return -EFAULT;
}
+ box.x2--; /* Hardware expects inclusive bottom-right corner */
+ box.y2--;
+
if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV515) {
box.x1 = (box.x1) &
R300_CLIPRECT_MASK;
R300_CLIPRECT_MASK;
box.y2 = (box.y2 + R300_CLIPRECT_OFFSET) &
R300_CLIPRECT_MASK;
-
}
+
OUT_RING((box.x1 << R300_CLIPRECT_X_SHIFT) |
(box.y1 << R300_CLIPRECT_Y_SHIFT));
OUT_RING((box.x2 << R300_CLIPRECT_X_SHIFT) |
ADVANCE_RING();
}
+ /* flus cache and wait idle clean after cliprect change */
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(R300_RB3D_DC_FLUSH);
+ ADVANCE_RING();
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+ OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+ ADVANCE_RING();
+ /* set flush flag */
+ dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
return 0;
}
ADD_RANGE(0x21DC, 1);
ADD_RANGE(R300_VAP_UNKNOWN_221C, 1);
ADD_RANGE(R300_VAP_CLIP_X_0, 4);
- ADD_RANGE(R300_VAP_PVS_WAITIDLE, 1);
+ ADD_RANGE(R300_VAP_PVS_STATE_FLUSH_REG, 1);
ADD_RANGE(R300_VAP_UNKNOWN_2288, 1);
ADD_RANGE(R300_VAP_OUTPUT_VTX_FMT_0, 2);
ADD_RANGE(R300_VAP_PVS_CNTL_1, 3);
ADD_RANGE(R300_GB_ENABLE, 1);
ADD_RANGE(R300_GB_MSPOS0, 5);
- ADD_RANGE(R300_TX_CNTL, 1);
+ ADD_RANGE(R300_TX_INVALTAGS, 1);
ADD_RANGE(R300_TX_ENABLE, 1);
ADD_RANGE(0x4200, 4);
ADD_RANGE(0x4214, 1);
if (sz * 16 > cmdbuf->bufsz)
return -EINVAL;
- BEGIN_RING(5 + sz * 4);
- /* Wait for VAP to come to senses.. */
- /* there is no need to emit it multiple times, (only once before VAP is programmed,
- but this optimization is for later */
- OUT_RING_REG(R300_VAP_PVS_WAITIDLE, 0);
+ /* VAP is very sensitive so we purge cache before we program it
+ * and we also flush its state before & after */
+ BEGIN_RING(6);
+ OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(R300_RB3D_DC_FLUSH);
+ OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+ OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+ OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+ OUT_RING(0);
+ ADVANCE_RING();
+ /* set flush flag */
+ dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
+ BEGIN_RING(3 + sz * 4);
OUT_RING_REG(R300_VAP_PVS_UPLOAD_ADDRESS, addr);
OUT_RING(CP_PACKET0_TABLE(R300_VAP_PVS_UPLOAD_DATA, sz * 4 - 1));
OUT_RING_TABLE((int *)cmdbuf->buf, sz * 4);
+ ADVANCE_RING();
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(R300_VAP_PVS_STATE_FLUSH_REG, 0));
+ OUT_RING(0);
ADVANCE_RING();
cmdbuf->buf += sz * 16;
OUT_RING_TABLE((int *)cmdbuf->buf, 8);
ADVANCE_RING();
+ BEGIN_RING(4);
+ OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(R300_RB3D_DC_FLUSH);
+ OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+ OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+ ADVANCE_RING();
+ /* set flush flag */
+ dev_priv->track_flush |= RADEON_FLUSH_EMITED;
+
cmdbuf->buf += 8 * 4;
cmdbuf->bufsz -= 8 * 4;
return 0;
}
-static __inline__ int r300_emit_indx_buffer(drm_radeon_private_t *dev_priv,
- drm_radeon_kcmd_buffer_t *cmdbuf)
+static __inline__ int r300_emit_draw_indx_2(drm_radeon_private_t *dev_priv,
+ drm_radeon_kcmd_buffer_t *cmdbuf)
{
- u32 *cmd = (u32 *) cmdbuf->buf;
- int count, ret;
+ u32 *cmd;
+ int count;
+ int expected_count;
RING_LOCALS;
- count=(cmd[0]>>16) & 0x3fff;
+ cmd = (u32 *) cmdbuf->buf;
+ count = (cmd[0]>>16) & 0x3fff;
+ expected_count = cmd[1] >> 16;
+ if (!(cmd[1] & R300_VAP_VF_CNTL__INDEX_SIZE_32bit))
+ expected_count = (expected_count+1)/2;
- if ((cmd[1] & 0x8000ffff) != 0x80000810) {
- DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
- return -EINVAL;
- }
- ret = !radeon_check_offset(dev_priv, cmd[2]);
- if (ret) {
- DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+ if (count && count != expected_count) {
+ DRM_ERROR("3D_DRAW_INDX_2: packet size %i, expected %i\n",
+ count, expected_count);
return -EINVAL;
}
cmdbuf->buf += (count+2)*4;
cmdbuf->bufsz -= (count+2)*4;
+ if (!count) {
+ drm_r300_cmd_header_t header;
+
+ if (cmdbuf->bufsz < 4*4 + sizeof(header)) {
+ DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER, but stream is too short.\n");
+ return -EINVAL;
+ }
+
+ header.u = *(unsigned int *)cmdbuf->buf;
+
+ cmdbuf->buf += sizeof(header);
+ cmdbuf->bufsz -= sizeof(header);
+ cmd = (u32 *) cmdbuf->buf;
+
+ if (header.header.cmd_type != R300_CMD_PACKET3 ||
+ header.packet3.packet != R300_CMD_PACKET3_RAW ||
+ cmd[0] != CP_PACKET3(RADEON_CP_INDX_BUFFER, 2)) {
+ DRM_ERROR("3D_DRAW_INDX_2: expect subsequent INDX_BUFFER.\n");
+ return -EINVAL;
+ }
+
+ if ((cmd[1] & 0x8000ffff) != 0x80000810) {
+ DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
+ return -EINVAL;
+ }
+ if (!radeon_check_offset(dev_priv, cmd[2])) {
+ DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
+ return -EINVAL;
+ }
+ if (cmd[3] != expected_count) {
+ DRM_ERROR("INDX_BUFFER: buffer size %i, expected %i\n",
+ cmd[3], expected_count);
+ return -EINVAL;
+ }
+
+ BEGIN_RING(4);
+ OUT_RING(cmd[0]);
+ OUT_RING_TABLE((int *)(cmdbuf->buf + 4), 3);
+ ADVANCE_RING();
+
+ cmdbuf->buf += 4*4;
+ cmdbuf->bufsz -= 4*4;
+ }
+
return 0;
}
case RADEON_CNTL_BITBLT_MULTI:
return r300_emit_bitblt_multi(dev_priv, cmdbuf);
- case RADEON_CP_INDX_BUFFER: /* DRAW_INDX_2 without INDX_BUFFER seems to lock up the gpu */
- return r300_emit_indx_buffer(dev_priv, cmdbuf);
- case RADEON_CP_3D_DRAW_IMMD_2: /* triggers drawing using in-packet vertex data */
- case RADEON_CP_3D_DRAW_VBUF_2: /* triggers drawing of vertex buffers setup elsewhere */
- case RADEON_CP_3D_DRAW_INDX_2: /* triggers drawing using indices to vertex buffer */
+ case RADEON_CP_INDX_BUFFER:
+ DRM_ERROR("packet3 INDX_BUFFER without preceding 3D_DRAW_INDX_2 is illegal.\n");
+ return -EINVAL;
+ case RADEON_CP_3D_DRAW_IMMD_2:
+ /* triggers drawing using in-packet vertex data */
+ case RADEON_CP_3D_DRAW_VBUF_2:
+ /* triggers drawing of vertex buffers setup elsewhere */
+ dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+ RADEON_PURGE_EMITED);
+ break;
+ case RADEON_CP_3D_DRAW_INDX_2:
+ /* triggers drawing using indices to vertex buffer */
+ /* whenever we send vertex we clear flush & purge */
+ dev_priv->track_flush &= ~(RADEON_FLUSH_EMITED |
+ RADEON_PURGE_EMITED);
+ return r300_emit_draw_indx_2(dev_priv, cmdbuf);
case RADEON_WAIT_FOR_IDLE:
case RADEON_CP_NOP:
/* these packets are safe */
*/
static __inline__ void r300_pacify(drm_radeon_private_t *dev_priv)
{
+ uint32_t cache_z, cache_3d, cache_2d;
RING_LOCALS;
- BEGIN_RING(6);
- OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
- OUT_RING(R300_RB3D_DSTCACHE_UNKNOWN_0A);
+ cache_z = R300_ZC_FLUSH;
+ cache_2d = R300_RB2D_DC_FLUSH;
+ cache_3d = R300_RB3D_DC_FLUSH;
+ if (!(dev_priv->track_flush & RADEON_PURGE_EMITED)) {
+ /* we can purge, primitive where draw since last purge */
+ cache_z |= R300_ZC_FREE;
+ cache_2d |= R300_RB2D_DC_FREE;
+ cache_3d |= R300_RB3D_DC_FREE;
+ }
+
+ /* flush & purge zbuffer */
+ BEGIN_RING(2);
OUT_RING(CP_PACKET0(R300_ZB_ZCACHE_CTLSTAT, 0));
- OUT_RING(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE|
- R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
- OUT_RING(CP_PACKET3(RADEON_CP_NOP, 0));
- OUT_RING(0x0);
+ OUT_RING(cache_z);
+ ADVANCE_RING();
+ /* flush & purge 3d */
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(cache_3d);
+ ADVANCE_RING();
+ /* flush & purge texture */
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(R300_TX_INVALTAGS, 0));
+ OUT_RING(0);
+ ADVANCE_RING();
+ /* FIXME: is this one really needed ? */
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(R300_RB3D_AARESOLVE_CTL, 0));
+ OUT_RING(0);
+ ADVANCE_RING();
+ BEGIN_RING(2);
+ OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+ OUT_RING(RADEON_WAIT_3D_IDLECLEAN);
+ ADVANCE_RING();
+ /* flush & purge 2d through E2 as RB2D will trigger lockup */
+ BEGIN_RING(4);
+ OUT_RING(CP_PACKET0(R300_DSTCACHE_CTLSTAT, 0));
+ OUT_RING(cache_2d);
+ OUT_RING(CP_PACKET0(RADEON_WAIT_UNTIL, 0));
+ OUT_RING(RADEON_WAIT_2D_IDLECLEAN |
+ RADEON_WAIT_HOST_IDLECLEAN);
ADVANCE_RING();
+ /* set flush & purge flags */
+ dev_priv->track_flush |= RADEON_FLUSH_EMITED | RADEON_PURGE_EMITED;
}
/**
DRM_DEBUG("\n");
- /* See the comment above r300_emit_begin3d for why this call must be here,
- * and what the cleanup gotos are for. */
+ /* pacify */
r300_pacify(dev_priv);
if (cmdbuf->nbox <= R300_SIMULTANEOUS_CLIPRECTS) {
projects / linux-2.6 / blobdiff