From fbdcc8a2d4c2b7f145b034ee7bde7105e66e0a4e Mon Sep 17 00:00:00 2001 From: David Nieto Date: Thu, 7 Dec 2017 16:48:09 -0800 Subject: gpu: nvgpu: Initial Nvlink driver skeleton Adds the skeleton and integration of the GV100 endpoint driver to NVGPU (1) Adds a OS abstraction layer for the internal nvlink structure. (2) Adds linux specific integration with Nvlink core driver. (3) Adds function pointers for nvlink api, initialization and isr process. (4) Adds initial support for minion. (5) Adds new GPU enable properties to handle NVLINK presence (6) Adds new GPU enable properties for SG_PHY bypass (required for NVLINK over PCI) (7) Adds parsing of nvlink vbios structures. (8) Adds logging defines for NVGPU JIRA: EVLR-2328 Change-Id: I0720a165a15c7187892c8c1a0662ec598354ac06 Signed-off-by: David Nieto Reviewed-on: https://git-master.nvidia.com/r/1644708 Reviewed-by: mobile promotions Tested-by: mobile promotions --- drivers/gpu/nvgpu/Makefile | 2 +- drivers/gpu/nvgpu/common/falcon/falcon.c | 6 +- drivers/gpu/nvgpu/common/linux/nvlink.c | 531 ++++++++++++++++++++++++++++++ drivers/gpu/nvgpu/common/linux/pci.c | 13 + drivers/gpu/nvgpu/common/vbios/bios.c | 53 ++- drivers/gpu/nvgpu/gk20a/gk20a.c | 12 + drivers/gpu/nvgpu/gk20a/gk20a.h | 28 +- drivers/gpu/nvgpu/gp10b/mc_gp10b.c | 5 +- drivers/gpu/nvgpu/include/nvgpu/bios.h | 4 +- drivers/gpu/nvgpu/include/nvgpu/enabled.h | 8 +- drivers/gpu/nvgpu/include/nvgpu/falcon.h | 3 +- drivers/gpu/nvgpu/include/nvgpu/log.h | 3 +- drivers/gpu/nvgpu/include/nvgpu/nvlink.h | 206 ++++++++++++ 13 files changed, 864 insertions(+), 10 deletions(-) create mode 100644 drivers/gpu/nvgpu/common/linux/nvlink.c create mode 100644 drivers/gpu/nvgpu/include/nvgpu/nvlink.h (limited to 'drivers/gpu') diff --git a/drivers/gpu/nvgpu/Makefile b/drivers/gpu/nvgpu/Makefile index d70c44e8..626fea24 100644 --- a/drivers/gpu/nvgpu/Makefile +++ b/drivers/gpu/nvgpu/Makefile @@ -55,6 +55,7 @@ nvgpu-y := \ common/linux/ce2.o \ common/linux/sim.o \ common/linux/os_sched.o \ + common/linux/nvlink.o \ common/mm/nvgpu_allocator.o \ common/mm/bitmap_allocator.o \ common/mm/buddy_allocator.o \ @@ -291,7 +292,6 @@ nvgpu-$(CONFIG_TEGRA_GR_VIRTUALIZATION) += \ common/linux/vgpu/gp10b/vgpu_fuse_gp10b.o \ common/linux/vgpu/gp10b/vgpu_mm_gp10b.o - nvgpu-$(CONFIG_TEGRA_GR_VIRTUALIZATION) += \ common/linux/vgpu/gv11b/platform_gv11b_vgpu_tegra.o \ common/linux/vgpu/gv11b/vgpu_gv11b.o \ diff --git a/drivers/gpu/nvgpu/common/falcon/falcon.c b/drivers/gpu/nvgpu/common/falcon/falcon.c index 42b33c27..41e394f9 100644 --- a/drivers/gpu/nvgpu/common/falcon/falcon.c +++ b/drivers/gpu/nvgpu/common/falcon/falcon.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -385,6 +385,10 @@ void nvgpu_flcn_sw_init(struct gk20a *g, u32 flcn_id) flcn = &g->nvdec_flcn; flcn->flcn_id = flcn_id; break; + case FALCON_ID_MINION: + flcn = &g->minion_flcn; + flcn->flcn_id = flcn_id; + break; default: nvgpu_err(g, "Invalid/Unsupported falcon ID %x", flcn_id); break; diff --git a/drivers/gpu/nvgpu/common/linux/nvlink.c b/drivers/gpu/nvgpu/common/linux/nvlink.c new file mode 100644 index 00000000..eea6eda7 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/nvlink.c @@ -0,0 +1,531 @@ +/* + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + */ + +#include +#include +#include +#include "module.h" + +#ifdef CONFIG_TEGRA_NVLINK +#include +#endif + +#ifdef CONFIG_TEGRA_NVLINK + +/* + * WAR: use this function to find detault link, as only one is supported + * on the library for now + * Returns NVLINK_MAX_LINKS_SW on failure + */ +static u32 __nvgpu_nvlink_get_link(struct nvlink_device *ndev) +{ + u32 link_id; + struct gk20a *g = (struct gk20a *) ndev->priv; + + if (!g) + return NVLINK_MAX_LINKS_SW; + + /* Lets find the detected link */ + if (g->nvlink.initialized_links) + link_id = fls(g->nvlink.initialized_links); + else + return NVLINK_MAX_LINKS_SW; + + if (g->nvlink.links[link_id].remote_info.is_connected) + return link_id; + + return NVLINK_MAX_LINKS_SW; +} +static int nvgpu_nvlink_early_init(struct nvlink_device *ndev) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + int err; + + /* For now master topology is the only one supported */ + if (!ndev->is_master) { + nvgpu_log(g, gpu_dbg_info | gpu_dbg_nvlink, + "dGPU is not master of Nvlink link"); + return -EINVAL; + } + + err = g->ops.nvlink.early_init(g); + return err; +} + +static int nvgpu_nvlink_link_early_init(struct nvlink_device *ndev) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + int err; + u32 link_id; + + /* + * First check the topology and setup connectivity + * HACK: we are only enabling one link for now!!! + */ + link_id = fls(g->nvlink.discovered_links); + g->nvlink.links[link_id].remote_info.is_connected = true; + + err = g->ops.nvlink.link_early_init(g, BIT(link_id)); + + if (err == 0) { + g->nvlink.links[link_id].priv = (void *) &(ndev->link); + ndev->link.priv = (void *) g; + } + return err; +} + +static int nvgpu_nvlink_interface_init(struct nvlink_device *ndev) +{ + int err; + struct gk20a *g = (struct gk20a *) ndev->priv; + + err = g->ops.nvlink.interface_init(g); + return err; +} + +static int nvgpu_nvlink_shutdown(struct nvlink_device *ndev) +{ + int err; + struct gk20a *g = (struct gk20a *) ndev->priv; + + err = g->ops.nvlink.shutdown(g); + return 0; +} + +static int nvgpu_nvlink_reg_init(struct nvlink_device *ndev) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + int err; + + err = g->ops.nvlink.reg_init(g); + + return err; +} + +static u32 nvgpu_nvlink_get_link_mode(struct nvlink_device *ndev) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + u32 mode; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return -EINVAL; + + mode = g->ops.nvlink.link_get_mode(g, link_id); + + switch (mode) { + case nvgpu_nvlink_link_off: + return NVLINK_LINK_OFF; + case nvgpu_nvlink_link_hs: + return NVLINK_LINK_HS; + case nvgpu_nvlink_link_safe: + return NVLINK_LINK_SAFE; + case nvgpu_nvlink_link_fault: + return NVLINK_LINK_FAULT; + case nvgpu_nvlink_link_recovery: + return NVLINK_LINK_RECOVERY; + case nvgpu_nvlink_link_detect: + return NVLINK_LINK_DETECT; + case nvgpu_nvlink_link_reset: + return NVLINK_LINK_RESET; + case nvgpu_nvlink_link_enable_pm: + return NVLINK_LINK_ENABLE_PM; + case nvgpu_nvlink_link_disable_pm: + return NVLINK_LINK_DISABLE_PM; + case nvgpu_nvlink_link_disable_err_detect: + return NVLINK_LINK_DISABLE_ERR_DETECT; + case nvgpu_nvlink_link_lane_disable: + return NVLINK_LINK_LANE_DISABLE; + case nvgpu_nvlink_link_lane_shutdown: + return NVLINK_LINK_LANE_SHUTDOWN; + default: + nvgpu_log(g, gpu_dbg_info | gpu_dbg_nvlink, + "unsupported mode %u", mode); + } + + return NVLINK_LINK_OFF; +} + +static u32 nvgpu_nvlink_get_link_state(struct nvlink_device *ndev) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return -EINVAL; + + return g->ops.nvlink.link_get_state(g, link_id); +} + +static int nvgpu_nvlink_set_link_mode(struct nvlink_device *ndev, u32 mode) +{ + + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + u32 mode_sw; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return -EINVAL; + + switch (mode) { + case NVLINK_LINK_OFF: + mode_sw = nvgpu_nvlink_link_off; + break; + case NVLINK_LINK_HS: + mode_sw = nvgpu_nvlink_link_hs; + break; + case NVLINK_LINK_SAFE: + mode_sw = nvgpu_nvlink_link_safe; + break; + case NVLINK_LINK_FAULT: + mode_sw = nvgpu_nvlink_link_fault; + break; + case NVLINK_LINK_RECOVERY: + mode_sw = nvgpu_nvlink_link_recovery; + break; + case NVLINK_LINK_DETECT: + mode_sw = nvgpu_nvlink_link_detect; + break; + case NVLINK_LINK_RESET: + mode_sw = nvgpu_nvlink_link_reset; + break; + case NVLINK_LINK_ENABLE_PM: + mode_sw = nvgpu_nvlink_link_enable_pm; + break; + case NVLINK_LINK_DISABLE_PM: + mode_sw = nvgpu_nvlink_link_disable_pm; + break; + case NVLINK_LINK_LANE_DISABLE: + mode_sw = nvgpu_nvlink_link_lane_disable; + break; + case NVLINK_LINK_LANE_SHUTDOWN: + mode_sw = nvgpu_nvlink_link_lane_shutdown; + break; + default: + mode_sw = nvgpu_nvlink_link_off; + } + + return g->ops.nvlink.link_set_mode(g, link_id, mode_sw); +} + +static void nvgpu_nvlink_get_tx_sublink_state(struct nvlink_device *ndev, + u32 *state) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return; + if (state) + *state = g->ops.nvlink.get_tx_sublink_state(g, link_id); +} + +static void nvgpu_nvlink_get_rx_sublink_state(struct nvlink_device *ndev, + u32 *state) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return; + if (state) + *state = g->ops.nvlink.get_rx_sublink_state(g, link_id); +} + +static u32 nvgpu_nvlink_get_sublink_mode(struct nvlink_device *ndev, + bool is_rx_sublink) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + u32 mode; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return -EINVAL; + + mode = g->ops.nvlink.get_sublink_mode(g, link_id, is_rx_sublink); + switch (mode) { + case nvgpu_nvlink_sublink_tx_hs: + return NVLINK_TX_HS; + case nvgpu_nvlink_sublink_tx_off: + return NVLINK_TX_OFF; + case nvgpu_nvlink_sublink_tx_single_lane: + return NVLINK_TX_SINGLE_LANE; + case nvgpu_nvlink_sublink_tx_safe: + return NVLINK_TX_SAFE; + case nvgpu_nvlink_sublink_tx_enable_pm: + return NVLINK_TX_ENABLE_PM; + case nvgpu_nvlink_sublink_tx_disable_pm: + return NVLINK_TX_DISABLE_PM; + case nvgpu_nvlink_sublink_tx_common: + return NVLINK_TX_COMMON; + case nvgpu_nvlink_sublink_tx_common_disable: + return NVLINK_TX_COMMON_DISABLE; + case nvgpu_nvlink_sublink_tx_data_ready: + return NVLINK_TX_DATA_READY; + case nvgpu_nvlink_sublink_tx_prbs_en: + return NVLINK_TX_PRBS_EN; + case nvgpu_nvlink_sublink_rx_hs: + return NVLINK_RX_HS; + case nvgpu_nvlink_sublink_rx_enable_pm: + return NVLINK_RX_ENABLE_PM; + case nvgpu_nvlink_sublink_rx_disable_pm: + return NVLINK_RX_DISABLE_PM; + case nvgpu_nvlink_sublink_rx_single_lane: + return NVLINK_RX_SINGLE_LANE; + case nvgpu_nvlink_sublink_rx_safe: + return NVLINK_RX_SAFE; + case nvgpu_nvlink_sublink_rx_off: + return NVLINK_RX_OFF; + case nvgpu_nvlink_sublink_rx_rxcal: + return NVLINK_RX_RXCAL; + default: + nvgpu_log(g, gpu_dbg_nvlink, "Unsupported mode: %u", mode); + break; + } + + if (is_rx_sublink) + return NVLINK_RX_OFF; + return NVLINK_TX_OFF; +} + +static int nvgpu_nvlink_set_sublink_mode(struct nvlink_device *ndev, + bool is_rx_sublink, u32 mode) +{ + struct gk20a *g = (struct gk20a *) ndev->priv; + u32 link_id; + u32 mode_sw; + + link_id = __nvgpu_nvlink_get_link(ndev); + if (link_id == NVLINK_MAX_LINKS_SW) + return -EINVAL; + + if (!is_rx_sublink) { + switch (mode) { + case NVLINK_TX_HS: + mode_sw = nvgpu_nvlink_sublink_tx_hs; + break; + case NVLINK_TX_ENABLE_PM: + mode_sw = nvgpu_nvlink_sublink_tx_enable_pm; + break; + case NVLINK_TX_DISABLE_PM: + mode_sw = nvgpu_nvlink_sublink_tx_disable_pm; + break; + case NVLINK_TX_SINGLE_LANE: + mode_sw = nvgpu_nvlink_sublink_tx_single_lane; + break; + case NVLINK_TX_SAFE: + mode_sw = nvgpu_nvlink_sublink_tx_safe; + break; + case NVLINK_TX_OFF: + mode_sw = nvgpu_nvlink_sublink_tx_off; + break; + case NVLINK_TX_COMMON: + mode_sw = nvgpu_nvlink_sublink_tx_common; + break; + case NVLINK_TX_COMMON_DISABLE: + mode_sw = nvgpu_nvlink_sublink_tx_common_disable; + break; + case NVLINK_TX_DATA_READY: + mode_sw = nvgpu_nvlink_sublink_tx_data_ready; + break; + case NVLINK_TX_PRBS_EN: + mode_sw = nvgpu_nvlink_sublink_tx_prbs_en; + break; + default: + return -EINVAL; + } + } else { + switch (mode) { + case NVLINK_RX_HS: + mode_sw = nvgpu_nvlink_sublink_rx_hs; + break; + case NVLINK_RX_ENABLE_PM: + mode_sw = nvgpu_nvlink_sublink_rx_enable_pm; + break; + case NVLINK_RX_DISABLE_PM: + mode_sw = nvgpu_nvlink_sublink_rx_disable_pm; + break; + case NVLINK_RX_SINGLE_LANE: + mode_sw = nvgpu_nvlink_sublink_rx_single_lane; + break; + case NVLINK_RX_SAFE: + mode_sw = nvgpu_nvlink_sublink_rx_safe; + break; + case NVLINK_RX_OFF: + mode_sw = nvgpu_nvlink_sublink_rx_off; + break; + case NVLINK_RX_RXCAL: + mode_sw = nvgpu_nvlink_sublink_rx_rxcal; + break; + default: + return -EINVAL; + } + } + + return g->ops.nvlink.set_sublink_mode(g, link_id, is_rx_sublink, + mode_sw); +} +#endif + +u32 nvgpu_nvlink_enumerate(struct gk20a *g) +{ + u32 err = -ENODEV; +#ifdef CONFIG_TEGRA_NVLINK + struct nvlink_device *ndev; + + ndev = (struct nvlink_device *) g->nvlink.priv; + + if (ndev) + err = nvlink_enumerate(ndev); +#endif + return err; +} + +u32 nvgpu_nvlink_train(struct gk20a *g, u32 link_id, bool from_off) +{ + u32 err = -ENODEV; +#ifdef CONFIG_TEGRA_NVLINK + struct nvlink_device *ndev; + + ndev = (struct nvlink_device *) g->nvlink.priv; + + if (!ndev) + return -ENODEV; + + /* Check if the link is connected */ + if (!g->nvlink.links[link_id].remote_info.is_connected) + return -ENODEV; + + if (from_off) + return nvlink_transition_intranode_conn_off_to_safe(ndev); + return nvlink_train_intranode_conn_safe_to_hs(ndev); + +#endif + return err; +} + +u32 nvgpu_nvlink_probe(struct gk20a *g) +{ +#ifdef CONFIG_TEGRA_NVLINK + u32 err = 0; + struct device_node *np = nvgpu_get_node(g); + struct device_node *nvlink_np = NULL, *endp_np = NULL; + struct nvlink_device *ndev; + u32 phys_link_id; + + /* Parse DT */ + if (np) { + nvlink_np = of_get_child_by_name(np, "nvidia,nvlink"); + if (nvlink_np) + endp_np = of_get_child_by_name(np, "endpoint"); + } + + if (!endp_np) { + nvgpu_log(g, gpu_dbg_info | gpu_dbg_nvlink, + "No Nvlink DT detected"); + return -ENODEV; + } + + /* Allocating structures */ + ndev = nvgpu_kzalloc(g, sizeof(struct nvlink_device)); + if (!ndev) { + nvgpu_err(g, "OOM while allocating nvlink device struct"); + return -ENOMEM; + } + + ndev->priv = (void *) g; + g->nvlink.priv = (void *) ndev; + + /* Parse DT structure to detect endpoint topology */ + of_property_read_u32(endp_np, "local_dev_id", &ndev->device_id); + of_property_read_u32(endp_np, "local_link_id", &ndev->link.link_id); + ndev->is_master = of_property_read_bool(endp_np, "is_master"); + of_property_read_u32(endp_np, "remote_dev_id", + &ndev->link.remote_dev_info.device_id); + of_property_read_u32(endp_np, "remote_link_id", + &ndev->link.remote_dev_info.link_id); + of_property_read_u32(endp_np, "physical_link", + &phys_link_id); + + g->nvlink.topology_connected_links = BIT(phys_link_id); + + mutex_init(&ndev->init_state_mutex); + + /* Check that we are in dGPU mode */ + if (ndev->device_id != NVLINK_ENDPT_GV100) { + nvgpu_err(g, "Local nvlink device is not dGPU"); + err = -EINVAL; + goto free_nvlink; + } + + err = nvlink_set_init_state(ndev, NVLINK_DEV_OFF); + if (err) { + nvgpu_err(g, "Error initalizing device state to OFF"); + goto free_nvlink; + } + + /* Fill in device struct */ + ndev->dev_ops.dev_early_init = nvgpu_nvlink_early_init; + ndev->dev_ops.dev_interface_init = nvgpu_nvlink_interface_init; + ndev->dev_ops.dev_reg_init = nvgpu_nvlink_reg_init; + ndev->dev_ops.dev_shutdown = nvgpu_nvlink_shutdown; + + /* Fill in the link struct */ + ndev->link.device_id = ndev->device_id; + ndev->link.mode = NVLINK_LINK_OFF; + ndev->link.link_ops.get_link_mode = nvgpu_nvlink_get_link_mode; + ndev->link.link_ops.set_link_mode = nvgpu_nvlink_set_link_mode; + ndev->link.link_ops.get_sublink_mode = nvgpu_nvlink_get_sublink_mode; + ndev->link.link_ops.set_sublink_mode = nvgpu_nvlink_set_sublink_mode; + ndev->link.link_ops.get_link_state = nvgpu_nvlink_get_link_state; + ndev->link.link_ops.get_tx_sublink_state = + nvgpu_nvlink_get_tx_sublink_state; + ndev->link.link_ops.get_rx_sublink_state = + nvgpu_nvlink_get_rx_sublink_state; + ndev->link.link_ops.link_early_init = + nvgpu_nvlink_link_early_init; + + /* Register device with core driver*/ + err = nvlink_register_device(ndev); + if (err) { + nvgpu_err(g, "failed on nvlink device registration"); + goto free_nvlink; + } + + /* Register link with core driver */ + err = nvlink_register_link(&ndev->link); + if (err) { + nvgpu_err(g, "failed on nvlink link registration"); + goto free_nvlink; + } + + /* Enable NVLINK support */ + __nvgpu_set_enabled(g, NVGPU_SUPPORT_NVLINK, true); +free_nvlink: + nvgpu_kfree(g, ndev); + return err; + +#else + return -ENODEV; +#endif +} + diff --git a/drivers/gpu/nvgpu/common/linux/pci.c b/drivers/gpu/nvgpu/common/linux/pci.c index 9c18fbc9..905a3d39 100644 --- a/drivers/gpu/nvgpu/common/linux/pci.c +++ b/drivers/gpu/nvgpu/common/linux/pci.c @@ -21,6 +21,7 @@ #include #include #include +#include #include #include @@ -629,6 +630,18 @@ static int nvgpu_pci_probe(struct pci_dev *pdev, return err; } + err = nvgpu_nvlink_probe(g); + /* + * ENODEV is a legal error which means there is no NVLINK + * any other error is fatal + */ + if (err) { + if (err != -ENODEV) { + nvgpu_err(g, "fatal error probing nvlink, bailing out"); + return err; + } + } + g->mm.has_physical_mode = false; np = nvgpu_get_node(g); diff --git a/drivers/gpu/nvgpu/common/vbios/bios.c b/drivers/gpu/nvgpu/common/vbios/bios.c index fa700a66..52c0a798 100644 --- a/drivers/gpu/nvgpu/common/vbios/bios.c +++ b/drivers/gpu/nvgpu/common/vbios/bios.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2015-2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2015-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -74,6 +74,22 @@ struct bit { #define TOKEN_ID_VIRT_PTRS 0x56 #define TOKEN_ID_MEMORY_PTRS 0x4D +#define NVLINK_CONFIG_DATA_HDR_VER_10 0x1 +#define NVLINK_CONFIG_DATA_HDR_10_SIZE 16 +#define NVLINK_CONFIG_DATA_HDR_11_SIZE 17 +#define NVLINK_CONFIG_DATA_HDR_12_SIZE 21 + +struct nvlink_config_data_hdr_v1 { + u8 version; + u8 hdr_size; + u16 rsvd0; + u32 link_disable_mask; + u32 link_mode_mask; + u32 link_refclk_mask; + u8 train_at_boot; + u32 ac_coupling_mask; +} __packed; + #define MEMORY_PTRS_V1 1 #define MEMORY_PTRS_V2 2 @@ -369,6 +385,41 @@ static void nvgpu_bios_parse_nvinit_ptrs(struct gk20a *g, int offset) g->bios.bootscripts = &g->bios.data[nvinit_ptrs.bootscripts_ptr]; g->bios.bootscripts_size = nvinit_ptrs.bootscripts_size; g->bios.condition_table_ptr = nvinit_ptrs.condition_table_ptr; + g->bios.nvlink_config_data_offset = nvinit_ptrs.nvlink_config_data_ptr; +} + +u32 nvgpu_bios_get_nvlink_config_data(struct gk20a *g) +{ + struct nvlink_config_data_hdr_v1 config; + + if (g->bios.nvlink_config_data_offset == 0) + return -EINVAL; + + memcpy(&config, &g->bios.data[g->bios.nvlink_config_data_offset], + sizeof(config)); + + if (config.version != NVLINK_CONFIG_DATA_HDR_VER_10) { + nvgpu_err(g, "unsupported nvlink bios version: 0x%x", + config.version); + return -EINVAL; + } + + switch (config.hdr_size) { + case NVLINK_CONFIG_DATA_HDR_12_SIZE: + g->nvlink.ac_coupling_mask = config.ac_coupling_mask; + case NVLINK_CONFIG_DATA_HDR_11_SIZE: + g->nvlink.train_at_boot = config.train_at_boot; + case NVLINK_CONFIG_DATA_HDR_10_SIZE: + g->nvlink.link_disable_mask = config.link_disable_mask; + g->nvlink.link_mode_mask = config.link_mode_mask; + g->nvlink.link_refclk_mask = config.link_refclk_mask; + break; + default: + nvgpu_err(g, "invalid nvlink bios config size"); + return -EINVAL; + } + + return 0; } static void nvgpu_bios_parse_memory_ptrs(struct gk20a *g, int offset, u8 version) diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c index b4886e31..868792c0 100644 --- a/drivers/gpu/nvgpu/gk20a/gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/gk20a.c @@ -186,6 +186,18 @@ int gk20a_finalize_poweron(struct gk20a *g) } } + if (nvgpu_is_enabled(g, NVGPU_SUPPORT_NVLINK)) { + if (g->ops.nvlink.init) { + err = g->ops.nvlink.init(g); + if (err) { + nvgpu_err(g, "failed to init nvlink"); + __nvgpu_set_enabled(g, NVGPU_SUPPORT_NVLINK, + false); + } + } else + __nvgpu_set_enabled(g, NVGPU_SUPPORT_NVLINK, false); + } + if (g->ops.fb.mem_unlock) { err = g->ops.fb.mem_unlock(g); if (err) { diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h index cc62865c..f187f730 100644 --- a/drivers/gpu/nvgpu/gk20a/gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/gk20a.h @@ -62,6 +62,7 @@ struct nvgpu_ctxsw_trace_filter; #include #include #include +#include #include "clk_gk20a.h" #include "ce2_gk20a.h" @@ -961,6 +962,7 @@ struct gpu_ops { bool enable, bool is_stalling, u32 unit); void (*isr_stall)(struct gk20a *g); bool (*is_intr_hub_pending)(struct gk20a *g, u32 mc_intr); + bool (*is_intr_nvlink_pending)(struct gk20a *g, u32 mc_intr); u32 (*intr_stall)(struct gk20a *g); void (*intr_stall_pause)(struct gk20a *g); void (*intr_stall_resume)(struct gk20a *g); @@ -1057,7 +1059,27 @@ struct gpu_ops { struct { int (*check_priv_security)(struct gk20a *g); } fuse; - + struct { + u32 (*init)(struct gk20a *g); + u32 (*discover_ioctrl)(struct gk20a *g); + u32 (*discover_link)(struct gk20a *g); + u32 (*isr)(struct gk20a *g); + /* API */ + int (*link_early_init)(struct gk20a *g, unsigned long mask); + u32 (*link_get_mode)(struct gk20a *g, u32 link_id); + u32 (*link_get_state)(struct gk20a *g, u32 link_id); + int (*link_set_mode)(struct gk20a *g, u32 link_id, u32 mode); + u32 (*get_sublink_mode)(struct gk20a *g, u32 link_id, + bool is_rx_sublink); + u32 (*get_rx_sublink_state)(struct gk20a *g, u32 link_id); + u32 (*get_tx_sublink_state)(struct gk20a *g, u32 link_id); + int (*set_sublink_mode)(struct gk20a *g, u32 link_id, + bool is_rx_sublink, u32 mode); + int (*interface_init)(struct gk20a *g); + int (*reg_init)(struct gk20a *g); + int (*shutdown)(struct gk20a *g); + int (*early_init)(struct gk20a *g); + } nvlink; }; struct nvgpu_bios_ucode { @@ -1100,6 +1122,8 @@ struct nvgpu_bios { struct bit_token *clock_token; struct bit_token *virt_token; u32 expansion_rom_offset; + + u32 nvlink_config_data_offset; }; struct nvgpu_gpu_params { @@ -1153,8 +1177,10 @@ struct gk20a { struct nvgpu_falcon fecs_flcn; struct nvgpu_falcon gpccs_flcn; struct nvgpu_falcon nvdec_flcn; + struct nvgpu_falcon minion_flcn; struct clk_gk20a clk; struct fifo_gk20a fifo; + struct nvgpu_nvlink_dev nvlink; struct gr_gk20a gr; struct sim_gk20a *sim; struct mm_gk20a mm; diff --git a/drivers/gpu/nvgpu/gp10b/mc_gp10b.c b/drivers/gpu/nvgpu/gp10b/mc_gp10b.c index 9aea76f9..dde12854 100644 --- a/drivers/gpu/nvgpu/gp10b/mc_gp10b.c +++ b/drivers/gpu/nvgpu/gp10b/mc_gp10b.c @@ -1,7 +1,7 @@ /* * GP10B master * - * Copyright (c) 2014-2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2014-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -122,6 +122,9 @@ void mc_gp10b_isr_stall(struct gk20a *g) g->ops.ltc.isr(g); if (mc_intr_0 & mc_intr_pbus_pending_f()) g->ops.bus.isr(g); + if (g->ops.mc.is_intr_nvlink_pending && + g->ops.mc.is_intr_nvlink_pending(g, mc_intr_0)) + g->ops.nvlink.isr(g); gk20a_dbg(gpu_dbg_intr, "stall intr done 0x%08x\n", mc_intr_0); diff --git a/drivers/gpu/nvgpu/include/nvgpu/bios.h b/drivers/gpu/nvgpu/include/nvgpu/bios.h index a4823caa..fb0a313f 100644 --- a/drivers/gpu/nvgpu/include/nvgpu/bios.h +++ b/drivers/gpu/nvgpu/include/nvgpu/bios.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2014-2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2014-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -1051,5 +1051,5 @@ u32 nvgpu_bios_read_u32(struct gk20a *g, u32 offset); void *nvgpu_bios_get_perf_table_ptrs(struct gk20a *g, struct bit_token *ptoken, u8 table_id); int nvgpu_bios_execute_script(struct gk20a *g, u32 offset); - +u32 nvgpu_bios_get_nvlink_config_data(struct gk20a *g); #endif diff --git a/drivers/gpu/nvgpu/include/nvgpu/enabled.h b/drivers/gpu/nvgpu/include/nvgpu/enabled.h index c614ce4d..4d8dbb08 100644 --- a/drivers/gpu/nvgpu/include/nvgpu/enabled.h +++ b/drivers/gpu/nvgpu/include/nvgpu/enabled.h @@ -77,7 +77,8 @@ struct gk20a; #define NVGPU_SUPPORT_MAP_BUFFER_BATCH 25 /* Support DMA coherence */ #define NVGPU_DMA_COHERENT 26 - +/* Use physical scatter tables instead of IOMMU */ +#define NVGPU_MM_USE_PHYSICAL_SG 27 /* * Host flags @@ -113,6 +114,11 @@ struct gk20a; #define NVGPU_SEC_SECUREGPCCS 41 #define NVGPU_SEC_PRIVSECURITY 42 +/* + * Nvlink flags + */ + +#define NVGPU_SUPPORT_NVLINK 45 /* * PMU flags. */ diff --git a/drivers/gpu/nvgpu/include/nvgpu/falcon.h b/drivers/gpu/nvgpu/include/nvgpu/falcon.h index 1f104fa1..31587ee7 100644 --- a/drivers/gpu/nvgpu/include/nvgpu/falcon.h +++ b/drivers/gpu/nvgpu/include/nvgpu/falcon.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -34,6 +34,7 @@ #define FALCON_ID_GPCCS (3) #define FALCON_ID_NVDEC (4) #define FALCON_ID_SEC2 (7) +#define FALCON_ID_MINION (10) /* * Falcon Base address Defines diff --git a/drivers/gpu/nvgpu/include/nvgpu/log.h b/drivers/gpu/nvgpu/include/nvgpu/log.h index 65f86198..7d7a41e3 100644 --- a/drivers/gpu/nvgpu/include/nvgpu/log.h +++ b/drivers/gpu/nvgpu/include/nvgpu/log.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -79,6 +79,7 @@ enum nvgpu_log_categories { gpu_dbg_dma = BIT(22), /* DMA allocation prints. */ gpu_dbg_sgl = BIT(23), /* SGL related traces. */ gpu_dbg_vidmem = BIT(24), /* VIDMEM tracing. */ + gpu_dbg_nvlink = BIT(25), /* nvlink Operation tracing. */ gpu_dbg_mem = BIT(31), /* memory accesses; very verbose. */ }; diff --git a/drivers/gpu/nvgpu/include/nvgpu/nvlink.h b/drivers/gpu/nvgpu/include/nvgpu/nvlink.h new file mode 100644 index 00000000..48851ff1 --- /dev/null +++ b/drivers/gpu/nvgpu/include/nvgpu/nvlink.h @@ -0,0 +1,206 @@ +/* + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#ifndef __NVGPU_NVLINK_H__ +#define __NVGPU_NVLINK_H__ + +#include + +struct gk20a; + +struct nvgpu_nvlink_ioctrl_list { + bool valid; + u32 pri_base_addr; + u8 intr_enum; + u8 reset_enum; +}; + +struct nvgpu_nvlink_device_list { + bool valid; + u8 device_type; + u8 device_id; + u8 device_version; + u32 pri_base_addr; + u8 intr_enum; + u8 reset_enum; + u8 num_tx; + u8 num_rx; + u8 pll_master; + u8 pll_master_id; +}; + +enum nvgpu_nvlink_endp { + nvgpu_nvlink_endp_gpu, + nvgpu_nvlink_endp_tegra, + nvgpu_nvlink_endp__last, +}; + +enum nvgpu_nvlink_link_mode { + nvgpu_nvlink_link_off, + nvgpu_nvlink_link_hs, + nvgpu_nvlink_link_safe, + nvgpu_nvlink_link_fault, + nvgpu_nvlink_link_recovery, + nvgpu_nvlink_link_detect, + nvgpu_nvlink_link_reset, + nvgpu_nvlink_link_enable_pm, + nvgpu_nvlink_link_disable_pm, + nvgpu_nvlink_link_disable_err_detect, + nvgpu_nvlink_link_lane_disable, + nvgpu_nvlink_link_lane_shutdown, + nvgpu_nvlink_link__last, +}; + +enum nvgpu_nvlink_sublink_mode { + nvgpu_nvlink_sublink_tx_hs, + nvgpu_nvlink_sublink_tx_enable_pm, + nvgpu_nvlink_sublink_tx_disable_pm, + nvgpu_nvlink_sublink_tx_single_lane, + nvgpu_nvlink_sublink_tx_safe, + nvgpu_nvlink_sublink_tx_off, + nvgpu_nvlink_sublink_tx_common, + nvgpu_nvlink_sublink_tx_common_disable, + nvgpu_nvlink_sublink_tx_data_ready, + nvgpu_nvlink_sublink_tx_prbs_en, + nvgpu_nvlink_sublink_tx__last, + /* RX */ + nvgpu_nvlink_sublink_rx_hs, + nvgpu_nvlink_sublink_rx_enable_pm, + nvgpu_nvlink_sublink_rx_disable_pm, + nvgpu_nvlink_sublink_rx_single_lane, + nvgpu_nvlink_sublink_rx_safe, + nvgpu_nvlink_sublink_rx_off, + nvgpu_nvlink_sublink_rx_rxcal, + nvgpu_nvlink_sublink_rx__last, +}; + +struct nvgpu_nvlink_conn_info { + enum nvgpu_nvlink_endp device_type; + u32 link_number; + bool is_connected; +}; + +struct nvgpu_nvlink_link { + bool valid; + struct gk20a *g; + u8 link_id; + + u32 dlpl_base; + u8 dlpl_version; + + u32 tl_base; + u8 tl_version; + + u32 mif_base; + u8 mif_version; + + u8 intr_enum; + u8 reset_enum; + + bool dl_init_done; + + u8 pll_master_link_id; + u8 pll_slave_link_id; + + struct nvgpu_nvlink_conn_info remote_info; + void *priv; +}; + +#define NVLINK_MAX_LINKS_SW 6 + +enum nvgpu_nvlink_speed { + nvgpu_nvlink_speed_default, + nvgpu_nvlink_speed__last, +}; + +struct nvgpu_nvlink_dev { + struct nvgpu_nvlink_ioctrl_list *ioctrl_table; + u32 io_num_entries; + + struct nvgpu_nvlink_device_list *device_table; + u32 num_devices; + + struct nvgpu_nvlink_link links[NVLINK_MAX_LINKS_SW]; + + u8 dlpl_type; + u32 dlpl_base[NVLINK_MAX_LINKS_SW]; + + u8 tl_type; + u32 tl_base[NVLINK_MAX_LINKS_SW]; + + u8 mif_type; + u32 mif_base[NVLINK_MAX_LINKS_SW]; + + u8 ipt_type; + u32 ipt_base; + u8 ipt_version; + + u8 dlpl_multicast_type; + u8 dlpl_multicast_version; + u32 dlpl_multicast_base; + + u8 tl_multicast_type; + u8 tl_multicast_version; + u32 tl_multicast_base; + + u8 mif_multicast_type; + u8 mif_multicast_version; + u32 mif_multicast_base; + + u8 ioctrl_type; + u32 ioctrl_base; + + u8 minion_type; + u32 minion_base; + u8 minion_version; + + u32 discovered_links; + + /* VBIOS settings */ + u32 link_disable_mask; + u32 link_mode_mask; + u32 link_refclk_mask; + u8 train_at_boot; + u32 ac_coupling_mask; + + u32 init_disabled_links; + u32 connected_links; + u32 initialized_links; + u32 enabled_links; + u32 topology_connected_links; + + enum nvgpu_nvlink_speed speed; + + /* hshub settings */ + u32 hshub_config0; + u32 hshub_config1; + u32 hshub_config2; + u32 hshub_config6; + /* priv struct */ + void *priv; +}; + + +u32 nvgpu_nvlink_enumerate(struct gk20a *g); +u32 nvgpu_nvlink_train(struct gk20a *g, u32 link_id, bool from_off); +u32 nvgpu_nvlink_probe(struct gk20a *g); +#endif -- cgit v1.2.2