Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / gpu / drm / nouveau / nvkm / engine / disp / dport.c
diff --git a/kernel/drivers/gpu/drm/nouveau/nvkm/engine/disp/dport.c b/kernel/drivers/gpu/drm/nouveau/nvkm/engine/disp/dport.c
new file mode 100644 (file)
index 0000000..6834766
--- /dev/null
@@ -0,0 +1,398 @@
+/*
+ * Copyright 2013 Red Hat Inc.
+ *
+ * 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, sublicense,
+ * 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 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) 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.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "dport.h"
+#include "outpdp.h"
+#include "nv50.h"
+
+#include <subdev/bios.h>
+#include <subdev/bios/init.h>
+#include <subdev/i2c.h>
+
+#include <nvif/class.h>
+
+/******************************************************************************
+ * link training
+ *****************************************************************************/
+struct dp_state {
+       struct nvkm_output_dp *outp;
+       int link_nr;
+       u32 link_bw;
+       u8  stat[6];
+       u8  conf[4];
+       bool pc2;
+       u8  pc2stat;
+       u8  pc2conf[2];
+};
+
+static int
+dp_set_link_config(struct dp_state *dp)
+{
+       struct nvkm_output_dp_impl *impl = (void *)nv_oclass(dp->outp);
+       struct nvkm_output_dp *outp = dp->outp;
+       struct nvkm_disp *disp = nvkm_disp(outp);
+       struct nvkm_bios *bios = nvkm_bios(disp);
+       struct nvbios_init init = {
+               .subdev = nv_subdev(disp),
+               .bios = bios,
+               .offset = 0x0000,
+               .outp = &outp->base.info,
+               .crtc = -1,
+               .execute = 1,
+       };
+       u32 lnkcmp;
+       u8 sink[2];
+       int ret;
+
+       DBG("%d lanes at %d KB/s\n", dp->link_nr, dp->link_bw);
+
+       /* set desired link configuration on the source */
+       if ((lnkcmp = dp->outp->info.lnkcmp)) {
+               if (outp->version < 0x30) {
+                       while ((dp->link_bw / 10) < nv_ro16(bios, lnkcmp))
+                               lnkcmp += 4;
+                       init.offset = nv_ro16(bios, lnkcmp + 2);
+               } else {
+                       while ((dp->link_bw / 27000) < nv_ro08(bios, lnkcmp))
+                               lnkcmp += 3;
+                       init.offset = nv_ro16(bios, lnkcmp + 1);
+               }
+
+               nvbios_exec(&init);
+       }
+
+       ret = impl->lnk_ctl(outp, dp->link_nr, dp->link_bw / 27000,
+                           outp->dpcd[DPCD_RC02] &
+                                      DPCD_RC02_ENHANCED_FRAME_CAP);
+       if (ret) {
+               if (ret < 0)
+                       ERR("lnk_ctl failed with %d\n", ret);
+               return ret;
+       }
+
+       impl->lnk_pwr(outp, dp->link_nr);
+
+       /* set desired link configuration on the sink */
+       sink[0] = dp->link_bw / 27000;
+       sink[1] = dp->link_nr;
+       if (outp->dpcd[DPCD_RC02] & DPCD_RC02_ENHANCED_FRAME_CAP)
+               sink[1] |= DPCD_LC01_ENHANCED_FRAME_EN;
+
+       return nv_wraux(outp->base.edid, DPCD_LC00_LINK_BW_SET, sink, 2);
+}
+
+static void
+dp_set_training_pattern(struct dp_state *dp, u8 pattern)
+{
+       struct nvkm_output_dp_impl *impl = (void *)nv_oclass(dp->outp);
+       struct nvkm_output_dp *outp = dp->outp;
+       u8 sink_tp;
+
+       DBG("training pattern %d\n", pattern);
+       impl->pattern(outp, pattern);
+
+       nv_rdaux(outp->base.edid, DPCD_LC02, &sink_tp, 1);
+       sink_tp &= ~DPCD_LC02_TRAINING_PATTERN_SET;
+       sink_tp |= pattern;
+       nv_wraux(outp->base.edid, DPCD_LC02, &sink_tp, 1);
+}
+
+static int
+dp_link_train_commit(struct dp_state *dp, bool pc)
+{
+       struct nvkm_output_dp_impl *impl = (void *)nv_oclass(dp->outp);
+       struct nvkm_output_dp *outp = dp->outp;
+       int ret, i;
+
+       for (i = 0; i < dp->link_nr; i++) {
+               u8 lane = (dp->stat[4 + (i >> 1)] >> ((i & 1) * 4)) & 0xf;
+               u8 lpc2 = (dp->pc2stat >> (i * 2)) & 0x3;
+               u8 lpre = (lane & 0x0c) >> 2;
+               u8 lvsw = (lane & 0x03) >> 0;
+               u8 hivs = 3 - lpre;
+               u8 hipe = 3;
+               u8 hipc = 3;
+
+               if (lpc2 >= hipc)
+                       lpc2 = hipc | DPCD_LC0F_LANE0_MAX_POST_CURSOR2_REACHED;
+               if (lpre >= hipe) {
+                       lpre = hipe | DPCD_LC03_MAX_SWING_REACHED; /* yes. */
+                       lvsw = hivs = 3 - (lpre & 3);
+               } else
+               if (lvsw >= hivs) {
+                       lvsw = hivs | DPCD_LC03_MAX_SWING_REACHED;
+               }
+
+               dp->conf[i] = (lpre << 3) | lvsw;
+               dp->pc2conf[i >> 1] |= lpc2 << ((i & 1) * 4);
+
+               DBG("config lane %d %02x %02x\n", i, dp->conf[i], lpc2);
+               impl->drv_ctl(outp, i, lvsw & 3, lpre & 3, lpc2 & 3);
+       }
+
+       ret = nv_wraux(outp->base.edid, DPCD_LC03(0), dp->conf, 4);
+       if (ret)
+               return ret;
+
+       if (pc) {
+               ret = nv_wraux(outp->base.edid, DPCD_LC0F, dp->pc2conf, 2);
+               if (ret)
+                       return ret;
+       }
+
+       return 0;
+}
+
+static int
+dp_link_train_update(struct dp_state *dp, bool pc, u32 delay)
+{
+       struct nvkm_output_dp *outp = dp->outp;
+       int ret;
+
+       if (outp->dpcd[DPCD_RC0E_AUX_RD_INTERVAL])
+               mdelay(outp->dpcd[DPCD_RC0E_AUX_RD_INTERVAL] * 4);
+       else
+               udelay(delay);
+
+       ret = nv_rdaux(outp->base.edid, DPCD_LS02, dp->stat, 6);
+       if (ret)
+               return ret;
+
+       if (pc) {
+               ret = nv_rdaux(outp->base.edid, DPCD_LS0C, &dp->pc2stat, 1);
+               if (ret)
+                       dp->pc2stat = 0x00;
+               DBG("status %6ph pc2 %02x\n", dp->stat, dp->pc2stat);
+       } else {
+               DBG("status %6ph\n", dp->stat);
+       }
+
+       return 0;
+}
+
+static int
+dp_link_train_cr(struct dp_state *dp)
+{
+       bool cr_done = false, abort = false;
+       int voltage = dp->conf[0] & DPCD_LC03_VOLTAGE_SWING_SET;
+       int tries = 0, i;
+
+       dp_set_training_pattern(dp, 1);
+
+       do {
+               if (dp_link_train_commit(dp, false) ||
+                   dp_link_train_update(dp, false, 100))
+                       break;
+
+               cr_done = true;
+               for (i = 0; i < dp->link_nr; i++) {
+                       u8 lane = (dp->stat[i >> 1] >> ((i & 1) * 4)) & 0xf;
+                       if (!(lane & DPCD_LS02_LANE0_CR_DONE)) {
+                               cr_done = false;
+                               if (dp->conf[i] & DPCD_LC03_MAX_SWING_REACHED)
+                                       abort = true;
+                               break;
+                       }
+               }
+
+               if ((dp->conf[0] & DPCD_LC03_VOLTAGE_SWING_SET) != voltage) {
+                       voltage = dp->conf[0] & DPCD_LC03_VOLTAGE_SWING_SET;
+                       tries = 0;
+               }
+       } while (!cr_done && !abort && ++tries < 5);
+
+       return cr_done ? 0 : -1;
+}
+
+static int
+dp_link_train_eq(struct dp_state *dp)
+{
+       struct nvkm_output_dp *outp = dp->outp;
+       bool eq_done = false, cr_done = true;
+       int tries = 0, i;
+
+       if (outp->dpcd[2] & DPCD_RC02_TPS3_SUPPORTED)
+               dp_set_training_pattern(dp, 3);
+       else
+               dp_set_training_pattern(dp, 2);
+
+       do {
+               if ((tries &&
+                   dp_link_train_commit(dp, dp->pc2)) ||
+                   dp_link_train_update(dp, dp->pc2, 400))
+                       break;
+
+               eq_done = !!(dp->stat[2] & DPCD_LS04_INTERLANE_ALIGN_DONE);
+               for (i = 0; i < dp->link_nr && eq_done; i++) {
+                       u8 lane = (dp->stat[i >> 1] >> ((i & 1) * 4)) & 0xf;
+                       if (!(lane & DPCD_LS02_LANE0_CR_DONE))
+                               cr_done = false;
+                       if (!(lane & DPCD_LS02_LANE0_CHANNEL_EQ_DONE) ||
+                           !(lane & DPCD_LS02_LANE0_SYMBOL_LOCKED))
+                               eq_done = false;
+               }
+       } while (!eq_done && cr_done && ++tries <= 5);
+
+       return eq_done ? 0 : -1;
+}
+
+static void
+dp_link_train_init(struct dp_state *dp, bool spread)
+{
+       struct nvkm_output_dp *outp = dp->outp;
+       struct nvkm_disp *disp = nvkm_disp(outp);
+       struct nvkm_bios *bios = nvkm_bios(disp);
+       struct nvbios_init init = {
+               .subdev = nv_subdev(disp),
+               .bios = bios,
+               .outp = &outp->base.info,
+               .crtc = -1,
+               .execute = 1,
+       };
+
+       /* set desired spread */
+       if (spread)
+               init.offset = outp->info.script[2];
+       else
+               init.offset = outp->info.script[3];
+       nvbios_exec(&init);
+
+       /* pre-train script */
+       init.offset = outp->info.script[0];
+       nvbios_exec(&init);
+}
+
+static void
+dp_link_train_fini(struct dp_state *dp)
+{
+       struct nvkm_output_dp *outp = dp->outp;
+       struct nvkm_disp *disp = nvkm_disp(outp);
+       struct nvkm_bios *bios = nvkm_bios(disp);
+       struct nvbios_init init = {
+               .subdev = nv_subdev(disp),
+               .bios = bios,
+               .outp = &outp->base.info,
+               .crtc = -1,
+               .execute = 1,
+       };
+
+       /* post-train script */
+       init.offset = outp->info.script[1],
+       nvbios_exec(&init);
+}
+
+static const struct dp_rates {
+       u32 rate;
+       u8  bw;
+       u8  nr;
+} nvkm_dp_rates[] = {
+       { 2160000, 0x14, 4 },
+       { 1080000, 0x0a, 4 },
+       { 1080000, 0x14, 2 },
+       {  648000, 0x06, 4 },
+       {  540000, 0x0a, 2 },
+       {  540000, 0x14, 1 },
+       {  324000, 0x06, 2 },
+       {  270000, 0x0a, 1 },
+       {  162000, 0x06, 1 },
+       {}
+};
+
+void
+nvkm_dp_train(struct work_struct *w)
+{
+       struct nvkm_output_dp *outp = container_of(w, typeof(*outp), lt.work);
+       struct nv50_disp_priv *priv = (void *)nvkm_disp(outp);
+       const struct dp_rates *cfg = nvkm_dp_rates;
+       struct dp_state _dp = {
+               .outp = outp,
+       }, *dp = &_dp;
+       u32 datarate = 0;
+       int ret;
+
+       if (!outp->base.info.location && priv->sor.magic)
+               priv->sor.magic(&outp->base);
+
+       /* bring capabilities within encoder limits */
+       if (nv_mclass(priv) < GF110_DISP)
+               outp->dpcd[2] &= ~DPCD_RC02_TPS3_SUPPORTED;
+       if ((outp->dpcd[2] & 0x1f) > outp->base.info.dpconf.link_nr) {
+               outp->dpcd[2] &= ~DPCD_RC02_MAX_LANE_COUNT;
+               outp->dpcd[2] |= outp->base.info.dpconf.link_nr;
+       }
+       if (outp->dpcd[1] > outp->base.info.dpconf.link_bw)
+               outp->dpcd[1] = outp->base.info.dpconf.link_bw;
+       dp->pc2 = outp->dpcd[2] & DPCD_RC02_TPS3_SUPPORTED;
+
+       /* restrict link config to the lowest required rate, if requested */
+       if (datarate) {
+               datarate = (datarate / 8) * 10; /* 8B/10B coding overhead */
+               while (cfg[1].rate >= datarate)
+                       cfg++;
+       }
+       cfg--;
+
+       /* disable link interrupt handling during link training */
+       nvkm_notify_put(&outp->irq);
+
+       /* enable down-spreading and execute pre-train script from vbios */
+       dp_link_train_init(dp, outp->dpcd[3] & 0x01);
+
+       while (ret = -EIO, (++cfg)->rate) {
+               /* select next configuration supported by encoder and sink */
+               while (cfg->nr > (outp->dpcd[2] & DPCD_RC02_MAX_LANE_COUNT) ||
+                      cfg->bw > (outp->dpcd[DPCD_RC01_MAX_LINK_RATE]))
+                       cfg++;
+               dp->link_bw = cfg->bw * 27000;
+               dp->link_nr = cfg->nr;
+
+               /* program selected link configuration */
+               ret = dp_set_link_config(dp);
+               if (ret == 0) {
+                       /* attempt to train the link at this configuration */
+                       memset(dp->stat, 0x00, sizeof(dp->stat));
+                       if (!dp_link_train_cr(dp) &&
+                           !dp_link_train_eq(dp))
+                               break;
+               } else
+               if (ret) {
+                       /* dp_set_link_config() handled training, or
+                        * we failed to communicate with the sink.
+                        */
+                       break;
+               }
+       }
+
+       /* finish link training and execute post-train script from vbios */
+       dp_set_training_pattern(dp, 0);
+       if (ret < 0)
+               ERR("link training failed\n");
+
+       dp_link_train_fini(dp);
+
+       /* signal completion and enable link interrupt handling */
+       DBG("training complete\n");
+       atomic_set(&outp->lt.done, 1);
+       wake_up(&outp->lt.wait);
+       nvkm_notify_get(&outp->irq);
+}