drm/nv50: Improve PGRAPH interrupt handling.
Marcin Koƛcielnicki [Mon, 1 Mar 2010 00:18:39 +0000 (00:18 +0000)]
This makes nouveau recognise and report more kinds of PGRAPH errors, as
well as prevent GPU lockups resulting from some of them.

Lots of guesswork was involved and some part of this is probably
incorrect. Some potential-lockuop situations are handled by just
resetting a whole PGRAPH subunit, which doesn't sound like a "proper"
solution, but seems to work just fine... for now.

Signed-off-by: Marcin Koƛcielnicki <koriakin@0x04.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/Makefile
drivers/gpu/drm/nouveau/nouveau_drv.h
drivers/gpu/drm/nouveau/nouveau_irq.c
drivers/gpu/drm/nouveau/nouveau_state.c
drivers/gpu/drm/nouveau/nv50_fb.c [new file with mode: 0644]
drivers/gpu/drm/nouveau/nv50_graph.c
drivers/gpu/drm/nouveau/nv50_grctx.c

index 32db806..7f0d807 100644 (file)
@@ -12,7 +12,7 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nouveau_dp.o nouveau_grctx.o \
              nv04_timer.o \
              nv04_mc.o nv40_mc.o nv50_mc.o \
-             nv04_fb.o nv10_fb.o nv40_fb.o \
+             nv04_fb.o nv10_fb.o nv40_fb.o nv50_fb.o \
              nv04_fifo.o nv10_fifo.o nv40_fifo.o nv50_fifo.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o \
index 2f8ce42..ad2d75d 100644 (file)
@@ -930,6 +930,10 @@ extern void nv40_fb_takedown(struct drm_device *);
 extern void nv40_fb_set_region_tiling(struct drm_device *, int, uint32_t,
                                      uint32_t, uint32_t);
 
+/* nv50_fb.c */
+extern int  nv50_fb_init(struct drm_device *);
+extern void nv50_fb_takedown(struct drm_device *);
+
 /* nv04_fifo.c */
 extern int  nv04_fifo_init(struct drm_device *);
 extern void nv04_fifo_disable(struct drm_device *);
index 95220dd..2bd59a9 100644 (file)
@@ -311,6 +311,31 @@ nouveau_print_bitfield_names_(uint32_t value,
 #define nouveau_print_bitfield_names(val, namelist) \
        nouveau_print_bitfield_names_((val), (namelist), ARRAY_SIZE(namelist))
 
+struct nouveau_enum_names {
+       uint32_t value;
+       const char *name;
+};
+
+static void
+nouveau_print_enum_names_(uint32_t value,
+                               const struct nouveau_enum_names *namelist,
+                               const int namelist_len)
+{
+       /*
+        * Caller must have already printed the KERN_* log level for us.
+        * Also the caller is responsible for adding the newline.
+        */
+       int i;
+       for (i = 0; i < namelist_len; ++i) {
+               if (value == namelist[i].value) {
+                       printk("%s", namelist[i].name);
+                       return;
+               }
+       }
+       printk("unknown value 0x%08x", value);
+}
+#define nouveau_print_enum_names(val, namelist) \
+       nouveau_print_enum_names_((val), (namelist), ARRAY_SIZE(namelist))
 
 static int
 nouveau_graph_chid_from_grctx(struct drm_device *dev)
@@ -427,14 +452,16 @@ nouveau_graph_dump_trap_info(struct drm_device *dev, const char *id,
        struct drm_nouveau_private *dev_priv = dev->dev_private;
        uint32_t nsource = trap->nsource, nstatus = trap->nstatus;
 
-       NV_INFO(dev, "%s - nSource:", id);
-       nouveau_print_bitfield_names(nsource, nsource_names);
-       printk(", nStatus:");
-       if (dev_priv->card_type < NV_10)
-               nouveau_print_bitfield_names(nstatus, nstatus_names);
-       else
-               nouveau_print_bitfield_names(nstatus, nstatus_names_nv10);
-       printk("\n");
+       if (dev_priv->card_type < NV_50) {
+               NV_INFO(dev, "%s - nSource:", id);
+               nouveau_print_bitfield_names(nsource, nsource_names);
+               printk(", nStatus:");
+               if (dev_priv->card_type < NV_10)
+                       nouveau_print_bitfield_names(nstatus, nstatus_names);
+               else
+                       nouveau_print_bitfield_names(nstatus, nstatus_names_nv10);
+               printk("\n");
+       }
 
        NV_INFO(dev, "%s - Ch %d/%d Class 0x%04x Mthd 0x%04x "
                                        "Data 0x%08x:0x%08x\n",
@@ -578,27 +605,502 @@ nouveau_pgraph_irq_handler(struct drm_device *dev)
 }
 
 static void
+nv50_pfb_vm_trap(struct drm_device *dev, int display, const char *name)
+{
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       uint32_t trap[6];
+       int i, ch;
+       uint32_t idx = nv_rd32(dev, 0x100c90);
+       if (idx & 0x80000000) {
+               idx &= 0xffffff;
+               if (display) {
+                       for (i = 0; i < 6; i++) {
+                               nv_wr32(dev, 0x100c90, idx | i << 24);
+                               trap[i] = nv_rd32(dev, 0x100c94);
+                       }
+                       for (ch = 0; ch < dev_priv->engine.fifo.channels; ch++) {
+                               struct nouveau_channel *chan = dev_priv->fifos[ch];
+
+                               if (!chan || !chan->ramin)
+                                       continue;
+
+                               if (trap[1] == chan->ramin->instance >> 12)
+                                       break;
+                       }
+                       NV_INFO(dev, "%s - VM: Trapped %s at %02x%04x%04x status %08x %08x channel %d\n",
+                                       name, (trap[5]&0x100?"read":"write"),
+                                       trap[5]&0xff, trap[4]&0xffff,
+                                       trap[3]&0xffff, trap[0], trap[2], ch);
+               }
+               nv_wr32(dev, 0x100c90, idx | 0x80000000);
+       } else if (display) {
+               NV_INFO(dev, "%s - no VM fault?\n", name);
+       }
+}
+
+static struct nouveau_enum_names nv50_mp_exec_error_names[] =
+{
+       { 3, "STACK_UNDERFLOW" },
+       { 4, "QUADON_ACTIVE" },
+       { 8, "TIMEOUT" },
+       { 0x10, "INVALID_OPCODE" },
+       { 0x40, "BREAKPOINT" },
+};
+
+static void
+nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display)
+{
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       uint32_t units = nv_rd32(dev, 0x1540);
+       uint32_t addr, mp10, status, pc, oplow, ophigh;
+       int i;
+       int mps = 0;
+       for (i = 0; i < 4; i++) {
+               if (!(units & 1 << (i+24)))
+                       continue;
+               if (dev_priv->chipset < 0xa0)
+                       addr = 0x408200 + (tpid << 12) + (i << 7);
+               else
+                       addr = 0x408100 + (tpid << 11) + (i << 7);
+               mp10 = nv_rd32(dev, addr + 0x10);
+               status = nv_rd32(dev, addr + 0x14);
+               if (!status)
+                       continue;
+               if (display) {
+                       nv_rd32(dev, addr + 0x20);
+                       pc = nv_rd32(dev, addr + 0x24);
+                       oplow = nv_rd32(dev, addr + 0x70);
+                       ophigh= nv_rd32(dev, addr + 0x74);
+                       NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - "
+                                       "TP %d MP %d: ", tpid, i);
+                       nouveau_print_enum_names(status,
+                                       nv50_mp_exec_error_names);
+                       printk(" at %06x warp %d, opcode %08x %08x\n",
+                                       pc&0xffffff, pc >> 24,
+                                       oplow, ophigh);
+               }
+               nv_wr32(dev, addr + 0x10, mp10);
+               nv_wr32(dev, addr + 0x14, 0);
+               mps++;
+       }
+       if (!mps && display)
+               NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - TP %d: "
+                               "No MPs claiming errors?\n", tpid);
+}
+
+static void
+nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
+               uint32_t ustatus_new, int display, const char *name)
+{
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       int tps = 0;
+       uint32_t units = nv_rd32(dev, 0x1540);
+       int i, r;
+       uint32_t ustatus_addr, ustatus;
+       for (i = 0; i < 16; i++) {
+               if (!(units & (1 << i)))
+                       continue;
+               if (dev_priv->chipset < 0xa0)
+                       ustatus_addr = ustatus_old + (i << 12);
+               else
+                       ustatus_addr = ustatus_new + (i << 11);
+               ustatus = nv_rd32(dev, ustatus_addr) & 0x7fffffff;
+               if (!ustatus)
+                       continue;
+               tps++;
+               switch (type) {
+               case 6: /* texture error... unknown for now */
+                       nv50_pfb_vm_trap(dev, display, name);
+                       if (display) {
+                               NV_ERROR(dev, "magic set %d:\n", i);
+                               for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
+                                       NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
+                                               nv_rd32(dev, r));
+                       }
+                       break;
+               case 7: /* MP error */
+                       if (ustatus & 0x00010000) {
+                               nv50_pgraph_mp_trap(dev, i, display);
+                               ustatus &= ~0x00010000;
+                       }
+                       break;
+               case 8: /* TPDMA error */
+                       {
+                       uint32_t e0c = nv_rd32(dev, ustatus_addr + 4);
+                       uint32_t e10 = nv_rd32(dev, ustatus_addr + 8);
+                       uint32_t e14 = nv_rd32(dev, ustatus_addr + 0xc);
+                       uint32_t e18 = nv_rd32(dev, ustatus_addr + 0x10);
+                       uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14);
+                       uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18);
+                       uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c);
+                       nv50_pfb_vm_trap(dev, display, name);
+                       /* 2d engine destination */
+                       if (ustatus & 0x00000010) {
+                               if (display) {
+                                       NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n",
+                                                       i, e14, e10);
+                                       NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+                                                       i, e0c, e18, e1c, e20, e24);
+                               }
+                               ustatus &= ~0x00000010;
+                       }
+                       /* Render target */
+                       if (ustatus & 0x00000040) {
+                               if (display) {
+                                       NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n",
+                                                       i, e14, e10);
+                                       NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+                                                       i, e0c, e18, e1c, e20, e24);
+                               }
+                               ustatus &= ~0x00000040;
+                       }
+                       /* CUDA memory: l[], g[] or stack. */
+                       if (ustatus & 0x00000080) {
+                               if (display) {
+                                       if (e18 & 0x80000000) {
+                                               /* g[] read fault? */
+                                               NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n",
+                                                               i, e14, e10 | ((e18 >> 24) & 0x1f));
+                                               e18 &= ~0x1f000000;
+                                       } else if (e18 & 0xc) {
+                                               /* g[] write fault? */
+                                               NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n",
+                                                               i, e14, e10 | ((e18 >> 7) & 0x1f));
+                                               e18 &= ~0x00000f80;
+                                       } else {
+                                               NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n",
+                                                               i, e14, e10);
+                                       }
+                                       NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+                                                       i, e0c, e18, e1c, e20, e24);
+                               }
+                               ustatus &= ~0x00000080;
+                       }
+                       }
+                       break;
+               }
+               if (ustatus) {
+                       if (display)
+                               NV_INFO(dev, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
+               }
+               nv_wr32(dev, ustatus_addr, 0xc0000000);
+       }
+
+       if (!tps && display)
+               NV_INFO(dev, "%s - No TPs claiming errors?\n", name);
+}
+
+static void
+nv50_pgraph_trap_handler(struct drm_device *dev)
+{
+       struct nouveau_pgraph_trap trap;
+       uint32_t status = nv_rd32(dev, 0x400108);
+       uint32_t ustatus;
+       int display = nouveau_ratelimit();
+
+
+       if (!status && display) {
+               nouveau_graph_trap_info(dev, &trap);
+               nouveau_graph_dump_trap_info(dev, "PGRAPH_TRAP", &trap);
+               NV_INFO(dev, "PGRAPH_TRAP - no units reporting traps?\n");
+       }
+
+       /* DISPATCH: Relays commands to other units and handles NOTIFY,
+        * COND, QUERY. If you get a trap from it, the command is still stuck
+        * in DISPATCH and you need to do something about it. */
+       if (status & 0x001) {
+               ustatus = nv_rd32(dev, 0x400804) & 0x7fffffff;
+               if (!ustatus && display) {
+                       NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - no ustatus?\n");
+               }
+
+               /* Known to be triggered by screwed up NOTIFY and COND... */
+               if (ustatus & 0x00000001) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_DISPATCH_FAULT");
+                       nv_wr32(dev, 0x400500, 0);
+                       if (nv_rd32(dev, 0x400808) & 0x80000000) {
+                               if (display) {
+                                       if (nouveau_graph_trapped_channel(dev, &trap.channel))
+                                               trap.channel = -1;
+                                       trap.class = nv_rd32(dev, 0x400814);
+                                       trap.mthd = nv_rd32(dev, 0x400808) & 0x1ffc;
+                                       trap.subc = (nv_rd32(dev, 0x400808) >> 16) & 0x7;
+                                       trap.data = nv_rd32(dev, 0x40080c);
+                                       trap.data2 = nv_rd32(dev, 0x400810);
+                                       nouveau_graph_dump_trap_info(dev,
+                                                       "PGRAPH_TRAP_DISPATCH_FAULT", &trap);
+                                       NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - 400808: %08x\n", nv_rd32(dev, 0x400808));
+                                       NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - 400848: %08x\n", nv_rd32(dev, 0x400848));
+                               }
+                               nv_wr32(dev, 0x400808, 0);
+                       } else if (display) {
+                               NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - No stuck command?\n");
+                       }
+                       nv_wr32(dev, 0x4008e8, nv_rd32(dev, 0x4008e8) & 3);
+                       nv_wr32(dev, 0x400848, 0);
+                       ustatus &= ~0x00000001;
+               }
+               if (ustatus & 0x00000002) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_DISPATCH_QUERY");
+                       nv_wr32(dev, 0x400500, 0);
+                       if (nv_rd32(dev, 0x40084c) & 0x80000000) {
+                               if (display) {
+                                       if (nouveau_graph_trapped_channel(dev, &trap.channel))
+                                               trap.channel = -1;
+                                       trap.class = nv_rd32(dev, 0x400814);
+                                       trap.mthd = nv_rd32(dev, 0x40084c) & 0x1ffc;
+                                       trap.subc = (nv_rd32(dev, 0x40084c) >> 16) & 0x7;
+                                       trap.data = nv_rd32(dev, 0x40085c);
+                                       trap.data2 = 0;
+                                       nouveau_graph_dump_trap_info(dev,
+                                                       "PGRAPH_TRAP_DISPATCH_QUERY", &trap);
+                                       NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_QUERY - 40084c: %08x\n", nv_rd32(dev, 0x40084c));
+                               }
+                               nv_wr32(dev, 0x40084c, 0);
+                       } else if (display) {
+                               NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_QUERY - No stuck command?\n");
+                       }
+                       ustatus &= ~0x00000002;
+               }
+               if (ustatus && display)
+                       NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - Unhandled ustatus 0x%08x\n", ustatus);
+               nv_wr32(dev, 0x400804, 0xc0000000);
+               nv_wr32(dev, 0x400108, 0x001);
+               status &= ~0x001;
+       }
+
+       /* TRAPs other than dispatch use the "normal" trap regs. */
+       if (status && display) {
+               nouveau_graph_trap_info(dev, &trap);
+               nouveau_graph_dump_trap_info(dev,
+                               "PGRAPH_TRAP", &trap);
+       }
+
+       /* M2MF: Memory to memory copy engine. */
+       if (status & 0x002) {
+               ustatus = nv_rd32(dev, 0x406800) & 0x7fffffff;
+               if (!ustatus && display) {
+                       NV_INFO(dev, "PGRAPH_TRAP_M2MF - no ustatus?\n");
+               }
+               if (ustatus & 0x00000001) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_NOTIFY");
+                       ustatus &= ~0x00000001;
+               }
+               if (ustatus & 0x00000002) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_IN");
+                       ustatus &= ~0x00000002;
+               }
+               if (ustatus & 0x00000004) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_OUT");
+                       ustatus &= ~0x00000004;
+               }
+               NV_INFO (dev, "PGRAPH_TRAP_M2MF - %08x %08x %08x %08x\n",
+                               nv_rd32(dev, 0x406804),
+                               nv_rd32(dev, 0x406808),
+                               nv_rd32(dev, 0x40680c),
+                               nv_rd32(dev, 0x406810));
+               if (ustatus && display)
+                       NV_INFO(dev, "PGRAPH_TRAP_M2MF - Unhandled ustatus 0x%08x\n", ustatus);
+               /* No sane way found yet -- just reset the bugger. */
+               nv_wr32(dev, 0x400040, 2);
+               nv_wr32(dev, 0x400040, 0);
+               nv_wr32(dev, 0x406800, 0xc0000000);
+               nv_wr32(dev, 0x400108, 0x002);
+               status &= ~0x002;
+       }
+
+       /* VFETCH: Fetches data from vertex buffers. */
+       if (status & 0x004) {
+               ustatus = nv_rd32(dev, 0x400c04) & 0x7fffffff;
+               if (!ustatus && display) {
+                       NV_INFO(dev, "PGRAPH_TRAP_VFETCH - no ustatus?\n");
+               }
+               if (ustatus & 0x00000001) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_VFETCH_FAULT");
+                       NV_INFO (dev, "PGRAPH_TRAP_VFETCH_FAULT - %08x %08x %08x %08x\n",
+                                       nv_rd32(dev, 0x400c00),
+                                       nv_rd32(dev, 0x400c08),
+                                       nv_rd32(dev, 0x400c0c),
+                                       nv_rd32(dev, 0x400c10));
+                       ustatus &= ~0x00000001;
+               }
+               if (ustatus && display)
+                       NV_INFO(dev, "PGRAPH_TRAP_VFETCH - Unhandled ustatus 0x%08x\n", ustatus);
+               nv_wr32(dev, 0x400c04, 0xc0000000);
+               nv_wr32(dev, 0x400108, 0x004);
+               status &= ~0x004;
+       }
+
+       /* STRMOUT: DirectX streamout / OpenGL transform feedback. */
+       if (status & 0x008) {
+               ustatus = nv_rd32(dev, 0x401800) & 0x7fffffff;
+               if (!ustatus && display) {
+                       NV_INFO(dev, "PGRAPH_TRAP_STRMOUT - no ustatus?\n");
+               }
+               if (ustatus & 0x00000001) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_STRMOUT_FAULT");
+                       NV_INFO (dev, "PGRAPH_TRAP_STRMOUT_FAULT - %08x %08x %08x %08x\n",
+                                       nv_rd32(dev, 0x401804),
+                                       nv_rd32(dev, 0x401808),
+                                       nv_rd32(dev, 0x40180c),
+                                       nv_rd32(dev, 0x401810));
+                       ustatus &= ~0x00000001;
+               }
+               if (ustatus && display)
+                       NV_INFO(dev, "PGRAPH_TRAP_STRMOUT - Unhandled ustatus 0x%08x\n", ustatus);
+               /* No sane way found yet -- just reset the bugger. */
+               nv_wr32(dev, 0x400040, 0x80);
+               nv_wr32(dev, 0x400040, 0);
+               nv_wr32(dev, 0x401800, 0xc0000000);
+               nv_wr32(dev, 0x400108, 0x008);
+               status &= ~0x008;
+       }
+
+       /* CCACHE: Handles code and c[] caches and fills them. */
+       if (status & 0x010) {
+               ustatus = nv_rd32(dev, 0x405018) & 0x7fffffff;
+               if (!ustatus && display) {
+                       NV_INFO(dev, "PGRAPH_TRAP_CCACHE - no ustatus?\n");
+               }
+               if (ustatus & 0x00000001) {
+                       nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_CCACHE_FAULT");
+                       NV_INFO (dev, "PGRAPH_TRAP_CCACHE_FAULT - %08x %08x %08x %08x %08x %08x %08x\n",
+                                       nv_rd32(dev, 0x405800),
+                                       nv_rd32(dev, 0x405804),
+                                       nv_rd32(dev, 0x405808),
+                                       nv_rd32(dev, 0x40580c),
+                                       nv_rd32(dev, 0x405810),
+                                       nv_rd32(dev, 0x405814),
+                                       nv_rd32(dev, 0x40581c));
+                       ustatus &= ~0x00000001;
+               }
+               if (ustatus && display)
+                       NV_INFO(dev, "PGRAPH_TRAP_CCACHE - Unhandled ustatus 0x%08x\n", ustatus);
+               nv_wr32(dev, 0x405018, 0xc0000000);
+               nv_wr32(dev, 0x400108, 0x010);
+               status &= ~0x010;
+       }
+
+       /* Unknown, not seen yet... 0x402000 is the only trap status reg
+        * remaining, so try to handle it anyway. Perhaps related to that
+        * unknown DMA slot on tesla? */
+       if (status & 0x20) {
+               nv50_pfb_vm_trap(dev, display, "PGRAPH_TRAP_UNKC04");
+               ustatus = nv_rd32(dev, 0x402000) & 0x7fffffff;
+               if (display)
+                       NV_INFO(dev, "PGRAPH_TRAP_UNKC04 - Unhandled ustatus 0x%08x\n", ustatus);
+               nv_wr32(dev, 0x402000, 0xc0000000);
+               /* no status modifiction on purpose */
+       }
+
+       /* TEXTURE: CUDA texturing units */
+       if (status & 0x040) {
+               nv50_pgraph_tp_trap (dev, 6, 0x408900, 0x408600, display,
+                               "PGRAPH_TRAP_TEXTURE");
+               nv_wr32(dev, 0x400108, 0x040);
+               status &= ~0x040;
+       }
+
+       /* MP: CUDA execution engines. */
+       if (status & 0x080) {
+               nv50_pgraph_tp_trap (dev, 7, 0x408314, 0x40831c, display,
+                               "PGRAPH_TRAP_MP");
+               nv_wr32(dev, 0x400108, 0x080);
+               status &= ~0x080;
+       }
+
+       /* TPDMA:  Handles TP-initiated uncached memory accesses:
+        * l[], g[], stack, 2d surfaces, render targets. */
+       if (status & 0x100) {
+               nv50_pgraph_tp_trap (dev, 8, 0x408e08, 0x408708, display,
+                               "PGRAPH_TRAP_TPDMA");
+               nv_wr32(dev, 0x400108, 0x100);
+               status &= ~0x100;
+       }
+
+       if (status) {
+               if (display)
+                       NV_INFO(dev, "PGRAPH_TRAP - Unknown trap 0x%08x\n",
+                               status);
+               nv_wr32(dev, 0x400108, status);
+       }
+}
+
+/* There must be a *lot* of these. Will take some time to gather them up. */
+static struct nouveau_enum_names nv50_data_error_names[] =
+{
+       { 4,    "INVALID_VALUE" },
+       { 5,    "INVALID_ENUM" },
+       { 8,    "INVALID_OBJECT" },
+       { 0xc,  "INVALID_BITFIELD" },
+       { 0x28, "MP_NO_REG_SPACE" },
+       { 0x2b, "MP_BLOCK_SIZE_MISMATCH" },
+};
+
+static void
 nv50_pgraph_irq_handler(struct drm_device *dev)
 {
+       struct nouveau_pgraph_trap trap;
+       int unhandled = 0;
        uint32_t status;
 
        while ((status = nv_rd32(dev, NV03_PGRAPH_INTR))) {
-               uint32_t nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
-
+               /* NOTIFY: You've set a NOTIFY an a command and it's done. */
                if (status & 0x00000001) {
-                       nouveau_pgraph_intr_notify(dev, nsource);
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_NOTIFY", &trap);
                        status &= ~0x00000001;
                        nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000001);
                }
 
-               if (status & 0x00000010) {
-                       nouveau_pgraph_intr_error(dev, nsource |
-                                       NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD);
+               /* COMPUTE_QUERY: Purpose and exact cause unknown, happens
+                * when you write 0x200 to 0x50c0 method 0x31c. */
+               if (status & 0x00000002) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_COMPUTE_QUERY", &trap);
+                       status &= ~0x00000002;
+                       nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000002);
+               }
 
+               /* Unknown, never seen: 0x4 */
+
+               /* ILLEGAL_MTHD: You used a wrong method for this class. */
+               if (status & 0x00000010) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_pgraph_intr_swmthd(dev, &trap))
+                               unhandled = 1;
+                       if (unhandled && nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_ILLEGAL_MTHD", &trap);
                        status &= ~0x00000010;
                        nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000010);
                }
 
+               /* ILLEGAL_CLASS: You used a wrong class. */
+               if (status & 0x00000020) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_ILLEGAL_CLASS", &trap);
+                       status &= ~0x00000020;
+                       nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000020);
+               }
+
+               /* DOUBLE_NOTIFY: You tried to set a NOTIFY on another NOTIFY. */
+               if (status & 0x00000040) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_DOUBLE_NOTIFY", &trap);
+                       status &= ~0x00000040;
+                       nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000040);
+               }
+
+               /* CONTEXT_SWITCH: PGRAPH needs us to load a new context */
                if (status & 0x00001000) {
                        nv_wr32(dev, 0x400500, 0x00000000);
                        nv_wr32(dev, NV03_PGRAPH_INTR,
@@ -613,49 +1115,59 @@ nv50_pgraph_irq_handler(struct drm_device *dev)
                        status &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
                }
 
-               if (status & 0x00100000) {
-                       nouveau_pgraph_intr_error(dev, nsource |
-                                       NV03_PGRAPH_NSOURCE_DATA_ERROR);
+               /* BUFFER_NOTIFY: Your m2mf transfer finished */
+               if (status & 0x00010000) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_BUFFER_NOTIFY", &trap);
+                       status &= ~0x00010000;
+                       nv_wr32(dev, NV03_PGRAPH_INTR, 0x00010000);
+               }
 
+               /* DATA_ERROR: Invalid value for this method, or invalid
+                * state in current PGRAPH context for this operation */
+               if (status & 0x00100000) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit()) {
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_DATA_ERROR", &trap);
+                               NV_INFO (dev, "PGRAPH_DATA_ERROR - ");
+                               nouveau_print_enum_names(nv_rd32(dev, 0x400110),
+                                               nv50_data_error_names);
+                               printk("\n");
+                       }
                        status &= ~0x00100000;
                        nv_wr32(dev, NV03_PGRAPH_INTR, 0x00100000);
                }
 
+               /* TRAP: Something bad happened in the middle of command
+                * execution.  Has a billion types, subtypes, and even
+                * subsubtypes. */
                if (status & 0x00200000) {
-                       int r;
-
-                       nouveau_pgraph_intr_error(dev, nsource |
-                                       NV03_PGRAPH_NSOURCE_PROTECTION_ERROR);
-
-                       NV_ERROR(dev, "magic set 1:\n");
-                       for (r = 0x408900; r <= 0x408910; r += 4)
-                               NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
-                                       nv_rd32(dev, r));
-                       nv_wr32(dev, 0x408900,
-                               nv_rd32(dev, 0x408904) | 0xc0000000);
-                       for (r = 0x408e08; r <= 0x408e24; r += 4)
-                               NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
-                                                       nv_rd32(dev, r));
-                       nv_wr32(dev, 0x408e08,
-                               nv_rd32(dev, 0x408e08) | 0xc0000000);
-
-                       NV_ERROR(dev, "magic set 2:\n");
-                       for (r = 0x409900; r <= 0x409910; r += 4)
-                               NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
-                                       nv_rd32(dev, r));
-                       nv_wr32(dev, 0x409900,
-                               nv_rd32(dev, 0x409904) | 0xc0000000);
-                       for (r = 0x409e08; r <= 0x409e24; r += 4)
-                               NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
-                                       nv_rd32(dev, r));
-                       nv_wr32(dev, 0x409e08,
-                               nv_rd32(dev, 0x409e08) | 0xc0000000);
-
+                       nv50_pgraph_trap_handler(dev);
                        status &= ~0x00200000;
-                       nv_wr32(dev, NV03_PGRAPH_NSOURCE, nsource);
                        nv_wr32(dev, NV03_PGRAPH_INTR, 0x00200000);
                }
 
+               /* Unknown, never seen: 0x00400000 */
+
+               /* SINGLE_STEP: Happens on every method if you turned on
+                * single stepping in 40008c */
+               if (status & 0x01000000) {
+                       nouveau_graph_trap_info(dev, &trap);
+                       if (nouveau_ratelimit())
+                               nouveau_graph_dump_trap_info(dev,
+                                               "PGRAPH_SINGLE_STEP", &trap);
+                       status &= ~0x01000000;
+                       nv_wr32(dev, NV03_PGRAPH_INTR, 0x01000000);
+               }
+
+               /* 0x02000000 happens when you pause a ctxprog...
+                * but the only way this can happen that I know is by
+                * poking the relevant MMIO register, and we don't
+                * do that. */
+
                if (status) {
                        NV_INFO(dev, "Unhandled PGRAPH_INTR - 0x%08x\n",
                                status);
@@ -672,7 +1184,8 @@ nv50_pgraph_irq_handler(struct drm_device *dev)
        }
 
        nv_wr32(dev, NV03_PMC_INTR_0, NV_PMC_INTR_0_PGRAPH_PENDING);
-       nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
+       if (nv_rd32(dev, 0x400824) & (1 << 31))
+               nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
 }
 
 static void
index 516a8d3..f4ea3e6 100644 (file)
@@ -34,7 +34,6 @@
 #include "nouveau_drm.h"
 #include "nv50_display.h"
 
-static int nouveau_stub_init(struct drm_device *dev) { return 0; }
 static void nouveau_stub_takedown(struct drm_device *dev) {}
 
 static int nouveau_init_engine_ptrs(struct drm_device *dev)
@@ -276,8 +275,8 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
                engine->timer.init              = nv04_timer_init;
                engine->timer.read              = nv04_timer_read;
                engine->timer.takedown          = nv04_timer_takedown;
-               engine->fb.init                 = nouveau_stub_init;
-               engine->fb.takedown             = nouveau_stub_takedown;
+               engine->fb.init                 = nv50_fb_init;
+               engine->fb.takedown             = nv50_fb_takedown;
                engine->graph.grclass           = nv50_graph_grclass;
                engine->graph.init              = nv50_graph_init;
                engine->graph.takedown          = nv50_graph_takedown;
diff --git a/drivers/gpu/drm/nouveau/nv50_fb.c b/drivers/gpu/drm/nouveau/nv50_fb.c
new file mode 100644 (file)
index 0000000..a95e694
--- /dev/null
@@ -0,0 +1,32 @@
+#include "drmP.h"
+#include "drm.h"
+#include "nouveau_drv.h"
+#include "nouveau_drm.h"
+
+int
+nv50_fb_init(struct drm_device *dev)
+{
+       /* This is needed to get meaningful information from 100c90
+        * on traps. No idea what these values mean exactly. */
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+       switch (dev_priv->chipset) {
+       case 0x50:
+               nv_wr32(dev, 0x100c90, 0x0707ff);
+               break;
+       case 0xa5:
+       case 0xa8:
+               nv_wr32(dev, 0x100c90, 0x0d0fff);
+               break;
+       default:
+               nv_wr32(dev, 0x100c90, 0x1d07ff);
+               break;
+       }
+
+       return 0;
+}
+
+void
+nv50_fb_takedown(struct drm_device *dev)
+{
+}
index 08d87b7..c62b33a 100644 (file)
@@ -56,6 +56,10 @@ nv50_graph_init_intr(struct drm_device *dev)
 static void
 nv50_graph_init_regs__nv(struct drm_device *dev)
 {
+       struct drm_nouveau_private *dev_priv = dev->dev_private;
+       uint32_t units = nv_rd32(dev, 0x1540);
+       int i;
+
        NV_DEBUG(dev, "\n");
 
        nv_wr32(dev, 0x400804, 0xc0000000);
@@ -65,6 +69,20 @@ nv50_graph_init_regs__nv(struct drm_device *dev)
        nv_wr32(dev, 0x405018, 0xc0000000);
        nv_wr32(dev, 0x402000, 0xc0000000);
 
+       for (i = 0; i < 16; i++) {
+               if (units & 1 << i) {
+                       if (dev_priv->chipset < 0xa0) {
+                               nv_wr32(dev, 0x408900 + (i << 12), 0xc0000000);
+                               nv_wr32(dev, 0x408e08 + (i << 12), 0xc0000000);
+                               nv_wr32(dev, 0x408314 + (i << 12), 0xc0000000);
+                       } else {
+                               nv_wr32(dev, 0x408600 + (i << 11), 0xc0000000);
+                               nv_wr32(dev, 0x408708 + (i << 11), 0xc0000000);
+                               nv_wr32(dev, 0x40831c + (i << 11), 0xc0000000);
+                       }
+               }
+       }
+
        nv_wr32(dev, 0x400108, 0xffffffff);
 
        nv_wr32(dev, 0x400824, 0x00004000);
index 9f909ab..546b319 100644 (file)
@@ -274,7 +274,7 @@ nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
        int offset, base;
        uint32_t units = nv_rd32 (ctx->dev, 0x1540);
 
-       /* 0800 */
+       /* 0800: DISPATCH */
        cp_ctx(ctx, 0x400808, 7);
        gr_def(ctx, 0x400814, 0x00000030);
        cp_ctx(ctx, 0x400834, 0x32);
@@ -305,7 +305,7 @@ nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
                gr_def(ctx, 0x400b20, 0x0001629d);
        }
 
-       /* 0C00 */
+       /* 0C00: VFETCH */
        cp_ctx(ctx, 0x400c08, 0x2);
        gr_def(ctx, 0x400c08, 0x0000fe0c);
 
@@ -331,7 +331,7 @@ nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
        cp_ctx(ctx, 0x401540, 0x5);
        gr_def(ctx, 0x401550, 0x00001018);
 
-       /* 1800 */
+       /* 1800: STREAMOUT */
        cp_ctx(ctx, 0x401814, 0x1);
        gr_def(ctx, 0x401814, 0x000000ff);
        if (dev_priv->chipset == 0x50) {
@@ -646,7 +646,7 @@ nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
        if (dev_priv->chipset == 0x50)
                cp_ctx(ctx, 0x4063e0, 0x1);
 
-       /* 6800 */
+       /* 6800: M2MF */
        if (dev_priv->chipset < 0x90) {
                cp_ctx(ctx, 0x406814, 0x2b);
                gr_def(ctx, 0x406818, 0x00000f80);