static ssize_t tpc_fs_mask_read(struct device *device, struct device_attribute *attr, char *buf) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); struct gr_gk20a *gr = &g->gr; u32 gpc_index; u32 tpc_fs_mask = 0; int err = 0; err = gk20a_busy(g->dev); if (err) return err; for (gpc_index = 0; gpc_index < gr->gpc_count; gpc_index++) { if (g->ops.gr.get_gpc_tpc_mask) tpc_fs_mask |= g->ops.gr.get_gpc_tpc_mask(g, gpc_index) << (gr->max_tpc_per_gpc_count * gpc_index); } gk20a_idle(g->dev); return sprintf(buf, "0x%x\n", tpc_fs_mask); }
static ssize_t elpg_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; /* * Since elpg is refcounted, we should not unnecessarily call * enable/disable if it is already so. */ err = gk20a_busy(g->dev); if (err) return -EAGAIN; if (val && !g->elpg_enabled) { g->elpg_enabled = true; gk20a_pmu_enable_elpg(g); } else if (!val && g->elpg_enabled) { g->elpg_enabled = false; gk20a_pmu_disable_elpg(g); } gk20a_idle(g->dev); dev_info(device, "ELPG is %s.\n", g->elpg_enabled ? "enabled" : "disabled"); return count; }
static ssize_t elcg_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; err = gk20a_busy(g->dev); if (err) return err; if (val) { g->elcg_enabled = true; gr_gk20a_init_elcg_mode(g, ELCG_AUTO, ENGINE_GR_GK20A); gr_gk20a_init_elcg_mode(g, ELCG_AUTO, ENGINE_CE2_GK20A); } else { g->elcg_enabled = false; gr_gk20a_init_elcg_mode(g, ELCG_RUN, ENGINE_GR_GK20A); gr_gk20a_init_elcg_mode(g, ELCG_RUN, ENGINE_CE2_GK20A); } gk20a_idle(g->dev); dev_info(device, "ELCG is %s.\n", g->elcg_enabled ? "enabled" : "disabled"); return count; }
static ssize_t gk20a_load_show(struct device *dev, struct device_attribute *attr, char *buf) { struct platform_device *pdev = to_platform_device(dev); struct gk20a *g = get_gk20a(pdev); u32 busy_time; ssize_t res; int err; if (!g->power_on) { busy_time = 0; } else { err = gk20a_busy(g->dev); if (err) return err; gk20a_pmu_load_update(g); gk20a_pmu_load_norm(g, &busy_time); gk20a_idle(g->dev); } res = snprintf(buf, PAGE_SIZE, "%u\n", busy_time); return res; }
static int monitor_get(void *data, u64 *val) { struct gk20a *g = (struct gk20a *)data; struct clk_gk20a *clk = &g->clk; u32 ncycle = 100; /* count GPCCLK for ncycle of clkin */ u32 clkin = clk->gpc_pll.clk_in; u32 count1, count2; gk20a_busy(g->dev); gk20a_writel(g, trim_gpc_clk_cntr_ncgpcclk_cfg_r(0), trim_gpc_clk_cntr_ncgpcclk_cfg_reset_asserted_f()); gk20a_writel(g, trim_gpc_clk_cntr_ncgpcclk_cfg_r(0), trim_gpc_clk_cntr_ncgpcclk_cfg_enable_asserted_f() | trim_gpc_clk_cntr_ncgpcclk_cfg_write_en_asserted_f() | trim_gpc_clk_cntr_ncgpcclk_cfg_noofipclks_f(ncycle)); /* start */ /* It should take about 8us to finish 100 cycle of 12MHz. But longer than 100us delay is required here. */ gk20a_readl(g, trim_gpc_clk_cntr_ncgpcclk_cfg_r(0)); udelay(2000); count1 = gk20a_readl(g, trim_gpc_clk_cntr_ncgpcclk_cnt_r(0)); udelay(100); count2 = gk20a_readl(g, trim_gpc_clk_cntr_ncgpcclk_cnt_r(0)); *val = (u64)(trim_gpc_clk_cntr_ncgpcclk_cnt_value_v(count2) * clkin / ncycle); gk20a_idle(g->dev); if (count1 != count2) return -EBUSY; return 0; }
static ssize_t slcg_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; if (val) g->slcg_enabled = true; else g->slcg_enabled = false; /* * TODO: slcg_therm_load_gating is not enabled anywhere during * init. Therefore, it would be incongruous to add it here. Once * it is added to init, we should add it here too. */ err = gk20a_busy(g->dev); if (err) return err; if (g->ops.clock_gating.slcg_bus_load_gating_prod) g->ops.clock_gating.slcg_bus_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_ce2_load_gating_prod) g->ops.clock_gating.slcg_ce2_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_chiplet_load_gating_prod) g->ops.clock_gating.slcg_chiplet_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_ctxsw_firmware_load_gating_prod) g->ops.clock_gating.slcg_ctxsw_firmware_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_fb_load_gating_prod) g->ops.clock_gating.slcg_fb_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_fifo_load_gating_prod) g->ops.clock_gating.slcg_fifo_load_gating_prod(g, g->slcg_enabled); g->ops.clock_gating.slcg_gr_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_ltc_load_gating_prod) g->ops.clock_gating.slcg_ltc_load_gating_prod(g, g->slcg_enabled); g->ops.clock_gating.slcg_perf_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_priring_load_gating_prod) g->ops.clock_gating.slcg_priring_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_pmu_load_gating_prod) g->ops.clock_gating.slcg_pmu_load_gating_prod(g, g->slcg_enabled); if (g->ops.clock_gating.slcg_xbar_load_gating_prod) g->ops.clock_gating.slcg_xbar_load_gating_prod(g, g->slcg_enabled); gk20a_idle(g->dev); dev_info(device, "SLCG is %s.\n", g->slcg_enabled ? "enabled" : "disabled"); return count; }
static int gk20a_ctrl_prepare_compressible_read( struct gk20a *g, struct nvgpu_gpu_prepare_compressible_read_args *args) { struct nvgpu_fence fence; struct gk20a_fence *fence_out = NULL; int ret = 0; int flags = args->submit_flags; fence.id = args->fence.syncpt_id; fence.value = args->fence.syncpt_value; ret = gk20a_busy(g->dev); if (ret) return ret; ret = gk20a_prepare_compressible_read(g, args->handle, args->request_compbits, args->offset, args->compbits_hoffset, args->compbits_voffset, args->width, args->height, args->block_height_log2, flags, &fence, &args->valid_compbits, &args->zbc_color, &fence_out); gk20a_idle(g->dev); if (ret) return ret; /* Convert fence_out to something we can pass back to user space. */ if (flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET) { if (flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE) { if (fence_out) { int fd = gk20a_fence_install_fd(fence_out); if (fd < 0) ret = fd; else args->fence.fd = fd; } else { args->fence.fd = -1; } } else { if (fence_out) { args->fence.syncpt_id = fence_out->syncpt_id; args->fence.syncpt_value = fence_out->syncpt_value; } else { args->fence.syncpt_id = -1; args->fence.syncpt_value = 0; } } } gk20a_fence_put(fence_out); return 0; }
static int gk20a_ctrl_mark_compressible_write( struct gk20a *g, struct nvgpu_gpu_mark_compressible_write_args *args) { int ret; ret = gk20a_busy(g->dev); if (ret) return ret; ret = gk20a_mark_compressible_write(g, args->handle, args->valid_compbits, args->offset, args->zbc_color); gk20a_idle(g->dev); return ret; }
static ssize_t allow_all_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; err = gk20a_busy(g->dev); g->allow_all = (val ? true : false); gk20a_idle(g->dev); return count; }
static ssize_t blcg_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; if (val) g->blcg_enabled = true; else g->blcg_enabled = false; err = gk20a_busy(g->dev); if (err) return err; if (g->ops.clock_gating.blcg_bus_load_gating_prod) g->ops.clock_gating.blcg_bus_load_gating_prod(g, g->blcg_enabled); if (g->ops.clock_gating.blcg_ctxsw_firmware_load_gating_prod) g->ops.clock_gating.blcg_ctxsw_firmware_load_gating_prod(g, g->blcg_enabled); if (g->ops.clock_gating.blcg_fb_load_gating_prod) g->ops.clock_gating.blcg_fb_load_gating_prod(g, g->blcg_enabled); if (g->ops.clock_gating.blcg_fifo_load_gating_prod) g->ops.clock_gating.blcg_fifo_load_gating_prod(g, g->blcg_enabled); g->ops.clock_gating.blcg_gr_load_gating_prod(g, g->blcg_enabled); if (g->ops.clock_gating.blcg_ltc_load_gating_prod) g->ops.clock_gating.blcg_ltc_load_gating_prod(g, g->blcg_enabled); if (g->ops.clock_gating.blcg_pmu_load_gating_prod) g->ops.clock_gating.blcg_pmu_load_gating_prod(g, g->blcg_enabled); gk20a_idle(g->dev); dev_info(device, "BLCG is %s.\n", g->blcg_enabled ? "enabled" : "disabled"); return count; }
static ssize_t aelpg_enable_store(struct device *device, struct device_attribute *attr, const char *buf, size_t count) { struct platform_device *ndev = to_platform_device(device); struct gk20a *g = get_gk20a(ndev); unsigned long val = 0; int status = 0; union pmu_ap_cmd ap_cmd; int err; if (kstrtoul(buf, 10, &val) < 0) return -EINVAL; err = gk20a_busy(g->dev); if (err) return err; if (g->pmu.pmu_ready) { if (val && !g->aelpg_enabled) { g->aelpg_enabled = true; /* Enable AELPG */ ap_cmd.init.cmd_id = PMU_AP_CMD_ID_ENABLE_CTRL; status = gk20a_pmu_ap_send_command(g, &ap_cmd, false); } else if (!val && g->aelpg_enabled) { g->aelpg_enabled = false; /* Disable AELPG */ ap_cmd.init.cmd_id = PMU_AP_CMD_ID_DISABLE_CTRL; status = gk20a_pmu_ap_send_command(g, &ap_cmd, false); } } else { dev_info(device, "PMU is not ready, AELPG request failed\n"); } gk20a_idle(g->dev); dev_info(device, "AELPG is %s.\n", g->aelpg_enabled ? "enabled" : "disabled"); return count; }
long gk20a_ctrl_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) { struct platform_device *dev = filp->private_data; struct gk20a *g = get_gk20a(dev); struct nvhost_gpu_zcull_get_ctx_size_args *get_ctx_size_args; struct nvhost_gpu_zcull_get_info_args *get_info_args; struct nvhost_gpu_zbc_set_table_args *set_table_args; struct nvhost_gpu_zbc_query_table_args *query_table_args; u8 buf[NVHOST_GPU_IOCTL_MAX_ARG_SIZE]; struct gr_zcull_info *zcull_info; struct zbc_entry *zbc_val; struct zbc_query_params *zbc_tbl; int i, err = 0; nvhost_dbg_fn(""); if ((_IOC_TYPE(cmd) != NVHOST_GPU_IOCTL_MAGIC) || (_IOC_NR(cmd) == 0) || (_IOC_NR(cmd) > NVHOST_GPU_IOCTL_LAST)) return -EFAULT; BUG_ON(_IOC_SIZE(cmd) > NVHOST_GPU_IOCTL_MAX_ARG_SIZE); if (_IOC_DIR(cmd) & _IOC_WRITE) { if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd))) return -EFAULT; } if (!g->gr.sw_ready) { gk20a_busy(g->dev); gk20a_idle(g->dev); } switch (cmd) { case NVHOST_GPU_IOCTL_ZCULL_GET_CTX_SIZE: get_ctx_size_args = (struct nvhost_gpu_zcull_get_ctx_size_args *)buf; get_ctx_size_args->size = gr_gk20a_get_ctxsw_zcull_size(g, &g->gr); break; case NVHOST_GPU_IOCTL_ZCULL_GET_INFO: get_info_args = (struct nvhost_gpu_zcull_get_info_args *)buf; memset(get_info_args, 0, sizeof(struct nvhost_gpu_zcull_get_info_args)); zcull_info = kzalloc(sizeof(struct gr_zcull_info), GFP_KERNEL); if (zcull_info == NULL) return -ENOMEM; err = gr_gk20a_get_zcull_info(g, &g->gr, zcull_info); if (err) break; get_info_args->width_align_pixels = zcull_info->width_align_pixels; get_info_args->height_align_pixels = zcull_info->height_align_pixels; get_info_args->pixel_squares_by_aliquots = zcull_info->pixel_squares_by_aliquots; get_info_args->aliquot_total = zcull_info->aliquot_total; get_info_args->region_byte_multiplier = zcull_info->region_byte_multiplier; get_info_args->region_header_size = zcull_info->region_header_size; get_info_args->subregion_header_size = zcull_info->subregion_header_size; get_info_args->subregion_width_align_pixels = zcull_info->subregion_width_align_pixels; get_info_args->subregion_height_align_pixels = zcull_info->subregion_height_align_pixels; get_info_args->subregion_count = zcull_info->subregion_count; if (zcull_info) kfree(zcull_info); break; case NVHOST_GPU_IOCTL_ZBC_SET_TABLE: set_table_args = (struct nvhost_gpu_zbc_set_table_args *)buf; zbc_val = kzalloc(sizeof(struct zbc_entry), GFP_KERNEL); if (zbc_val == NULL) return -ENOMEM; zbc_val->format = set_table_args->format; zbc_val->type = set_table_args->type; switch (zbc_val->type) { case GK20A_ZBC_TYPE_COLOR: for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) { zbc_val->color_ds[i] = set_table_args->color_ds[i]; zbc_val->color_l2[i] = set_table_args->color_l2[i]; } break; case GK20A_ZBC_TYPE_DEPTH: zbc_val->depth = set_table_args->depth; break; default: err = -EINVAL; } if (!err) { gk20a_busy(dev); err = gk20a_gr_zbc_set_table(g, &g->gr, zbc_val); gk20a_idle(dev); } if (zbc_val) kfree(zbc_val); break; case NVHOST_GPU_IOCTL_ZBC_QUERY_TABLE: query_table_args = (struct nvhost_gpu_zbc_query_table_args *)buf; zbc_tbl = kzalloc(sizeof(struct zbc_query_params), GFP_KERNEL); if (zbc_tbl == NULL) return -ENOMEM; zbc_tbl->type = query_table_args->type; zbc_tbl->index_size = query_table_args->index_size; err = gr_gk20a_query_zbc(g, &g->gr, zbc_tbl); if (!err) { switch (zbc_tbl->type) { case GK20A_ZBC_TYPE_COLOR: for (i = 0; i < GK20A_ZBC_COLOR_VALUE_SIZE; i++) { query_table_args->color_ds[i] = zbc_tbl->color_ds[i]; query_table_args->color_l2[i] = zbc_tbl->color_l2[i]; } break; case GK20A_ZBC_TYPE_DEPTH: query_table_args->depth = zbc_tbl->depth; break; case GK20A_ZBC_TYPE_INVALID: query_table_args->index_size = zbc_tbl->index_size; break; default: err = -EINVAL; } if (!err) { query_table_args->format = zbc_tbl->format; query_table_args->ref_cnt = zbc_tbl->ref_cnt; } } if (zbc_tbl) kfree(zbc_tbl); break; case NVHOST_GPU_IOCTL_GET_CHARACTERISTICS: err = gk20a_ctrl_ioctl_gpu_characteristics( g, (struct nvhost_gpu_get_characteristics *)buf); break; default: nvhost_err(dev_from_gk20a(g), "unrecognized gpu ioctl cmd: 0x%x", cmd); err = -ENOTTY; break; } if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ)) err = copy_to_user((void __user *)arg, buf, _IOC_SIZE(cmd)); return err; }
static void gk20a_debug_show_dump(struct platform_device *pdev, struct gk20a_debug_output *o) { struct gk20a_platform *platform = gk20a_get_platform(pdev); struct gk20a *g = platform->g; struct fifo_gk20a *f = &g->fifo; u32 chid; int i, err; err = gk20a_busy(g->dev); if (err) { gk20a_debug_output(o, "failed to power on gpu: %d\n", err); return; } for (i = 0; i < fifo_pbdma_status__size_1_v(); i++) { u32 status = gk20a_readl(g, fifo_pbdma_status_r(i)); u32 chan_status = fifo_pbdma_status_chan_status_v(status); gk20a_debug_output(o, "%s pbdma %d: ", g->dev->name, i); gk20a_debug_output(o, "id: %d (%s), next_id: %d (%s) status: %s\n", fifo_pbdma_status_id_v(status), fifo_pbdma_status_id_type_v(status) ? "tsg" : "channel", fifo_pbdma_status_next_id_v(status), fifo_pbdma_status_next_id_type_v(status) ? "tsg" : "channel", chan_status_str[chan_status]); gk20a_debug_output(o, "PUT: %016llx GET: %016llx " "FETCH: %08x HEADER: %08x\n", (u64)gk20a_readl(g, pbdma_put_r(i)) + ((u64)gk20a_readl(g, pbdma_put_hi_r(i)) << 32ULL), (u64)gk20a_readl(g, pbdma_get_r(i)) + ((u64)gk20a_readl(g, pbdma_get_hi_r(i)) << 32ULL), gk20a_readl(g, pbdma_gp_fetch_r(i)), gk20a_readl(g, pbdma_pb_header_r(i))); } gk20a_debug_output(o, "\n"); for (i = 0; i < fifo_engine_status__size_1_v(); i++) { u32 status = gk20a_readl(g, fifo_engine_status_r(i)); u32 ctx_status = fifo_engine_status_ctx_status_v(status); gk20a_debug_output(o, "%s eng %d: ", g->dev->name, i); gk20a_debug_output(o, "id: %d (%s), next_id: %d (%s), ctx: %s ", fifo_engine_status_id_v(status), fifo_engine_status_id_type_v(status) ? "tsg" : "channel", fifo_engine_status_next_id_v(status), fifo_engine_status_next_id_type_v(status) ? "tsg" : "channel", ctx_status_str[ctx_status]); if (fifo_engine_status_faulted_v(status)) gk20a_debug_output(o, "faulted "); if (fifo_engine_status_engine_v(status)) gk20a_debug_output(o, "busy "); gk20a_debug_output(o, "\n"); } gk20a_debug_output(o, "\n"); for (chid = 0; chid < f->num_channels; chid++) { if (f->channel[chid].in_use) { struct channel_gk20a *gpu_ch = &f->channel[chid]; gk20a_debug_show_channel(g, o, gpu_ch); } } gk20a_idle(g->dev); }
long gk20a_as_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) { int err = 0; struct gk20a_as_share *as_share = filp->private_data; struct gk20a *g = gk20a_from_as(as_share->as); u8 buf[NVHOST_AS_IOCTL_MAX_ARG_SIZE]; if ((_IOC_TYPE(cmd) != NVHOST_AS_IOCTL_MAGIC) || (_IOC_NR(cmd) == 0) || (_IOC_NR(cmd) > NVHOST_AS_IOCTL_LAST)) return -EFAULT; BUG_ON(_IOC_SIZE(cmd) > NVHOST_AS_IOCTL_MAX_ARG_SIZE); if (_IOC_DIR(cmd) & _IOC_WRITE) { if (copy_from_user(buf, (void __user *)arg, _IOC_SIZE(cmd))) return -EFAULT; } err = gk20a_busy(g->dev); if (err) return err; switch (cmd) { case NVHOST_AS_IOCTL_BIND_CHANNEL: trace_gk20a_as_ioctl_bind_channel(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_bind_channel(as_share, (struct nvhost_as_bind_channel_args *)buf); break; case NVHOST32_AS_IOCTL_ALLOC_SPACE: { struct nvhost32_as_alloc_space_args *args32 = (struct nvhost32_as_alloc_space_args *)buf; struct nvhost_as_alloc_space_args args; args.pages = args32->pages; args.page_size = args32->page_size; args.flags = args32->flags; args.o_a.offset = args32->o_a.offset; trace_gk20a_as_ioctl_alloc_space(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_alloc_space(as_share, &args); args32->o_a.offset = args.o_a.offset; break; } case NVHOST_AS_IOCTL_ALLOC_SPACE: trace_gk20a_as_ioctl_alloc_space(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_alloc_space(as_share, (struct nvhost_as_alloc_space_args *)buf); break; case NVHOST_AS_IOCTL_FREE_SPACE: trace_gk20a_as_ioctl_free_space(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_free_space(as_share, (struct nvhost_as_free_space_args *)buf); break; case NVHOST_AS_IOCTL_MAP_BUFFER: trace_gk20a_as_ioctl_map_buffer(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_map_buffer(as_share, (struct nvhost_as_map_buffer_args *)buf); break; case NVHOST_AS_IOCTL_MAP_BUFFER_EX: trace_gk20a_as_ioctl_map_buffer(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_map_buffer_ex(as_share, (struct nvhost_as_map_buffer_ex_args *)buf); break; case NVHOST_AS_IOCTL_UNMAP_BUFFER: trace_gk20a_as_ioctl_unmap_buffer(dev_name(dev_from_gk20a(g))); err = gk20a_as_ioctl_unmap_buffer(as_share, (struct nvhost_as_unmap_buffer_args *)buf); break; default: dev_err(dev_from_gk20a(g), "unrecognized as ioctl: 0x%x", cmd); err = -ENOTTY; break; } gk20a_idle(g->dev); if ((err == 0) && (_IOC_DIR(cmd) & _IOC_READ)) err = copy_to_user((void __user *)arg, buf, _IOC_SIZE(cmd)); return err; }