diff options
Diffstat (limited to 'drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c')
| -rw-r--r-- | drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c | 877 |
1 files changed, 336 insertions, 541 deletions
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c index 270d7cd63fc7..b19b912d5787 100644 --- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c +++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/nv50.c @@ -24,27 +24,13 @@ #include "nv50.h" #include <core/client.h> -#include <core/device.h> -#include <core/handle.h> +#include <core/gpuobj.h> #include <engine/fifo.h> -#include <subdev/timer.h> -struct nv50_gr_priv { - struct nvkm_gr base; - spinlock_t lock; - u32 size; -}; - -struct nv50_gr_chan { - struct nvkm_gr_chan base; -}; - -static u64 +u64 nv50_gr_units(struct nvkm_gr *gr) { - struct nv50_gr_priv *priv = (void *)gr; - - return nv_rd32(priv, 0x1540); + return nvkm_rd32(gr->engine.subdev.device, 0x1540); } /******************************************************************************* @@ -52,86 +38,25 @@ nv50_gr_units(struct nvkm_gr *gr) ******************************************************************************/ static int -nv50_gr_object_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv50_gr_object_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent, + int align, struct nvkm_gpuobj **pgpuobj) { - struct nvkm_gpuobj *obj; - int ret; - - ret = nvkm_gpuobj_create(parent, engine, oclass, 0, parent, - 16, 16, 0, &obj); - *pobject = nv_object(obj); - if (ret) - return ret; - - nv_wo32(obj, 0x00, nv_mclass(obj)); - nv_wo32(obj, 0x04, 0x00000000); - nv_wo32(obj, 0x08, 0x00000000); - nv_wo32(obj, 0x0c, 0x00000000); - return 0; + int ret = nvkm_gpuobj_new(object->engine->subdev.device, 16, + align, false, parent, pgpuobj); + if (ret == 0) { + nvkm_kmap(*pgpuobj); + nvkm_wo32(*pgpuobj, 0x00, object->oclass); + nvkm_wo32(*pgpuobj, 0x04, 0x00000000); + nvkm_wo32(*pgpuobj, 0x08, 0x00000000); + nvkm_wo32(*pgpuobj, 0x0c, 0x00000000); + nvkm_done(*pgpuobj); + } + return ret; } -static struct nvkm_ofuncs -nv50_gr_ofuncs = { - .ctor = nv50_gr_object_ctor, - .dtor = _nvkm_gpuobj_dtor, - .init = _nvkm_gpuobj_init, - .fini = _nvkm_gpuobj_fini, - .rd32 = _nvkm_gpuobj_rd32, - .wr32 = _nvkm_gpuobj_wr32, -}; - -static struct nvkm_oclass -nv50_gr_sclass[] = { - { 0x0030, &nv50_gr_ofuncs }, - { 0x502d, &nv50_gr_ofuncs }, - { 0x5039, &nv50_gr_ofuncs }, - { 0x5097, &nv50_gr_ofuncs }, - { 0x50c0, &nv50_gr_ofuncs }, - {} -}; - -static struct nvkm_oclass -g84_gr_sclass[] = { - { 0x0030, &nv50_gr_ofuncs }, - { 0x502d, &nv50_gr_ofuncs }, - { 0x5039, &nv50_gr_ofuncs }, - { 0x50c0, &nv50_gr_ofuncs }, - { 0x8297, &nv50_gr_ofuncs }, - {} -}; - -static struct nvkm_oclass -gt200_gr_sclass[] = { - { 0x0030, &nv50_gr_ofuncs }, - { 0x502d, &nv50_gr_ofuncs }, - { 0x5039, &nv50_gr_ofuncs }, - { 0x50c0, &nv50_gr_ofuncs }, - { 0x8397, &nv50_gr_ofuncs }, - {} -}; - -static struct nvkm_oclass -gt215_gr_sclass[] = { - { 0x0030, &nv50_gr_ofuncs }, - { 0x502d, &nv50_gr_ofuncs }, - { 0x5039, &nv50_gr_ofuncs }, - { 0x50c0, &nv50_gr_ofuncs }, - { 0x8597, &nv50_gr_ofuncs }, - { 0x85c0, &nv50_gr_ofuncs }, - {} -}; - -static struct nvkm_oclass -mcp89_gr_sclass[] = { - { 0x0030, &nv50_gr_ofuncs }, - { 0x502d, &nv50_gr_ofuncs }, - { 0x5039, &nv50_gr_ofuncs }, - { 0x50c0, &nv50_gr_ofuncs }, - { 0x85c0, &nv50_gr_ofuncs }, - { 0x8697, &nv50_gr_ofuncs }, - {} +const struct nvkm_object_func +nv50_gr_object = { + .bind = nv50_gr_object_bind, }; /******************************************************************************* @@ -139,160 +64,43 @@ mcp89_gr_sclass[] = { ******************************************************************************/ static int -nv50_gr_context_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +nv50_gr_chan_bind(struct nvkm_object *object, struct nvkm_gpuobj *parent, + int align, struct nvkm_gpuobj **pgpuobj) { - struct nv50_gr_priv *priv = (void *)engine; - struct nv50_gr_chan *chan; - int ret; - - ret = nvkm_gr_context_create(parent, engine, oclass, NULL, priv->size, - 0, NVOBJ_FLAG_ZERO_ALLOC, &chan); - *pobject = nv_object(chan); - if (ret) - return ret; - - nv50_grctx_fill(nv_device(priv), nv_gpuobj(chan)); - return 0; + struct nv50_gr *gr = nv50_gr_chan(object)->gr; + int ret = nvkm_gpuobj_new(gr->base.engine.subdev.device, gr->size, + align, true, parent, pgpuobj); + if (ret == 0) { + nvkm_kmap(*pgpuobj); + nv50_grctx_fill(gr->base.engine.subdev.device, *pgpuobj); + nvkm_done(*pgpuobj); + } + return ret; } -static struct nvkm_oclass -nv50_gr_cclass = { - .handle = NV_ENGCTX(GR, 0x50), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv50_gr_context_ctor, - .dtor = _nvkm_gr_context_dtor, - .init = _nvkm_gr_context_init, - .fini = _nvkm_gr_context_fini, - .rd32 = _nvkm_gr_context_rd32, - .wr32 = _nvkm_gr_context_wr32, - }, -}; - -/******************************************************************************* - * PGRAPH engine/subdev functions - ******************************************************************************/ - -static const struct nvkm_bitfield nv50_pgr_status[] = { - { 0x00000001, "BUSY" }, /* set when any bit is set */ - { 0x00000002, "DISPATCH" }, - { 0x00000004, "UNK2" }, - { 0x00000008, "UNK3" }, - { 0x00000010, "UNK4" }, - { 0x00000020, "UNK5" }, - { 0x00000040, "M2MF" }, - { 0x00000080, "UNK7" }, - { 0x00000100, "CTXPROG" }, - { 0x00000200, "VFETCH" }, - { 0x00000400, "CCACHE_PREGEOM" }, - { 0x00000800, "STRMOUT_VATTR_POSTGEOM" }, - { 0x00001000, "VCLIP" }, - { 0x00002000, "RATTR_APLANE" }, - { 0x00004000, "TRAST" }, - { 0x00008000, "CLIPID" }, - { 0x00010000, "ZCULL" }, - { 0x00020000, "ENG2D" }, - { 0x00040000, "RMASK" }, - { 0x00080000, "TPC_RAST" }, - { 0x00100000, "TPC_PROP" }, - { 0x00200000, "TPC_TEX" }, - { 0x00400000, "TPC_GEOM" }, - { 0x00800000, "TPC_MP" }, - { 0x01000000, "ROP" }, - {} -}; - -static const char *const nv50_pgr_vstatus_0[] = { - "VFETCH", "CCACHE", "PREGEOM", "POSTGEOM", "VATTR", "STRMOUT", "VCLIP", - NULL -}; - -static const char *const nv50_pgr_vstatus_1[] = { - "TPC_RAST", "TPC_PROP", "TPC_TEX", "TPC_GEOM", "TPC_MP", NULL -}; - -static const char *const nv50_pgr_vstatus_2[] = { - "RATTR", "APLANE", "TRAST", "CLIPID", "ZCULL", "ENG2D", "RMASK", - "ROP", NULL +static const struct nvkm_object_func +nv50_gr_chan = { + .bind = nv50_gr_chan_bind, }; -static void -nvkm_pgr_vstatus_print(struct nv50_gr_priv *priv, int r, - const char *const units[], u32 status) +int +nv50_gr_chan_new(struct nvkm_gr *base, struct nvkm_fifo_chan *fifoch, + const struct nvkm_oclass *oclass, struct nvkm_object **pobject) { - int i; - - nv_error(priv, "PGRAPH_VSTATUS%d: 0x%08x", r, status); + struct nv50_gr *gr = nv50_gr(base); + struct nv50_gr_chan *chan; - for (i = 0; units[i] && status; i++) { - if ((status & 7) == 1) - pr_cont(" %s", units[i]); - status >>= 3; - } - if (status) - pr_cont(" (invalid: 0x%x)", status); - pr_cont("\n"); + if (!(chan = kzalloc(sizeof(*chan), GFP_KERNEL))) + return -ENOMEM; + nvkm_object_ctor(&nv50_gr_chan, oclass, &chan->object); + chan->gr = gr; + *pobject = &chan->object; + return 0; } -static int -g84_gr_tlb_flush(struct nvkm_engine *engine) -{ - struct nvkm_timer *ptimer = nvkm_timer(engine); - struct nv50_gr_priv *priv = (void *)engine; - bool idle, timeout = false; - unsigned long flags; - u64 start; - u32 tmp; - - spin_lock_irqsave(&priv->lock, flags); - nv_mask(priv, 0x400500, 0x00000001, 0x00000000); - - start = ptimer->read(ptimer); - do { - idle = true; - - for (tmp = nv_rd32(priv, 0x400380); tmp && idle; tmp >>= 3) { - if ((tmp & 7) == 1) - idle = false; - } - - for (tmp = nv_rd32(priv, 0x400384); tmp && idle; tmp >>= 3) { - if ((tmp & 7) == 1) - idle = false; - } - - for (tmp = nv_rd32(priv, 0x400388); tmp && idle; tmp >>= 3) { - if ((tmp & 7) == 1) - idle = false; - } - } while (!idle && - !(timeout = ptimer->read(ptimer) - start > 2000000000)); - - if (timeout) { - nv_error(priv, "PGRAPH TLB flush idle timeout fail\n"); - - tmp = nv_rd32(priv, 0x400700); - nv_error(priv, "PGRAPH_STATUS : 0x%08x", tmp); - nvkm_bitfield_print(nv50_pgr_status, tmp); - pr_cont("\n"); - - nvkm_pgr_vstatus_print(priv, 0, nv50_pgr_vstatus_0, - nv_rd32(priv, 0x400380)); - nvkm_pgr_vstatus_print(priv, 1, nv50_pgr_vstatus_1, - nv_rd32(priv, 0x400384)); - nvkm_pgr_vstatus_print(priv, 2, nv50_pgr_vstatus_2, - nv_rd32(priv, 0x400388)); - } - - - nv_wr32(priv, 0x100c80, 0x00000001); - if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000)) - nv_error(priv, "vm flush timeout\n"); - nv_mask(priv, 0x400500, 0x00000001, 0x00000001); - spin_unlock_irqrestore(&priv->lock, flags); - return timeout ? -EBUSY : 0; -} +/******************************************************************************* + * PGRAPH engine/subdev functions + ******************************************************************************/ static const struct nvkm_bitfield nv50_mp_exec_errors[] = { { 0x01, "STACK_UNDERFLOW" }, @@ -427,157 +235,172 @@ static const struct nvkm_bitfield nv50_gr_trap_prop[] = { }; static void -nv50_priv_prop_trap(struct nv50_gr_priv *priv, - u32 ustatus_addr, u32 ustatus, u32 tp) +nv50_gr_prop_trap(struct nv50_gr *gr, u32 ustatus_addr, u32 ustatus, u32 tp) { - u32 e0c = nv_rd32(priv, ustatus_addr + 0x04); - u32 e10 = nv_rd32(priv, ustatus_addr + 0x08); - u32 e14 = nv_rd32(priv, ustatus_addr + 0x0c); - u32 e18 = nv_rd32(priv, ustatus_addr + 0x10); - u32 e1c = nv_rd32(priv, ustatus_addr + 0x14); - u32 e20 = nv_rd32(priv, ustatus_addr + 0x18); - u32 e24 = nv_rd32(priv, ustatus_addr + 0x1c); + struct nvkm_subdev *subdev = &gr->base.engine.subdev; + struct nvkm_device *device = subdev->device; + u32 e0c = nvkm_rd32(device, ustatus_addr + 0x04); + u32 e10 = nvkm_rd32(device, ustatus_addr + 0x08); + u32 e14 = nvkm_rd32(device, ustatus_addr + 0x0c); + u32 e18 = nvkm_rd32(device, ustatus_addr + 0x10); + u32 e1c = nvkm_rd32(device, ustatus_addr + 0x14); + u32 e20 = nvkm_rd32(device, ustatus_addr + 0x18); + u32 e24 = nvkm_rd32(device, ustatus_addr + 0x1c); + char msg[128]; /* CUDA memory: l[], g[] or stack. */ if (ustatus & 0x00000080) { if (e18 & 0x80000000) { /* g[] read fault? */ - nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n", + nvkm_error(subdev, "TRAP_PROP - TP %d - CUDA_FAULT - Global read fault at address %02x%08x\n", tp, e14, e10 | ((e18 >> 24) & 0x1f)); e18 &= ~0x1f000000; } else if (e18 & 0xc) { /* g[] write fault? */ - nv_error(priv, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n", + nvkm_error(subdev, "TRAP_PROP - TP %d - CUDA_FAULT - Global write fault at address %02x%08x\n", tp, e14, e10 | ((e18 >> 7) & 0x1f)); e18 &= ~0x00000f80; } else { - nv_error(priv, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n", + nvkm_error(subdev, "TRAP_PROP - TP %d - Unknown CUDA fault at address %02x%08x\n", tp, e14, e10); } ustatus &= ~0x00000080; } if (ustatus) { - nv_error(priv, "TRAP_PROP - TP %d -", tp); - nvkm_bitfield_print(nv50_gr_trap_prop, ustatus); - pr_cont(" - Address %02x%08x\n", e14, e10); + nvkm_snprintbf(msg, sizeof(msg), nv50_gr_trap_prop, ustatus); + nvkm_error(subdev, "TRAP_PROP - TP %d - %08x [%s] - " + "Address %02x%08x\n", + tp, ustatus, msg, e14, e10); } - nv_error(priv, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n", + nvkm_error(subdev, "TRAP_PROP - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n", tp, e0c, e18, e1c, e20, e24); } static void -nv50_priv_mp_trap(struct nv50_gr_priv *priv, int tpid, int display) +nv50_gr_mp_trap(struct nv50_gr *gr, int tpid, int display) { - u32 units = nv_rd32(priv, 0x1540); + struct nvkm_subdev *subdev = &gr->base.engine.subdev; + struct nvkm_device *device = subdev->device; + u32 units = nvkm_rd32(device, 0x1540); u32 addr, mp10, status, pc, oplow, ophigh; + char msg[128]; int i; int mps = 0; for (i = 0; i < 4; i++) { if (!(units & 1 << (i+24))) continue; - if (nv_device(priv)->chipset < 0xa0) + if (device->chipset < 0xa0) addr = 0x408200 + (tpid << 12) + (i << 7); else addr = 0x408100 + (tpid << 11) + (i << 7); - mp10 = nv_rd32(priv, addr + 0x10); - status = nv_rd32(priv, addr + 0x14); + mp10 = nvkm_rd32(device, addr + 0x10); + status = nvkm_rd32(device, addr + 0x14); if (!status) continue; if (display) { - nv_rd32(priv, addr + 0x20); - pc = nv_rd32(priv, addr + 0x24); - oplow = nv_rd32(priv, addr + 0x70); - ophigh = nv_rd32(priv, addr + 0x74); - nv_error(priv, "TRAP_MP_EXEC - " - "TP %d MP %d:", tpid, i); - nvkm_bitfield_print(nv50_mp_exec_errors, status); - pr_cont(" at %06x warp %d, opcode %08x %08x\n", - pc&0xffffff, pc >> 24, - oplow, ophigh); + nvkm_rd32(device, addr + 0x20); + pc = nvkm_rd32(device, addr + 0x24); + oplow = nvkm_rd32(device, addr + 0x70); + ophigh = nvkm_rd32(device, addr + 0x74); + nvkm_snprintbf(msg, sizeof(msg), + nv50_mp_exec_errors, status); + nvkm_error(subdev, "TRAP_MP_EXEC - TP %d MP %d: " + "%08x [%s] at %06x warp %d, " + "opcode %08x %08x\n", + tpid, i, status, msg, pc & 0xffffff, + pc >> 24, oplow, ophigh); } - nv_wr32(priv, addr + 0x10, mp10); - nv_wr32(priv, addr + 0x14, 0); + nvkm_wr32(device, addr + 0x10, mp10); + nvkm_wr32(device, addr + 0x14, 0); mps++; } if (!mps && display) - nv_error(priv, "TRAP_MP_EXEC - TP %d: " + nvkm_error(subdev, "TRAP_MP_EXEC - TP %d: " "No MPs claiming errors?\n", tpid); } static void -nv50_priv_tp_trap(struct nv50_gr_priv *priv, int type, u32 ustatus_old, +nv50_gr_tp_trap(struct nv50_gr *gr, int type, u32 ustatus_old, u32 ustatus_new, int display, const char *name) { + struct nvkm_subdev *subdev = &gr->base.engine.subdev; + struct nvkm_device *device = subdev->device; + u32 units = nvkm_rd32(device, 0x1540); int tps = 0; - u32 units = nv_rd32(priv, 0x1540); int i, r; + char msg[128]; u32 ustatus_addr, ustatus; for (i = 0; i < 16; i++) { if (!(units & (1 << i))) continue; - if (nv_device(priv)->chipset < 0xa0) + if (device->chipset < 0xa0) ustatus_addr = ustatus_old + (i << 12); else ustatus_addr = ustatus_new + (i << 11); - ustatus = nv_rd32(priv, ustatus_addr) & 0x7fffffff; + ustatus = nvkm_rd32(device, ustatus_addr) & 0x7fffffff; if (!ustatus) continue; tps++; switch (type) { case 6: /* texture error... unknown for now */ if (display) { - nv_error(priv, "magic set %d:\n", i); + nvkm_error(subdev, "magic set %d:\n", i); for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4) - nv_error(priv, "\t0x%08x: 0x%08x\n", r, - nv_rd32(priv, r)); + nvkm_error(subdev, "\t%08x: %08x\n", r, + nvkm_rd32(device, r)); if (ustatus) { - nv_error(priv, "%s - TP%d:", name, i); - nvkm_bitfield_print(nv50_tex_traps, - ustatus); - pr_cont("\n"); + nvkm_snprintbf(msg, sizeof(msg), + nv50_tex_traps, ustatus); + nvkm_error(subdev, + "%s - TP%d: %08x [%s]\n", + name, i, ustatus, msg); ustatus = 0; } } break; case 7: /* MP error */ if (ustatus & 0x04030000) { - nv50_priv_mp_trap(priv, i, display); + nv50_gr_mp_trap(gr, i, display); ustatus &= ~0x04030000; } if (ustatus && display) { - nv_error(priv, "%s - TP%d:", name, i); - nvkm_bitfield_print(nv50_mpc_traps, ustatus); - pr_cont("\n"); + nvkm_snprintbf(msg, sizeof(msg), + nv50_mpc_traps, ustatus); + nvkm_error(subdev, "%s - TP%d: %08x [%s]\n", + name, i, ustatus, msg); ustatus = 0; } break; case 8: /* PROP error */ if (display) - nv50_priv_prop_trap( - priv, ustatus_addr, ustatus, i); + nv50_gr_prop_trap( + gr, ustatus_addr, ustatus, i); ustatus = 0; break; } if (ustatus) { if (display) - nv_error(priv, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus); + nvkm_error(subdev, "%s - TP%d: Unhandled ustatus %08x\n", name, i, ustatus); } - nv_wr32(priv, ustatus_addr, 0xc0000000); + nvkm_wr32(device, ustatus_addr, 0xc0000000); } if (!tps && display) - nv_warn(priv, "%s - No TPs claiming errors?\n", name); + nvkm_warn(subdev, "%s - No TPs claiming errors?\n", name); } static int -nv50_gr_trap_handler(struct nv50_gr_priv *priv, u32 display, - int chid, u64 inst, struct nvkm_object *engctx) +nv50_gr_trap_handler(struct nv50_gr *gr, u32 display, + int chid, u64 inst, const char *name) { - u32 status = nv_rd32(priv, 0x400108); + struct nvkm_subdev *subdev = &gr->base.engine.subdev; + struct nvkm_device *device = subdev->device; + u32 status = nvkm_rd32(device, 0x400108); u32 ustatus; + char msg[128]; if (!status && display) { - nv_error(priv, "TRAP: no units reporting traps?\n"); + nvkm_error(subdev, "TRAP: no units reporting traps?\n"); return 1; } @@ -585,71 +408,72 @@ nv50_gr_trap_handler(struct nv50_gr_priv *priv, u32 display, * 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(priv, 0x400804) & 0x7fffffff; + ustatus = nvkm_rd32(device, 0x400804) & 0x7fffffff; if (!ustatus && display) { - nv_error(priv, "TRAP_DISPATCH - no ustatus?\n"); + nvkm_error(subdev, "TRAP_DISPATCH - no ustatus?\n"); } - nv_wr32(priv, 0x400500, 0x00000000); + nvkm_wr32(device, 0x400500, 0x00000000); /* Known to be triggered by screwed up NOTIFY and COND... */ if (ustatus & 0x00000001) { - u32 addr = nv_rd32(priv, 0x400808); + u32 addr = nvkm_rd32(device, 0x400808); u32 subc = (addr & 0x00070000) >> 16; u32 mthd = (addr & 0x00001ffc); - u32 datal = nv_rd32(priv, 0x40080c); - u32 datah = nv_rd32(priv, 0x400810); - u32 class = nv_rd32(priv, 0x400814); - u32 r848 = nv_rd32(priv, 0x400848); + u32 datal = nvkm_rd32(device, 0x40080c); + u32 datah = nvkm_rd32(device, 0x400810); + u32 class = nvkm_rd32(device, 0x400814); + u32 r848 = nvkm_rd32(device, 0x400848); - nv_error(priv, "TRAP DISPATCH_FAULT\n"); + nvkm_error(subdev, "TRAP DISPATCH_FAULT\n"); if (display && (addr & 0x80000000)) { - nv_error(priv, - "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x%08x 400808 0x%08x 400848 0x%08x\n", - chid, inst, - nvkm_client_name(engctx), subc, - class, mthd, datah, datal, addr, r848); + nvkm_error(subdev, + "ch %d [%010llx %s] subc %d " + "class %04x mthd %04x data %08x%08x " + "400808 %08x 400848 %08x\n", + chid, inst, name, subc, class, mthd, + datah, datal, addr, r848); } else if (display) { - nv_error(priv, "no stuck command?\n"); + nvkm_error(subdev, "no stuck command?\n"); } - nv_wr32(priv, 0x400808, 0); - nv_wr32(priv, 0x4008e8, nv_rd32(priv, 0x4008e8) & 3); - nv_wr32(priv, 0x400848, 0); + nvkm_wr32(device, 0x400808, 0); + nvkm_wr32(device, 0x4008e8, nvkm_rd32(device, 0x4008e8) & 3); + nvkm_wr32(device, 0x400848, 0); ustatus &= ~0x00000001; } if (ustatus & 0x00000002) { - u32 addr = nv_rd32(priv, 0x40084c); + u32 addr = nvkm_rd32(device, 0x40084c); u32 subc = (addr & 0x00070000) >> 16; u32 mthd = (addr & 0x00001ffc); - u32 data = nv_rd32(priv, 0x40085c); - u32 class = nv_rd32(priv, 0x400814); + u32 data = nvkm_rd32(device, 0x40085c); + u32 class = nvkm_rd32(device, 0x400814); - nv_error(priv, "TRAP DISPATCH_QUERY\n"); + nvkm_error(subdev, "TRAP DISPATCH_QUERY\n"); if (display && (addr & 0x80000000)) { - nv_error(priv, - "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x 40084c 0x%08x\n", - chid, inst, - nvkm_client_name(engctx), subc, - class, mthd, data, addr); + nvkm_error(subdev, + "ch %d [%010llx %s] subc %d " + "class %04x mthd %04x data %08x " + "40084c %08x\n", chid, inst, name, + subc, class, mthd, data, addr); } else if (display) { - nv_error(priv, "no stuck command?\n"); + nvkm_error(subdev, "no stuck command?\n"); } - nv_wr32(priv, 0x40084c, 0); + nvkm_wr32(device, 0x40084c, 0); ustatus &= ~0x00000002; } if (ustatus && display) { - nv_error(priv, "TRAP_DISPATCH (unknown " - "0x%08x)\n", ustatus); + nvkm_error(subdev, "TRAP_DISPATCH " + "(unknown %08x)\n", ustatus); } - nv_wr32(priv, 0x400804, 0xc0000000); - nv_wr32(priv, 0x400108, 0x001); + nvkm_wr32(device, 0x400804, 0xc0000000); + nvkm_wr32(device, 0x400108, 0x001); status &= ~0x001; if (!status) return 0; @@ -657,81 +481,91 @@ nv50_gr_trap_handler(struct nv50_gr_priv *priv, u32 display, /* M2MF: Memory to memory copy engine. */ if (status & 0x002) { - u32 ustatus = nv_rd32(priv, 0x406800) & 0x7fffffff; + u32 ustatus = nvkm_rd32(device, 0x406800) & 0x7fffffff; if (display) { - nv_error(priv, "TRAP_M2MF"); - nvkm_bitfield_print(nv50_gr_trap_m2mf, ustatus); - pr_cont("\n"); - nv_error(priv, "TRAP_M2MF %08x %08x %08x %08x\n", - nv_rd32(priv, 0x406804), nv_rd32(priv, 0x406808), - nv_rd32(priv, 0x40680c), nv_rd32(priv, 0x406810)); - + nvkm_snprintbf(msg, sizeof(msg), + nv50_gr_trap_m2mf, ustatus); + nvkm_error(subdev, "TRAP_M2MF %08x [%s]\n", + ustatus, msg); + nvkm_error(subdev, "TRAP_M2MF %08x %08x %08x %08x\n", + nvkm_rd32(device, 0x406804), + nvkm_rd32(device, 0x406808), + nvkm_rd32(device, 0x40680c), + nvkm_rd32(device, 0x406810)); } /* No sane way found yet -- just reset the bugger. */ - nv_wr32(priv, 0x400040, 2); - nv_wr32(priv, 0x400040, 0); - nv_wr32(priv, 0x406800, 0xc0000000); - nv_wr32(priv, 0x400108, 0x002); + nvkm_wr32(device, 0x400040, 2); + nvkm_wr32(device, 0x400040, 0); + nvkm_wr32(device, 0x406800, 0xc0000000); + nvkm_wr32(device, 0x400108, 0x002); status &= ~0x002; } /* VFETCH: Fetches data from vertex buffers. */ if (status & 0x004) { - u32 ustatus = nv_rd32(priv, 0x400c04) & 0x7fffffff; + u32 ustatus = nvkm_rd32(device, 0x400c04) & 0x7fffffff; if (display) { - nv_error(priv, "TRAP_VFETCH"); - nvkm_bitfield_print(nv50_gr_trap_vfetch, ustatus); - pr_cont("\n"); - nv_error(priv, "TRAP_VFETCH %08x %08x %08x %08x\n", - nv_rd32(priv, 0x400c00), nv_rd32(priv, 0x400c08), - nv_rd32(priv, 0x400c0c), nv_rd32(priv, 0x400c10)); + nvkm_snprintbf(msg, sizeof(msg), + nv50_gr_trap_vfetch, ustatus); + nvkm_error(subdev, "TRAP_VFETCH %08x [%s]\n", + ustatus, msg); + nvkm_error(subdev, "TRAP_VFETCH %08x %08x %08x %08x\n", + nvkm_rd32(device, 0x400c00), + nvkm_rd32(device, 0x400c08), + nvkm_rd32(device, 0x400c0c), + nvkm_rd32(device, 0x400c10)); } - nv_wr32(priv, 0x400c04, 0xc0000000); - nv_wr32(priv, 0x400108, 0x004); + nvkm_wr32(device, 0x400c04, 0xc0000000); + nvkm_wr32(device, 0x400108, 0x004); status &= ~0x004; } /* STRMOUT: DirectX streamout / OpenGL transform feedback. */ if (status & 0x008) { - ustatus = nv_rd32(priv, 0x401800) & 0x7fffffff; + ustatus = nvkm_rd32(device, 0x401800) & 0x7fffffff; if (display) { - nv_error(priv, "TRAP_STRMOUT"); - nvkm_bitfield_print(nv50_gr_trap_strmout, ustatus); - pr_cont("\n"); - nv_error(priv, "TRAP_STRMOUT %08x %08x %08x %08x\n", - nv_rd32(priv, 0x401804), nv_rd32(priv, 0x401808), - nv_rd32(priv, 0x40180c), nv_rd32(priv, 0x401810)); - + nvkm_snprintbf(msg, sizeof(msg), + nv50_gr_trap_strmout, ustatus); + nvkm_error(subdev, "TRAP_STRMOUT %08x [%s]\n", + ustatus, msg); + nvkm_error(subdev, "TRAP_STRMOUT %08x %08x %08x %08x\n", + nvkm_rd32(device, 0x401804), + nvkm_rd32(device, 0x401808), + nvkm_rd32(device, 0x40180c), + nvkm_rd32(device, 0x401810)); } /* No sane way found yet -- just reset the bugger. */ - nv_wr32(priv, 0x400040, 0x80); - nv_wr32(priv, 0x400040, 0); - nv_wr32(priv, 0x401800, 0xc0000000); - nv_wr32(priv, 0x400108, 0x008); + nvkm_wr32(device, 0x400040, 0x80); + nvkm_wr32(device, 0x400040, 0); + nvkm_wr32(device, 0x401800, 0xc0000000); + nvkm_wr32(device, 0x400108, 0x008); status &= ~0x008; } /* CCACHE: Handles code and c[] caches and fills them. */ if (status & 0x010) { - ustatus = nv_rd32(priv, 0x405018) & 0x7fffffff; + ustatus = nvkm_rd32(device, 0x405018) & 0x7fffffff; if (display) { - nv_error(priv, "TRAP_CCACHE"); - nvkm_bitfield_print(nv50_gr_trap_ccache, ustatus); - pr_cont("\n"); - nv_error(priv, "TRAP_CCACHE %08x %08x %08x %08x" - " %08x %08x %08x\n", - nv_rd32(priv, 0x405000), nv_rd32(priv, 0x405004), - nv_rd32(priv, 0x405008), nv_rd32(priv, 0x40500c), - nv_rd32(priv, 0x405010), nv_rd32(priv, 0x405014), - nv_rd32(priv, 0x40501c)); - + nvkm_snprintbf(msg, sizeof(msg), + nv50_gr_trap_ccache, ustatus); + nvkm_error(subdev, "TRAP_CCACHE %08x [%s]\n", + ustatus, msg); + nvkm_error(subdev, "TRAP_CCACHE %08x %08x %08x %08x " + "%08x %08x %08x\n", + nvkm_rd32(device, 0x405000), + nvkm_rd32(device, 0x405004), + nvkm_rd32(device, 0x405008), + nvkm_rd32(device, 0x40500c), + nvkm_rd32(device, 0x405010), + nvkm_rd32(device, 0x405014), + nvkm_rd32(device, 0x40501c)); } - nv_wr32(priv, 0x405018, 0xc0000000); - nv_wr32(priv, 0x400108, 0x010); + nvkm_wr32(device, 0x405018, 0xc0000000); + nvkm_wr32(device, 0x400108, 0x010); status &= ~0x010; } @@ -739,239 +573,174 @@ nv50_gr_trap_handler(struct nv50_gr_priv *priv, u32 display, * remaining, so try to handle it anyway. Perhaps related to that * unknown DMA slot on tesla? */ if (status & 0x20) { - ustatus = nv_rd32(priv, 0x402000) & 0x7fffffff; + ustatus = nvkm_rd32(device, 0x402000) & 0x7fffffff; if (display) - nv_error(priv, "TRAP_UNKC04 0x%08x\n", ustatus); - nv_wr32(priv, 0x402000, 0xc0000000); + nvkm_error(subdev, "TRAP_UNKC04 %08x\n", ustatus); + nvkm_wr32(device, 0x402000, 0xc0000000); /* no status modifiction on purpose */ } /* TEXTURE: CUDA texturing units */ if (status & 0x040) { - nv50_priv_tp_trap(priv, 6, 0x408900, 0x408600, display, + nv50_gr_tp_trap(gr, 6, 0x408900, 0x408600, display, "TRAP_TEXTURE"); - nv_wr32(priv, 0x400108, 0x040); + nvkm_wr32(device, 0x400108, 0x040); status &= ~0x040; } /* MP: CUDA execution engines. */ if (status & 0x080) { - nv50_priv_tp_trap(priv, 7, 0x408314, 0x40831c, display, + nv50_gr_tp_trap(gr, 7, 0x408314, 0x40831c, display, "TRAP_MP"); - nv_wr32(priv, 0x400108, 0x080); + nvkm_wr32(device, 0x400108, 0x080); status &= ~0x080; } /* PROP: Handles TP-initiated uncached memory accesses: * l[], g[], stack, 2d surfaces, render targets. */ if (status & 0x100) { - nv50_priv_tp_trap(priv, 8, 0x408e08, 0x408708, display, + nv50_gr_tp_trap(gr, 8, 0x408e08, 0x408708, display, "TRAP_PROP"); - nv_wr32(priv, 0x400108, 0x100); + nvkm_wr32(device, 0x400108, 0x100); status &= ~0x100; } if (status) { if (display) - nv_error(priv, "TRAP: unknown 0x%08x\n", status); - nv_wr32(priv, 0x400108, status); + nvkm_error(subdev, "TRAP: unknown %08x\n", status); + nvkm_wr32(device, 0x400108, status); } return 1; } -static void -nv50_gr_intr(struct nvkm_subdev *subdev) +void +nv50_gr_intr(struct nvkm_gr *base) { - struct nvkm_fifo *pfifo = nvkm_fifo(subdev); - struct nvkm_engine *engine = nv_engine(subdev); - struct nvkm_object *engctx; - struct nvkm_handle *handle = NULL; - struct nv50_gr_priv *priv = (void *)subdev; - u32 stat = nv_rd32(priv, 0x400100); - u32 inst = nv_rd32(priv, 0x40032c) & 0x0fffffff; - u32 addr = nv_rd32(priv, 0x400704); + struct nv50_gr *gr = nv50_gr(base); + struct nvkm_subdev *subdev = &gr->base.engine.subdev; + struct nvkm_device *device = subdev->device; + struct nvkm_fifo_chan *chan; + u32 stat = nvkm_rd32(device, 0x400100); + u32 inst = nvkm_rd32(device, 0x40032c) & 0x0fffffff; + u32 addr = nvkm_rd32(device, 0x400704); u32 subc = (addr & 0x00070000) >> 16; u32 mthd = (addr & 0x00001ffc); - u32 data = nv_rd32(priv, 0x400708); - u32 class = nv_rd32(priv, 0x400814); + u32 data = nvkm_rd32(device, 0x400708); + u32 class = nvkm_rd32(device, 0x400814); u32 show = stat, show_bitfield = stat; - int chid; - - engctx = nvkm_engctx_get(engine, inst); - chid = pfifo->chid(pfifo, engctx); - - if (stat & 0x00000010) { - handle = nvkm_handle_get_class(engctx, class); - if (handle && !nv_call(handle->object, mthd, data)) - show &= ~0x00000010; - nvkm_handle_put(handle); + const struct nvkm_enum *en; + unsigned long flags; + const char *name = "unknown"; + char msg[128]; + int chid = -1; + + chan = nvkm_fifo_chan_inst(device->fifo, (u64)inst << 12, &flags); + if (chan) { + name = chan->object.client->name; + chid = chan->chid; } if (show & 0x00100000) { - u32 ecode = nv_rd32(priv, 0x400110); - nv_error(priv, "DATA_ERROR "); - nvkm_enum_print(nv50_data_error_names, ecode); - pr_cont("\n"); + u32 ecode = nvkm_rd32(device, 0x400110); + en = nvkm_enum_find(nv50_data_error_names, ecode); + nvkm_error(subdev, "DATA_ERROR %08x [%s]\n", + ecode, en ? en->name : ""); show_bitfield &= ~0x00100000; } if (stat & 0x00200000) { - if (!nv50_gr_trap_handler(priv, show, chid, (u64)inst << 12, - engctx)) + if (!nv50_gr_trap_handler(gr, show, chid, (u64)inst << 12, name)) show &= ~0x00200000; show_bitfield &= ~0x00200000; } - nv_wr32(priv, 0x400100, stat); - nv_wr32(priv, 0x400500, 0x00010001); + nvkm_wr32(device, 0x400100, stat); + nvkm_wr32(device, 0x400500, 0x00010001); if (show) { show &= show_bitfield; - if (show) { - nv_error(priv, "%s", ""); - nvkm_bitfield_print(nv50_gr_intr_name, show); - pr_cont("\n"); - } - nv_error(priv, - "ch %d [0x%010llx %s] subc %d class 0x%04x mthd 0x%04x data 0x%08x\n", - chid, (u64)inst << 12, nvkm_client_name(engctx), - subc, class, mthd, data); + nvkm_snprintbf(msg, sizeof(msg), nv50_gr_intr_name, show); + nvkm_error(subdev, "%08x [%s] ch %d [%010llx %s] subc %d " + "class %04x mthd %04x data %08x\n", + stat, msg, chid, (u64)inst << 12, name, + subc, class, mthd, data); } - if (nv_rd32(priv, 0x400824) & (1 << 31)) - nv_wr32(priv, 0x400824, nv_rd32(priv, 0x400824) & ~(1 << 31)); + if (nvkm_rd32(device, 0x400824) & (1 << 31)) + nvkm_wr32(device, 0x400824, nvkm_rd32(device, 0x400824) & ~(1 << 31)); - nvkm_engctx_put(engctx); + nvkm_fifo_chan_put(device->fifo, flags, &chan); } -static int -nv50_gr_ctor(struct nvkm_object *parent, struct nvkm_object *engine, - struct nvkm_oclass *oclass, void *data, u32 size, - struct nvkm_object **pobject) +int +nv50_gr_init(struct nvkm_gr *base) { - struct nv50_gr_priv *priv; - int ret; - - ret = nvkm_gr_create(parent, engine, oclass, true, &priv); - *pobject = nv_object(priv); - if (ret) - return ret; - - nv_subdev(priv)->unit = 0x00201000; - nv_subdev(priv)->intr = nv50_gr_intr; - nv_engine(priv)->cclass = &nv50_gr_cclass; - - priv->base.units = nv50_gr_units; - - switch (nv_device(priv)->chipset) { - case 0x50: - nv_engine(priv)->sclass = nv50_gr_sclass; - break; - case 0x84: - case 0x86: - case 0x92: - case 0x94: - case 0x96: - case 0x98: - nv_engine(priv)->sclass = g84_gr_sclass; - break; - case 0xa0: - case 0xaa: - case 0xac: - nv_engine(priv)->sclass = gt200_gr_sclass; - break; - case 0xa3: - case 0xa5: - case 0xa8: - nv_engine(priv)->sclass = gt215_gr_sclass; - break; - case 0xaf: - nv_engine(priv)->sclass = mcp89_gr_sclass; - break; - - } - - /* unfortunate hw bug workaround... */ - if (nv_device(priv)->chipset != 0x50 && - nv_device(priv)->chipset != 0xac) - nv_engine(priv)->tlb_flush = g84_gr_tlb_flush; - - spin_lock_init(&priv->lock); - return 0; -} - -static int -nv50_gr_init(struct nvkm_object *object) -{ - struct nv50_gr_priv *priv = (void *)object; + struct nv50_gr *gr = nv50_gr(base); + struct nvkm_device *device = gr->base.engine.subdev.device; int ret, units, i; - ret = nvkm_gr_init(&priv->base); - if (ret) - return ret; - /* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */ - nv_wr32(priv, 0x40008c, 0x00000004); + nvkm_wr32(device, 0x40008c, 0x00000004); /* reset/enable traps and interrupts */ - nv_wr32(priv, 0x400804, 0xc0000000); - nv_wr32(priv, 0x406800, 0xc0000000); - nv_wr32(priv, 0x400c04, 0xc0000000); - nv_wr32(priv, 0x401800, 0xc0000000); - nv_wr32(priv, 0x405018, 0xc0000000); - nv_wr32(priv, 0x402000, 0xc0000000); - - units = nv_rd32(priv, 0x001540); + nvkm_wr32(device, 0x400804, 0xc0000000); + nvkm_wr32(device, 0x406800, 0xc0000000); + nvkm_wr32(device, 0x400c04, 0xc0000000); + nvkm_wr32(device, 0x401800, 0xc0000000); + nvkm_wr32(device, 0x405018, 0xc0000000); + nvkm_wr32(device, 0x402000, 0xc0000000); + + units = nvkm_rd32(device, 0x001540); for (i = 0; i < 16; i++) { if (!(units & (1 << i))) continue; - if (nv_device(priv)->chipset < 0xa0) { - nv_wr32(priv, 0x408900 + (i << 12), 0xc0000000); - nv_wr32(priv, 0x408e08 + (i << 12), 0xc0000000); - nv_wr32(priv, 0x408314 + (i << 12), 0xc0000000); + if (device->chipset < 0xa0) { + nvkm_wr32(device, 0x408900 + (i << 12), 0xc0000000); + nvkm_wr32(device, 0x408e08 + (i << 12), 0xc0000000); + nvkm_wr32(device, 0x408314 + (i << 12), 0xc0000000); } else { - nv_wr32(priv, 0x408600 + (i << 11), 0xc0000000); - nv_wr32(priv, 0x408708 + (i << 11), 0xc0000000); - nv_wr32(priv, 0x40831c + (i << 11), 0xc0000000); + nvkm_wr32(device, 0x408600 + (i << 11), 0xc0000000); + nvkm_wr32(device, 0x408708 + (i << 11), 0xc0000000); + nvkm_wr32(device, 0x40831c + (i << 11), 0xc0000000); } } - nv_wr32(priv, 0x400108, 0xffffffff); - nv_wr32(priv, 0x400138, 0xffffffff); - nv_wr32(priv, 0x400100, 0xffffffff); - nv_wr32(priv, 0x40013c, 0xffffffff); - nv_wr32(priv, 0x400500, 0x00010001); + nvkm_wr32(device, 0x400108, 0xffffffff); + nvkm_wr32(device, 0x400138, 0xffffffff); + nvkm_wr32(device, 0x400100, 0xffffffff); + nvkm_wr32(device, 0x40013c, 0xffffffff); + nvkm_wr32(device, 0x400500, 0x00010001); /* upload context program, initialise ctxctl defaults */ - ret = nv50_grctx_init(nv_device(priv), &priv->size); + ret = nv50_grctx_init(device, &gr->size); if (ret) return ret; - nv_wr32(priv, 0x400824, 0x00000000); - nv_wr32(priv, 0x400828, 0x00000000); - nv_wr32(priv, 0x40082c, 0x00000000); - nv_wr32(priv, 0x400830, 0x00000000); - nv_wr32(priv, 0x40032c, 0x00000000); - nv_wr32(priv, 0x400330, 0x00000000); + nvkm_wr32(device, 0x400824, 0x00000000); + nvkm_wr32(device, 0x400828, 0x00000000); + nvkm_wr32(device, 0x40082c, 0x00000000); + nvkm_wr32(device, 0x400830, 0x00000000); + nvkm_wr32(device, 0x40032c, 0x00000000); + nvkm_wr32(device, 0x400330, 0x00000000); /* some unknown zcull magic */ - switch (nv_device(priv)->chipset & 0xf0) { + switch (device->chipset & 0xf0) { case 0x50: case 0x80: case 0x90: - nv_wr32(priv, 0x402ca8, 0x00000800); + nvkm_wr32(device, 0x402ca8, 0x00000800); break; case 0xa0: default: - if (nv_device(priv)->chipset == 0xa0 || - nv_device(priv)->chipset == 0xaa || - nv_device(priv)->chipset == 0xac) { - nv_wr32(priv, 0x402ca8, 0x00000802); + if (device->chipset == 0xa0 || + device->chipset == 0xaa || + device->chipset == 0xac) { + nvkm_wr32(device, 0x402ca8, 0x00000802); } else { - nv_wr32(priv, 0x402cc0, 0x00000000); - nv_wr32(priv, 0x402ca8, 0x00000002); + nvkm_wr32(device, 0x402cc0, 0x00000000); + nvkm_wr32(device, 0x402ca8, 0x00000002); } break; @@ -979,21 +748,47 @@ nv50_gr_init(struct nvkm_object *object) /* zero out zcull regions */ for (i = 0; i < 8; i++) { - nv_wr32(priv, 0x402c20 + (i * 0x10), 0x00000000); - nv_wr32(priv, 0x402c24 + (i * 0x10), 0x00000000); - nv_wr32(priv, 0x402c28 + (i * 0x10), 0x00000000); - nv_wr32(priv, 0x402c2c + (i * 0x10), 0x00000000); + nvkm_wr32(device, 0x402c20 + (i * 0x10), 0x00000000); + nvkm_wr32(device, 0x402c24 + (i * 0x10), 0x00000000); + nvkm_wr32(device, 0x402c28 + (i * 0x10), 0x00000000); + nvkm_wr32(device, 0x402c2c + (i * 0x10), 0x00000000); } + return 0; } -struct nvkm_oclass -nv50_gr_oclass = { - .handle = NV_ENGINE(GR, 0x50), - .ofuncs = &(struct nvkm_ofuncs) { - .ctor = nv50_gr_ctor, - .dtor = _nvkm_gr_dtor, - .init = nv50_gr_init, - .fini = _nvkm_gr_fini, - }, +int +nv50_gr_new_(const struct nvkm_gr_func *func, struct nvkm_device *device, + int index, struct nvkm_gr **pgr) +{ + struct nv50_gr *gr; + + if (!(gr = kzalloc(sizeof(*gr), GFP_KERNEL))) + return -ENOMEM; + spin_lock_init(&gr->lock); + *pgr = &gr->base; + + return nvkm_gr_ctor(func, device, index, 0x00201000, true, &gr->base); +} + +static const struct nvkm_gr_func +nv50_gr = { + .init = nv50_gr_init, + .intr = nv50_gr_intr, + .chan_new = nv50_gr_chan_new, + .units = nv50_gr_units, + .sclass = { + { -1, -1, 0x0030, &nv50_gr_object }, + { -1, -1, 0x502d, &nv50_gr_object }, + { -1, -1, 0x5039, &nv50_gr_object }, + { -1, -1, 0x5097, &nv50_gr_object }, + { -1, -1, 0x50c0, &nv50_gr_object }, + {} + } }; + +int +nv50_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr) +{ + return nv50_gr_new_(&nv50_gr, device, index, pgr); +} |