radeon: Use request_firmware()
[linux-2.6.git] / drivers / gpu / drm / radeon / r100.c
index 69bd7cb..639d5b2 100644 (file)
 #include "drmP.h"
 #include "drm.h"
 #include "radeon_drm.h"
-#include "radeon_microcode.h"
 #include "radeon_reg.h"
 #include "radeon.h"
+#include <linux/firmware.h>
+#include <linux/platform_device.h>
+
+/* Firmware Names */
+#define FIRMWARE_R100          "radeon/R100_cp.bin"
+#define FIRMWARE_R200          "radeon/R200_cp.bin"
+#define FIRMWARE_R300          "radeon/R300_cp.bin"
+#define FIRMWARE_R420          "radeon/R420_cp.bin"
+#define FIRMWARE_RS690         "radeon/RS690_cp.bin"
+#define FIRMWARE_RS600         "radeon/RS600_cp.bin"
+#define FIRMWARE_R520          "radeon/R520_cp.bin"
+
+MODULE_FIRMWARE(FIRMWARE_R100);
+MODULE_FIRMWARE(FIRMWARE_R200);
+MODULE_FIRMWARE(FIRMWARE_R300);
+MODULE_FIRMWARE(FIRMWARE_R420);
+MODULE_FIRMWARE(FIRMWARE_RS690);
+MODULE_FIRMWARE(FIRMWARE_RS600);
+MODULE_FIRMWARE(FIRMWARE_R520);
 
 /* This files gather functions specifics to:
  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
@@ -173,8 +191,12 @@ void r100_mc_setup(struct radeon_device *rdev)
                DRM_ERROR("Failed to register debugfs file for R100 MC !\n");
        }
        /* Write VRAM size in case we are limiting it */
-       WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size);
-       tmp = rdev->mc.vram_location + rdev->mc.vram_size - 1;
+       WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
+       /* Novell bug 204882 for RN50/M6/M7 with 8/16/32MB VRAM,
+        * if the aperture is 64MB but we have 32MB VRAM
+        * we report only 32MB VRAM but we have to set MC_FB_LOCATION
+        * to 64MB, otherwise the gpu accidentially dies */
+       tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1;
        tmp = REG_SET(RADEON_MC_FB_TOP, tmp >> 16);
        tmp |= REG_SET(RADEON_MC_FB_START, rdev->mc.vram_location >> 16);
        WREG32(RADEON_MC_FB_LOCATION, tmp);
@@ -250,6 +272,72 @@ void r100_mc_fini(struct radeon_device *rdev)
 
 
 /*
+ * Interrupts
+ */
+int r100_irq_set(struct radeon_device *rdev)
+{
+       uint32_t tmp = 0;
+
+       if (rdev->irq.sw_int) {
+               tmp |= RADEON_SW_INT_ENABLE;
+       }
+       if (rdev->irq.crtc_vblank_int[0]) {
+               tmp |= RADEON_CRTC_VBLANK_MASK;
+       }
+       if (rdev->irq.crtc_vblank_int[1]) {
+               tmp |= RADEON_CRTC2_VBLANK_MASK;
+       }
+       WREG32(RADEON_GEN_INT_CNTL, tmp);
+       return 0;
+}
+
+static inline uint32_t r100_irq_ack(struct radeon_device *rdev)
+{
+       uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
+       uint32_t irq_mask = RADEON_SW_INT_TEST | RADEON_CRTC_VBLANK_STAT |
+               RADEON_CRTC2_VBLANK_STAT;
+
+       if (irqs) {
+               WREG32(RADEON_GEN_INT_STATUS, irqs);
+       }
+       return irqs & irq_mask;
+}
+
+int r100_irq_process(struct radeon_device *rdev)
+{
+       uint32_t status;
+
+       status = r100_irq_ack(rdev);
+       if (!status) {
+               return IRQ_NONE;
+       }
+       while (status) {
+               /* SW interrupt */
+               if (status & RADEON_SW_INT_TEST) {
+                       radeon_fence_process(rdev);
+               }
+               /* Vertical blank interrupts */
+               if (status & RADEON_CRTC_VBLANK_STAT) {
+                       drm_handle_vblank(rdev->ddev, 0);
+               }
+               if (status & RADEON_CRTC2_VBLANK_STAT) {
+                       drm_handle_vblank(rdev->ddev, 1);
+               }
+               status = r100_irq_ack(rdev);
+       }
+       return IRQ_HANDLED;
+}
+
+u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
+{
+       if (crtc == 0)
+               return RREG32(RADEON_CRTC_CRNT_FRAME);
+       else
+               return RREG32(RADEON_CRTC2_CRNT_FRAME);
+}
+
+
+/*
  * Fence emission
  */
 void r100_fence_ring_emit(struct radeon_device *rdev,
@@ -408,33 +496,33 @@ void r100_ring_start(struct radeon_device *rdev)
        radeon_ring_unlock_commit(rdev);
 }
 
-static void r100_cp_load_microcode(struct radeon_device *rdev)
+
+/* Load the microcode for the CP */
+static int r100_cp_init_microcode(struct radeon_device *rdev)
 {
-       int i;
+       struct platform_device *pdev;
+       const char *fw_name = NULL;
+       int err;
 
-       if (r100_gui_wait_for_idle(rdev)) {
-               printk(KERN_WARNING "Failed to wait GUI idle while "
-                      "programming pipes. Bad things might happen.\n");
-       }
+       DRM_DEBUG("\n");
 
-       WREG32(RADEON_CP_ME_RAM_ADDR, 0);
+       pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
+       err = IS_ERR(pdev);
+       if (err) {
+               printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
+               return -EINVAL;
+       }
        if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
            (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
            (rdev->family == CHIP_RS200)) {
                DRM_INFO("Loading R100 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, R100_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, R100_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_R100;
        } else if ((rdev->family == CHIP_R200) ||
                   (rdev->family == CHIP_RV250) ||
                   (rdev->family == CHIP_RV280) ||
                   (rdev->family == CHIP_RS300)) {
                DRM_INFO("Loading R200 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, R200_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, R200_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_R200;
        } else if ((rdev->family == CHIP_R300) ||
                   (rdev->family == CHIP_R350) ||
                   (rdev->family == CHIP_RV350) ||
@@ -442,31 +530,19 @@ static void r100_cp_load_microcode(struct radeon_device *rdev)
                   (rdev->family == CHIP_RS400) ||
                   (rdev->family == CHIP_RS480)) {
                DRM_INFO("Loading R300 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, R300_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, R300_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_R300;
        } else if ((rdev->family == CHIP_R420) ||
                   (rdev->family == CHIP_R423) ||
                   (rdev->family == CHIP_RV410)) {
                DRM_INFO("Loading R400 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, R420_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, R420_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_R420;
        } else if ((rdev->family == CHIP_RS690) ||
                   (rdev->family == CHIP_RS740)) {
                DRM_INFO("Loading RS690/RS740 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, RS690_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, RS690_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_RS690;
        } else if (rdev->family == CHIP_RS600) {
                DRM_INFO("Loading RS600 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, RS600_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, RS600_cp_microcode[i][0]);
-               }
+               fw_name = FIRMWARE_RS600;
        } else if ((rdev->family == CHIP_RV515) ||
                   (rdev->family == CHIP_R520) ||
                   (rdev->family == CHIP_RV530) ||
@@ -474,9 +550,43 @@ static void r100_cp_load_microcode(struct radeon_device *rdev)
                   (rdev->family == CHIP_RV560) ||
                   (rdev->family == CHIP_RV570)) {
                DRM_INFO("Loading R500 Microcode\n");
-               for (i = 0; i < 256; i++) {
-                       WREG32(RADEON_CP_ME_RAM_DATAH, R520_cp_microcode[i][1]);
-                       WREG32(RADEON_CP_ME_RAM_DATAL, R520_cp_microcode[i][0]);
+               fw_name = FIRMWARE_R520;
+       }
+
+       err = request_firmware(&rdev->fw, fw_name, &pdev->dev);
+       platform_device_unregister(pdev);
+       if (err) {
+               printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
+                      fw_name);
+       } else if (rdev->fw->size % 8) {
+               printk(KERN_ERR
+                      "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
+                      rdev->fw->size, fw_name);
+               err = -EINVAL;
+               release_firmware(rdev->fw);
+               rdev->fw = NULL;
+       }
+       return err;
+}
+static void r100_cp_load_microcode(struct radeon_device *rdev)
+{
+       const __be32 *fw_data;
+       int i, size;
+
+       if (r100_gui_wait_for_idle(rdev)) {
+               printk(KERN_WARNING "Failed to wait GUI idle while "
+                      "programming pipes. Bad things might happen.\n");
+       }
+
+       if (rdev->fw) {
+               size = rdev->fw->size / 4;
+               fw_data = (const __be32 *)&rdev->fw->data[0];
+               WREG32(RADEON_CP_ME_RAM_ADDR, 0);
+               for (i = 0; i < size; i += 2) {
+                       WREG32(RADEON_CP_ME_RAM_DATAH,
+                              be32_to_cpup(&fw_data[i]));
+                       WREG32(RADEON_CP_ME_RAM_DATAL,
+                              be32_to_cpup(&fw_data[i + 1]));
                }
        }
 }
@@ -515,6 +625,15 @@ int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
        } else {
                DRM_INFO("radeon: cp idle (0x%08X)\n", tmp);
        }
+
+       if (!rdev->fw) {
+               r = r100_cp_init_microcode(rdev);
+               if (r) {
+                       DRM_ERROR("Failed to load firmware!\n");
+                       return r;
+               }
+       }
+
        /* Align ring size */
        rb_bufsz = drm_order(ring_size / 8);
        ring_size = (1 << (rb_bufsz + 1)) * 4;
@@ -718,13 +837,14 @@ int r100_cs_packet_parse(struct radeon_cs_parser *p,
                         unsigned idx)
 {
        struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
-       uint32_t header = ib_chunk->kdata[idx];
+       uint32_t header;
 
        if (idx >= ib_chunk->length_dw) {
                DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
                          idx, ib_chunk->length_dw);
                return -EINVAL;
        }
+       header = ib_chunk->kdata[idx];
        pkt->idx = idx;
        pkt->type = CP_PACKET_GET_TYPE(header);
        pkt->count = CP_PACKET_GET_COUNT(header);
@@ -1447,25 +1567,28 @@ void r100_vram_init_sizes(struct radeon_device *rdev)
                uint32_t tom;
                /* read NB_TOM to get the amount of ram stolen for the GPU */
                tom = RREG32(RADEON_NB_TOM);
-               rdev->mc.vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
+               rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
                /* for IGPs we need to keep VRAM where it was put by the BIOS */
                rdev->mc.vram_location = (tom & 0xffff) << 16;
-               WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size);
+               WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
+               rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
        } else {
-               rdev->mc.vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
+               rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
                /* Some production boards of m6 will report 0
                 * if it's 8 MB
                 */
-               if (rdev->mc.vram_size == 0) {
-                       rdev->mc.vram_size = 8192 * 1024;
-                       WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size);
+               if (rdev->mc.real_vram_size == 0) {
+                       rdev->mc.real_vram_size = 8192 * 1024;
+                       WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
                }
                /* let driver place VRAM */
                rdev->mc.vram_location = 0xFFFFFFFFUL;
                 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM - 
                  * Novell bug 204882 + along with lots of ubuntu ones */
-               if (config_aper_size > rdev->mc.vram_size)
-                       rdev->mc.vram_size = config_aper_size;
+               if (config_aper_size > rdev->mc.real_vram_size)
+                       rdev->mc.mc_vram_size = config_aper_size;
+               else
+                       rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
        }
 
        /* work out accessible VRAM */
@@ -1477,8 +1600,11 @@ void r100_vram_init_sizes(struct radeon_device *rdev)
        if (accessible > rdev->mc.aper_size)
                accessible = rdev->mc.aper_size;
 
-       if (rdev->mc.vram_size > rdev->mc.aper_size)
-               rdev->mc.vram_size = rdev->mc.aper_size;
+       if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
+               rdev->mc.mc_vram_size = rdev->mc.aper_size;
+
+       if (rdev->mc.real_vram_size > rdev->mc.aper_size)
+               rdev->mc.real_vram_size = rdev->mc.aper_size;
 }
 
 void r100_vram_info(struct radeon_device *rdev)
@@ -1545,26 +1671,6 @@ void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
        r100_pll_errata_after_data(rdev);
 }
 
-uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
-{
-       if (reg < 0x10000)
-               return readl(((void __iomem *)rdev->rmmio) + reg);
-       else {
-               writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
-               return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
-       }
-}
-
-void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
-{
-       if (reg < 0x10000)
-               writel(v, ((void __iomem *)rdev->rmmio) + reg);
-       else {
-               writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
-               writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
-       }
-}
-
 int r100_init(struct radeon_device *rdev)
 {
        return 0;
@@ -1784,3 +1890,486 @@ void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
        int surf_index = reg * 16;
        WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
 }
+
+void r100_bandwidth_update(struct radeon_device *rdev)
+{
+       fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
+       fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
+       fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
+       uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
+       fixed20_12 memtcas_ff[8] = {
+               fixed_init(1),
+               fixed_init(2),
+               fixed_init(3),
+               fixed_init(0),
+               fixed_init_half(1),
+               fixed_init_half(2),
+               fixed_init(0),
+       };
+       fixed20_12 memtcas_rs480_ff[8] = {
+               fixed_init(0),
+               fixed_init(1),
+               fixed_init(2),
+               fixed_init(3),
+               fixed_init(0),
+               fixed_init_half(1),
+               fixed_init_half(2),
+               fixed_init_half(3),
+       };
+       fixed20_12 memtcas2_ff[8] = {
+               fixed_init(0),
+               fixed_init(1),
+               fixed_init(2),
+               fixed_init(3),
+               fixed_init(4),
+               fixed_init(5),
+               fixed_init(6),
+               fixed_init(7),
+       };
+       fixed20_12 memtrbs[8] = {
+               fixed_init(1),
+               fixed_init_half(1),
+               fixed_init(2),
+               fixed_init_half(2),
+               fixed_init(3),
+               fixed_init_half(3),
+               fixed_init(4),
+               fixed_init_half(4)
+       };
+       fixed20_12 memtrbs_r4xx[8] = {
+               fixed_init(4),
+               fixed_init(5),
+               fixed_init(6),
+               fixed_init(7),
+               fixed_init(8),
+               fixed_init(9),
+               fixed_init(10),
+               fixed_init(11)
+       };
+       fixed20_12 min_mem_eff;
+       fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
+       fixed20_12 cur_latency_mclk, cur_latency_sclk;
+       fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
+               disp_drain_rate2, read_return_rate;
+       fixed20_12 time_disp1_drop_priority;
+       int c;
+       int cur_size = 16;       /* in octawords */
+       int critical_point = 0, critical_point2;
+/*     uint32_t read_return_rate, time_disp1_drop_priority; */
+       int stop_req, max_stop_req;
+       struct drm_display_mode *mode1 = NULL;
+       struct drm_display_mode *mode2 = NULL;
+       uint32_t pixel_bytes1 = 0;
+       uint32_t pixel_bytes2 = 0;
+
+       if (rdev->mode_info.crtcs[0]->base.enabled) {
+               mode1 = &rdev->mode_info.crtcs[0]->base.mode;
+               pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
+       }
+       if (rdev->mode_info.crtcs[1]->base.enabled) {
+               mode2 = &rdev->mode_info.crtcs[1]->base.mode;
+               pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
+       }
+
+       min_mem_eff.full = rfixed_const_8(0);
+       /* get modes */
+       if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
+               uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
+               mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
+               mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
+               /* check crtc enables */
+               if (mode2)
+                       mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
+               if (mode1)
+                       mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
+               WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
+       }
+
+       /*
+        * determine is there is enough bw for current mode
+        */
+       mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
+       temp_ff.full = rfixed_const(100);
+       mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
+       sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
+       sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
+
+       temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
+       temp_ff.full = rfixed_const(temp);
+       mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
+
+       pix_clk.full = 0;
+       pix_clk2.full = 0;
+       peak_disp_bw.full = 0;
+       if (mode1) {
+               temp_ff.full = rfixed_const(1000);
+               pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
+               pix_clk.full = rfixed_div(pix_clk, temp_ff);
+               temp_ff.full = rfixed_const(pixel_bytes1);
+               peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
+       }
+       if (mode2) {
+               temp_ff.full = rfixed_const(1000);
+               pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
+               pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
+               temp_ff.full = rfixed_const(pixel_bytes2);
+               peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
+       }
+
+       mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
+       if (peak_disp_bw.full >= mem_bw.full) {
+               DRM_ERROR("You may not have enough display bandwidth for current mode\n"
+                         "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
+       }
+
+       /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
+       temp = RREG32(RADEON_MEM_TIMING_CNTL);
+       if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
+               mem_trcd = ((temp >> 2) & 0x3) + 1;
+               mem_trp  = ((temp & 0x3)) + 1;
+               mem_tras = ((temp & 0x70) >> 4) + 1;
+       } else if (rdev->family == CHIP_R300 ||
+                  rdev->family == CHIP_R350) { /* r300, r350 */
+               mem_trcd = (temp & 0x7) + 1;
+               mem_trp = ((temp >> 8) & 0x7) + 1;
+               mem_tras = ((temp >> 11) & 0xf) + 4;
+       } else if (rdev->family == CHIP_RV350 ||
+                  rdev->family <= CHIP_RV380) {
+               /* rv3x0 */
+               mem_trcd = (temp & 0x7) + 3;
+               mem_trp = ((temp >> 8) & 0x7) + 3;
+               mem_tras = ((temp >> 11) & 0xf) + 6;
+       } else if (rdev->family == CHIP_R420 ||
+                  rdev->family == CHIP_R423 ||
+                  rdev->family == CHIP_RV410) {
+               /* r4xx */
+               mem_trcd = (temp & 0xf) + 3;
+               if (mem_trcd > 15)
+                       mem_trcd = 15;
+               mem_trp = ((temp >> 8) & 0xf) + 3;
+               if (mem_trp > 15)
+                       mem_trp = 15;
+               mem_tras = ((temp >> 12) & 0x1f) + 6;
+               if (mem_tras > 31)
+                       mem_tras = 31;
+       } else { /* RV200, R200 */
+               mem_trcd = (temp & 0x7) + 1;
+               mem_trp = ((temp >> 8) & 0x7) + 1;
+               mem_tras = ((temp >> 12) & 0xf) + 4;
+       }
+       /* convert to FF */
+       trcd_ff.full = rfixed_const(mem_trcd);
+       trp_ff.full = rfixed_const(mem_trp);
+       tras_ff.full = rfixed_const(mem_tras);
+
+       /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
+       temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
+       data = (temp & (7 << 20)) >> 20;
+       if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
+               if (rdev->family == CHIP_RS480) /* don't think rs400 */
+                       tcas_ff = memtcas_rs480_ff[data];
+               else
+                       tcas_ff = memtcas_ff[data];
+       } else
+               tcas_ff = memtcas2_ff[data];
+
+       if (rdev->family == CHIP_RS400 ||
+           rdev->family == CHIP_RS480) {
+               /* extra cas latency stored in bits 23-25 0-4 clocks */
+               data = (temp >> 23) & 0x7;
+               if (data < 5)
+                       tcas_ff.full += rfixed_const(data);
+       }
+
+       if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
+               /* on the R300, Tcas is included in Trbs.
+                */
+               temp = RREG32(RADEON_MEM_CNTL);
+               data = (R300_MEM_NUM_CHANNELS_MASK & temp);
+               if (data == 1) {
+                       if (R300_MEM_USE_CD_CH_ONLY & temp) {
+                               temp = RREG32(R300_MC_IND_INDEX);
+                               temp &= ~R300_MC_IND_ADDR_MASK;
+                               temp |= R300_MC_READ_CNTL_CD_mcind;
+                               WREG32(R300_MC_IND_INDEX, temp);
+                               temp = RREG32(R300_MC_IND_DATA);
+                               data = (R300_MEM_RBS_POSITION_C_MASK & temp);
+                       } else {
+                               temp = RREG32(R300_MC_READ_CNTL_AB);
+                               data = (R300_MEM_RBS_POSITION_A_MASK & temp);
+                       }
+               } else {
+                       temp = RREG32(R300_MC_READ_CNTL_AB);
+                       data = (R300_MEM_RBS_POSITION_A_MASK & temp);
+               }
+               if (rdev->family == CHIP_RV410 ||
+                   rdev->family == CHIP_R420 ||
+                   rdev->family == CHIP_R423)
+                       trbs_ff = memtrbs_r4xx[data];
+               else
+                       trbs_ff = memtrbs[data];
+               tcas_ff.full += trbs_ff.full;
+       }
+
+       sclk_eff_ff.full = sclk_ff.full;
+
+       if (rdev->flags & RADEON_IS_AGP) {
+               fixed20_12 agpmode_ff;
+               agpmode_ff.full = rfixed_const(radeon_agpmode);
+               temp_ff.full = rfixed_const_666(16);
+               sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
+       }
+       /* TODO PCIE lanes may affect this - agpmode == 16?? */
+
+       if (ASIC_IS_R300(rdev)) {
+               sclk_delay_ff.full = rfixed_const(250);
+       } else {
+               if ((rdev->family == CHIP_RV100) ||
+                   rdev->flags & RADEON_IS_IGP) {
+                       if (rdev->mc.vram_is_ddr)
+                               sclk_delay_ff.full = rfixed_const(41);
+                       else
+                               sclk_delay_ff.full = rfixed_const(33);
+               } else {
+                       if (rdev->mc.vram_width == 128)
+                               sclk_delay_ff.full = rfixed_const(57);
+                       else
+                               sclk_delay_ff.full = rfixed_const(41);
+               }
+       }
+
+       mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
+
+       if (rdev->mc.vram_is_ddr) {
+               if (rdev->mc.vram_width == 32) {
+                       k1.full = rfixed_const(40);
+                       c  = 3;
+               } else {
+                       k1.full = rfixed_const(20);
+                       c  = 1;
+               }
+       } else {
+               k1.full = rfixed_const(40);
+               c  = 3;
+       }
+
+       temp_ff.full = rfixed_const(2);
+       mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
+       temp_ff.full = rfixed_const(c);
+       mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
+       temp_ff.full = rfixed_const(4);
+       mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
+       mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
+       mc_latency_mclk.full += k1.full;
+
+       mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
+       mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
+
+       /*
+         HW cursor time assuming worst case of full size colour cursor.
+       */
+       temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
+       temp_ff.full += trcd_ff.full;
+       if (temp_ff.full < tras_ff.full)
+               temp_ff.full = tras_ff.full;
+       cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
+
+       temp_ff.full = rfixed_const(cur_size);
+       cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
+       /*
+         Find the total latency for the display data.
+       */
+       disp_latency_overhead.full = rfixed_const(80);
+       disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
+       mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
+       mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
+
+       if (mc_latency_mclk.full > mc_latency_sclk.full)
+               disp_latency.full = mc_latency_mclk.full;
+       else
+               disp_latency.full = mc_latency_sclk.full;
+
+       /* setup Max GRPH_STOP_REQ default value */
+       if (ASIC_IS_RV100(rdev))
+               max_stop_req = 0x5c;
+       else
+               max_stop_req = 0x7c;
+
+       if (mode1) {
+               /*  CRTC1
+                   Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
+                   GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
+               */
+               stop_req = mode1->hdisplay * pixel_bytes1 / 16;
+
+               if (stop_req > max_stop_req)
+                       stop_req = max_stop_req;
+
+               /*
+                 Find the drain rate of the display buffer.
+               */
+               temp_ff.full = rfixed_const((16/pixel_bytes1));
+               disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
+
+               /*
+                 Find the critical point of the display buffer.
+               */
+               crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
+               crit_point_ff.full += rfixed_const_half(0);
+
+               critical_point = rfixed_trunc(crit_point_ff);
+
+               if (rdev->disp_priority == 2) {
+                       critical_point = 0;
+               }
+
+               /*
+                 The critical point should never be above max_stop_req-4.  Setting
+                 GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
+               */
+               if (max_stop_req - critical_point < 4)
+                       critical_point = 0;
+
+               if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
+                       /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
+                       critical_point = 0x10;
+               }
+
+               temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
+               temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
+               temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
+               temp &= ~(RADEON_GRPH_START_REQ_MASK);
+               if ((rdev->family == CHIP_R350) &&
+                   (stop_req > 0x15)) {
+                       stop_req -= 0x10;
+               }
+               temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
+               temp |= RADEON_GRPH_BUFFER_SIZE;
+               temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
+                         RADEON_GRPH_CRITICAL_AT_SOF |
+                         RADEON_GRPH_STOP_CNTL);
+               /*
+                 Write the result into the register.
+               */
+               WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
+                                                      (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
+
+#if 0
+               if ((rdev->family == CHIP_RS400) ||
+                   (rdev->family == CHIP_RS480)) {
+                       /* attempt to program RS400 disp regs correctly ??? */
+                       temp = RREG32(RS400_DISP1_REG_CNTL);
+                       temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
+                                 RS400_DISP1_STOP_REQ_LEVEL_MASK);
+                       WREG32(RS400_DISP1_REQ_CNTL1, (temp |
+                                                      (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
+                                                      (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
+                       temp = RREG32(RS400_DMIF_MEM_CNTL1);
+                       temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
+                                 RS400_DISP1_CRITICAL_POINT_STOP_MASK);
+                       WREG32(RS400_DMIF_MEM_CNTL1, (temp |
+                                                     (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
+                                                     (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
+               }
+#endif
+
+               DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
+                         /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
+                         (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
+       }
+
+       if (mode2) {
+               u32 grph2_cntl;
+               stop_req = mode2->hdisplay * pixel_bytes2 / 16;
+
+               if (stop_req > max_stop_req)
+                       stop_req = max_stop_req;
+
+               /*
+                 Find the drain rate of the display buffer.
+               */
+               temp_ff.full = rfixed_const((16/pixel_bytes2));
+               disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
+
+               grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
+               grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
+               grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
+               grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
+               if ((rdev->family == CHIP_R350) &&
+                   (stop_req > 0x15)) {
+                       stop_req -= 0x10;
+               }
+               grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
+               grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
+               grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
+                         RADEON_GRPH_CRITICAL_AT_SOF |
+                         RADEON_GRPH_STOP_CNTL);
+
+               if ((rdev->family == CHIP_RS100) ||
+                   (rdev->family == CHIP_RS200))
+                       critical_point2 = 0;
+               else {
+                       temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
+                       temp_ff.full = rfixed_const(temp);
+                       temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
+                       if (sclk_ff.full < temp_ff.full)
+                               temp_ff.full = sclk_ff.full;
+
+                       read_return_rate.full = temp_ff.full;
+
+                       if (mode1) {
+                               temp_ff.full = read_return_rate.full - disp_drain_rate.full;
+                               time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
+                       } else {
+                               time_disp1_drop_priority.full = 0;
+                       }
+                       crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
+                       crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
+                       crit_point_ff.full += rfixed_const_half(0);
+
+                       critical_point2 = rfixed_trunc(crit_point_ff);
+
+                       if (rdev->disp_priority == 2) {
+                               critical_point2 = 0;
+                       }
+
+                       if (max_stop_req - critical_point2 < 4)
+                               critical_point2 = 0;
+
+               }
+
+               if (critical_point2 == 0 && rdev->family == CHIP_R300) {
+                       /* some R300 cards have problem with this set to 0 */
+                       critical_point2 = 0x10;
+               }
+
+               WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
+                                                 (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
+
+               if ((rdev->family == CHIP_RS400) ||
+                   (rdev->family == CHIP_RS480)) {
+#if 0
+                       /* attempt to program RS400 disp2 regs correctly ??? */
+                       temp = RREG32(RS400_DISP2_REQ_CNTL1);
+                       temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
+                                 RS400_DISP2_STOP_REQ_LEVEL_MASK);
+                       WREG32(RS400_DISP2_REQ_CNTL1, (temp |
+                                                      (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
+                                                      (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
+                       temp = RREG32(RS400_DISP2_REQ_CNTL2);
+                       temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
+                                 RS400_DISP2_CRITICAL_POINT_STOP_MASK);
+                       WREG32(RS400_DISP2_REQ_CNTL2, (temp |
+                                                      (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
+                                                      (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
+#endif
+                       WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
+                       WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
+                       WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
+                       WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
+               }
+
+               DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
+                         (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
+       }
+}