTGReg);
+
+extern int CreateOverlaySurface(volatile STG4000REG __iomem *pSTGReg,
+				u32 ulWidth, u32 ulHeight,
+				int bLinear,
+				u32 ulOverlayOffset,
+				u32 * retStride, u32 * retUVStride);
+
+extern int SetOverlayBlendMode(volatile STG4000REG __iomem *pSTGReg,
+			       OVRL_BLEND_MODE mode,
+			       u32 ulAlpha, u32 ulColorKey);
+
+extern int SetOverlayViewPort(volatile STG4000REG __iomem *pSTGReg,
+			      u32 left, u32 top,
+			      u32 right, u32 bottom);
+
+extern void EnableOverlayPlane(volatile STG4000REG __iomem *pSTGReg);
+
+#endif /* _STG4000INTERFACE_H */
diff --git a/drivers/video/fbdev/kyro/STG4000OverlayDevice.c b/drivers/video/fbdev/kyro/STG4000OverlayDevice.c
new file mode 100644
index 000000000..9fde0e3b6
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000OverlayDevice.c
@@ -0,0 +1,600 @@
+/*
+ *  linux/drivers/video/kyro/STG4000OverlayDevice.c
+ *
+ *  Copyright (C) 2000 Imagination Technologies Ltd
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+
+#include "STG4000Reg.h"
+#include "STG4000Interface.h"
+
+/* HW Defines */
+
+#define STG4000_NO_SCALING    0x800
+#define STG4000_NO_DECIMATION 0xFFFFFFFF
+
+/* Primary surface */
+#define STG4000_PRIM_NUM_PIX   5
+#define STG4000_PRIM_ALIGN     4
+#define STG4000_PRIM_ADDR_BITS 20
+
+#define STG4000_PRIM_MIN_WIDTH  640
+#define STG4000_PRIM_MAX_WIDTH  1600
+#define STG4000_PRIM_MIN_HEIGHT 480
+#define STG4000_PRIM_MAX_HEIGHT 1200
+
+/* Overlay surface */
+#define STG4000_OVRL_NUM_PIX   4
+#define STG4000_OVRL_ALIGN     2
+#define STG4000_OVRL_ADDR_BITS 20
+#define STG4000_OVRL_NUM_MODES 5
+
+#define STG4000_OVRL_MIN_WIDTH  0
+#define STG4000_OVRL_MAX_WIDTH  720
+#define STG4000_OVRL_MIN_HEIGHT 0
+#define STG4000_OVRL_MAX_HEIGHT 576
+
+/* Decimation and Scaling */
+static u32 adwDecim8[33] = {
+	    0xffffffff, 0xfffeffff, 0xffdffbff, 0xfefefeff, 0xfdf7efbf,
+	    0xfbdf7bdf, 0xf7bbddef, 0xeeeeeeef, 0xeeddbb77, 0xedb76db7,
+	    0xdb6db6db, 0xdb5b5b5b, 0xdab5ad6b, 0xd5ab55ab, 0xd555aaab,
+	    0xaaaaaaab, 0xaaaa5555, 0xaa952a55, 0xa94a5295, 0xa5252525,
+	    0xa4924925, 0x92491249, 0x91224489, 0x91111111, 0x90884211,
+	    0x88410821, 0x88102041, 0x81010101, 0x80800801, 0x80010001,
+	    0x80000001, 0x00000001, 0x00000000
+};
+
+typedef struct _OVRL_SRC_DEST {
+	/*clipped on-screen pixel position of overlay */
+	u32 ulDstX1;
+	u32 ulDstY1;
+	u32 ulDstX2;
+	u32 ulDstY2;
+
+	/*clipped pixel pos of source data within buffer thses need to be 128 bit word aligned */
+	u32 ulSrcX1;
+	u32 ulSrcY1;
+	u32 ulSrcX2;
+	u32 ulSrcY2;
+
+	/* on-screen pixel position of overlay */
+	s32 lDstX1;
+	s32 lDstY1;
+	s32 lDstX2;
+	s32 lDstY2;
+} OVRL_SRC_DEST;
+
+static u32 ovlWidth, ovlHeight, ovlStride;
+static int ovlLinear;
+
+void ResetOverlayRegisters(volatile STG4000REG __iomem *pSTGReg)
+{
+	u32 tmp;
+
+	/* Set Overlay address to default */
+	tmp = STG_READ_REG(DACOverlayAddr);
+	CLEAR_BITS_FRM_TO(0, 20);
+	CLEAR_BIT(31);
+	STG_WRITE_REG(DACOverlayAddr, tmp);
+
+	/* Set Overlay U address */
+	tmp = STG_READ_REG(DACOverlayUAddr);
+	CLEAR_BITS_FRM_TO(0, 20);
+	STG_WRITE_REG(DACOverlayUAddr, tmp);
+
+	/* Set Overlay V address */
+	tmp = STG_READ_REG(DACOverlayVAddr);
+	CLEAR_BITS_FRM_TO(0, 20);
+	STG_WRITE_REG(DACOverlayVAddr, tmp);
+
+	/* Set Overlay Size */
+	tmp = STG_READ_REG(DACOverlaySize);
+	CLEAR_BITS_FRM_TO(0, 10);
+	CLEAR_BITS_FRM_TO(12, 31);
+	STG_WRITE_REG(DACOverlaySize, tmp);
+
+	/* Set Overlay Vt Decimation */
+	tmp = STG4000_NO_DECIMATION;
+	STG_WRITE_REG(DACOverlayVtDec, tmp);
+
+	/* Set Overlay format to default value */
+	tmp = STG_READ_REG(DACPixelFormat);
+	CLEAR_BITS_FRM_TO(4, 7);
+	CLEAR_BITS_FRM_TO(16, 22);
+	STG_WRITE_REG(DACPixelFormat, tmp);
+
+	/* Set Vertical scaling to default */
+	tmp = STG_READ_REG(DACVerticalScal);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 22);
+	tmp |= STG4000_NO_SCALING;	/* Set to no scaling */
+	STG_WRITE_REG(DACVerticalScal, tmp);
+
+	/* Set Horizontal Scaling to default */
+	tmp = STG_READ_REG(DACHorizontalScal);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 17);
+	tmp |= STG4000_NO_SCALING;	/* Set to no scaling */
+	STG_WRITE_REG(DACHorizontalScal, tmp);
+
+	/* Set Blend mode to Alpha Blend */
+	/* ????? SG 08/11/2001 Surely this isn't the alpha blend mode,
+	   hopefully its overwrite
+	 */
+	tmp = STG_READ_REG(DACBlendCtrl);
+	CLEAR_BITS_FRM_TO(0, 30);
+	tmp = (GRAPHICS_MODE << 28);
+	STG_WRITE_REG(DACBlendCtrl, tmp);
+
+}
+
+int CreateOverlaySurface(volatile STG4000REG __iomem *pSTGReg,
+			 u32 inWidth,
+			 u32 inHeight,
+			 int bLinear,
+			 u32 ulOverlayOffset,
+			 u32 * retStride, u32 * retUVStride)
+{
+	u32 tmp;
+	u32 ulStride;
+
+	if (inWidth > STG4000_OVRL_MAX_WIDTH ||
+	    inHeight > STG4000_OVRL_MAX_HEIGHT) {
+		return -EINVAL;
+	}
+
+	/* Stride in 16 byte words - 16Bpp */
+	if (bLinear) {
+		/* Format is 16bits so num 16 byte words is width/8 */
+		if ((inWidth & 0x7) == 0) {	/* inWidth % 8 */
+			ulStride = (inWidth / 8);
+		} else {
+			/* Round up to next 16byte boundary */
+			ulStride = ((inWidth + 8) / 8);
+		}
+	} else {
+		/* Y component is 8bits so num 16 byte words is width/16 */
+		if ((inWidth & 0xf) == 0) {	/* inWidth % 16 */
+			ulStride = (inWidth / 16);
+		} else {
+			/* Round up to next 16byte boundary */
+			ulStride = ((inWidth + 16) / 16);
+		}
+	}
+
+
+	/* Set Overlay address and Format mode */
+	tmp = STG_READ_REG(DACOverlayAddr);
+	CLEAR_BITS_FRM_TO(0, 20);
+	if (bLinear) {
+		CLEAR_BIT(31);	/* Overlay format to Linear */
+	} else {
+		tmp |= SET_BIT(31);	/* Overlay format to Planer */
+	}
+
+	/* Only bits 24:4 of the Overlay address */
+	tmp |= (ulOverlayOffset >> 4);
+	STG_WRITE_REG(DACOverlayAddr, tmp);
+
+	if (!bLinear) {
+		u32 uvSize =
+		    (inWidth & 0x1) ? (inWidth + 1 / 2) : (inWidth / 2);
+		u32 uvStride;
+		u32 ulOffset;
+		/* Y component is 8bits so num 32 byte words is width/32 */
+		if ((uvSize & 0xf) == 0) {	/* inWidth % 16 */
+			uvStride = (uvSize / 16);
+		} else {
+			/* Round up to next 32byte boundary */
+			uvStride = ((uvSize + 16) / 16);
+		}
+
+		ulOffset = ulOverlayOffset + (inHeight * (ulStride * 16));
+		/* Align U,V data to 32byte boundary */
+		if ((ulOffset & 0x1f) != 0)
+			ulOffset = (ulOffset + 32L) & 0xffffffE0L;
+
+		tmp = STG_READ_REG(DACOverlayUAddr);
+		CLEAR_BITS_FRM_TO(0, 20);
+		tmp |= (ulOffset >> 4);
+		STG_WRITE_REG(DACOverlayUAddr, tmp);
+
+		ulOffset += (inHeight / 2) * (uvStride * 16);
+		/* Align U,V data to 32byte boundary */
+		if ((ulOffset & 0x1f) != 0)
+			ulOffset = (ulOffset + 32L) & 0xffffffE0L;
+
+		tmp = STG_READ_REG(DACOverlayVAddr);
+		CLEAR_BITS_FRM_TO(0, 20);
+		tmp |= (ulOffset >> 4);
+		STG_WRITE_REG(DACOverlayVAddr, tmp);
+
+		*retUVStride = uvStride * 16;
+	}
+
+
+	/* Set Overlay YUV pixel format
+	 * Make sure that LUT not used - ??????
+	 */
+	tmp = STG_READ_REG(DACPixelFormat);
+	/* Only support Planer or UYVY linear formats */
+	CLEAR_BITS_FRM_TO(4, 9);
+	STG_WRITE_REG(DACPixelFormat, tmp);
+
+	ovlWidth = inWidth;
+	ovlHeight = inHeight;
+	ovlStride = ulStride;
+	ovlLinear = bLinear;
+	*retStride = ulStride << 4;	/* In bytes */
+
+	return 0;
+}
+
+int SetOverlayBlendMode(volatile STG4000REG __iomem *pSTGReg,
+			OVRL_BLEND_MODE mode,
+			u32 ulAlpha, u32 ulColorKey)
+{
+	u32 tmp;
+
+	tmp = STG_READ_REG(DACBlendCtrl);
+	CLEAR_BITS_FRM_TO(28, 30);
+	tmp |= (mode << 28);
+
+	switch (mode) {
+	case COLOR_KEY:
+		CLEAR_BITS_FRM_TO(0, 23);
+		tmp |= (ulColorKey & 0x00FFFFFF);
+		break;
+
+	case GLOBAL_ALPHA:
+		CLEAR_BITS_FRM_TO(24, 27);
+		tmp |= ((ulAlpha & 0xF) << 24);
+		break;
+
+	case CK_PIXEL_ALPHA:
+		CLEAR_BITS_FRM_TO(0, 23);
+		tmp |= (ulColorKey & 0x00FFFFFF);
+		break;
+
+	case CK_GLOBAL_ALPHA:
+		CLEAR_BITS_FRM_TO(0, 23);
+		tmp |= (ulColorKey & 0x00FFFFFF);
+		CLEAR_BITS_FRM_TO(24, 27);
+		tmp |= ((ulAlpha & 0xF) << 24);
+		break;
+
+	case GRAPHICS_MODE:
+	case PER_PIXEL_ALPHA:
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	STG_WRITE_REG(DACBlendCtrl, tmp);
+
+	return 0;
+}
+
+void EnableOverlayPlane(volatile STG4000REG __iomem *pSTGReg)
+{
+	u32 tmp;
+	/* Enable Overlay */
+	tmp = STG_READ_REG(DACPixelFormat);
+	tmp |= SET_BIT(7);
+	STG_WRITE_REG(DACPixelFormat, tmp);
+
+	/* Set video stream control */
+	tmp = STG_READ_REG(DACStreamCtrl);
+	tmp |= SET_BIT(1);	/* video stream */
+	STG_WRITE_REG(DACStreamCtrl, tmp);
+}
+
+static u32 Overlap(u32 ulBits, u32 ulPattern)
+{
+	u32 ulCount = 0;
+
+	while (ulBits) {
+		if (!(ulPattern & 1))
+			ulCount++;
+		ulBits--;
+		ulPattern = ulPattern >> 1;
+	}
+
+	return ulCount;
+
+}
+
+int SetOverlayViewPort(volatile STG4000REG __iomem *pSTGReg,
+		       u32 left, u32 top,
+		       u32 right, u32 bottom)
+{
+	OVRL_SRC_DEST srcDest;
+
+	u32 ulSrcTop, ulSrcBottom;
+	u32 ulSrc, ulDest;
+	u32 ulFxScale, ulFxOffset;
+	u32 ulHeight, ulWidth;
+	u32 ulPattern;
+	u32 ulDecimate, ulDecimated;
+	u32 ulApplied;
+	u32 ulDacXScale, ulDacYScale;
+	u32 ulScale;
+	u32 ulLeft, ulRight;
+	u32 ulSrcLeft, ulSrcRight;
+	u32 ulScaleLeft;
+	u32 ulhDecim;
+	u32 ulsVal;
+	u32 ulVertDecFactor;
+	int bResult;
+	u32 ulClipOff = 0;
+	u32 ulBits = 0;
+	u32 ulsAdd = 0;
+	u32 tmp, ulStride;
+	u32 ulExcessPixels, ulClip, ulExtraLines;
+
+
+	srcDest.ulSrcX1 = 0;
+	srcDest.ulSrcY1 = 0;
+	srcDest.ulSrcX2 = ovlWidth - 1;
+	srcDest.ulSrcY2 = ovlHeight - 1;
+
+	srcDest.ulDstX1 = left;
+	srcDest.ulDstY1 = top;
+	srcDest.ulDstX2 = right;
+	srcDest.ulDstY2 = bottom;
+
+	srcDest.lDstX1 = srcDest.ulDstX1;
+	srcDest.lDstY1 = srcDest.ulDstY1;
+	srcDest.lDstX2 = srcDest.ulDstX2;
+	srcDest.lDstY2 = srcDest.ulDstY2;
+
+    /************* Vertical decimation/scaling ******************/
+
+	/* Get Src Top and Bottom */
+	ulSrcTop = srcDest.ulSrcY1;
+	ulSrcBottom = srcDest.ulSrcY2;
+
+	ulSrc = ulSrcBottom - ulSrcTop;
+	ulDest = srcDest.lDstY2 - srcDest.lDstY1;	/* on-screen overlay */
+
+	if (ulSrc <= 1)
+		return -EINVAL;
+
+	/* First work out the position we are to display as offset from the
+	 * source of the buffer
+	 */
+	ulFxScale = (ulDest << 11) / ulSrc;	/* fixed point scale factor */
+	ulFxOffset = (srcDest.lDstY2 - srcDest.ulDstY2) << 11;
+
+	ulSrcBottom = ulSrcBottom - (ulFxOffset / ulFxScale);
+	ulSrc = ulSrcBottom - ulSrcTop;
+	ulHeight = ulSrc;
+
+	ulDest = srcDest.ulDstY2 - (srcDest.ulDstY1 - 1);
+	ulPattern = adwDecim8[ulBits];
+
+	/* At this point ulSrc represents the input decimator */
+	if (ulSrc > ulDest) {
+		ulDecimate = ulSrc - ulDest;
+		ulBits = 0;
+		ulApplied = ulSrc / 32;
+
+		while (((ulBits * ulApplied) +
+			Overlap((ulSrc % 32),
+				adwDecim8[ulBits])) < ulDecimate)
+			ulBits++;
+
+		ulPattern = adwDecim8[ulBits];
+		ulDecimated =
+		    (ulBits * ulApplied) + Overlap((ulSrc % 32),
+						   ulPattern);
+		ulSrc = ulSrc - ulDecimated;	/* the number number of lines that will go into the scaler */
+	}
+
+	if (ulBits && (ulBits != 32)) {
+		ulVertDecFactor = (63 - ulBits) / (32 - ulBits);	/* vertical decimation factor scaled up to nearest integer */
+	} else {
+		ulVertDecFactor = 1;
+	}
+
+	ulDacYScale = ((ulSrc - 1) * 2048) / (ulDest + 1);
+
+	tmp = STG_READ_REG(DACOverlayVtDec);	/* Decimation */
+	CLEAR_BITS_FRM_TO(0, 31);
+	tmp = ulPattern;
+	STG_WRITE_REG(DACOverlayVtDec, tmp);
+
+	/***************** Horizontal decimation/scaling ***************************/
+
+	/*
+	 * Now we handle the horizontal case, this is a simplified version of
+	 * the vertical case in that we decimate by factors of 2.  as we are
+	 * working in words we should always be able to decimate by these
+	 * factors.  as we always have to have a buffer which is aligned to a
+	 * whole number of 128 bit words, we must align the left side to the
+	 * lowest to the next lowest 128 bit boundary, and the right hand edge
+	 * to the next largets boundary, (in a similar way to how we didi it in
+	 * PMX1) as the left and right hand edges are aligned to these
+	 * boundaries normally this only becomes an issue when we are chopping
+	 * of one of the sides We shall work out vertical stuff first
+	 */
+	ulSrc = srcDest.ulSrcX2 - srcDest.ulSrcX1;
+	ulDest = srcDest.lDstX2 - srcDest.lDstX1;
+#ifdef _OLDCODE
+	ulLeft = srcDest.ulDstX1;
+	ulRight = srcDest.ulDstX2;
+#else
+	if (srcDest.ulDstX1 > 2) {
+		ulLeft = srcDest.ulDstX1 + 2;
+		ulRight = srcDest.ulDstX2 + 1;
+	} else {
+		ulLeft = srcDest.ulDstX1;
+		ulRight = srcDest.ulDstX2 + 1;
+	}
+#endif
+	/* first work out the position we are to display as offset from the source of the buffer */
+	bResult = 1;
+
+	do {
+		if (ulDest == 0)
+			return -EINVAL;
+
+		/* source pixels per dest pixel <<11 */
+		ulFxScale = ((ulSrc - 1) << 11) / (ulDest);
+
+		/* then number of destination pixels out we are */
+		ulFxOffset = ulFxScale * ((srcDest.ulDstX1 - srcDest.lDstX1) + ulClipOff);
+		ulFxOffset >>= 11;
+
+		/* this replaces the code which was making a decision as to use either ulFxOffset or ulSrcX1 */
+		ulSrcLeft = srcDest.ulSrcX1 + ulFxOffset;
+
+		/* then number of destination pixels out we are */
+		ulFxOffset = ulFxScale * (srcDest.lDstX2 - srcDest.ulDstX2);
+		ulFxOffset >>= 11;
+
+		ulSrcRight = srcDest.ulSrcX2 - ulFxOffset;
+
+		/*
+		 * we must align these to our 128 bit boundaries. we shall
+		 * round down the pixel pos to the nearest 8 pixels.
+		 */
+		ulScaleLeft = ulSrcLeft;
+
+		/* shift fxscale until it is in the range of the scaler */
+		ulhDecim = 0;
+		ulScale = (((ulSrcRight - ulSrcLeft) - 1) << (11 - ulhDecim)) / (ulRight - ulLeft + 2);
+
+		while (ulScale > 0x800) {
+			ulhDecim++;
+			ulScale = (((ulSrcRight - ulSrcLeft) - 1) << (11 - ulhDecim)) / (ulRight - ulLeft + 2);
+		}
+
+		/*
+		 * to try and get the best values We first try and use
+		 * src/dwdest for the scale factor, then we move onto src-1
+		 *
+		 * we want to check to see if we will need to clip data, if so
+		 * then we should clip our source so that we don't need to
+		 */
+		if (!ovlLinear) {
+			ulSrcLeft &= ~0x1f;
+
+			/*
+			 * we must align the right hand edge to the next 32
+			 * pixel` boundary, must be on a 256 boundary so u, and
+			 * v are 128 bit aligned
+			 */
+			ulSrcRight = (ulSrcRight + 0x1f) & ~0x1f;
+		} else {
+			ulSrcLeft &= ~0x7;
+
+			/*
+			 * we must align the right hand edge to the next
+			 * 8pixel` boundary
+			 */
+			ulSrcRight = (ulSrcRight + 0x7) & ~0x7;
+		}
+
+		/* this is the input size line store needs to cope with */
+		ulWidth = ulSrcRight - ulSrcLeft;
+
+		/*
+		 * use unclipped value to work out scale factror this is the
+		 * scale factor we want we shall now work out the horizonal
+		 * decimation and scaling
+		 */
+		ulsVal = ((ulWidth / 8) >> ulhDecim);
+
+		if ((ulWidth != (ulsVal << ulhDecim) * 8))
+			ulsAdd = 1;
+
+		/* input pixels to scaler; */
+		ulSrc = ulWidth >> ulhDecim;
+
+		if (ulSrc <= 2)
+			return -EINVAL;
+
+		ulExcessPixels = ((((ulScaleLeft - ulSrcLeft)) << (11 - ulhDecim)) / ulScale);
+
+		ulClip = (ulSrc << 11) / ulScale;
+		ulClip -= (ulRight - ulLeft);
+		ulClip += ulExcessPixels;
+
+		if (ulClip)
+			ulClip--;
+
+		/* We may need to do more here if we really have a HW rev < 5 */
+	} while (!bResult);
+
+	ulExtraLines = (1 << ulhDecim) * ulVertDecFactor;
+	ulExtraLines += 64;
+	ulHeight += ulExtraLines;
+
+	ulDacXScale = ulScale;
+
+
+	tmp = STG_READ_REG(DACVerticalScal);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 22);	/* Vertical Scaling */
+
+	/* Calculate new output line stride, this is always the number of 422
+	   words in the line buffer, so it doesn't matter if the
+	   mode is 420. Then set the vertical scale register.
+	 */
+	ulStride = (ulWidth >> (ulhDecim + 3)) + ulsAdd;
+	tmp |= ((ulStride << 16) | (ulDacYScale));	/* DAC_LS_CTRL = stride */
+	STG_WRITE_REG(DACVerticalScal, tmp);
+
+	/* Now set up the overlay size using the modified width and height
+	   from decimate and scaling calculations
+	 */
+	tmp = STG_READ_REG(DACOverlaySize);
+	CLEAR_BITS_FRM_TO(0, 10);
+	CLEAR_BITS_FRM_TO(12, 31);
+
+	if (ovlLinear) {
+		tmp |=
+		    (ovlStride | ((ulHeight + 1) << 12) |
+		     (((ulWidth / 8) - 1) << 23));
+	} else {
+		tmp |=
+		    (ovlStride | ((ulHeight + 1) << 12) |
+		     (((ulWidth / 32) - 1) << 23));
+	}
+
+	STG_WRITE_REG(DACOverlaySize, tmp);
+
+	/* Set Video Window Start */
+	tmp = ((ulLeft << 16)) | (srcDest.ulDstY1);
+	STG_WRITE_REG(DACVidWinStart, tmp);
+
+	/* Set Video Window End */
+	tmp = ((ulRight) << 16) | (srcDest.ulDstY2);
+	STG_WRITE_REG(DACVidWinEnd, tmp);
+
+	/* Finally set up the rest of the overlay regs in the order
+	   done in the IMG driver
+	 */
+	tmp = STG_READ_REG(DACPixelFormat);
+	tmp = ((ulExcessPixels << 16) | tmp) & 0x7fffffff;
+	STG_WRITE_REG(DACPixelFormat, tmp);
+
+	tmp = STG_READ_REG(DACHorizontalScal);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 17);
+	tmp |= ((ulhDecim << 16) | (ulDacXScale));
+	STG_WRITE_REG(DACHorizontalScal, tmp);
+
+	return 0;
+}
diff --git a/drivers/video/fbdev/kyro/STG4000Ramdac.c b/drivers/video/fbdev/kyro/STG4000Ramdac.c
new file mode 100644
index 000000000..e6ad037e4
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000Ramdac.c
@@ -0,0 +1,163 @@
+/*
+ *  linux/drivers/video/kyro/STG4000Ramdac.c
+ *
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/types.h>
+#include <video/kyro.h>
+
+#include "STG4000Reg.h"
+#include "STG4000Interface.h"
+
+static u32 STG_PIXEL_BUS_WIDTH = 128;	/* 128 bit bus width      */
+static u32 REF_CLOCK = 14318;
+
+int InitialiseRamdac(volatile STG4000REG __iomem * pSTGReg,
+		     u32 displayDepth,
+		     u32 displayWidth,
+		     u32 displayHeight,
+		     s32 HSyncPolarity,
+		     s32 VSyncPolarity, u32 * pixelClock)
+{
+	u32 tmp = 0;
+	u32 F = 0, R = 0, P = 0;
+	u32 stride = 0;
+	u32 ulPdiv = 0;
+	u32 physicalPixelDepth = 0;
+	/* Make sure DAC is in Reset */
+	tmp = STG_READ_REG(SoftwareReset);
+
+	if (tmp & 0x1) {
+		CLEAR_BIT(1);
+		STG_WRITE_REG(SoftwareReset, tmp);
+	}
+
+	/* Set Pixel Format */
+	tmp = STG_READ_REG(DACPixelFormat);
+	CLEAR_BITS_FRM_TO(0, 2);
+
+	/* Set LUT not used from 16bpp to 32 bpp ??? */
+	CLEAR_BITS_FRM_TO(8, 9);
+
+	switch (displayDepth) {
+	case 16:
+		{
+			physicalPixelDepth = 16;
+			tmp |= _16BPP;
+			break;
+		}
+	case 32:
+		{
+			/* Set for 32 bits per pixel */
+			physicalPixelDepth = 32;
+			tmp |= _32BPP;
+			break;
+		}
+	default:
+		return -EINVAL;
+	}
+
+	STG_WRITE_REG(DACPixelFormat, tmp);
+
+	/* Workout Bus transfer bandwidth according to pixel format */
+	ulPdiv = STG_PIXEL_BUS_WIDTH / physicalPixelDepth;
+
+	/* Get Screen Stride in pixels */
+	stride = displayWidth;
+
+	/* Set Primary size info */
+	tmp = STG_READ_REG(DACPrimSize);
+	CLEAR_BITS_FRM_TO(0, 10);
+	CLEAR_BITS_FRM_TO(12, 31);
+	tmp |=
+	    ((((displayHeight - 1) << 12) | (((displayWidth / ulPdiv) -
+					      1) << 23))
+	     | (stride / ulPdiv));
+	STG_WRITE_REG(DACPrimSize, tmp);
+
+
+	/* Set Pixel Clock */
+	*pixelClock = ProgramClock(REF_CLOCK, *pixelClock, &F, &R, &P);
+
+	/* Set DAC PLL Mode */
+	tmp = STG_READ_REG(DACPLLMode);
+	CLEAR_BITS_FRM_TO(0, 15);
+	/* tmp |= ((P-1) | ((F-2) << 2) | ((R-2) << 11)); */
+	tmp |= ((P) | ((F - 2) << 2) | ((R - 2) << 11));
+	STG_WRITE_REG(DACPLLMode, tmp);
+
+	/* Set Prim Address */
+	tmp = STG_READ_REG(DACPrimAddress);
+	CLEAR_BITS_FRM_TO(0, 20);
+	CLEAR_BITS_FRM_TO(20, 31);
+	STG_WRITE_REG(DACPrimAddress, tmp);
+
+	/* Set Cursor details with HW Cursor disabled */
+	tmp = STG_READ_REG(DACCursorCtrl);
+	tmp &= ~SET_BIT(31);
+	STG_WRITE_REG(DACCursorCtrl, tmp);
+
+	tmp = STG_READ_REG(DACCursorAddr);
+	CLEAR_BITS_FRM_TO(0, 20);
+	STG_WRITE_REG(DACCursorAddr, tmp);
+
+	/* Set Video Window */
+	tmp = STG_READ_REG(DACVidWinStart);
+	CLEAR_BITS_FRM_TO(0, 10);
+	CLEAR_BITS_FRM_TO(16, 26);
+	STG_WRITE_REG(DACVidWinStart, tmp);
+
+	tmp = STG_READ_REG(DACVidWinEnd);
+	CLEAR_BITS_FRM_TO(0, 10);
+	CLEAR_BITS_FRM_TO(16, 26);
+	STG_WRITE_REG(DACVidWinEnd, tmp);
+
+	/* Set DAC Border Color to default */
+	tmp = STG_READ_REG(DACBorderColor);
+	CLEAR_BITS_FRM_TO(0, 23);
+	STG_WRITE_REG(DACBorderColor, tmp);
+
+	/* Set Graphics and Overlay Burst Control */
+	STG_WRITE_REG(DACBurstCtrl, 0x0404);
+
+	/* Set CRC Trigger to default */
+	tmp = STG_READ_REG(DACCrcTrigger);
+	CLEAR_BIT(0);
+	STG_WRITE_REG(DACCrcTrigger, tmp);
+
+	/* Set Video Port Control to default */
+	tmp = STG_READ_REG(DigVidPortCtrl);
+	CLEAR_BIT(8);
+	CLEAR_BITS_FRM_TO(16, 27);
+	CLEAR_BITS_FRM_TO(1, 3);
+	CLEAR_BITS_FRM_TO(10, 11);
+	STG_WRITE_REG(DigVidPortCtrl, tmp);
+
+	return 0;
+}
+
+/* Ramdac control, turning output to the screen on and off */
+void DisableRamdacOutput(volatile STG4000REG __iomem * pSTGReg)
+{
+	u32 tmp;
+
+	/* Disable DAC for Graphics Stream Control */
+	tmp = (STG_READ_REG(DACStreamCtrl)) & ~SET_BIT(0);
+	STG_WRITE_REG(DACStreamCtrl, tmp);
+}
+
+void EnableRamdacOutput(volatile STG4000REG __iomem * pSTGReg)
+{
+	u32 tmp;
+
+	/* Enable DAC for Graphics Stream Control */
+	tmp = (STG_READ_REG(DACStreamCtrl)) | SET_BIT(0);
+	STG_WRITE_REG(DACStreamCtrl, tmp);
+}
diff --git a/drivers/video/fbdev/kyro/STG4000Reg.h b/drivers/video/fbdev/kyro/STG4000Reg.h
new file mode 100644
index 000000000..50f4670e9
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000Reg.h
@@ -0,0 +1,283 @@
+/*
+ *  linux/drivers/video/kyro/STG4000Reg.h
+ *
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#ifndef _STG4000REG_H
+#define _STG4000REG_H
+
+#define DWFILL unsigned long :32
+#define WFILL unsigned short :16
+
+/*
+ * Macros that access memory mapped card registers in PCI space
+ * Add an appropriate section for your OS or processor architecture.
+ */
+#if defined(__KERNEL__)
+#include <asm/page.h>
+#include <asm/io.h>
+#define STG_WRITE_REG(reg,data) (writel(data,&pSTGReg->reg))
+#define STG_READ_REG(reg)      (readl(&pSTGReg->reg))
+#else
+#define STG_WRITE_REG(reg,data) (pSTGReg->reg = data)
+#define STG_READ_REG(reg)      (pSTGReg->reg)
+#endif /* __KERNEL__ */
+
+#define SET_BIT(n) (1<<(n))
+#define CLEAR_BIT(n) (tmp &= ~(1<<n))
+#define CLEAR_BITS_FRM_TO(frm, to) \
+{\
+int i; \
+    for(i = frm; i<= to; i++) \
+	{ \
+	    tmp &= ~(1<<i); \
+	} \
+}
+
+#define CLEAR_BIT_2(n) (usTemp &= ~(1<<n))
+#define CLEAR_BITS_FRM_TO_2(frm, to) \
+{\
+int i; \
+    for(i = frm; i<= to; i++) \
+	{ \
+	    usTemp &= ~(1<<i); \
+	} \
+}
+
+/* LUT select */
+typedef enum _LUT_USES {
+	NO_LUT = 0, RESERVED, GRAPHICS, OVERLAY
+} LUT_USES;
+
+/* Primary surface pixel format select */
+typedef enum _PIXEL_FORMAT {
+	_8BPP = 0, _15BPP, _16BPP, _24BPP, _32BPP
+} PIXEL_FORMAT;
+
+/* Overlay blending mode select */
+typedef enum _BLEND_MODE {
+	GRAPHICS_MODE = 0, COLOR_KEY, PER_PIXEL_ALPHA, GLOBAL_ALPHA,
+	CK_PIXEL_ALPHA, CK_GLOBAL_ALPHA
+} OVRL_BLEND_MODE;
+
+/* Overlay Pixel format select */
+typedef enum _OVRL_PIX_FORMAT {
+	UYVY, VYUY, YUYV, YVYU
+} OVRL_PIX_FORMAT;
+
+/* Register Table */
+typedef struct {
+	/* 0h  */
+	volatile u32 Thread0Enable;	/* 0x0000 */
+	volatile u32 Thread1Enable;	/* 0x0004 */
+	volatile u32 Thread0Recover;	/* 0x0008 */
+	volatile u32 Thread1Recover;	/* 0x000C */
+	volatile u32 Thread0Step;	/* 0x0010 */
+	volatile u32 Thread1Step;	/* 0x0014 */
+	volatile u32 VideoInStatus;	/* 0x0018 */
+	volatile u32 Core2InSignStart;	/* 0x001C */
+	volatile u32 Core1ResetVector;	/* 0x0020 */
+	volatile u32 Core1ROMOffset;	/* 0x0024 */
+	volatile u32 Core1ArbiterPriority;	/* 0x0028 */
+	volatile u32 VideoInControl;	/* 0x002C */
+	volatile u32 VideoInReg0CtrlA;	/* 0x0030 */
+	volatile u32 VideoInReg0CtrlB;	/* 0x0034 */
+	volatile u32 VideoInReg1CtrlA;	/* 0x0038 */
+	volatile u32 VideoInReg1CtrlB;	/* 0x003C */
+	volatile u32 Thread0Kicker;	/* 0x0040 */
+	volatile u32 Core2InputSign;	/* 0x0044 */
+	volatile u32 Thread0ProgCtr;	/* 0x0048 */
+	volatile u32 Thread1ProgCtr;	/* 0x004C */
+	volatile u32 Thread1Kicker;	/* 0x0050 */
+	volatile u32 GPRegister1;	/* 0x0054 */
+	volatile u32 GPRegister2;	/* 0x0058 */
+	volatile u32 GPRegister3;	/* 0x005C */
+	volatile u32 GPRegister4;	/* 0x0060 */
+	volatile u32 SerialIntA;	/* 0x0064 */
+
+	volatile u32 Fill0[6];	/* GAP 0x0068 - 0x007C */
+
+	volatile u32 SoftwareReset;	/* 0x0080 */
+	volatile u32 SerialIntB;	/* 0x0084 */
+
+	volatile u32 Fill1[37];	/* GAP 0x0088 - 0x011C */
+
+	volatile u32 ROMELQV;	/* 0x011C */
+	volatile u32 WLWH;	/* 0x0120 */
+	volatile u32 ROMELWL;	/* 0x0124 */
+
+	volatile u32 dwFill_1;	/* GAP 0x0128 */
+
+	volatile u32 IntStatus;	/* 0x012C */
+	volatile u32 IntMask;	/* 0x0130 */
+	volatile u32 IntClear;	/* 0x0134 */
+
+	volatile u32 Fill2[6];	/* GAP 0x0138 - 0x014C */
+
+	volatile u32 ROMGPIOA;	/* 0x0150 */
+	volatile u32 ROMGPIOB;	/* 0x0154 */
+	volatile u32 ROMGPIOC;	/* 0x0158 */
+	volatile u32 ROMGPIOD;	/* 0x015C */
+
+	volatile u32 Fill3[2];	/* GAP 0x0160 - 0x0168 */
+
+	volatile u32 AGPIntID;	/* 0x0168 */
+	volatile u32 AGPIntClassCode;	/* 0x016C */
+	volatile u32 AGPIntBIST;	/* 0x0170 */
+	volatile u32 AGPIntSSID;	/* 0x0174 */
+	volatile u32 AGPIntPMCSR;	/* 0x0178 */
+	volatile u32 VGAFrameBufBase;	/* 0x017C */
+	volatile u32 VGANotify;	/* 0x0180 */
+	volatile u32 DACPLLMode;	/* 0x0184 */
+	volatile u32 Core1VideoClockDiv;	/* 0x0188 */
+	volatile u32 AGPIntStat;	/* 0x018C */
+
+	/*
+	   volatile u32 Fill4[0x0400/4 - 0x0190/4]; //GAP 0x0190 - 0x0400
+	   volatile u32 Fill5[0x05FC/4 - 0x0400/4]; //GAP 0x0400 - 0x05FC Fog Table
+	   volatile u32 Fill6[0x0604/4 - 0x0600/4]; //GAP 0x0600 - 0x0604
+	   volatile u32 Fill7[0x0680/4 - 0x0608/4]; //GAP 0x0608 - 0x0680
+	   volatile u32 Fill8[0x07FC/4 - 0x0684/4]; //GAP 0x0684 - 0x07FC
+	 */
+	volatile u32 Fill4[412];	/* 0x0190 - 0x07FC */
+
+	volatile u32 TACtrlStreamBase;	/* 0x0800 */
+	volatile u32 TAObjDataBase;	/* 0x0804 */
+	volatile u32 TAPtrDataBase;	/* 0x0808 */
+	volatile u32 TARegionDataBase;	/* 0x080C */
+	volatile u32 TATailPtrBase;	/* 0x0810 */
+	volatile u32 TAPtrRegionSize;	/* 0x0814 */
+	volatile u32 TAConfiguration;	/* 0x0818 */
+	volatile u32 TAObjDataStartAddr;	/* 0x081C */
+	volatile u32 TAObjDataEndAddr;	/* 0x0820 */
+	volatile u32 TAXScreenClip;	/* 0x0824 */
+	volatile u32 TAYScreenClip;	/* 0x0828 */
+	volatile u32 TARHWClamp;	/* 0x082C */
+	volatile u32 TARHWCompare;	/* 0x0830 */
+	volatile u32 TAStart;	/* 0x0834 */
+	volatile u32 TAObjReStart;	/* 0x0838 */
+	volatile u32 TAPtrReStart;	/* 0x083C */
+	volatile u32 TAStatus1;	/* 0x0840 */
+	volatile u32 TAStatus2;	/* 0x0844 */
+	volatile u32 TAIntStatus;	/* 0x0848 */
+	volatile u32 TAIntMask;	/* 0x084C */
+
+	volatile u32 Fill5[235];	/* GAP 0x0850 - 0x0BF8 */
+
+	volatile u32 TextureAddrThresh;	/* 0x0BFC */
+	volatile u32 Core1Translation;	/* 0x0C00 */
+	volatile u32 TextureAddrReMap;	/* 0x0C04 */
+	volatile u32 RenderOutAGPRemap;	/* 0x0C08 */
+	volatile u32 _3DRegionReadTrans;	/* 0x0C0C */
+	volatile u32 _3DPtrReadTrans;	/* 0x0C10 */
+	volatile u32 _3DParamReadTrans;	/* 0x0C14 */
+	volatile u32 _3DRegionReadThresh;	/* 0x0C18 */
+	volatile u32 _3DPtrReadThresh;	/* 0x0C1C */
+	volatile u32 _3DParamReadThresh;	/* 0x0C20 */
+	volatile u32 _3DRegionReadAGPRemap;	/* 0x0C24 */
+	volatile u32 _3DPtrReadAGPRemap;	/* 0x0C28 */
+	volatile u32 _3DParamReadAGPRemap;	/* 0x0C2C */
+	volatile u32 ZBufferAGPRemap;	/* 0x0C30 */
+	volatile u32 TAIndexAGPRemap;	/* 0x0C34 */
+	volatile u32 TAVertexAGPRemap;	/* 0x0C38 */
+	volatile u32 TAUVAddrTrans;	/* 0x0C3C */
+	volatile u32 TATailPtrCacheTrans;	/* 0x0C40 */
+	volatile u32 TAParamWriteTrans;	/* 0x0C44 */
+	volatile u32 TAPtrWriteTrans;	/* 0x0C48 */
+	volatile u32 TAParamWriteThresh;	/* 0x0C4C */
+	volatile u32 TAPtrWriteThresh;	/* 0x0C50 */
+	volatile u32 TATailPtrCacheAGPRe;	/* 0x0C54 */
+	volatile u32 TAParamWriteAGPRe;	/* 0x0C58 */
+	volatile u32 TAPtrWriteAGPRe;	/* 0x0C5C */
+	volatile u32 SDRAMArbiterConf;	/* 0x0C60 */
+	volatile u32 SDRAMConf0;	/* 0x0C64 */
+	volatile u32 SDRAMConf1;	/* 0x0C68 */
+	volatile u32 SDRAMConf2;	/* 0x0C6C */
+	volatile u32 SDRAMRefresh;	/* 0x0C70 */
+	volatile u32 SDRAMPowerStat;	/* 0x0C74 */
+
+	volatile u32 Fill6[2];	/* GAP 0x0C78 - 0x0C7C */
+
+	volatile u32 RAMBistData;	/* 0x0C80 */
+	volatile u32 RAMBistCtrl;	/* 0x0C84 */
+	volatile u32 FIFOBistKey;	/* 0x0C88 */
+	volatile u32 RAMBistResult;	/* 0x0C8C */
+	volatile u32 FIFOBistResult;	/* 0x0C90 */
+
+	/*
+	   volatile u32 Fill11[0x0CBC/4 - 0x0C94/4]; //GAP 0x0C94 - 0x0CBC
+	   volatile u32 Fill12[0x0CD0/4 - 0x0CC0/4]; //GAP 0x0CC0 - 0x0CD0 3DRegisters
+	 */
+
+	volatile u32 Fill7[16];	/* 0x0c94 - 0x0cd0 */
+
+	volatile u32 SDRAMAddrSign;	/* 0x0CD4 */
+	volatile u32 SDRAMDataSign;	/* 0x0CD8 */
+	volatile u32 SDRAMSignConf;	/* 0x0CDC */
+
+	/* DWFILL; //GAP 0x0CE0 */
+	volatile u32 dwFill_2;
+
+	volatile u32 ISPSignature;	/* 0x0CE4 */
+
+	volatile u32 Fill8[454];	/*GAP 0x0CE8 - 0x13FC */
+
+	volatile u32 DACPrimAddress;	/* 0x1400 */
+	volatile u32 DACPrimSize;	/* 0x1404 */
+	volatile u32 DACCursorAddr;	/* 0x1408 */
+	volatile u32 DACCursorCtrl;	/* 0x140C */
+	volatile u32 DACOverlayAddr;	/* 0x1410 */
+	volatile u32 DACOverlayUAddr;	/* 0x1414 */
+	volatile u32 DACOverlayVAddr;	/* 0x1418 */
+	volatile u32 DACOverlaySize;	/* 0x141C */
+	volatile u32 DACOverlayVtDec;	/* 0x1420 */
+
+	volatile u32 Fill9[9];	/* GAP 0x1424 - 0x1444 */
+
+	volatile u32 DACVerticalScal;	/* 0x1448 */
+	volatile u32 DACPixelFormat;	/* 0x144C */
+	volatile u32 DACHorizontalScal;	/* 0x1450 */
+	volatile u32 DACVidWinStart;	/* 0x1454 */
+	volatile u32 DACVidWinEnd;	/* 0x1458 */
+	volatile u32 DACBlendCtrl;	/* 0x145C */
+	volatile u32 DACHorTim1;	/* 0x1460 */
+	volatile u32 DACHorTim2;	/* 0x1464 */
+	volatile u32 DACHorTim3;	/* 0x1468 */
+	volatile u32 DACVerTim1;	/* 0x146C */
+	volatile u32 DACVerTim2;	/* 0x1470 */
+	volatile u32 DACVerTim3;	/* 0x1474 */
+	volatile u32 DACBorderColor;	/* 0x1478 */
+	volatile u32 DACSyncCtrl;	/* 0x147C */
+	volatile u32 DACStreamCtrl;	/* 0x1480 */
+	volatile u32 DACLUTAddress;	/* 0x1484 */
+	volatile u32 DACLUTData;	/* 0x1488 */
+	volatile u32 DACBurstCtrl;	/* 0x148C */
+	volatile u32 DACCrcTrigger;	/* 0x1490 */
+	volatile u32 DACCrcDone;	/* 0x1494 */
+	volatile u32 DACCrcResult1;	/* 0x1498 */
+	volatile u32 DACCrcResult2;	/* 0x149C */
+	volatile u32 DACLinecount;	/* 0x14A0 */
+
+	volatile u32 Fill10[151];	/*GAP 0x14A4 - 0x16FC */
+
+	volatile u32 DigVidPortCtrl;	/* 0x1700 */
+	volatile u32 DigVidPortStat;	/* 0x1704 */
+
+	/*
+	   volatile u32 Fill11[0x1FFC/4 - 0x1708/4]; //GAP 0x1708 - 0x1FFC
+	   volatile u32 Fill17[0x3000/4 - 0x2FFC/4]; //GAP 0x2000 - 0x2FFC ALUT
+	 */
+
+	volatile u32 Fill11[1598];
+
+	/* DWFILL; //GAP 0x3000          ALUT 256MB offset */
+	volatile u32 Fill_3;
+
+} STG4000REG;
+
+#endif /* _STG4000REG_H */
diff --git a/drivers/video/fbdev/kyro/STG4000VTG.c b/drivers/video/fbdev/kyro/STG4000VTG.c
new file mode 100644
index 000000000..bd389709d
--- /dev/null
+++ b/drivers/video/fbdev/kyro/STG4000VTG.c
@@ -0,0 +1,170 @@
+/*
+ *  linux/drivers/video/kyro/STG4000VTG.c
+ *
+ *  Copyright (C) 2002 STMicroelectronics
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/types.h>
+#include <video/kyro.h>
+
+#include "STG4000Reg.h"
+#include "STG4000Interface.h"
+
+void DisableVGA(volatile STG4000REG __iomem *pSTGReg)
+{
+	u32 tmp;
+	volatile u32 count = 0, i;
+
+	/* Reset the VGA registers */
+	tmp = STG_READ_REG(SoftwareReset);
+	CLEAR_BIT(8);
+	STG_WRITE_REG(SoftwareReset, tmp);
+
+	/* Just for Delay */
+	for (i = 0; i < 1000; i++) {
+		count++;
+	}
+
+	/* Pull-out the VGA registers from reset */
+	tmp = STG_READ_REG(SoftwareReset);
+	tmp |= SET_BIT(8);
+	STG_WRITE_REG(SoftwareReset, tmp);
+}
+
+void StopVTG(volatile STG4000REG __iomem *pSTGReg)
+{
+	u32 tmp = 0;
+
+	/* Stop Ver and Hor Sync Generator */
+	tmp = (STG_READ_REG(DACSyncCtrl)) | SET_BIT(0) | SET_BIT(2);
+	CLEAR_BIT(31);
+	STG_WRITE_REG(DACSyncCtrl, tmp);
+}
+
+void StartVTG(volatile STG4000REG __iomem *pSTGReg)
+{
+	u32 tmp = 0;
+
+	/* Start Ver and Hor Sync Generator */
+	tmp = ((STG_READ_REG(DACSyncCtrl)) | SET_BIT(31));
+	CLEAR_BIT(0);
+	CLEAR_BIT(2);
+	STG_WRITE_REG(DACSyncCtrl, tmp);
+}
+
+void SetupVTG(volatile STG4000REG __iomem *pSTGReg,
+	      const struct kyrofb_info * pTiming)
+{
+	u32 tmp = 0;
+	u32 margins = 0;
+	u32 ulBorder;
+	u32 xRes = pTiming->XRES;
+	u32 yRes = pTiming->YRES;
+
+	/* Horizontal */
+	u32 HAddrTime, HRightBorder, HLeftBorder;
+	u32 HBackPorcStrt, HFrontPorchStrt, HTotal,
+	    HLeftBorderStrt, HRightBorderStrt, HDisplayStrt;
+
+	/* Vertical */
+	u32 VDisplayStrt, VBottomBorder, VTopBorder;
+	u32 VBackPorchStrt, VTotal, VTopBorderStrt,
+	    VFrontPorchStrt, VBottomBorderStrt, VAddrTime;
+
+	/* Need to calculate the right border */
+	if ((xRes == 640) && (yRes == 480)) {
+		if ((pTiming->VFREQ == 60) || (pTiming->VFREQ == 72)) {
+			margins = 8;
+		}
+	}
+
+	/* Work out the Border */
+	ulBorder =
+	    (pTiming->HTot -
+	     (pTiming->HST + (pTiming->HBP - margins) + xRes +
+	      (pTiming->HFP - margins))) >> 1;
+
+	/* Border the same for Vertical and Horizontal */
+	VBottomBorder = HLeftBorder = VTopBorder = HRightBorder = ulBorder;
+
+    /************ Get Timing values for Horizontal ******************/
+	HAddrTime = xRes;
+	HBackPorcStrt = pTiming->HST;
+	HTotal = pTiming->HTot;
+	HDisplayStrt =
+	    pTiming->HST + (pTiming->HBP - margins) + HLeftBorder;
+	HLeftBorderStrt = HDisplayStrt - HLeftBorder;
+	HFrontPorchStrt =
+	    pTiming->HST + (pTiming->HBP - margins) + HLeftBorder +
+	    HAddrTime + HRightBorder;
+	HRightBorderStrt = HFrontPorchStrt - HRightBorder;
+
+    /************ Get Timing values for Vertical ******************/
+	VAddrTime = yRes;
+	VBackPorchStrt = pTiming->VST;
+	VTotal = pTiming->VTot;
+	VDisplayStrt =
+	    pTiming->VST + (pTiming->VBP - margins) + VTopBorder;
+	VTopBorderStrt = VDisplayStrt - VTopBorder;
+	VFrontPorchStrt =
+	    pTiming->VST + (pTiming->VBP - margins) + VTopBorder +
+	    VAddrTime + VBottomBorder;
+	VBottomBorderStrt = VFrontPorchStrt - VBottomBorder;
+
+	/* Set Hor Timing 1, 2, 3 */
+	tmp = STG_READ_REG(DACHorTim1);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (HTotal) | (HBackPorcStrt << 16);
+	STG_WRITE_REG(DACHorTim1, tmp);
+
+	tmp = STG_READ_REG(DACHorTim2);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (HDisplayStrt << 16) | HLeftBorderStrt;
+	STG_WRITE_REG(DACHorTim2, tmp);
+
+	tmp = STG_READ_REG(DACHorTim3);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (HFrontPorchStrt << 16) | HRightBorderStrt;
+	STG_WRITE_REG(DACHorTim3, tmp);
+
+	/* Set Ver Timing 1, 2, 3 */
+	tmp = STG_READ_REG(DACVerTim1);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (VBackPorchStrt << 16) | (VTotal);
+	STG_WRITE_REG(DACVerTim1, tmp);
+
+	tmp = STG_READ_REG(DACVerTim2);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (VDisplayStrt << 16) | VTopBorderStrt;
+	STG_WRITE_REG(DACVerTim2, tmp);
+
+	tmp = STG_READ_REG(DACVerTim3);
+	CLEAR_BITS_FRM_TO(0, 11);
+	CLEAR_BITS_FRM_TO(16, 27);
+	tmp |= (VFrontPorchStrt << 16) | VBottomBorderStrt;
+	STG_WRITE_REG(DACVerTim3, tmp);
+
+	/* Set Verical and Horizontal Polarity */
+	tmp = STG_READ_REG(DACSyncCtrl) | SET_BIT(3) | SET_BIT(1);
+
+	if ((pTiming->HSP > 0) && (pTiming->VSP < 0)) {	/* +hsync -vsync */
+		tmp &= ~0x8;
+	} else if ((pTiming->HSP < 0) && (pTiming->VSP > 0)) {	/* -hsync +vsync */
+		tmp &= ~0x2;
+	} else if ((pTiming->HSP < 0) && (pTiming->VSP < 0)) {	/* -hsync -vsync */
+		tmp &= ~0xA;
+	} else if ((pTiming->HSP > 0) && (pTiming->VSP > 0)) {	/* +hsync -vsync */
+		tmp &= ~0x0;
+	}
+
+	STG_WRITE_REG(DACSyncCtrl, tmp);
+}
diff --git a/drivers/video/fbdev/kyro/fbdev.c b/drivers/video/fbdev/kyro/fbdev.c
new file mode 100644
index 000000000..c23738988
--- /dev/null
+++ b/drivers/video/fbdev/kyro/fbdev.c
@@ -0,0 +1,836 @@
+/*
+ *  linux/drivers/video/kyro/fbdev.c
+ *
+ *  Copyright (C) 2002 STMicroelectronics
+ *  Copyright (C) 2003, 2004 Paul Mundt
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/aperture.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/delay.h>
+#include <linux/fb.h>
+#include <linux/ioctl.h>
+#include <linux/init.h>
+#include <linux/pci.h>
+#include <asm/io.h>
+#include <linux/uaccess.h>
+
+#include <video/kyro.h>
+
+#include "STG4000Reg.h"
+#include "STG4000Interface.h"
+
+/*
+ * PCI Definitions
+ */
+#define PCI_VENDOR_ID_ST	0x104a
+#define PCI_DEVICE_ID_STG4000	0x0010
+
+#define KHZ2PICOS(a) (1000000000UL/(a))
+
+/****************************************************************************/
+static struct fb_fix_screeninfo kyro_fix = {
+	.id		= "ST Kyro",
+	.type		= FB_TYPE_PACKED_PIXELS,
+	.visual		= FB_VISUAL_TRUECOLOR,
+	.accel		= FB_ACCEL_NONE,
+};
+
+static const struct fb_var_screeninfo kyro_var = {
+	/* 640x480, 16bpp @ 60 Hz */
+	.xres		= 640,
+	.yres		= 480,
+	.xres_virtual	= 640,
+	.yres_virtual	= 480,
+	.bits_per_pixel	= 16,
+	.red		= { 11, 5, 0 },
+	.green		= {  5, 6, 0 },
+	.blue		= {  0, 5, 0 },
+	.activate	= FB_ACTIVATE_NOW,
+	.height		= -1,
+	.width		= -1,
+	.pixclock	= KHZ2PICOS(25175),
+	.left_margin	= 48,
+	.right_margin	= 16,
+	.upper_margin	= 33,
+	.lower_margin	= 10,
+	.hsync_len	= 96,
+	.vsync_len	= 2,
+	.vmode		= FB_VMODE_NONINTERLACED,
+};
+
+typedef struct {
+	STG4000REG __iomem *pSTGReg;	/* Virtual address of PCI register region */
+	u32 ulNextFreeVidMem;	/* Offset from start of vid mem to next free region */
+	u32 ulOverlayOffset;	/* Offset from start of vid mem to overlay */
+	u32 ulOverlayStride;	/* Interleaved YUV and 422 mode Y stride */
+	u32 ulOverlayUVStride;	/* 422 mode U & V stride */
+} device_info_t;
+
+/* global graphics card info structure (one per card) */
+static device_info_t deviceInfo;
+
+static char *mode_option = NULL;
+static int nopan = 0;
+static int nowrap = 1;
+static int nomtrr = 0;
+
+/* PCI driver prototypes */
+static int kyrofb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
+static void kyrofb_remove(struct pci_dev *pdev);
+
+static struct fb_videomode kyro_modedb[] = {
+	{
+		/* 640x350 @ 85Hz */
+		NULL, 85, 640, 350, KHZ2PICOS(31500),
+		96, 32, 60, 32, 64, 3,
+		FB_SYNC_HOR_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 640x400 @ 85Hz */
+		NULL, 85, 640, 400, KHZ2PICOS(31500),
+		96, 32, 41, 1, 64, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 720x400 @ 85Hz */
+		NULL, 85, 720, 400, KHZ2PICOS(35500),
+		108, 36, 42, 1, 72, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 640x480 @ 60Hz */
+		NULL, 60, 640, 480, KHZ2PICOS(25175),
+		48, 16, 33, 10, 96, 2,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 640x480 @ 72Hz */
+		NULL, 72, 640, 480, KHZ2PICOS(31500),
+		128, 24, 28, 9, 40, 3,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 640x480 @ 75Hz */
+		NULL, 75, 640, 480, KHZ2PICOS(31500),
+		120, 16, 16, 1, 64, 3,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 640x480 @ 85Hz */
+		NULL, 85, 640, 480, KHZ2PICOS(36000),
+		80, 56, 25, 1, 56, 3,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 800x600 @ 56Hz */
+		NULL, 56, 800, 600, KHZ2PICOS(36000),
+		128, 24, 22, 1, 72, 2,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 800x600 @ 60Hz */
+		NULL, 60, 800, 600, KHZ2PICOS(40000),
+		88, 40, 23, 1, 128, 4,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 800x600 @ 72Hz */
+		NULL, 72, 800, 600, KHZ2PICOS(50000),
+		64, 56, 23, 37, 120, 6,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 800x600 @ 75Hz */
+		NULL, 75, 800, 600, KHZ2PICOS(49500),
+		160, 16, 21, 1, 80, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 800x600 @ 85Hz */
+		NULL, 85, 800, 600, KHZ2PICOS(56250),
+		152, 32, 27, 1, 64, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1024x768 @ 60Hz */
+		NULL, 60, 1024, 768, KHZ2PICOS(65000),
+		160, 24, 29, 3, 136, 6,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1024x768 @ 70Hz */
+		NULL, 70, 1024, 768, KHZ2PICOS(75000),
+		144, 24, 29, 3, 136, 6,
+		0, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1024x768 @ 75Hz */
+		NULL, 75, 1024, 768, KHZ2PICOS(78750),
+		176, 16, 28, 1, 96, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1024x768 @ 85Hz */
+		NULL, 85, 1024, 768, KHZ2PICOS(94500),
+		208, 48, 36, 1, 96, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1152x864 @ 75Hz */
+		NULL, 75, 1152, 864, KHZ2PICOS(108000),
+		256, 64, 32, 1, 128, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1280x960 @ 60Hz */
+		NULL, 60, 1280, 960, KHZ2PICOS(108000),
+		312, 96, 36, 1, 112, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1280x960 @ 85Hz */
+		NULL, 85, 1280, 960, KHZ2PICOS(148500),
+		224, 64, 47, 1, 160, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1280x1024 @ 60Hz */
+		NULL, 60, 1280, 1024, KHZ2PICOS(108000),
+		248, 48, 38, 1, 112, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1280x1024 @ 75Hz */
+		NULL, 75, 1280, 1024, KHZ2PICOS(135000),
+		248, 16, 38, 1, 144, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1280x1024 @ 85Hz */
+		NULL, 85, 1280, 1024, KHZ2PICOS(157500),
+		224, 64, 44, 1, 160, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1600x1200 @ 60Hz */
+		NULL, 60, 1600, 1200, KHZ2PICOS(162000),
+		304, 64, 46, 1, 192, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1600x1200 @ 65Hz */
+		NULL, 65, 1600, 1200, KHZ2PICOS(175500),
+		304, 64, 46, 1, 192, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1600x1200 @ 70Hz */
+		NULL, 70, 1600, 1200, KHZ2PICOS(189000),
+		304, 64, 46, 1, 192, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1600x1200 @ 75Hz */
+		NULL, 75, 1600, 1200, KHZ2PICOS(202500),
+		304, 64, 46, 1, 192, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1600x1200 @ 85Hz */
+		NULL, 85, 1600, 1200, KHZ2PICOS(229500),
+		304, 64, 46, 1, 192, 3,
+		FB_SYNC_HOR_HIGH_ACT | FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1792x1344 @ 60Hz */
+		NULL, 60, 1792, 1344, KHZ2PICOS(204750),
+		328, 128, 46, 1, 200, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1792x1344 @ 75Hz */
+		NULL, 75, 1792, 1344, KHZ2PICOS(261000),
+		352, 96, 69, 1, 216, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1856x1392 @ 60Hz */
+		NULL, 60, 1856, 1392, KHZ2PICOS(218250),
+		352, 96, 43, 1, 224, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1856x1392 @ 75Hz */
+		NULL, 75, 1856, 1392, KHZ2PICOS(288000),
+		352, 128, 104, 1, 224, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1920x1440 @ 60Hz */
+		NULL, 60, 1920, 1440, KHZ2PICOS(234000),
+		344, 128, 56, 1, 208, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	}, {
+		/* 1920x1440 @ 75Hz */
+		NULL, 75, 1920, 1440, KHZ2PICOS(297000),
+		352, 144, 56, 1, 224, 3,
+		FB_SYNC_VERT_HIGH_ACT, FB_VMODE_NONINTERLACED
+	},
+};
+#define NUM_TOTAL_MODES	ARRAY_SIZE(kyro_modedb)
+
+/*
+ * This needs to be kept ordered corresponding to kyro_modedb.
+ */
+enum {
+	VMODE_640_350_85,
+	VMODE_640_400_85,
+	VMODE_720_400_85,
+	VMODE_640_480_60,
+	VMODE_640_480_72,
+	VMODE_640_480_75,
+	VMODE_640_480_85,
+	VMODE_800_600_56,
+	VMODE_800_600_60,
+	VMODE_800_600_72,
+	VMODE_800_600_75,
+	VMODE_800_600_85,
+	VMODE_1024_768_60,
+	VMODE_1024_768_70,
+	VMODE_1024_768_75,
+	VMODE_1024_768_85,
+	VMODE_1152_864_75,
+	VMODE_1280_960_60,
+	VMODE_1280_960_85,
+	VMODE_1280_1024_60,
+	VMODE_1280_1024_75,
+	VMODE_1280_1024_85,
+	VMODE_1600_1200_60,
+	VMODE_1600_1200_65,
+	VMODE_1600_1200_70,
+	VMODE_1600_1200_75,
+	VMODE_1600_1200_85,
+	VMODE_1792_1344_60,
+	VMODE_1792_1344_75,
+	VMODE_1856_1392_60,
+	VMODE_1856_1392_75,
+	VMODE_1920_1440_60,
+	VMODE_1920_1440_75,
+};
+
+/* Accessors */
+static int kyro_dev_video_mode_set(struct fb_info *info)
+{
+	struct kyrofb_info *par = info->par;
+
+	/* Turn off display */
+	StopVTG(deviceInfo.pSTGReg);
+	DisableRamdacOutput(deviceInfo.pSTGReg);
+
+	/* Bring us out of VGA and into Hi-Res mode, if not already. */
+	DisableVGA(deviceInfo.pSTGReg);
+
+	if (InitialiseRamdac(deviceInfo.pSTGReg,
+			     info->var.bits_per_pixel,
+			     info->var.xres, info->var.yres,
+			     par->HSP, par->VSP, &par->PIXCLK) < 0)
+		return -EINVAL;
+
+	SetupVTG(deviceInfo.pSTGReg, par);
+
+	ResetOverlayRegisters(deviceInfo.pSTGReg);
+
+	/* Turn on display in new mode */
+	EnableRamdacOutput(deviceInfo.pSTGReg);
+	StartVTG(deviceInfo.pSTGReg);
+
+	deviceInfo.ulNextFreeVidMem = info->var.xres * info->var.yres *
+				      info->var.bits_per_pixel;
+	deviceInfo.ulOverlayOffset = 0;
+
+	return 0;
+}
+
+static int kyro_dev_overlay_create(u32 ulWidth,
+				   u32 ulHeight, int bLinear)
+{
+	u32 offset;
+	u32 stride, uvStride;
+
+	if (deviceInfo.ulOverlayOffset != 0)
+		/*
+		 * Can only create one overlay without resetting the card or
+		 * changing display mode
+		 */
+		return -EINVAL;
+
+	ResetOverlayRegisters(deviceInfo.pSTGReg);
+
+	/* Overlays are addressed in multiples of 16bytes or 32bytes, so make
+	 * sure the start offset is on an appropriate boundary.
+	 */
+	offset = deviceInfo.ulNextFreeVidMem;
+	if ((offset & 0x1f) != 0) {
+		offset = (offset + 32L) & 0xffffffE0L;
+	}
+
+	if (CreateOverlaySurface(deviceInfo.pSTGReg, ulWidth, ulHeight,
+				 bLinear, offset, &stride, &uvStride) < 0)
+		return -EINVAL;
+
+	deviceInfo.ulOverlayOffset = offset;
+	deviceInfo.ulOverlayStride = stride;
+	deviceInfo.ulOverlayUVStride = uvStride;
+	deviceInfo.ulNextFreeVidMem = offset + (ulHeight * stride) + (ulHeight * 2 * uvStride);
+
+	SetOverlayBlendMode(deviceInfo.pSTGReg, GLOBAL_ALPHA, 0xf, 0x0);
+
+	return 0;
+}
+
+static int kyro_dev_overlay_viewport_set(u32 x, u32 y, u32 ulWidth, u32 ulHeight)
+{
+	u32 right;
+	u32 bottom;
+
+	if (deviceInfo.ulOverlayOffset == 0)
+		/* probably haven't called CreateOverlay yet */
+		return -EINVAL;
+
+	if (ulWidth == 0 || ulWidth == 0xffffffff ||
+	    ulHeight == 0 || ulHeight == 0xffffffff ||
+	    (x < 2 && ulWidth + 2 == 0))
+		return -EINVAL;
+
+	/*
+	 * SetOverlayViewPort() adjusts X coordinates by +2 (left) and +1
+	 * (right) before packing them into 16-bit register fields.
+	 */
+	if (x > U16_MAX - 2 || y > U16_MAX)
+		return -EINVAL;
+
+	right = x + ulWidth;
+	bottom = y + ulHeight;
+
+	if (right < x || bottom < y)
+		return -EINVAL;
+
+	right--;
+	bottom--;
+
+	if (right > U16_MAX - 1 || bottom > U16_MAX)
+		return -EINVAL;
+
+	/* Stop Ramdac Output */
+	DisableRamdacOutput(deviceInfo.pSTGReg);
+
+	SetOverlayViewPort(deviceInfo.pSTGReg,
+			x, y, right, bottom);
+
+	EnableOverlayPlane(deviceInfo.pSTGReg);
+	/* Start Ramdac Output */
+	EnableRamdacOutput(deviceInfo.pSTGReg);
+
+	return 0;
+}
+
+static inline unsigned long get_line_length(int x, int bpp)
+{
+	return (unsigned long)((((x*bpp)+31)&~31) >> 3);
+}
+
+static int kyrofb_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
+{
+	struct kyrofb_info *par = info->par;
+
+	if (!var->pixclock)
+		return -EINVAL;
+
+	if (var->bits_per_pixel != 16 && var->bits_per_pixel != 32) {
+		printk(KERN_WARNING "kyrofb: depth not supported: %u\n", var->bits_per_pixel);
+		return -EINVAL;
+	}
+
+	switch (var->bits_per_pixel) {
+	case 16:
+		var->red.offset = 11;
+		var->red.length = 5;
+		var->green.offset = 5;
+		var->green.length = 6;
+		var->blue.length = 5;
+		break;
+	case 32:
+		var->transp.offset = 24;
+		var->red.offset = 16;
+		var->green.offset = 8;
+		var->blue.offset = 0;
+
+		var->red.length = 8;
+		var->green.length = 8;
+		var->blue.length = 8;
+		var->transp.length = 8;
+		break;
+	}
+
+	/* Height/Width of picture in mm */
+	var->height = var->width = -1;
+
+	/* Timing information. All values are in picoseconds */
+
+	/* par->PIXCLK is in 100Hz units. Convert to picoseconds -
+	 * ensuring we do not exceed 32 bit precision
+	 */
+	/*
+	 * XXX: Enabling this really screws over the pixclock value when we
+	 * read it back with fbset. As such, leaving this commented out appears
+	 * to do the right thing (at least for now) .. bearing in mind that we
+	 * have infact already done the KHZ2PICOS conversion in both the modedb
+	 * and kyro_var. -- PFM.
+	 */
+//	var->pixclock = 1000000000 / (par->PIXCLK / 10);
+
+	/* the header file claims we should use picoseconds
+	 * - nobody else does though, the all use pixels and lines
+	 * of h and v sizes. Both options here.
+	 */
+
+	/*
+	 * If we're being called by __fb_try_mode(), then we don't want to
+	 * override any of the var settings that we've already parsed
+	 * from our modedb. -- PFM.
+	 */
+	if ((var->activate & FB_ACTIVATE_MASK) == FB_ACTIVATE_TEST)
+		return 0;
+
+	var->left_margin = par->HBP;
+	var->hsync_len = par->HST;
+	var->right_margin = par->HFP;
+
+	var->upper_margin = par->VBP;
+	var->vsync_len = par->VST;
+	var->lower_margin = par->VFP;
+
+	if (par->HSP == 1)
+		var->sync |= FB_SYNC_HOR_HIGH_ACT;
+	if (par->VSP == 1)
+		var->sync |= FB_SYNC_VERT_HIGH_ACT;
+
+	return 0;
+}
+
+static int kyrofb_set_par(struct fb_info *info)
+{
+	struct kyrofb_info *par = info->par;
+	unsigned long lineclock;
+	unsigned long frameclock;
+
+	/* Actual resolution */
+	par->XRES = info->var.xres;
+	par->YRES = info->var.yres;
+
+	/* pixel depth */
+	par->PIXDEPTH = info->var.bits_per_pixel;
+
+	/* Refresh rate */
+	/* time for a line in ns */
+	lineclock = (info->var.pixclock * (info->var.xres +
+				    info->var.right_margin +
+				    info->var.hsync_len +
+				    info->var.left_margin)) / 1000;
+
+	if (!lineclock)
+		return -EINVAL;
+
+	/* time for a frame in ns (precision in 32bpp) */
+	frameclock = lineclock * (info->var.yres +
+				  info->var.lower_margin +
+				  info->var.vsync_len +
+				  info->var.upper_margin);
+
+	/* Calculate refresh rate and horrizontal clocks */
+	par->VFREQ = (1000000000 + (frameclock / 2)) / frameclock;
+	par->HCLK = (1000000000 + (lineclock / 2)) / lineclock;
+	par->PIXCLK = ((1000000000 + (info->var.pixclock / 2))
+					/ info->var.pixclock) * 10;
+
+	/* calculate horizontal timings */
+	par->HFP = info->var.right_margin;
+	par->HST = info->var.hsync_len;
+	par->HBP = info->var.left_margin;
+	par->HTot = par->XRES + par->HBP + par->HST + par->HFP;
+
+	/* calculate vertical timings */
+	par->VFP = info->var.lower_margin;
+	par->VST = info->var.vsync_len;
+	par->VBP = info->var.upper_margin;
+	par->VTot = par->YRES + par->VBP + par->VST + par->VFP;
+
+	par->HSP = (info->var.sync & FB_SYNC_HOR_HIGH_ACT) ? 1 : 0;
+	par->VSP = (info->var.sync & FB_SYNC_VERT_HIGH_ACT) ? 1 : 0;
+
+	kyro_dev_video_mode_set(info);
+
+	/* length of a line in bytes    */
+	info->fix.line_length = get_line_length(par->XRES, par->PIXDEPTH);
+	info->fix.visual = FB_VISUAL_TRUECOLOR;
+
+	return 0;
+}
+
+static int kyrofb_setcolreg(u_int regno, u_int red, u_int green,
+			    u_int blue, u_int transp, struct fb_info *info)
+{
+	struct kyrofb_info *par = info->par;
+
+	if (regno > 255)
+		return 1;	/* Invalid register */
+
+	if (regno < 16) {
+		switch (info->var.bits_per_pixel) {
+		case 16:
+			par->palette[regno] =
+			     (red   & 0xf800) |
+			    ((green & 0xfc00) >> 5) |
+			    ((blue  & 0xf800) >> 11);
+			break;
+		case 32:
+			red >>= 8; green >>= 8; blue >>= 8; transp >>= 8;
+			par->palette[regno] =
+			    (transp << 24) | (red << 16) | (green << 8) | blue;
+			break;
+		}
+	}
+
+	return 0;
+}
+
+#ifndef MODULE
+static int __init kyrofb_setup(char *options)
+{
+	char *this_opt;
+
+	if (!options || !*options)
+		return 0;
+
+	while ((this_opt = strsep(&options, ","))) {
+		if (!*this_opt)
+			continue;
+		if (strcmp(this_opt, "nopan") == 0) {
+			nopan = 1;
+		} else if (strcmp(this_opt, "nowrap") == 0) {
+			nowrap = 1;
+		} else if (strcmp(this_opt, "nomtrr") == 0) {
+			nomtrr = 1;
+		} else {
+			mode_option = this_opt;
+		}
+	}
+
+	return 0;
+}
+#endif
+
+static int kyrofb_ioctl(struct fb_info *info,
+			unsigned int cmd, unsigned long arg)
+{
+	overlay_create ol_create;
+	overlay_viewport_set ol_viewport_set;
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+	case KYRO_IOCTL_OVERLAY_CREATE:
+		if (copy_from_user(&ol_create, argp, sizeof(overlay_create)))
+			return -EFAULT;
+
+		if (kyro_dev_overlay_create(ol_create.ulWidth,
+					    ol_create.ulHeight, 0) < 0) {
+			printk(KERN_ERR "Kyro FB: failed to create overlay surface.\n");
+
+			return -EINVAL;
+		}
+		break;
+	case KYRO_IOCTL_OVERLAY_VIEWPORT_SET:
+		if (copy_from_user(&ol_viewport_set, argp,
+			       sizeof(overlay_viewport_set)))
+			return -EFAULT;
+
+		if (kyro_dev_overlay_viewport_set(ol_viewport_set.xOrgin,
+						  ol_viewport_set.yOrgin,
+						  ol_viewport_set.xSize,
+						  ol_viewport_set.ySize) != 0)
+		{
+			printk(KERN_ERR "Kyro FB: failed to create overlay viewport.\n");
+			return -EINVAL;
+		}
+		break;
+	case KYRO_IOCTL_SET_VIDEO_MODE:
+		{
+			printk(KERN_ERR "Kyro FB: KYRO_IOCTL_SET_VIDEO_MODE is"
+				"obsolete, use the appropriate fb_ioctl()"
+				"command instead.\n");
+			return -EINVAL;
+		}
+	case KYRO_IOCTL_UVSTRIDE:
+		if (copy_to_user(argp, &deviceInfo.ulOverlayUVStride, sizeof(deviceInfo.ulOverlayUVStride)))
+			return -EFAULT;
+		break;
+	case KYRO_IOCTL_STRIDE:
+		if (copy_to_user(argp, &deviceInfo.ulOverlayStride, sizeof(deviceInfo.ulOverlayStride)))
+			return -EFAULT;
+		break;
+	case KYRO_IOCTL_OVERLAY_OFFSET:
+		if (copy_to_user(argp, &deviceInfo.ulOverlayOffset, sizeof(deviceInfo.ulOverlayOffset)))
+			return -EFAULT;
+		break;
+	}
+
+	return 0;
+}
+
+static const struct pci_device_id kyrofb_pci_tbl[] = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_ST, PCI_DEVICE_ID_STG4000) },
+	{ }
+};
+
+MODULE_DEVICE_TABLE(pci, kyrofb_pci_tbl);
+
+static struct pci_driver kyrofb_pci_driver = {
+	.name		= "kyrofb",
+	.id_table	= kyrofb_pci_tbl,
+	.probe		= kyrofb_probe,
+	.remove		= kyrofb_remove,
+};
+
+static const struct fb_ops kyrofb_ops = {
+	.owner		= THIS_MODULE,
+	FB_DEFAULT_IOMEM_OPS,
+	.fb_check_var	= kyrofb_check_var,
+	.fb_set_par	= kyrofb_set_par,
+	.fb_setcolreg	= kyrofb_setcolreg,
+	.fb_ioctl	= kyrofb_ioctl,
+};
+
+static int kyrofb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+	struct fb_info *info;
+	struct kyrofb_info *currentpar;
+	unsigned long size;
+	int err;
+
+	err = aperture_remove_conflicting_pci_devices(pdev, "kyrofb");
+	if (err)
+		return err;
+
+	err = pcim_enable_device(pdev);
+	if (err) {
+		printk(KERN_WARNING "kyrofb: Can't enable pdev: %d\n", err);
+		return err;
+	}
+
+	info = framebuffer_alloc(sizeof(struct kyrofb_info), &pdev->dev);
+	if (!info)
+		return -ENOMEM;
+
+	err = pcim_request_all_regions(pdev, "kyrofb");
+	if (err)
+		goto out_free_fb;
+
+	currentpar = info->par;
+
+	kyro_fix.smem_start = pci_resource_start(pdev, 0);
+	kyro_fix.smem_len   = pci_resource_len(pdev, 0);
+	kyro_fix.mmio_start = pci_resource_start(pdev, 1);
+	kyro_fix.mmio_len   = pci_resource_len(pdev, 1);
+
+	currentpar->regbase = deviceInfo.pSTGReg =
+		devm_ioremap(&pdev->dev, kyro_fix.mmio_start,
+			     kyro_fix.mmio_len);
+	if (!currentpar->regbase)
+		goto out_free_fb;
+
+	info->screen_base = devm_ioremap_wc(&pdev->dev, kyro_fix.smem_start,
+					    kyro_fix.smem_len);
+	if (!info->screen_base)
+		goto out_free_fb;
+
+	if (!nomtrr)
+		currentpar->wc_cookie = arch_phys_wc_add(kyro_fix.smem_start,
+							 kyro_fix.smem_len);
+
+	kyro_fix.ypanstep	= nopan ? 0 : 1;
+	kyro_fix.ywrapstep	= nowrap ? 0 : 1;
+
+	info->fbops		= &kyrofb_ops;
+	info->fix		= kyro_fix;
+	info->pseudo_palette	= currentpar->palette;
+
+	SetCoreClockPLL(deviceInfo.pSTGReg, pdev);
+
+	deviceInfo.ulNextFreeVidMem = 0;
+	deviceInfo.ulOverlayOffset = 0;
+
+	/* This should give a reasonable default video mode */
+	if (!fb_find_mode(&info->var, info, mode_option, kyro_modedb,
+			  NUM_TOTAL_MODES, &kyro_modedb[VMODE_1024_768_75], 32))
+		info->var = kyro_var;
+
+	fb_alloc_cmap(&info->cmap, 256, 0);
+
+	kyrofb_set_par(info);
+	kyrofb_check_var(&info->var, info);
+
+	size = get_line_length(info->var.xres_virtual,
+			       info->var.bits_per_pixel);
+	size *= info->var.yres_virtual;
+
+	fb_memset_io(info->screen_base, 0, size);
+
+	if (register_framebuffer(info) < 0)
+		goto out_free_fb;
+
+	fb_info(info, "%s frame buffer device, at %dx%d@%d using %ldk/%ldk of VRAM\n",
+		info->fix.id,
+		info->var.xres, info->var.yres, info->var.bits_per_pixel,
+		size >> 10, (unsigned long)info->fix.smem_len >> 10);
+
+	pci_set_drvdata(pdev, info);
+
+	return 0;
+
+out_free_fb:
+	framebuffer_release(info);
+
+	return -EINVAL;
+}
+
+static void kyrofb_remove(struct pci_dev *pdev)
+{
+	struct fb_info *info = pci_get_drvdata(pdev);
+	struct kyrofb_info *par = info->par;
+
+	/* Reset the board */
+	StopVTG(deviceInfo.pSTGReg);
+	DisableRamdacOutput(deviceInfo.pSTGReg);
+
+	/* Sync up the PLL */
+	SetCoreClockPLL(deviceInfo.pSTGReg, pdev);
+
+	deviceInfo.ulNextFreeVidMem = 0;
+	deviceInfo.ulOverlayOffset = 0;
+
+	arch_phys_wc_del(par->wc_cookie);
+
+	unregister_framebuffer(info);
+	framebuffer_release(info);
+}
+
+static int __init kyrofb_init(void)
+{
+#ifndef MODULE
+	char *option = NULL;
+#endif
+
+	if (fb_modesetting_disabled("kyrofb"))
+		return -ENODEV;
+
+#ifndef MODULE
+	if (fb_get_options("kyrofb", &option))
+		return -ENODEV;
+	kyrofb_setup(option);
+#endif
+	return pci_register_driver(&kyrofb_pci_driver);
+}
+
+static void __exit kyrofb_exit(void)
+{
+	pci_unregister_driver(&kyrofb_pci_driver);
+}
+
+module_init(kyrofb_init);
+
+#ifdef MODULE
+module_exit(kyrofb_exit);
+#endif
+
+MODULE_AUTHOR("STMicroelectronics; Paul Mundt <lethal@linux-sh.org>");
+MODULE_DESCRIPTION("STG4000/Kyro/PowerVR 3 driver");
+MODULE_LICENSE("GPL");