index : renesas/drm.git	android live master
	libdrm, cloned from git://anongit.freedesktop.org/mesa/drm	Laurent Pinchart

summaryrefslogtreecommitdiff

log msg author committer range

path: root/shared-core/via_dma.c

Age	Commit message (Expand)	Author
2007-05-26	whitespace fixups from kernel	Dave Airlie
2007-04-17	via: Make sure we flush write-combining using a follow-up read.	Thomas Hellstrom
2007-02-16	Simple fence object sample driver for via, based on idling the GPU.	Thomas Hellstrom
2007-02-02	via: Try to improve command-buffer chaining.	Thomas Hellstrom
2007-02-02	Disable AGP DMA for chips with the new 3D engine.	Thomas Hellstrom
2006-12-20	Some via PCI posting flushes.	Thomas Hellstrom
2006-08-28	drm: lots of small cleanups and whitespace issues fixed up	Dave Airlie
2006-03-19	make some functions static in via driver	Dave Airlie
2006-01-02	Realign via driver with changes in Linux kernel (mainly whitespace)	Dave Airlie
2005-09-25	Add the via PCI DMA blit code.	Thomas Hellstrom
2005-09-03	convert ioctl flags to use flags instead of separate ints	Dave Airlie
2005-08-15	Port the VIA DRM to FreeBSD. Original patch by Jake, with some cleanup by	Eric Anholt
2005-07-15	VIA: Fix sparse warnings (Alexey Dobriyan)	Thomas Hellstrom
2005-05-08	VIA: Fix for oops when AGP ring-buffer initialization is called and there	Thomas Hellstrom
2005-04-18	VIA:	Thomas Hellstrom
2005-03-28	Via updates:	Thomas Hellstrom
2005-02-01	cleanup patch from Adrian Bunk <bunk@stusta.de>	Dave Airlie
2005-01-27	Fixed multiple devices DMA bug. Fixed PCI path FIRE command detection	Thomas Hellstrom
2005-01-07	via updates: moved the verifier state struct to dev_priv. Implemented AGP	Thomas Hellstrom
2005-01-05	VIA update:	Thomas Hellstrom
2004-12-31	VIA DRM: Stability enhancements and cleanups in via_dma.c Added explicit	Thomas Hellstrom
2004-12-21	via DRM: Tightened the security for some functions of the	Thomas Hellstrom
2004-12-19	via changes: Fixed typecasting bug in via_dma.c, and possible short-circuit	Thomas Hellstrom
2004-12-13	VIA drm updates:	Thomas Hellstrom
2004-12-06	Security and optimization fixes for the via drm:	Thomas Hellstrom
2004-12-03	Added 3D functionality to the via command verifier. Modified the via	Thomas Hellstrom
2004-11-27	Reworked PCI MMIO command buffer parser, and imported code from the Mesa	Thomas Hellstrom
2004-11-03	Fixed bug in via_dma.c. The code didn't check that the lock was held by the	Thomas Hellstrom
2004-11-01	Reworked the jump-code in via_dma.c. The command regulator now seems to	Thomas Hellstrom
2004-10-30	Some stabilizing work to the DMA ring-buffer code. Temporarily replaced the	Thomas Hellstrom
2004-10-12	Via updates. Fixed unlikely but possible uint32_t overflow in ring-buffer	Thomas Hellstrom
2004-10-09	Lindent the via stuff so I can include it in kernel	Dave Airlie
2004-10-08	Changed unsigned to uint32_t in some ioctl parameters. Introduced first	Thomas Hellstrom
2004-09-30	Lindent of core build. Drivers checked for no binary diffs. A few files	Jon Smirl
2004-09-27	First check in for DRM that splits core from personality modules	Jon Smirl
2004-09-12	Fix DRM to compile cleanly with recent kernel changes in PCI IO and	Jon Smirl
2004-09-07	Added IOCTL for writing 2D DMA command buffers over PCI. Bumped minor	Thomas Hellstrom
2004-08-30	Fix copy_from_user return value ignored warning at via_dma.c:168.	Erdi Chen
2004-08-27	__NO_VERSION__ hasn't been needed since 2.3 days ditch it...	Dave Airlie
2004-08-26	Rearrange things so that via_dma.c will compile with inline via_check_dma	Jon Smirl
2004-08-24	This patch adds three new ioctl's to the VIA Unichrome/Pro DRM driver:	Erdi Chen

generated by cgit v1.2.3 (git 2.25.1) at 2025-01-14 13:35:20 +0000

/* savage_state.c -- State and drawing support for Savage
 *
 * Copyright 2004  Felix Kuehling
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sub license,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
 * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 */
#include "drmP.h"
#include "savage_drm.h"
#include "savage_drv.h"

void savage_emit_clip_rect_s3d(drm_savage_private_t *dev_priv,
			       const struct drm_clip_rect *pbox)
{
	uint32_t scstart = dev_priv->state.s3d.new_scstart;
	uint32_t scend = dev_priv->state.s3d.new_scend;
	scstart = (scstart & ~SAVAGE_SCISSOR_MASK_S3D) |
		((uint32_t)pbox->x1 & 0x000007ff) |
		(((uint32_t)pbox->y1 << 16) & 0x07ff0000);
	scend   = (scend & ~SAVAGE_SCISSOR_MASK_S3D) |
		(((uint32_t)pbox->x2 - 1) & 0x000007ff) |
		((((uint32_t)pbox->y2 - 1) << 16) & 0x07ff0000);
	if (scstart != dev_priv->state.s3d.scstart ||
	    scend   != dev_priv->state.s3d.scend) {
		DMA_LOCALS;
		BEGIN_DMA(4);
		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
		DMA_SET_REGISTERS(SAVAGE_SCSTART_S3D, 2);
		DMA_WRITE(scstart);
		DMA_WRITE(scend);
		dev_priv->state.s3d.scstart = scstart;
		dev_priv->state.s3d.scend = scend;
		dev_priv->waiting = 1;
		DMA_COMMIT();
	}
}

void savage_emit_clip_rect_s4(drm_savage_private_t *dev_priv,
			      const struct drm_clip_rect *pbox)
{
	uint32_t drawctrl0 = dev_priv->state.s4.new_drawctrl0;
	uint32_t drawctrl1 = dev_priv->state.s4.new_drawctrl1;
	drawctrl0 = (drawctrl0 & ~SAVAGE_SCISSOR_MASK_S4) |
		((uint32_t)pbox->x1 & 0x000007ff) |
		(((uint32_t)pbox->y1 << 12) & 0x00fff000);
	drawctrl1 = (drawctrl1 & ~SAVAGE_SCISSOR_MASK_S4) |
		(((uint32_t)pbox->x2 - 1) & 0x000007ff) |
		((((uint32_t)pbox->y2 - 1) << 12) & 0x00fff000);
	if (drawctrl0 != dev_priv->state.s4.drawctrl0 ||
	    drawctrl1 != dev_priv->state.s4.drawctrl1) {
		DMA_LOCALS;
		BEGIN_DMA(4);
		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
		DMA_SET_REGISTERS(SAVAGE_DRAWCTRL0_S4, 2);
		DMA_WRITE(drawctrl0);
		DMA_WRITE(drawctrl1);
		dev_priv->state.s4.drawctrl0 = drawctrl0;
		dev_priv->state.s4.drawctrl1 = drawctrl1;
		dev_priv->waiting = 1;
		DMA_COMMIT();
	}
}

static int savage_verify_texaddr(drm_savage_private_t *dev_priv, int unit,
				 uint32_t addr)
{
	if ((addr & 6) != 2) { /* reserved bits */
		DRM_ERROR("bad texAddr%d %08x (reserved bits)\n", unit, addr);
		return -EINVAL;
	}
	if (!(addr & 1)) { /* local */
		addr &= ~7;
		if (addr < dev_priv->texture_offset ||
		    addr >= dev_priv->texture_offset + dev_priv->texture_size) {
			DRM_ERROR
			    ("bad texAddr%d %08x (local addr out of range)\n",
			     unit, addr);
			return -EINVAL;
		}
	} else { /* AGP */
		if (!dev_priv->agp_textures) {
			DRM_ERROR("bad texAddr%d %08x (AGP not available)\n",
				  unit, addr);
			return -EINVAL;
		}
		addr &= ~7;
		if (addr < dev_priv->agp_textures->offset ||
		    addr >= (dev_priv->agp_textures->offset +
			     dev_priv->agp_textures->size)) {
			DRM_ERROR
			    ("bad texAddr%d %08x (AGP addr out of range)\n",
			     unit, addr);
			return -EINVAL;
		}
	}
	return 0;
}

#define SAVE_STATE(reg,where)			\
	if(start <= reg && start + count > reg)	\
		dev_priv->state.where = regs[reg - start]
#define SAVE_STATE_MASK(reg,where,mask) do {			\
	if(start <= reg && start + count > reg) {			\
		uint32_t tmp;					\
		tmp = regs[reg - start];			\
		dev_priv->state.where = (tmp & (mask)) |	\
			(dev_priv->state.where & ~(mask));	\
	}							\
} while (0)
static int savage_verify_state_s3d(drm_savage_private_t *dev_priv,
				   unsigned int start, unsigned int count,
				   const uint32_t *regs)
{
	if (start < SAVAGE_TEXPALADDR_S3D ||
	    start + count - 1 > SAVAGE_DESTTEXRWWATERMARK_S3D) {
		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
			  start, start + count - 1);
		return -EINVAL;
	}

	SAVE_STATE_MASK(SAVAGE_SCSTART_S3D, s3d.new_scstart,
			~SAVAGE_SCISSOR_MASK_S3D);
	SAVE_STATE_MASK(SAVAGE_SCEND_S3D, s3d.new_scend,
			~SAVAGE_SCISSOR_MASK_S3D);

	/* if any texture regs were changed ... */
	if (start <= SAVAGE_TEXCTRL_S3D &&
	    start + count > SAVAGE_TEXPALADDR_S3D) {
		/* ... check texture state */
		SAVE_STATE(SAVAGE_TEXCTRL_S3D, s3d.texctrl);
		SAVE_STATE(SAVAGE_TEXADDR_S3D, s3d.texaddr);
		if (dev_priv->state.s3d.texctrl & SAVAGE_TEXCTRL_TEXEN_MASK)
			return savage_verify_texaddr(dev_priv, 0,
						dev_priv->state.s3d.texaddr);
	}

	return 0;
}

static int savage_verify_state_s4(drm_savage_private_t *dev_priv,
				  unsigned int start, unsigned int count,
				  const uint32_t *regs)
{
	int ret = 0;

	if (start < SAVAGE_DRAWLOCALCTRL_S4 ||
	    start + count - 1 > SAVAGE_TEXBLENDCOLOR_S4) {
		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
			  start, start + count - 1);
		return -EINVAL;
	}

	SAVE_STATE_MASK(SAVAGE_DRAWCTRL0_S4, s4.new_drawctrl0,
			~SAVAGE_SCISSOR_MASK_S4);
	SAVE_STATE_MASK(SAVAGE_DRAWCTRL1_S4, s4.new_drawctrl1,
			~SAVAGE_SCISSOR_MASK_S4);

	/* if any texture regs were changed ... */
	if (start <= SAVAGE_TEXDESCR_S4 &&
	    start + count > SAVAGE_TEXPALADDR_S4) {
		/* ... check texture state */
		SAVE_STATE(SAVAGE_TEXDESCR_S4, s4.texdescr);
		SAVE_STATE(SAVAGE_TEXADDR0_S4, s4.texaddr0);
		SAVE_STATE(SAVAGE_TEXADDR1_S4, s4.texaddr1);
		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX0EN_MASK)
			ret |= savage_verify_texaddr(dev_priv, 0,
						dev_priv->state.s4.texaddr0);
		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX1EN_MASK)
			ret |= savage_verify_texaddr(dev_priv, 1,
						dev_priv->state.s4.texaddr1);
	}

	return ret;
}
#undef SAVE_STATE
#undef SAVE_STATE_MASK

static int savage_dispatch_state(drm_savage_private_t *dev_priv,
				 const drm_savage_cmd_header_t *cmd_header,
				 const uint32_t *regs)
{
	unsigned int count = cmd_header->state.count;
	unsigned int start = cmd_header->state.start;
	unsigned int count2 = 0;
	unsigned int bci_size;
	int ret;
	DMA_LOCALS;

	if (!count)
		return 0;

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		ret = savage_verify_state_s3d(dev_priv, start, count, regs);
		if (ret != 0)
			return ret;
		/* scissor regs are emitted in savage_dispatch_draw */
		if (start < SAVAGE_SCSTART_S3D) {
			if (start + count > SAVAGE_SCEND_S3D + 1)
				count2 = count - (SAVAGE_SCEND_S3D + 1 - start);
			if (start + count > SAVAGE_SCSTART_S3D)
				count = SAVAGE_SCSTART_S3D - start;
		} else if (start <= SAVAGE_SCEND_S3D) {
			if (start + count > SAVAGE_SCEND_S3D + 1) {
				count -= SAVAGE_SCEND_S3D + 1 - start;
				start = SAVAGE_SCEND_S3D + 1;
			} else
				return 0;
		}
	} else {
		ret = savage_verify_state_s4(dev_priv, start, count, regs);
		if (ret != 0)
			return ret;
		/* scissor regs are emitted in savage_dispatch_draw */
		if (start < SAVAGE_DRAWCTRL0_S4) {
			if (start + count > SAVAGE_DRAWCTRL1_S4 + 1)
				count2 = count -
					 (SAVAGE_DRAWCTRL1_S4 + 1 - start);
			if (start + count > SAVAGE_DRAWCTRL0_S4)
				count = SAVAGE_DRAWCTRL0_S4 - start;
		} else if (start <= SAVAGE_DRAWCTRL1_S4) {
			if (start + count > SAVAGE_DRAWCTRL1_S4 + 1) {
				count -= SAVAGE_DRAWCTRL1_S4 + 1 - start;
				start = SAVAGE_DRAWCTRL1_S4 + 1;
			} else
				return 0;
		}
	}

	bci_size = count + (count + 254) / 255 + count2 + (count2 + 254) / 255;

	if (cmd_header->state.global) {
		BEGIN_DMA(bci_size + 1);
		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
		dev_priv->waiting = 1;
	} else {
		BEGIN_DMA(bci_size);
	}

	do {
		while (count > 0) {
			unsigned int n = count < 255 ? count : 255;
			DMA_SET_REGISTERS(start, n);
			DMA_COPY(regs, n);
			count -= n;
			start += n;
			regs += n;
		}
		start += 2;
		regs += 2;
		count = count2;
		count2 = 0;
	} while (count);

	DMA_COMMIT();

	return 0;
}

static int savage_dispatch_dma_prim(drm_savage_private_t *dev_priv,
				    const drm_savage_cmd_header_t *cmd_header,
				    const struct drm_buf *dmabuf)
{
	unsigned char reorder = 0;
	unsigned int prim = cmd_header->prim.prim;
	unsigned int skip = cmd_header->prim.skip;
	unsigned int n = cmd_header->prim.count;
	unsigned int start = cmd_header->prim.start;
	unsigned int i;
	BCI_LOCALS;

	if (!dmabuf) {
		DRM_ERROR("called without dma buffers!\n");
		return -EINVAL;
	}

	if (!n)
		return 0;

	switch (prim) {
	case SAVAGE_PRIM_TRILIST_201:
		reorder = 1;
		prim = SAVAGE_PRIM_TRILIST;
	case SAVAGE_PRIM_TRILIST:
		if (n % 3 != 0) {
			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
				  n);
			return -EINVAL;
		}
		break;
	case SAVAGE_PRIM_TRISTRIP:
	case SAVAGE_PRIM_TRIFAN:
		if (n < 3) {
			DRM_ERROR
			   ("wrong number of vertices %u in TRIFAN/STRIP\n",
			    n);
			return -EINVAL;
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return -EINVAL;
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip != 0) {
			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
			return -EINVAL;
		}
	} else {
		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
			return -EINVAL;
		}
		if (reorder) {
			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
			return -EINVAL;
		}
	}

	if (start + n > dmabuf->total / 32) {
		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
			  start, start + n - 1, dmabuf->total / 32);
		return -EINVAL;
	}

	/* Vertex DMA doesn't work with command DMA at the same time,
	 * so we use BCI_... to submit commands here. Flush buffered
	 * faked DMA first. */
	DMA_FLUSH();

	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
		BEGIN_BCI(2);
		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
		dev_priv->state.common.vbaddr = dmabuf->bus_address;
	}
	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
		/* Workaround for what looks like a hardware bug. If a
		 * WAIT_3D_IDLE was emitted some time before the
		 * indexed drawing command then the engine will lock
		 * up. There are two known workarounds:
		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
		BEGIN_BCI(63);
		for (i = 0; i < 63; ++i)
			BCI_WRITE(BCI_CMD_WAIT);
		dev_priv->waiting = 0;
	}

	prim <<= 25;
	while (n != 0) {
		/* Can emit up to 255 indices (85 triangles) at once. */
		unsigned int count = n > 255 ? 255 : n;
		if (reorder) {
			/* Need to reorder indices for correct flat
			 * shading while preserving the clock sense
			 * for correct culling. Only on Savage3D. */
			int reorder[3] = { -1, -1, -1 };
			reorder[start % 3] = 2;

			BEGIN_BCI((count + 1 + 1) / 2);
			BCI_DRAW_INDICES_S3D(count, prim, start + 2);

			for (i = start + 1; i + 1 < start + count; i += 2)
				BCI_WRITE((i + reorder[i % 3]) |
					  ((i + 1 +
					    reorder[(i + 1) % 3]) << 16));
			if (i < start + count)
				BCI_WRITE(i + reorder[i % 3]);
		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
			BEGIN_BCI((count + 1 + 1) / 2);
			BCI_DRAW_INDICES_S3D(count, prim, start);

			for (i = start + 1; i + 1 < start + count; i += 2)
				BCI_WRITE(i | ((i + 1) << 16));
			if (i < start + count)
				BCI_WRITE(i);
		} else {
			BEGIN_BCI((count + 2 + 1) / 2);
			BCI_DRAW_INDICES_S4(count, prim, skip);

			for (i = start; i + 1 < start + count; i += 2)
				BCI_WRITE(i | ((i + 1) << 16));
			if (i < start + count)
				BCI_WRITE(i);
		}

		start += count;
		n -= count;

		prim |= BCI_CMD_DRAW_CONT;
	}

	return 0;
}

static int savage_dispatch_vb_prim(drm_savage_private_t *dev_priv,
				   const drm_savage_cmd_header_t *cmd_header,
				   const uint32_t *vtxbuf, unsigned int vb_size,
				   unsigned int vb_stride)
{
	unsigned char reorder = 0;
	unsigned int prim = cmd_header->prim.prim;
	unsigned int skip = cmd_header->prim.skip;
	unsigned int n = cmd_header->prim.count;
	unsigned int start = cmd_header->prim.start;
	unsigned int vtx_size;
	unsigned int i;
	DMA_LOCALS;

	if (!n)
		return 0;

	switch (prim) {
	case SAVAGE_PRIM_TRILIST_201:
		reorder = 1;
		prim = SAVAGE_PRIM_TRILIST;
	case SAVAGE_PRIM_TRILIST:
		if (n % 3 != 0) {
			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
				  n);
			return -EINVAL;
		}
		break;
	case SAVAGE_PRIM_TRISTRIP:
	case SAVAGE_PRIM_TRIFAN:
		if (n < 3) {
			DRM_ERROR
			    ("wrong number of vertices %u in TRIFAN/STRIP\n",
			     n);
			return -EINVAL;
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return -EINVAL;
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip > SAVAGE_SKIP_ALL_S3D) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return -EINVAL;
		}
		vtx_size = 8; /* full vertex */
	} else {
		if (skip > SAVAGE_SKIP_ALL_S4) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return -EINVAL;
		}
		vtx_size = 10; /* full vertex */
	}

	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);

	if (vtx_size > vb_stride) {
		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
			  vtx_size, vb_stride);
		return -EINVAL;
	}

	if (start + n > vb_size / (vb_stride * 4)) {
		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
			  start, start + n - 1, vb_size / (vb_stride * 4));
		return -EINVAL;
	}

	prim <<= 25;
	while (n != 0) {
		/* Can emit up to 255 vertices (85 triangles) at once. */
		unsigned int count = n > 255 ? 255 : n;
		if (reorder) {
			/* Need to reorder vertices for correct flat
			 * shading while preserving the clock sense
			 * for correct culling. Only on Savage3D. */
			int reorder[3] = { -1, -1, -1 };
			reorder[start % 3] = 2;

			BEGIN_DMA(count * vtx_size + 1);
			DMA_DRAW_PRIMITIVE(count, prim, skip);

			for (i = start; i < start + count; ++i) {
				unsigned int j = i + reorder[i % 3];
				DMA_COPY(&vtxbuf[vb_stride * j], vtx_size);
			}

			DMA_COMMIT();
		} else {
			BEGIN_DMA(count * vtx_size + 1);
			DMA_DRAW_PRIMITIVE(count, prim, skip);

			if (vb_stride == vtx_size) {
				DMA_COPY(&vtxbuf[vb_stride * start],
					 vtx_size * count);
			} else {
				for (i = start; i < start + count; ++i) {
					DMA_COPY(&vtxbuf[vb_stride * i],
						 vtx_size);
				}
			}

			DMA_COMMIT();
		}

		start += count;
		n -= count;

		prim |= BCI_CMD_DRAW_CONT;
	}

	return 0;
}

static int savage_dispatch_dma_idx(drm_savage_private_t *dev_priv,
				   const drm_savage_cmd_header_t *cmd_header,
				   const uint16_t *idx,
				   const struct drm_buf *dmabuf)
{
	unsigned char reorder = 0;
	unsigned int prim = cmd_header->idx.prim;
	unsigned int skip = cmd_header->idx.skip;
	unsigned int n = cmd_header->idx.count;
	unsigned int i;
	BCI_LOCALS;

	if (!dmabuf) {
		DRM_ERROR("called without dma buffers!\n");
		return -EINVAL;
	}

	if (!n)
		return 0;

	switch (prim) {
	case SAVAGE_PRIM_TRILIST_201:
		reorder = 1;
		prim = SAVAGE_PRIM_TRILIST;
	case SAVAGE_PRIM_TRILIST:
		if (n % 3 != 0) {
			DRM_ERROR("wrong number of indices %u in TRILIST\n", n);
			return -EINVAL;
		}
		break;
	case SAVAGE_PRIM_TRISTRIP:
	case SAVAGE_PRIM_TRIFAN:
		if (n < 3) {
			DRM_ERROR
			    ("wrong number of indices %u in TRIFAN/STRIP\n", n);
			return -EINVAL;
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return -EINVAL;
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip != 0) {
			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
			return -EINVAL;
		}
	} else {
		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
			return -EINVAL;
		}
		if (reorder) {
			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
			return -EINVAL;
		}
	}

	/* Vertex DMA doesn't work with command DMA at the same time,
	 * so we use BCI_... to submit commands here. Flush buffered
	 * faked DMA first. */
	DMA_FLUSH();

	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
		BEGIN_BCI(2);
		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
		dev_priv->state.common.vbaddr = dmabuf->bus_address;
	}
	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
		/* Workaround for what looks like a hardware bug. If a
		 * WAIT_3D_IDLE was emitted some time before the
		 * indexed drawing command then the engine will lock
		 * up. There are two known workarounds:
		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
		BEGIN_BCI(63);
		for (i = 0; i < 63; ++i)
			BCI_WRITE(BCI_CMD_WAIT);
		dev_priv->waiting = 0;
	}

	prim <<= 25;
	while (n != 0) {
		/* Can emit up to 255 indices (85 triangles) at once. */
		unsigned int count = n > 255 ? 255 : n;

		/* check indices */
		for (i = 0; i < count; ++i) {
			if (idx[i] > dmabuf->total / 32) {
				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
					  i, idx[i], dmabuf->total / 32);
				return -EINVAL;
			}
		}

		if (reorder) {
			/* Need to reorder indices for correct flat
			 * shading while preserving the clock sense
			 * for correct culling. Only on Savage3D. */
			int reorder[3] = { 2, -1, -1 };

			BEGIN_BCI((count + 1 + 1) / 2);
			BCI_DRAW_INDICES_S3D(count, prim, idx[2]);

			for (i = 1; i + 1 < count; i += 2)
				BCI_WRITE(idx[i + reorder[i % 3]] |
					  (idx[i + 1 +
					   reorder[(i + 1) % 3]] << 16));
			if (i < count)
				BCI_WRITE(idx[i + reorder[i % 3]]);
		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
			BEGIN_BCI((count + 1 + 1) / 2);
			BCI_DRAW_INDICES_S3D(count, prim, idx[0]);

			for (i = 1; i + 1 < count; i += 2)
				BCI_WRITE(idx[i] | (idx[i + 1] << 16));
			if (i < count)
				BCI_WRITE(idx[i]);
		} else {
			BEGIN_BCI((count + 2 + 1) / 2);
			BCI_DRAW_INDICES_S4(count, prim, skip);

			for (i = 0; i + 1 < count; i += 2)
				BCI_WRITE(idx[i] | (idx[i + 1] << 16));
			if (i < count)
				BCI_WRITE(idx[i]);
		}

		idx += count;
		n -= count;

		prim |= BCI_CMD_DRAW_CONT;
	}

	return 0;
}

static int savage_dispatch_vb_idx(drm_savage_private_t *dev_priv,
				  const drm_savage_cmd_header_t *cmd_header,
				  const uint16_t *idx,
				  const uint32_t *vtxbuf,
				  unsigned int vb_size, unsigned int vb_stride)
{
	unsigned char reorder = 0;
	unsigned int prim = cmd_header->idx.prim;
	unsigned int skip = cmd_header->idx.skip;
	unsigned int n = cmd_header->idx.count;
	unsigned int vtx_size;
	unsigned int i;
	DMA_LOCALS;

	if (!n)
		return 0;

	switch (prim) {
	case SAVAGE_PRIM_TRILIST_201:
		reorder = 1;
		prim = SAVAGE_PRIM_TRILIST;
	case SAVAGE_PRIM_TRILIST:
		if (n % 3 != 0) {
			DRM_ERROR("wrong number of indices %u in TRILIST\n", n);
			return -EINVAL;
		}
		break;
	case SAVAGE_PRIM_TRISTRIP:
	case SAVAGE_PRIM_TRIFAN:
		if (n < 3) {
			DRM_ERROR
			    ("wrong number of indices %u in TRIFAN/STRIP\n", n);
			return -EINVAL;
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return -EINVAL;
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip > SAVAGE_SKIP_ALL_S3D) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return -EINVAL;
		}
		vtx_size = 8; /* full vertex */
	} else {
		if (skip > SAVAGE_SKIP_ALL_S4) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return -EINVAL;
		}
		vtx_size = 10; /* full vertex */
	}

	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);

	if (vtx_size > vb_stride) {
		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
			  vtx_size, vb_stride);
		return -EINVAL;
	}

	prim <<= 25;
	while (n != 0) {
		/* Can emit up to 255 vertices (85 triangles) at once. */
		unsigned int count = n > 255 ? 255 : n;

		/* Check indices */
		for (i = 0; i < count; ++i) {
			if (idx[i] > vb_size / (vb_stride * 4)) {
				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
					  i, idx[i],  vb_size / (vb_stride * 4));
				return -EINVAL;
			}
		}

		if (reorder) {
			/* Need to reorder vertices for correct flat
			 * shading while preserving the clock sense
			 * for correct culling. Only on Savage3D. */
			int reorder[3] = { 2, -1, -1 };

			BEGIN_DMA(count * vtx_size + 1);
			DMA_DRAW_PRIMITIVE(count, prim, skip);

			for (i = 0; i < count; ++i) {
				unsigned int j = idx[i + reorder[i % 3]];
				DMA_COPY(&vtxbuf[vb_stride * j], vtx_size);
			}