savage_state.c

linux-2.6.15.6

			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 __user * 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 DRM_ERR(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 DRM_ERR(EINVAL);
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return DRM_ERR(EINVAL);
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip > SAVAGE_SKIP_ALL_S3D) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return DRM_ERR(EINVAL);
		}
		vtx_size = 8;	/* full vertex */
	} else {
		if (skip > SAVAGE_SKIP_ALL_S4) {
			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
			return DRM_ERR(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 DRM_ERR(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 DRM_ERR(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_FROM_USER(&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_FROM_USER(&vtxbuf[vb_stride * start],
						   vtx_size * count);
			} else {
				for (i = start; i < start + count; ++i) {
					DMA_COPY_FROM_USER(&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 __user * usr_idx,
				   const drm_buf_t * 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 DRM_ERR(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 DRM_ERR(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 DRM_ERR(EINVAL);
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return DRM_ERR(EINVAL);
	}

	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
		if (skip != 0) {
			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
			return DRM_ERR(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 DRM_ERR(EINVAL);
		}
		if (reorder) {
			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
			return DRM_ERR(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;
		/* Is it ok to allocate 510 bytes on the stack in an ioctl? */
		uint16_t idx[255];

		/* Copy and check indices */
		DRM_COPY_FROM_USER_UNCHECKED(idx, usr_idx, count * 2);
		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 DRM_ERR(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]);
		}

		usr_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 __user * usr_idx,
				  const uint32_t __user * 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 DRM_ERR(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 DRM_ERR(EINVAL);
		}
		break;
	default:
		DRM_ERROR("invalid primitive type %u\n", prim);
		return DRM_ERR(EINVAL);
	}

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

	prim <<= 25;
	while (n != 0) {
		/* Can emit up to 255 vertices (85 triangles) at once. */
		unsigned int count = n > 255 ? 255 : n;
		/* Is it ok to allocate 510 bytes on the stack in an ioctl? */
		uint16_t idx[255];

		/* Copy and check indices */
		DRM_COPY_FROM_USER_UNCHECKED(idx, usr_idx, count * 2);
		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 DRM_ERR(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) {