Files
linux-nvgpu/drivers/gpu/nvgpu/include/nvgpu/engines.h
Santosh BS 2a865e3aad gpu: nvgpu: NVENC support on TU104
This patch adds nvenc support for TU104
- Fetch engine/dev info for nvenc
- Falcon NS boot (fw loading) support
- Engine context creation for nvenc
- Skip golden image for multimedia engines
- Avoid subctx for nvenc as it is a non-VEID engine
- Job submission/flow changes for nvenc
- Code refactoring to scale up the support for other multimedia
  engines in future.

Bug 3763551

Change-Id: I03d4e731ebcef456bcc5ce157f3aa39883270dc0
Signed-off-by: Santosh BS <santoshb@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2859416
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Rajesh Devaraj <rdevaraj@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
2023-03-24 17:07:49 -07:00

460 lines
18 KiB
C

/*
* Copyright (c) 2019-2023, 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_ENGINE_H
#define NVGPU_ENGINE_H
/**
* @file
*
* Abstract interface for engine related functionality.
*/
#include <nvgpu/types.h>
struct gk20a;
struct nvgpu_device;
/** @cond DOXYGEN_SHOULD_SKIP_THIS */
#define ENGINE_PBDMA_INSTANCE0 0U
int nvgpu_engine_init_one_dev_extra(struct gk20a *g,
const struct nvgpu_device *dev);
/** @endcond DOXYGEN_SHOULD_SKIP_THIS */
/**
* Invalid engine id value.
*/
#define NVGPU_INVALID_ENG_ID (~U32(0U))
struct gk20a;
struct nvgpu_fifo;
struct nvgpu_device;
/**
* Engine enum types used for s/w purpose. These enum values are
* different as compared to engine enum types defined by h/w.
* Refer device.h header file.
*/
enum nvgpu_fifo_engine {
/** GR engine enum */
NVGPU_ENGINE_GR = 0U,
/** GR CE engine enum */
NVGPU_ENGINE_GRCE = 1U,
/** Async CE engine enum */
NVGPU_ENGINE_ASYNC_CE = 2U,
/** NVENC engine enum */
NVGPU_ENGINE_NVENC = 3U,
/** Invalid engine enum */
NVGPU_ENGINE_INVAL = 4U,
};
/**
* @brief Get s/w defined engine enum type for engine enum type defined by h/w.
* See device.h for engine enum types defined by h/w.
*
* @param g [in] The GPU driver struct.
* @param dev [in] Device to check.
*
* This is used to map engine enum type defined by h/w to engine enum type
* defined by s/w.
*
* @return S/w defined valid engine enum type < #NVGPU_ENGINE_INVAL.
* @retval #NVGPU_ENGINE_INVAL in case h/w #engine_type
* does not support gr (graphics) and ce (copy engine) engine enum
* types or if #engine_type does not match with h/w defined engine enum
* types for gr and/or ce engines.
*/
enum nvgpu_fifo_engine nvgpu_engine_enum_from_dev(struct gk20a *g,
const struct nvgpu_device *dev);
/**
* @brief Get pointer to #nvgpu_device for the h/w engine id.
*
* @param g [in] The GPU driver struct.
* @param engine_id [in] Active (h/w) Engine id.
*
* If #engine_id is one of the supported h/w engine ids, get pointer to
* #nvgpu_engine_info from an array of structures that is indexed by h/w
* engine id.
*
* @return Pointer to #nvgpu_device.
* @retval NULL if #g is NULL.
* @retval NULL if engine_id is not less than max supported number of engines
* i.e. #nvgpu_fifo.max_engines or if #engine_id does not match with
* any engine id supported by h/w or number of available engines
* i.e. #nvgpu_fifo.num_engines is 0.
*/
const struct nvgpu_device *nvgpu_engine_get_active_eng_info(
struct gk20a *g, u32 engine_id);
/**
* @brief Check if engine id is one of the supported h/w engine ids.
*
* @param g [in] The GPU driver struct.
* @param engine_id [in] Engine id.
*
* Check if #engine_id is one of the supported active engine ids.
*
* @return True if #engine_id is valid.
* @retval False if engine_id is not less than maximum number of supported
* engines on the chip i.e. #nvgpu_fifo.max_engines or if engine id
* does not match with any of the engine ids supported by h/w.
*/
bool nvgpu_engine_check_valid_id(struct gk20a *g, u32 engine_id);
/**
* @brief Get h/w engine id based on engine's instance identification number
* #NVGPU_ENGINE_GR engine enum type.
*
* @param g [in] The GPU driver struct.
* @param inst_id [in] Engine's instance identification number.
*
* @return H/W engine id for #NVGPU_ENGINE_GR engine enum type.
* @retval #NVGPU_INVALID_ENG_ID if #NVGPU_ENGINE_GR engine enum type could not
* be found in the set of available h/w engine ids.
*/
u32 nvgpu_engine_get_gr_id_for_inst(struct gk20a *g, u32 inst_id);
/**
* @brief Get instance count and first available h/w engine id for
* #NVGPU_ENGINE_GR engine enum type.
*
* @param g [in] The GPU driver struct.
*
* Call #nvgpu_engine_get_ids to get first available #NVGPU_ENGINE_GR
* engine enum type.
*
* @return First available h/w engine id for #NVGPU_ENGINE_GR engine enum type.
* @retval #NVGPU_INVALID_ENG_ID if #NVGPU_ENGINE_GR engine enum type could not
* be found in the set of available h/w engine ids.
*/
u32 nvgpu_engine_get_gr_id(struct gk20a *g);
/**
* @brief Get intr mask for the GR engine supported by the chip.
*
* @param g[in] The GPU driver struct.
*
* Return bitmask of each GR engine's interrupt bit.
*
* @return Interrupt mask for GR engine.
*/
u32 nvgpu_gr_engine_interrupt_mask(struct gk20a *g);
/**
* @brief Get intr mask for the CE engines supported by the chip.
*
* @param g [in] The GPU driver struct.
*
* Query all types of copy engine devices and OR their interrupt bits into
* a CE interrupt mask.
*
* @return 0U if there is no CE support in the system.
* @return The logical OR of all interrupt bits for all CE devices present.
*/
u32 nvgpu_ce_engine_interrupt_mask(struct gk20a *g);
/**
* @brief Get intr mask for the device corresponding the provided engine_id.
*
* @param g [in] The GPU driver struct.
* @param engine_id [in] HW engine_id.
*
* Return the interrupt mask for the host device corresponding to \a engine_id.
*
* @return Intr mask for the #engine_id or 0 if the engine_id does not have a
* corresponding device.
*/
u32 nvgpu_engine_act_interrupt_mask(struct gk20a *g, u32 engine_id);
/**
* @brief Allocate and initialize s/w context for engine related info.
*
* @param g [in] The GPU driver struct.
*
* - Get max number of engines supported on the chip from h/w config register.
* - Allocate kernel memory area for storing engine info for max number of
* engines supported on the chip. This allocated area of memory is set to 0
* and then filled with value read from device info h/w registers.
* Also this area of memory is indexed by h/w engine id.
* - Allocate kernel memory area for max number of engines supported on the
* chip to map s/w defined engine ids starting from 0 to h/w (active) engine
* id read from device info h/w register. This allocated area of memory is
* set to 0xff and then filled with engine id read from device info h/w
* registers. This area of memory is indexed by s/w defined engine id starting
* from 0.
* - Initialize engine info related s/w context by calling hal that will read
* device info h/w registers and also initialize s/w variable
* #nvgpu_fifo.num_engines that is used to count total number of valid h/w
* engine ids read from device info h/w registers.
*
* @return 0 upon success, < 0 otherwise.
* @retval -ENOMEM upon failure to allocate memory for engine structure.
* @retval -EINVAL upon failure to get engine info from device info h/w
* registers.
*/
int nvgpu_engine_setup_sw(struct gk20a *g);
/**
* @brief Clean up s/w context for engine related info.
*
* @param g [in] The GPU driver struct.
*
* - Free up kernel memory area that is used for storing engine info read
* from device info h/w registers.
* - Free up kernel memory area to map s/w defined engine ids (starting with 0)
* to active (h/w) engine ids read from device info h/w register.
*/
void nvgpu_engine_cleanup_sw(struct gk20a *g);
#ifdef CONFIG_NVGPU_FIFO_ENGINE_ACTIVITY
void nvgpu_engine_enable_activity_all(struct gk20a *g);
int nvgpu_engine_disable_activity(struct gk20a *g,
const struct nvgpu_device *dev,
bool wait_for_idle);
int nvgpu_engine_disable_activity_all(struct gk20a *g,
bool wait_for_idle);
int nvgpu_engine_wait_for_idle(struct gk20a *g);
#endif
#ifdef CONFIG_NVGPU_ENGINE_RESET
/**
* Called from recovery. This will not be part of the safety build after
* recovery is not supported in the safety build.
*/
void nvgpu_engine_reset(struct gk20a *g, u32 engine_id);
#endif
/**
* @brief Get runlist id for the last available #NVGPU_ENGINE_ASYNC_CE
* engine enum type.
*
* @param g [in] The GPU driver struct.
*
* - Call #nvgpu_engine_get_gr_runlist_id to get runlist id for the first
* available #NVGPU_ENGINE_GR engine type.
* - Get #nvgpu_engine_info.runlist_id for last available #NVGPU_ENGINE_ASYNC_CE
* engine enum type.
* - Get pointer to #nvgpu_engine_info for each of #nvgpu_fifo.num_engines.
* - Get #nvgpu_engine_info.runlist_id for the #nvgpu_engine_info.engine_enum
* type matching with #NVGPU_ENGINE_ASYNC_CE.
*
* @return #nvgpu_engine_info.runlist_id for the last available
* #NVGPU_ENGINE_ASYNC_CE engine enum type.
* @retval Return value of #nvgpu_engine_get_gr_runlist_id if #g is NULL.
* @retval Return value of #nvgpu_engine_get_gr_runlist_id if
* #NVGPU_ENGINE_ASYNC_CE engine enum type is not available.
* @retval Return value of #nvgpu_engine_get_gr_runlist_id if
* #nvgpu_fifo.num_engines is 0.
*/
u32 nvgpu_engine_get_fast_ce_runlist_id(struct gk20a *g);
/**
* @brief Get runlist id for the first available #NVGPU_ENGINE_GR engine enum
* type.
*
* @param g [in] The GPU driver struct.
*
* - Get h/w engine id for the first available #NVGPU_ENGINE_GR engine enum
* type.
* -- Get #nvgpu_engine_info for the first available gr engine id.
* -- Get #nvgpu_engine_info.runlist_id for first available gr engine id.
*
* @return #nvgpu_engine_info.runlist_id for the first available gr engine id.
* @retval U32_MAX if #NVGPU_ENGINE_GR engine enum type is not available.
* @retval U32_MAX if pointer to #nvgpu_engine_info for the first available
* gr h/w engine id is NULL.
*/
u32 nvgpu_engine_get_gr_runlist_id(struct gk20a *g);
/**
* @brief Get runlist id for the first available #NVGPU_ENGINE_NVENC engine enum
* type.
*
* @param g [in] The GPU driver struct.
*
* - Get h/w engine id for the first available #NVGPU_ENGINE_NVENC engine enum
* type.
* -- Get #nvgpu_engine_info for the first available nvenc engine id.
* -- Get #nvgpu_engine_info.runlist_id for first available nvenc engine id.
*
* @return #nvgpu_engine_info.runlist_id for the first available nvenc engine id.
* @retval U32_MAX if #NVGPU_ENGINE_NVENC engine enum type is not available.
* @retval U32_MAX if pointer to #nvgpu_engine_info for the first available
* nvenc h/w engine id is NULL.
*/
u32 nvgpu_engine_get_nvenc_runlist_id(struct gk20a *g);
/**
* @brief Check if runlist id corresponds to runlist id of one of the
* engine ids supported by h/w.
*
* @param g [in] The GPU driver struct.
* @param runlist_id [in] Runlist id.
*
* Check if #runlist_id corresponds to runlist id of one of the engine
* ids supported by h/w by checking #nvgpu_engine_info for each of
* #nvgpu_fifo.num_engines engines.
*
* @return True if #runlist_id is valid.
* @return False if #nvgpu_engine_info is NULL for all the engine ids starting
* with 0 upto #nvgpu_fifo.num_engines or #runlist_id did not match with
* any of the runlist ids of engine ids supported by h/w.
*/
bool nvgpu_engine_is_valid_runlist_id(struct gk20a *g, u32 runlist_id);
/**
* @brief Check if the given runlist id belongs to one of the multimedia engines
* supported by the h/w.
*
* @param g [in] The GPU driver struct.
* @param runlist_id [in] Runlist id.
*
* Check if #runlist_id corresponds to runlist id of one of the multimedia engines
* supported by h/w, by checking #nvgpu_engine_info for each of the supported
* multimedia engines in #nvgpu_fifo.num_engines engines.
*
* @return True if #runlist_id is valid.
* @return False if #nvgpu_engine_info is NULL for all multimedia engines starting
* with 0 up to #nvgpu_fifo.num_engines or #runlist_id did not match with
* any of the runlist ids of multimedia engine ids supported by h/w.
*/
bool nvgpu_engine_is_multimedia_runlist_id(struct gk20a *g, u32 runlist_id);
/**
* @brief Get mmu fault id for the engine id.
*
* @param g [in] The GPU driver struct.
* @param engine_id [in] Engine id.
*
* Get pointer to #nvgpu_engine_info for the #engine_id. Use this pointer to
* get mmu fault id for the #engine_id.
*
* @return Mmu fault id for #engine_id.
* @retval Invalid mmu fault id, #NVGPU_INVALID_ENG_ID.
*/
u32 nvgpu_engine_id_to_mmu_fault_id(struct gk20a *g, u32 engine_id);
/**
* @brief Get engine id from mmu fault id.
*
* @param g [in] The GPU driver struct.
* @param fault_id [in] Mmu fault id.
*
* Check if #fault_id corresponds to fault id of one of the active engine
* ids supported by h/w by checking #nvgpu_engine_info for each of
* #nvgpu_fifo.num_engines engines.
*
* @return Valid engine id corresponding to #fault_id.
* @retval Invalid engine id, #NVGPU_INVALID_ENG_ID if pointer to
* #nvgpu_engine_info is NULL for the engine ids starting with
* 0 upto #nvgpu_fifo.num_engines or
* #fault_id did not match with any of the fault ids of h/w engine ids.
*/
u32 nvgpu_engine_mmu_fault_id_to_engine_id(struct gk20a *g, u32 fault_id);
/**
* Called from recovery. This will not be part of the safety build after
* recovery is not supported in the safety build.
*/
u32 nvgpu_engine_get_mask_on_id(struct gk20a *g, u32 id, bool is_tsg);
/**
* @brief Read device info h/w registers to get engine info.
*
* @param f [in] Pointer to #nvgpu_fifo struct.
*
* - Get device info related info for h/w engine enum type,
* #NVGPU_DEVTYPE_GRAPHICS.
* - Get PBDMA id serving the runlist id of the h/w engine enum type,
* #NVGPU_DEVTYPE_GRAPHICS.
* - Get s/w defined engine enum type for the h/w engine enum type read
* from device info registers.
* - Initialize #nvgpu_fifo.engine_info and #nvgpu_fifo.active_engines_list
* with the data from local device info struct and also initialize s/w
* variable.
* #nvgpu_fifo.num_engines that is used to count total number of valid h/w
* engine ids read from device info h/w registers.
* - Call function to initialize CE engine info.
*
* @return 0 upon success.
* @retval -EINVAL if call to function to get device info related info for
* h/w engine enum type, #NVGPU_DEVTYPE_GRAPHICS returned failure.
* @retval -EINVAL if call to function to get pbdma id for runlist id of
* h/w engine enum type, #NVGPU_DEVTYPE_GRAPHICS returned failure.
* @retval Return value of function called to initialize CE engine info.
*/
int nvgpu_engine_init_info(struct nvgpu_fifo *f);
/**
* Called from recovery handling for architectures before volta. This will
* not be part of safety build after recovery is not supported in the safety
* build.
*/
void nvgpu_engine_get_id_and_type(struct gk20a *g, u32 engine_id,
u32 *id, u32 *type);
/**
* Called from ctxsw timeout intr handling. This function will not be part
* of safety build after recovery is not supported in the safety build.
*/
u32 nvgpu_engine_find_busy_doing_ctxsw(struct gk20a *g,
u32 *id_ptr, bool *is_tsg_ptr);
/**
* Called from runlist update timeout handling. This function will not be part
* of safety build after recovery is not supported in safety build.
*/
u32 nvgpu_engine_get_runlist_busy_engines(struct gk20a *g, u32 runlist_id);
#ifdef CONFIG_NVGPU_DEBUGGER
bool nvgpu_engine_should_defer_reset(struct gk20a *g, u32 engine_id,
u32 engine_subid, bool fake_fault);
#endif
/**
* @brief Get veid from mmu fault id.
*
* @param g [in] The GPU driver struct.
* @param mmu_fault_id [in] Mmu fault id.
* @param gr_eng_fault_id [in] GR engine's mmu fault id.
*
* Get valid veid by subtracting #gr_eng_fault_id from #mmu_fault_id,
* if #mmu_fault_id is greater than or equal to #gr_eng_fault_id and less
* than #gr_eng_fault_id + #nvgpu_fifo.max_subctx_count.
*
* @return Veid.
* @retval Invalid veid, #INVAL_ID.
*/
u32 nvgpu_engine_mmu_fault_id_to_veid(struct gk20a *g, u32 mmu_fault_id,
u32 gr_eng_fault_id);
/**
* @brief Get engine id, veid and pbdma id from mmu fault id.
*
* @param g [in] The GPU driver struct.
* @param mmu_fault_id [in] Mmu fault id.
* @param engine_id [in,out] Pointer to store active engine id.
* @param veid [in,out] Pointer to store veid.
* @param pbdma_id [in,out] Pointer to store pbdma id.
*
* Calls function to get h/w engine id and veid for given #mmu_fault_id.
* If h/w (active) engine id is not #INVAL_ID, call function to get pbdma id for
* the engine having fault id as #mmu_fault_id.
*
* @return Updated #engine_id, #veid and #pbdma_id pointers
*/
void nvgpu_engine_mmu_fault_id_to_eng_ve_pbdma_id(struct gk20a *g,
u32 mmu_fault_id, u32 *engine_id, u32 *veid, u32 *pbdma_id);
/**
* @brief Remove a device entry from engine list.
*
* @param g [in] The GPU driver struct.
* @param dev [in] A device.
*
* Remove the device entry \a dev from fifo->host_engines, fifo->active_engines.
* The device entry is retained in g->devs->devlist_heads list to ensure device
* reset.
*/
void nvgpu_engine_remove_one_dev(struct nvgpu_fifo *f,
const struct nvgpu_device *dev);
#endif /*NVGPU_ENGINE_H*/