/*
 * NVGPU Public Interface Header
 *
 * Copyright (c) 2011-2018, 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.
 */

#ifndef _UAPI__LINUX_NVGPU_IOCTL_H
#define _UAPI__LINUX_NVGPU_IOCTL_H

#include <linux/ioctl.h>
#include <linux/types.h>

#if !defined(__KERNEL__)
#define __user
#endif

/*
 * /dev/nvhost-ctrl-gpu device
 *
 * Opening a '/dev/nvhost-ctrl-gpu' device node creates a way to send
 * ctrl ioctl to gpu driver.
 *
 * /dev/nvhost-gpu is for channel (context specific) operations. We use
 * /dev/nvhost-ctrl-gpu for global (context independent) operations on
 * gpu device.
 */

#define NVGPU_GPU_IOCTL_MAGIC 'G'

/* return zcull ctx size */
struct nvgpu_gpu_zcull_get_ctx_size_args {
	__u32 size;
} __packed;

/* return zcull info */
struct nvgpu_gpu_zcull_get_info_args {
	__u32 width_align_pixels;
	__u32 height_align_pixels;
	__u32 pixel_squares_by_aliquots;
	__u32 aliquot_total;
	__u32 region_byte_multiplier;
	__u32 region_header_size;
	__u32 subregion_header_size;
	__u32 subregion_width_align_pixels;
	__u32 subregion_height_align_pixels;
	__u32 subregion_count;
};

#define NVGPU_ZBC_COLOR_VALUE_SIZE	4
#define NVGPU_ZBC_TYPE_INVALID		0
#define NVGPU_ZBC_TYPE_COLOR		1
#define NVGPU_ZBC_TYPE_DEPTH		2

struct nvgpu_gpu_zbc_set_table_args {
	__u32 color_ds[NVGPU_ZBC_COLOR_VALUE_SIZE];
	__u32 color_l2[NVGPU_ZBC_COLOR_VALUE_SIZE];
	__u32 depth;
	__u32 format;
	__u32 type;	/* color or depth */
} __packed;

struct nvgpu_gpu_zbc_query_table_args {
	__u32 color_ds[NVGPU_ZBC_COLOR_VALUE_SIZE];
	__u32 color_l2[NVGPU_ZBC_COLOR_VALUE_SIZE];
	__u32 depth;
	__u32 ref_cnt;
	__u32 format;
	__u32 type;		/* color or depth */
	__u32 index_size;	/* [out] size, [in] index */
} __packed;


/* This contains the minimal set by which the userspace can
   determine all the properties of the GPU */
#define NVGPU_GPU_ARCH_GK100	0x000000E0
#define NVGPU_GPU_ARCH_GM200	0x00000120
#define NVGPU_GPU_ARCH_GP100	0x00000130
#define NVGPU_GPU_ARCH_GV110	0x00000150
#define NVGPU_GPU_ARCH_GV100	0x00000140

#define NVGPU_GPU_IMPL_GK20A	0x0000000A
#define NVGPU_GPU_IMPL_GM204	0x00000004
#define NVGPU_GPU_IMPL_GM206	0x00000006
#define NVGPU_GPU_IMPL_GM20B	0x0000000B
#define NVGPU_GPU_IMPL_GM20B_B	0x0000000E
#define NVGPU_GPU_IMPL_GP104	0x00000004
#define NVGPU_GPU_IMPL_GP106	0x00000006
#define NVGPU_GPU_IMPL_GP10B	0x0000000B
#define NVGPU_GPU_IMPL_GV11B	0x0000000B
#define NVGPU_GPU_IMPL_GV100	0x00000000

#define NVGPU_GPU_BUS_TYPE_NONE         0
#define NVGPU_GPU_BUS_TYPE_AXI         32

#define NVGPU_GPU_FLAGS_HAS_SYNCPOINTS			(1ULL << 0)
/* MAP_BUFFER_EX with partial mappings */
#define NVGPU_GPU_FLAGS_SUPPORT_PARTIAL_MAPPINGS	(1ULL << 1)
/* MAP_BUFFER_EX with sparse allocations */
#define NVGPU_GPU_FLAGS_SUPPORT_SPARSE_ALLOCS		(1ULL << 2)
/* sync fence FDs are available in, e.g., submit_gpfifo */
#define NVGPU_GPU_FLAGS_SUPPORT_SYNC_FENCE_FDS		(1ULL << 3)
/* NVGPU_IOCTL_CHANNEL_CYCLE_STATS is available */
#define NVGPU_GPU_FLAGS_SUPPORT_CYCLE_STATS		(1ULL << 4)
/* NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT is available */
#define NVGPU_GPU_FLAGS_SUPPORT_CYCLE_STATS_SNAPSHOT	(1ULL << 6)
/* User-space managed address spaces support */
#define NVGPU_GPU_FLAGS_SUPPORT_USERSPACE_MANAGED_AS	(1ULL << 7)
/* Both gpu driver and device support TSG */
#define NVGPU_GPU_FLAGS_SUPPORT_TSG			(1ULL << 8)
/* Clock control support */
#define NVGPU_GPU_FLAGS_SUPPORT_CLOCK_CONTROLS		(1ULL << 9)
/* NVGPU_GPU_IOCTL_GET_VOLTAGE is available */
#define NVGPU_GPU_FLAGS_SUPPORT_GET_VOLTAGE		(1ULL << 10)
/* NVGPU_GPU_IOCTL_GET_CURRENT is available */
#define NVGPU_GPU_FLAGS_SUPPORT_GET_CURRENT		(1ULL << 11)
/* NVGPU_GPU_IOCTL_GET_POWER is available */
#define NVGPU_GPU_FLAGS_SUPPORT_GET_POWER		(1ULL << 12)
/* NVGPU_GPU_IOCTL_GET_TEMPERATURE is available */
#define NVGPU_GPU_FLAGS_SUPPORT_GET_TEMPERATURE		(1ULL << 13)
/* NVGPU_GPU_IOCTL_SET_THERM_ALERT_LIMIT is available */
#define NVGPU_GPU_FLAGS_SUPPORT_SET_THERM_ALERT_LIMIT	(1ULL << 14)
/* NVGPU_GPU_IOCTL_GET_EVENT_FD is available */
#define NVGPU_GPU_FLAGS_SUPPORT_DEVICE_EVENTS		(1ULL << 15)
/* FECS context switch tracing is available */
#define NVGPU_GPU_FLAGS_SUPPORT_FECS_CTXSW_TRACE	(1ULL << 16)
/* NVGPU_AS_IOCTL_MAP_BUFFER_COMPBITS is available */
#define NVGPU_GPU_FLAGS_SUPPORT_MAP_COMPBITS		(1ULL << 17)
/* Fast deterministic submits with no job tracking are supported */
#define NVGPU_GPU_FLAGS_SUPPORT_DETERMINISTIC_SUBMIT_NO_JOBTRACKING (1ULL << 18)
/* Deterministic submits are supported even with job tracking */
#define NVGPU_GPU_FLAGS_SUPPORT_DETERMINISTIC_SUBMIT_FULL (1ULL << 19)
/* IO coherence support is available */
#define NVGPU_GPU_FLAGS_SUPPORT_IO_COHERENCE		(1ULL << 20)
/* NVGPU_SUBMIT_GPFIFO_FLAGS_RESCHEDULE_RUNLIST is available */
#define NVGPU_GPU_FLAGS_SUPPORT_RESCHEDULE_RUNLIST	(1ULL << 21)
/*  subcontexts are available */
#define NVGPU_GPU_FLAGS_SUPPORT_TSG_SUBCONTEXTS         (1ULL << 22)
/* Direct PTE kind control is supported (map_buffer_ex) */
#define NVGPU_GPU_FLAGS_SUPPORT_MAP_DIRECT_KIND_CTRL	(1ULL << 23)
/* NVGPU_GPU_IOCTL_SET_DETERMINISTIC_OPTS is available */
#define NVGPU_GPU_FLAGS_SUPPORT_DETERMINISTIC_OPTS	(1ULL << 24)
/* SCG support is available */
#define NVGPU_GPU_FLAGS_SUPPORT_SCG			(1ULL << 25)
/* GPU_VA address of a syncpoint is supported */
#define NVGPU_GPU_FLAGS_SUPPORT_SYNCPOINT_ADDRESS	(1ULL << 26)
/* VPR is supported */
#define NVGPU_GPU_FLAGS_SUPPORT_VPR			(1ULL << 27)
/* Allocating per-channel syncpoint in user space is supported */
#define NVGPU_GPU_FLAGS_SUPPORT_USER_SYNCPOINT		(1ULL << 28)
/* SM LRF ECC is enabled */
#define NVGPU_GPU_FLAGS_ECC_ENABLED_SM_LRF	(1ULL << 60)
/* SM SHM ECC is enabled */
#define NVGPU_GPU_FLAGS_ECC_ENABLED_SM_SHM	(1ULL << 61)
/* TEX ECC is enabled */
#define NVGPU_GPU_FLAGS_ECC_ENABLED_TEX		(1ULL << 62)
/* L2 ECC is enabled */
#define NVGPU_GPU_FLAGS_ECC_ENABLED_LTC		(1ULL << 63)
/* All types of ECC are enabled */
#define NVGPU_GPU_FLAGS_ALL_ECC_ENABLED	\
				(NVGPU_GPU_FLAGS_ECC_ENABLED_SM_LRF |	\
				NVGPU_GPU_FLAGS_ECC_ENABLED_SM_SHM |	\
				NVGPU_GPU_FLAGS_ECC_ENABLED_TEX    |	\
				NVGPU_GPU_FLAGS_ECC_ENABLED_LTC)

struct nvgpu_gpu_characteristics {
	__u32 arch;
	__u32 impl;
	__u32 rev;
	__u32 num_gpc;

	__u64 L2_cache_size;               /* bytes */
	__u64 on_board_video_memory_size;  /* bytes */

	__u32 num_tpc_per_gpc; /* the architectural maximum */
	__u32 bus_type;

	__u32 big_page_size; /* the default big page size */
	__u32 compression_page_size;

	__u32 pde_coverage_bit_count;

	/* bit N set ==> big page size 2^N is available in
	   NVGPU_GPU_IOCTL_ALLOC_AS. The default big page size is
	   always available regardless of this field. */
	__u32 available_big_page_sizes;

	__u64 flags;

	__u32 twod_class;
	__u32 threed_class;
	__u32 compute_class;
	__u32 gpfifo_class;
	__u32 inline_to_memory_class;
	__u32 dma_copy_class;

	__u32 gpc_mask; /* enabled GPCs */

	__u32 sm_arch_sm_version; /* sm version */
	__u32 sm_arch_spa_version; /* sm instruction set */
	__u32 sm_arch_warp_count;

	/* IOCTL interface levels by service. -1 if not supported */
	__s16 gpu_ioctl_nr_last;
	__s16 tsg_ioctl_nr_last;
	__s16 dbg_gpu_ioctl_nr_last;
	__s16 ioctl_channel_nr_last;
	__s16 as_ioctl_nr_last;

	__u8 gpu_va_bit_count;
	__u8 reserved;

	__u32 max_fbps_count;
	__u32 fbp_en_mask;
	__u32 max_ltc_per_fbp;
	__u32 max_lts_per_ltc;
	__u32 max_tex_per_tpc;
	__u32 max_gpc_count;
	/* mask of Rop_L2 for each FBP */
	__u32 rop_l2_en_mask_DEPRECATED[2];


	__u8 chipname[8];

	__u64 gr_compbit_store_base_hw;
	__u32 gr_gobs_per_comptagline_per_slice;
	__u32 num_ltc;
	__u32 lts_per_ltc;
	__u32 cbc_cache_line_size;
	__u32 cbc_comptags_per_line;

	/* MAP_BUFFER_BATCH: the upper limit for num_unmaps and
	 * num_maps */
	__u32 map_buffer_batch_limit;

	__u64 max_freq;

	/* supported preemption modes */
	__u32 graphics_preemption_mode_flags; /* NVGPU_GRAPHICS_PREEMPTION_MODE_* */
	__u32 compute_preemption_mode_flags; /* NVGPU_COMPUTE_PREEMPTION_MODE_* */
	/* default preemption modes */
	__u32 default_graphics_preempt_mode; /* NVGPU_GRAPHICS_PREEMPTION_MODE_* */
	__u32 default_compute_preempt_mode; /* NVGPU_COMPUTE_PREEMPTION_MODE_* */

	__u64 local_video_memory_size; /* in bytes, non-zero only for dGPUs */

	/* These are meaningful only for PCI devices */
	__u16 pci_vendor_id, pci_device_id;
	__u16 pci_subsystem_vendor_id, pci_subsystem_device_id;
	__u16 pci_class;
	__u8  pci_revision;
	__u8  vbios_oem_version;
	__u32 vbios_version;

	/* NVGPU_DBG_GPU_IOCTL_REG_OPS: the upper limit for the number
	 * of regops */
	__u32 reg_ops_limit;
	__u32 reserved1;

	__s16 event_ioctl_nr_last;
	__u16 pad;

	__u32 max_css_buffer_size;

	/* Notes:
	   - This struct can be safely appended with new fields. However, always
	     keep the structure size multiple of 8 and make sure that the binary
	     layout does not change between 32-bit and 64-bit architectures.
	   - If the last field is reserved/padding, it is not
	     generally safe to repurpose the field in future revisions.
	*/
};

struct nvgpu_gpu_get_characteristics {
	/* [in]  size reserved by the user space. Can be 0.
	   [out] full buffer size by kernel */
	__u64 gpu_characteristics_buf_size;

	/* [in]  address of nvgpu_gpu_characteristics buffer. Filled with field
	   values by exactly MIN(buf_size_in, buf_size_out) bytes. Ignored, if
	   buf_size_in is zero.  */
	__u64 gpu_characteristics_buf_addr;
};

#define NVGPU_GPU_COMPBITS_NONE		0
#define NVGPU_GPU_COMPBITS_GPU		(1 << 0)
#define NVGPU_GPU_COMPBITS_CDEH		(1 << 1)
#define NVGPU_GPU_COMPBITS_CDEV		(1 << 2)

struct nvgpu_gpu_prepare_compressible_read_args {
	__u32 handle;			/* in, dmabuf fd */
	union {
		__u32 request_compbits;	/* in */
		__u32 valid_compbits;	/* out */
	};
	__u64 offset;			/* in, within handle */
	__u64 compbits_hoffset;		/* in, within handle */
	__u64 compbits_voffset;		/* in, within handle */
	__u32 width;			/* in, in pixels */
	__u32 height;			/* in, in pixels */
	__u32 block_height_log2;	/* in */
	__u32 submit_flags;		/* in (NVGPU_SUBMIT_GPFIFO_FLAGS_) */
	union {
		struct {
			__u32 syncpt_id;
			__u32 syncpt_value;
		};
		__s32 fd;
	} fence;			/* in/out */
	__u32 zbc_color;		/* out */
	__u32 reserved;		/* must be zero */
	__u64 scatterbuffer_offset;	/* in, within handle */
	__u32 reserved2[2];		/* must be zero */
};

struct nvgpu_gpu_mark_compressible_write_args {
	__u32 handle;			/* in, dmabuf fd */
	__u32 valid_compbits;		/* in */
	__u64 offset;			/* in, within handle */
	__u32 zbc_color;		/* in */
	__u32 reserved[3];		/* must be zero */
};

struct nvgpu_alloc_as_args {
	__u32 big_page_size;
	__s32 as_fd;

/*
 * The GPU address space will be managed by the userspace. This has
 * the following changes in functionality:
 *   1. All non-fixed-offset user mappings are rejected (i.e.,
 *      fixed-offset only)
 *   2. Address space does not need to be allocated for fixed-offset
 *      mappings, except to mark sparse address space areas.
 *   3. Maps and unmaps are immediate. In particular, mapping ref
 *      increments at kickoffs and decrements at job completion are
 *      bypassed.
 */
#define NVGPU_GPU_IOCTL_ALLOC_AS_FLAGS_USERSPACE_MANAGED (1 << 0)
	__u32 flags;

	__u32 reserved;			/* must be zero */
};

struct nvgpu_gpu_open_tsg_args {
	__u32 tsg_fd;			/* out, tsg fd */
	__u32 reserved;			/* must be zero */
};

struct nvgpu_gpu_get_tpc_masks_args {
	/* [in]  TPC mask buffer size reserved by userspace. Should be
		 at least sizeof(__u32) * fls(gpc_mask) to receive TPC
		 mask for each GPC.
	   [out] full kernel buffer size
	*/
	__u32 mask_buf_size;
	__u32 reserved;

	/* [in]  pointer to TPC mask buffer. It will receive one
		 32-bit TPC mask per GPC or 0 if GPC is not enabled or
		 not present. This parameter is ignored if
		 mask_buf_size is 0. */
	__u64 mask_buf_addr;
};

struct nvgpu_gpu_open_channel_args {
	union {
		__s32 channel_fd; /* deprecated: use out.channel_fd instead */
		struct {
			 /* runlist_id is the runlist for the
			  * channel. Basically, the runlist specifies the target
			  * engine(s) for which the channel is
			  * opened. Runlist_id -1 is synonym for the primary
			  * graphics runlist. */
			__s32 runlist_id;
		} in;
		struct {
			__s32 channel_fd;
		} out;
	};
};

/* L2 cache writeback, optionally invalidate clean lines and flush fb */
struct nvgpu_gpu_l2_fb_args {
	__u32 l2_flush:1;
	__u32 l2_invalidate:1;
	__u32 fb_flush:1;
	__u32 reserved;
} __packed;

struct nvgpu_gpu_inval_icache_args {
	int channel_fd;
	__u32 reserved;
} __packed;

struct nvgpu_gpu_mmu_debug_mode_args {
	__u32 state;
	__u32 reserved;
} __packed;

struct nvgpu_gpu_sm_debug_mode_args {
	int channel_fd;
	__u32 enable;
	__u64 sms;
} __packed;

struct warpstate {
	__u64 valid_warps[2];
	__u64 trapped_warps[2];
	__u64 paused_warps[2];
};

struct nvgpu_gpu_wait_pause_args {
	__u64 pwarpstate;
};

struct nvgpu_gpu_tpc_exception_en_status_args {
	__u64 tpc_exception_en_sm_mask;
};

struct nvgpu_gpu_num_vsms {
	__u32 num_vsms;
	__u32 reserved;
};

struct nvgpu_gpu_vsms_mapping_entry {
	__u8 gpc_index;
	__u8 tpc_index;
};

struct nvgpu_gpu_vsms_mapping {
	__u64 vsms_map_buf_addr;
};

struct nvgpu_gpu_get_buffer_info_args {
	union {
		struct {
			__u32 dmabuf_fd;   /* dma-buf fd */
		} in;
		struct {
			__u64 id;          /* Unique within live
					    * buffers */
			__u64 length;      /* Allocated length of the
					    * buffer */
			__u64 reserved0;
			__u64 reserved1;
		} out;
	};
};

#define NVGPU_GPU_GET_CPU_TIME_CORRELATION_INFO_MAX_COUNT		16
#define NVGPU_GPU_GET_CPU_TIME_CORRELATION_INFO_SRC_ID_TSC		1

struct nvgpu_gpu_get_cpu_time_correlation_sample {
	/* gpu timestamp value */
	__u64 cpu_timestamp;
	/* raw GPU counter (PTIMER) value */
	__u64 gpu_timestamp;
};

struct nvgpu_gpu_get_cpu_time_correlation_info_args {
	/* timestamp pairs */
	struct nvgpu_gpu_get_cpu_time_correlation_sample samples[
		NVGPU_GPU_GET_CPU_TIME_CORRELATION_INFO_MAX_COUNT];
	/* number of pairs to read */
	__u32 count;
	/* cpu clock source id */
	__u32 source_id;
};

struct nvgpu_gpu_get_gpu_time_args {
	/* raw GPU counter (PTIMER) value */
	__u64 gpu_timestamp;

	/* reserved for future extensions */
	__u64 reserved;
};

struct nvgpu_gpu_get_engine_info_item {

#define NVGPU_GPU_ENGINE_ID_GR 0
#define NVGPU_GPU_ENGINE_ID_GR_COPY 1
#define NVGPU_GPU_ENGINE_ID_ASYNC_COPY 2
	__u32 engine_id;

	__u32 engine_instance;

	/* runlist id for opening channels to the engine, or -1 if
	 * channels are not supported */
	__s32 runlist_id;

	__u32 reserved;
};

struct nvgpu_gpu_get_engine_info_args {
	/* [in]  Buffer size reserved by userspace.
	 *
	 * [out] Full kernel buffer size. Multiple of sizeof(struct
	 *       nvgpu_gpu_get_engine_info_item)
	*/
	__u32 engine_info_buf_size;
	__u32 reserved;
	__u64 engine_info_buf_addr;
};

#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_CONTIGUOUS		(1U << 0)

/* CPU access and coherency flags (3 bits). Use CPU access with care,
 * BAR resources are scarce. */
#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_CPU_NOT_MAPPABLE	(0U << 1)
#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_CPU_WRITE_COMBINE	(1U << 1)
#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_CPU_CACHED		(2U << 1)
#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_CPU_MASK		(7U << 1)

#define NVGPU_GPU_ALLOC_VIDMEM_FLAG_VPR			(1U << 4)

/* Allocation of device-specific local video memory. Returns dmabuf fd
 * on success. */
struct nvgpu_gpu_alloc_vidmem_args {
	union {
		struct {
			/* Size for allocation. Must be a multiple of
			 * small page size. */
			__u64 size;

			/* NVGPU_GPU_ALLOC_VIDMEM_FLAG_* */
			__u32 flags;

			/* Informational mem tag for resource usage
			 * tracking. */
			__u16 memtag;

			__u16 reserved0;

			/* GPU-visible physical memory alignment in
			 * bytes.
			 *
			 * Alignment must be a power of two. Minimum
			 * alignment is the small page size, which 0
			 * also denotes.
			 *
			 * For contiguous and non-contiguous
			 * allocations, the start address of the
			 * physical memory allocation will be aligned
			 * by this value.
			 *
			 * For non-contiguous allocations, memory is
			 * internally allocated in round_up(size /
			 * alignment) contiguous blocks. The start
			 * address of each block is aligned by the
			 * alignment value. If the size is not a
			 * multiple of alignment (which is ok), the
			 * last allocation block size is (size %
			 * alignment).
			 *
			 * By specifying the big page size here and
			 * allocation size that is a multiple of big
			 * pages, it will be guaranteed that the
			 * allocated buffer is big page size mappable.
			 */
			__u32 alignment;

			__u32 reserved1[3];
		} in;

		struct {
			__s32 dmabuf_fd;
		} out;
	};
};

/* Memory clock */
#define NVGPU_GPU_CLK_DOMAIN_MCLK                                (0)
/* Main graphics core clock */
#define NVGPU_GPU_CLK_DOMAIN_GPCCLK	                         (1)

struct nvgpu_gpu_clk_range {

	/* Flags (not currently used) */
	__u32 flags;

	/* NVGPU_GPU_CLK_DOMAIN_* */
	__u32 clk_domain;
	__u64 min_hz;
	__u64 max_hz;
};

/* Request on specific clock domains */
#define NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS		(1UL << 0)

struct nvgpu_gpu_clk_range_args {

	/* Flags. If NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS the request will
	   apply only to domains specified in clock entries. In this case
	   caller must set clock domain in each entry. Otherwise, the
	   ioctl will return all clock domains.
	*/
	__u32 flags;

	__u16 pad0;

	/* in/out: Number of entries in clk_range_entries buffer. If zero,
	   NVGPU_GPU_IOCTL_CLK_GET_RANGE will return 0 and
	   num_entries will be set to number of clock domains.
	 */
	__u16 num_entries;

	/* in: Pointer to clock range entries in the caller's address space.
	   size must be >= max_entries * sizeof(struct nvgpu_gpu_clk_range)
	 */
	__u64 clk_range_entries;
};

struct nvgpu_gpu_clk_vf_point {
	__u64 freq_hz;
};

struct nvgpu_gpu_clk_vf_points_args {

	/* in: Flags (not currently used) */
	__u32 flags;

	/* in: NVGPU_GPU_CLK_DOMAIN_* */
	__u32 clk_domain;

	/* in/out: max number of nvgpu_gpu_clk_vf_point entries in
	   clk_vf_point_entries.  If max_entries is zero,
	   NVGPU_GPU_IOCTL_CLK_GET_VF_POINTS will return 0 and max_entries will
	   be set to the max number of VF entries for this clock domain. If
	   there are more entries than max_entries, then ioctl will return
	   -EINVAL.
	*/
	__u16 max_entries;

	/* out: Number of nvgpu_gpu_clk_vf_point entries returned in
	   clk_vf_point_entries. Number of entries might vary depending on
	   thermal conditions.
	*/
	__u16 num_entries;

	__u32 reserved;

	/* in: Pointer to clock VF point entries in the caller's address space.
	   size must be >= max_entries * sizeof(struct nvgpu_gpu_clk_vf_point).
	 */
	__u64 clk_vf_point_entries;
};

/* Target clock requested by application*/
#define NVGPU_GPU_CLK_TYPE_TARGET	1
/* Actual clock frequency for the domain.
   May deviate from desired target frequency due to PLL constraints. */
#define NVGPU_GPU_CLK_TYPE_ACTUAL	2
/* Effective clock, measured from hardware */
#define NVGPU_GPU_CLK_TYPE_EFFECTIVE	3

struct nvgpu_gpu_clk_info {

	/* Flags (not currently used) */
	__u16 flags;

	/* in: When NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS set, indicates
	   the type of clock info to be returned for this entry. It is
	   allowed to have several entries with different clock types in
	   the same request (for instance query both target and actual
	   clocks for a given clock domain). This field is ignored for a
	   SET operation. */
	__u16 clk_type;

	/* NVGPU_GPU_CLK_DOMAIN_xxx */
	__u32 clk_domain;

	__u64 freq_hz;
};

struct nvgpu_gpu_clk_get_info_args {

	/* Flags. If NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS the request will
	   apply only to domains specified in clock entries. In this case
	   caller must set clock domain in each entry. Otherwise, the
	   ioctl will return all clock domains.
	*/
	__u32 flags;

	/* in: indicates which type of clock info to be returned (see
	   NVGPU_GPU_CLK_TYPE_xxx). If NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS
	   is defined, clk_type is specified in each clock info entry instead.
	 */
	__u16 clk_type;

	/* in/out: Number of clock info entries contained in clk_info_entries.
	   If zero, NVGPU_GPU_IOCTL_CLK_GET_INFO will return 0 and
	   num_entries will be set to number of clock domains. Also,
	   last_req_nr will be updated, which allows checking if a given
	   request has completed. If there are more entries than max_entries,
	   then ioctl will return -EINVAL.
	 */
	__u16 num_entries;

	/* in: Pointer to nvgpu_gpu_clk_info entries in the caller's address
	   space. Buffer size must be at least:
		num_entries * sizeof(struct nvgpu_gpu_clk_info)
	   If NVGPU_GPU_CLK_FLAG_SPECIFIC_DOMAINS is set, caller should set
	   clk_domain to be queried in  each entry. With this flag,
	   clk_info_entries passed to an NVGPU_GPU_IOCTL_CLK_SET_INFO,
	   can be re-used on completion for a NVGPU_GPU_IOCTL_CLK_GET_INFO.
	   This allows checking actual_mhz.
	 */
	__u64 clk_info_entries;

};

struct nvgpu_gpu_clk_set_info_args {

	/* in: Flags (not currently used). */
	__u32 flags;

	__u16 pad0;

	/* Number of clock info entries contained in clk_info_entries.
	   Must be > 0.
	 */
	__u16 num_entries;

	/* Pointer to clock info entries in the caller's address space. Buffer
	   size must be at least
		num_entries * sizeof(struct nvgpu_gpu_clk_info)
	 */
	__u64 clk_info_entries;

	/* out: File descriptor for request completion. Application can poll
	   this file descriptor to determine when the request has completed.
	   The fd must be closed afterwards.
	 */
	__s32 completion_fd;
};

struct nvgpu_gpu_get_event_fd_args {

	/* in: Flags (not currently used). */
	__u32 flags;

	/* out: File descriptor for events, e.g. clock update.
	 * On successful polling of this event_fd, application is
	 * expected to read status (nvgpu_gpu_event_info),
	 * which provides detailed event information
	 * For a poll operation, alarms will be reported with POLLPRI,
	 * and GPU shutdown will be reported with POLLHUP.
	 */
	__s32 event_fd;
};

struct nvgpu_gpu_get_memory_state_args {
	/*
	 * Current free space for this device; may change even when any
	 * kernel-managed metadata (e.g., page tables or channels) is allocated
	 * or freed. For an idle gpu, an allocation of this size would succeed.
	 */
	__u64 total_free_bytes;

	/* For future use; must be set to 0. */
	__u64 reserved[4];
};

struct nvgpu_gpu_get_fbp_l2_masks_args {
	/* [in]  L2 mask buffer size reserved by userspace. Should be
		 at least sizeof(__u32) * fls(fbp_en_mask) to receive LTC
		 mask for each FBP.
	   [out] full kernel buffer size
	*/
	__u32 mask_buf_size;
	__u32 reserved;

	/* [in]  pointer to L2 mask buffer. It will receive one
		 32-bit L2 mask per FBP or 0 if FBP is not enabled or
		 not present. This parameter is ignored if
		 mask_buf_size is 0. */
	__u64 mask_buf_addr;
};

#define NVGPU_GPU_VOLTAGE_CORE		1
#define NVGPU_GPU_VOLTAGE_SRAM		2
#define NVGPU_GPU_VOLTAGE_BUS		3	/* input to regulator */

struct nvgpu_gpu_get_voltage_args {
	__u64 reserved;
	__u32 which;		/* in: NVGPU_GPU_VOLTAGE_* */
	__u32 voltage;		/* uV */
};

struct nvgpu_gpu_get_current_args {
	__u32 reserved[3];
	__u32 currnt;		/* mA */
};

struct nvgpu_gpu_get_power_args {
	__u32 reserved[3];
	__u32 power;		/* mW */
};

struct nvgpu_gpu_get_temperature_args {
	__u32 reserved[3];
	/* Temperature in signed fixed point format SFXP24.8
	 *    Celsius = temp_f24_8 / 256.
	 */
	__s32 temp_f24_8;
};

struct nvgpu_gpu_set_therm_alert_limit_args {
	__u32 reserved[3];
	/* Temperature in signed fixed point format SFXP24.8
	 *    Celsius = temp_f24_8 / 256.
	 */
	__s32 temp_f24_8;
};

/*
 * Adjust options of deterministic channels in channel batches.
 *
 * This supports only one option currently: relax railgate blocking by
 * "disabling" the channel.
 *
 * Open deterministic channels do not allow the GPU to railgate by default. It
 * may be preferable to hold preopened channel contexts open and idle and still
 * railgate the GPU, taking the channels back into use dynamically in userspace
 * as an optimization. This ioctl allows to drop or reacquire the requirement
 * to hold GPU power on for individual channels. If allow_railgate is set on a
 * channel, no work can be submitted to it.
 *
 * num_channels is updated to signify how many channels were updated
 * successfully. It can be used to test which was the first update to fail.
 */
struct nvgpu_gpu_set_deterministic_opts_args {
	__u32 num_channels; /* in/out */
/*
 * Set or unset the railgating reference held by deterministic channels. If
 * the channel status is already the same as the flag, this is a no-op. Both
 * of these flags cannot be set at the same time. If none are set, the state
 * is left as is.
 */
#define NVGPU_GPU_SET_DETERMINISTIC_OPTS_FLAGS_ALLOW_RAILGATING    (1 << 0)
#define NVGPU_GPU_SET_DETERMINISTIC_OPTS_FLAGS_DISALLOW_RAILGATING (1 << 1)
	__u32 flags;        /* in */
	/*
	 * This is a pointer to an array of size num_channels.
	 *
	 * The channels have to be valid fds and be previously set as
	 * deterministic.
	 */
	__u64 channels; /* in */
};

#define NVGPU_GPU_IOCTL_ZCULL_GET_CTX_SIZE \
	_IOR(NVGPU_GPU_IOCTL_MAGIC, 1, struct nvgpu_gpu_zcull_get_ctx_size_args)
#define NVGPU_GPU_IOCTL_ZCULL_GET_INFO \
	_IOR(NVGPU_GPU_IOCTL_MAGIC, 2, struct nvgpu_gpu_zcull_get_info_args)
#define NVGPU_GPU_IOCTL_ZBC_SET_TABLE	\
	_IOW(NVGPU_GPU_IOCTL_MAGIC, 3, struct nvgpu_gpu_zbc_set_table_args)
#define NVGPU_GPU_IOCTL_ZBC_QUERY_TABLE	\
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 4, struct nvgpu_gpu_zbc_query_table_args)
#define NVGPU_GPU_IOCTL_GET_CHARACTERISTICS   \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 5, struct nvgpu_gpu_get_characteristics)
#define NVGPU_GPU_IOCTL_PREPARE_COMPRESSIBLE_READ \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 6, struct nvgpu_gpu_prepare_compressible_read_args)
#define NVGPU_GPU_IOCTL_MARK_COMPRESSIBLE_WRITE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 7, struct nvgpu_gpu_mark_compressible_write_args)
#define NVGPU_GPU_IOCTL_ALLOC_AS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 8, struct nvgpu_alloc_as_args)
#define NVGPU_GPU_IOCTL_OPEN_TSG \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 9, struct nvgpu_gpu_open_tsg_args)
#define NVGPU_GPU_IOCTL_GET_TPC_MASKS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 10, struct nvgpu_gpu_get_tpc_masks_args)
#define NVGPU_GPU_IOCTL_OPEN_CHANNEL \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 11, struct nvgpu_gpu_open_channel_args)
#define NVGPU_GPU_IOCTL_FLUSH_L2 \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 12, struct nvgpu_gpu_l2_fb_args)
#define NVGPU_GPU_IOCTL_INVAL_ICACHE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 13, struct nvgpu_gpu_inval_icache_args)
#define NVGPU_GPU_IOCTL_SET_MMUDEBUG_MODE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 14, struct nvgpu_gpu_mmu_debug_mode_args)
#define NVGPU_GPU_IOCTL_SET_SM_DEBUG_MODE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 15, struct nvgpu_gpu_sm_debug_mode_args)
#define NVGPU_GPU_IOCTL_WAIT_FOR_PAUSE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 16, struct nvgpu_gpu_wait_pause_args)
#define NVGPU_GPU_IOCTL_GET_TPC_EXCEPTION_EN_STATUS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 17, struct nvgpu_gpu_tpc_exception_en_status_args)
#define NVGPU_GPU_IOCTL_NUM_VSMS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 18, struct nvgpu_gpu_num_vsms)
#define NVGPU_GPU_IOCTL_VSMS_MAPPING \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 19, struct nvgpu_gpu_vsms_mapping)
#define NVGPU_GPU_IOCTL_RESUME_FROM_PAUSE \
	_IO(NVGPU_GPU_IOCTL_MAGIC, 21)
#define NVGPU_GPU_IOCTL_TRIGGER_SUSPEND \
	_IO(NVGPU_GPU_IOCTL_MAGIC, 22)
#define NVGPU_GPU_IOCTL_CLEAR_SM_ERRORS \
	_IO(NVGPU_GPU_IOCTL_MAGIC, 23)
#define NVGPU_GPU_IOCTL_GET_CPU_TIME_CORRELATION_INFO \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 24, \
			struct nvgpu_gpu_get_cpu_time_correlation_info_args)
#define NVGPU_GPU_IOCTL_GET_GPU_TIME \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 25, \
			struct nvgpu_gpu_get_gpu_time_args)
#define NVGPU_GPU_IOCTL_GET_ENGINE_INFO \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 26, \
			struct nvgpu_gpu_get_engine_info_args)
#define NVGPU_GPU_IOCTL_ALLOC_VIDMEM \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 27, \
			struct nvgpu_gpu_alloc_vidmem_args)
#define NVGPU_GPU_IOCTL_CLK_GET_RANGE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 28, struct nvgpu_gpu_clk_range_args)
#define NVGPU_GPU_IOCTL_CLK_GET_VF_POINTS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 29, struct nvgpu_gpu_clk_vf_points_args)
#define NVGPU_GPU_IOCTL_CLK_GET_INFO \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 30, struct nvgpu_gpu_clk_get_info_args)
#define NVGPU_GPU_IOCTL_CLK_SET_INFO \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 31, struct nvgpu_gpu_clk_set_info_args)
#define NVGPU_GPU_IOCTL_GET_EVENT_FD \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 32, struct nvgpu_gpu_get_event_fd_args)
#define NVGPU_GPU_IOCTL_GET_MEMORY_STATE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 33, \
			struct nvgpu_gpu_get_memory_state_args)
#define NVGPU_GPU_IOCTL_GET_VOLTAGE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 34, struct nvgpu_gpu_get_voltage_args)
#define NVGPU_GPU_IOCTL_GET_CURRENT \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 35, struct nvgpu_gpu_get_current_args)
#define NVGPU_GPU_IOCTL_GET_POWER \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 36, struct nvgpu_gpu_get_power_args)
#define NVGPU_GPU_IOCTL_GET_TEMPERATURE \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 37, struct nvgpu_gpu_get_temperature_args)
#define NVGPU_GPU_IOCTL_GET_FBP_L2_MASKS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 38, struct nvgpu_gpu_get_fbp_l2_masks_args)
#define NVGPU_GPU_IOCTL_SET_THERM_ALERT_LIMIT \
		_IOWR(NVGPU_GPU_IOCTL_MAGIC, 39, \
			struct nvgpu_gpu_set_therm_alert_limit_args)
#define NVGPU_GPU_IOCTL_SET_DETERMINISTIC_OPTS \
	_IOWR(NVGPU_GPU_IOCTL_MAGIC, 40, \
			struct nvgpu_gpu_set_deterministic_opts_args)
#define NVGPU_GPU_IOCTL_LAST		\
	_IOC_NR(NVGPU_GPU_IOCTL_SET_DETERMINISTIC_OPTS)
#define NVGPU_GPU_IOCTL_MAX_ARG_SIZE	\
	sizeof(struct nvgpu_gpu_get_cpu_time_correlation_info_args)

/*
 * Event session
 *
 * NVGPU_GPU_IOCTL_GET_EVENT_FD opens an event session.
 * Below ioctls can be used on these sessions fds.
 */
#define NVGPU_EVENT_IOCTL_MAGIC	'E'

/* Normal events (POLLIN) */
/* Event associated to a VF update */
#define NVGPU_GPU_EVENT_VF_UPDATE				0

/* Recoverable alarms (POLLPRI) */
/* Alarm when target frequency on any session is not possible */
#define NVGPU_GPU_EVENT_ALARM_TARGET_VF_NOT_POSSIBLE		1
/* Alarm when target frequency on current session is not possible */
#define NVGPU_GPU_EVENT_ALARM_LOCAL_TARGET_VF_NOT_POSSIBLE	2
/* Alarm when Clock Arbiter failed */
#define NVGPU_GPU_EVENT_ALARM_CLOCK_ARBITER_FAILED		3
/* Alarm when VF table update failed */
#define NVGPU_GPU_EVENT_ALARM_VF_TABLE_UPDATE_FAILED		4
/* Alarm on thermal condition */
#define NVGPU_GPU_EVENT_ALARM_THERMAL_ABOVE_THRESHOLD		5
/* Alarm on power condition */
#define NVGPU_GPU_EVENT_ALARM_POWER_ABOVE_THRESHOLD		6

/* Non recoverable alarm (POLLHUP) */
/* Alarm on GPU shutdown/fall from bus */
#define NVGPU_GPU_EVENT_ALARM_GPU_LOST				7

#define NVGPU_GPU_EVENT_LAST	NVGPU_GPU_EVENT_ALARM_GPU_LOST

struct nvgpu_gpu_event_info {
	__u32 event_id;		/* NVGPU_GPU_EVENT_* */
	__u32 reserved;
	__u64 timestamp;	/* CPU timestamp (in nanoseconds) */
};

struct nvgpu_gpu_set_event_filter_args {

	/* in: Flags (not currently used). */
	__u32 flags;

	/* in: Size of event filter in 32-bit words */
	__u32 size;

	/* in: Address of buffer containing bit mask of events.
	 * Bit #n is set if event #n should be monitored.
	 */
	__u64 buffer;
};

#define NVGPU_EVENT_IOCTL_SET_FILTER \
	_IOW(NVGPU_EVENT_IOCTL_MAGIC, 1, struct nvgpu_gpu_set_event_filter_args)
#define NVGPU_EVENT_IOCTL_LAST		\
	_IOC_NR(NVGPU_EVENT_IOCTL_SET_FILTER)
#define NVGPU_EVENT_IOCTL_MAX_ARG_SIZE	\
	sizeof(struct nvgpu_gpu_set_event_filter_args)

/*
 * /dev/nvhost-tsg-gpu device
 *
 * Opening a '/dev/nvhost-tsg-gpu' device node creates a way to
 * bind/unbind a channel to/from TSG group
 */

#define NVGPU_TSG_IOCTL_MAGIC 'T'

struct nvgpu_tsg_bind_channel_ex_args {
	/* in: channel fd */
	__s32 channel_fd;

	/* in: VEID in Volta */
	__u32 subcontext_id;
	__u32 num_active_tpcs;
	__u8 tpc_pg_enabled;
	__u8 reserved[11];
};

#define NVGPU_TSG_IOCTL_BIND_CHANNEL \
	_IOW(NVGPU_TSG_IOCTL_MAGIC, 1, int)
#define NVGPU_TSG_IOCTL_UNBIND_CHANNEL \
	_IOW(NVGPU_TSG_IOCTL_MAGIC, 2, int)
#define NVGPU_IOCTL_TSG_ENABLE \
	_IO(NVGPU_TSG_IOCTL_MAGIC, 3)
#define NVGPU_IOCTL_TSG_DISABLE \
	_IO(NVGPU_TSG_IOCTL_MAGIC, 4)
#define NVGPU_IOCTL_TSG_PREEMPT \
	_IO(NVGPU_TSG_IOCTL_MAGIC, 5)
#define NVGPU_IOCTL_TSG_EVENT_ID_CTRL \
	_IOWR(NVGPU_TSG_IOCTL_MAGIC, 7, struct nvgpu_event_id_ctrl_args)
#define NVGPU_IOCTL_TSG_SET_RUNLIST_INTERLEAVE \
	_IOW(NVGPU_TSG_IOCTL_MAGIC, 8, struct nvgpu_runlist_interleave_args)
#define NVGPU_IOCTL_TSG_SET_TIMESLICE \
	_IOW(NVGPU_TSG_IOCTL_MAGIC, 9, struct nvgpu_timeslice_args)
#define NVGPU_IOCTL_TSG_GET_TIMESLICE \
	_IOR(NVGPU_TSG_IOCTL_MAGIC, 10, struct nvgpu_timeslice_args)
#define NVGPU_TSG_IOCTL_BIND_CHANNEL_EX \
	_IOWR(NVGPU_TSG_IOCTL_MAGIC, 11, struct nvgpu_tsg_bind_channel_ex_args)
#define NVGPU_TSG_IOCTL_MAX_ARG_SIZE	\
		sizeof(struct nvgpu_tsg_bind_channel_ex_args)
#define NVGPU_TSG_IOCTL_LAST		\
	_IOC_NR(NVGPU_TSG_IOCTL_BIND_CHANNEL_EX)

/*
 * /dev/nvhost-dbg-gpu device
 *
 * Opening a '/dev/nvhost-dbg-gpu' device node creates a new debugger
 * session.  nvgpu channels (for the same module) can then be bound to such a
 * session.
 *
 * One nvgpu channel can also be bound to multiple debug sessions
 *
 * As long as there is an open device file to the session, or any bound
 * nvgpu channels it will be valid.  Once all references to the session
 * are removed the session is deleted.
 *
 */

#define NVGPU_DBG_GPU_IOCTL_MAGIC 'D'

/*
 * Binding/attaching a debugger session to an nvgpu channel
 *
 * The 'channel_fd' given here is the fd used to allocate the
 * gpu channel context.
 *
 */
struct nvgpu_dbg_gpu_bind_channel_args {
	__u32 channel_fd; /* in */
	__u32 _pad0[1];
};

#define NVGPU_DBG_GPU_IOCTL_BIND_CHANNEL				\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 1, struct nvgpu_dbg_gpu_bind_channel_args)

/*
 * Register operations
 * All operations are targeted towards first channel
 * attached to debug session
 */
/* valid op values */
#define NVGPU_DBG_GPU_REG_OP_READ_32                             (0x00000000)
#define NVGPU_DBG_GPU_REG_OP_WRITE_32                            (0x00000001)
#define NVGPU_DBG_GPU_REG_OP_READ_64                             (0x00000002)
#define NVGPU_DBG_GPU_REG_OP_WRITE_64                            (0x00000003)
/* note: 8b ops are unsupported */
#define NVGPU_DBG_GPU_REG_OP_READ_08                             (0x00000004)
#define NVGPU_DBG_GPU_REG_OP_WRITE_08                            (0x00000005)

/* valid type values */
#define NVGPU_DBG_GPU_REG_OP_TYPE_GLOBAL                         (0x00000000)
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX                         (0x00000001)
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX_TPC                     (0x00000002)
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX_SM                      (0x00000004)
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX_CROP                    (0x00000008)
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX_ZROP                    (0x00000010)
/*#define NVGPU_DBG_GPU_REG_OP_TYPE_FB                           (0x00000020)*/
#define NVGPU_DBG_GPU_REG_OP_TYPE_GR_CTX_QUAD                    (0x00000040)

/* valid status values */
#define NVGPU_DBG_GPU_REG_OP_STATUS_SUCCESS                      (0x00000000)
#define NVGPU_DBG_GPU_REG_OP_STATUS_INVALID_OP                   (0x00000001)
#define NVGPU_DBG_GPU_REG_OP_STATUS_INVALID_TYPE                 (0x00000002)
#define NVGPU_DBG_GPU_REG_OP_STATUS_INVALID_OFFSET               (0x00000004)
#define NVGPU_DBG_GPU_REG_OP_STATUS_UNSUPPORTED_OP               (0x00000008)
#define NVGPU_DBG_GPU_REG_OP_STATUS_INVALID_MASK                 (0x00000010)

struct nvgpu_dbg_gpu_reg_op {
	__u8    op;
	__u8    type;
	__u8    status;
	__u8    quad;
	__u32   group_mask;
	__u32   sub_group_mask;
	__u32   offset;
	__u32   value_lo;
	__u32   value_hi;
	__u32   and_n_mask_lo;
	__u32   and_n_mask_hi;
};

struct nvgpu_dbg_gpu_exec_reg_ops_args {
	__u64 ops; /* pointer to nvgpu_reg_op operations */
	__u32 num_ops;
	__u32 _pad0[1];
};

#define NVGPU_DBG_GPU_IOCTL_REG_OPS					\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 2, struct nvgpu_dbg_gpu_exec_reg_ops_args)

/* Enable/disable/clear event notifications */
struct nvgpu_dbg_gpu_events_ctrl_args {
	__u32 cmd; /* in */
	__u32 _pad0[1];
};

/* valid event ctrl values */
#define NVGPU_DBG_GPU_EVENTS_CTRL_CMD_DISABLE                    (0x00000000)
#define NVGPU_DBG_GPU_EVENTS_CTRL_CMD_ENABLE                     (0x00000001)
#define NVGPU_DBG_GPU_EVENTS_CTRL_CMD_CLEAR                      (0x00000002)

#define NVGPU_DBG_GPU_IOCTL_EVENTS_CTRL				\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 3, struct nvgpu_dbg_gpu_events_ctrl_args)


/* Powergate/Unpowergate control */

#define NVGPU_DBG_GPU_POWERGATE_MODE_ENABLE                                 1
#define NVGPU_DBG_GPU_POWERGATE_MODE_DISABLE                                2

struct nvgpu_dbg_gpu_powergate_args {
	__u32 mode;
} __packed;

#define NVGPU_DBG_GPU_IOCTL_POWERGATE					\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 4, struct nvgpu_dbg_gpu_powergate_args)


/* SMPC Context Switch Mode */
#define NVGPU_DBG_GPU_SMPC_CTXSW_MODE_NO_CTXSW               (0x00000000)
#define NVGPU_DBG_GPU_SMPC_CTXSW_MODE_CTXSW                  (0x00000001)

struct nvgpu_dbg_gpu_smpc_ctxsw_mode_args {
	__u32 mode;
} __packed;

#define NVGPU_DBG_GPU_IOCTL_SMPC_CTXSW_MODE \
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 5, struct nvgpu_dbg_gpu_smpc_ctxsw_mode_args)

/* Suspend /Resume SM control */
#define NVGPU_DBG_GPU_SUSPEND_ALL_SMS 1
#define NVGPU_DBG_GPU_RESUME_ALL_SMS  2

struct nvgpu_dbg_gpu_suspend_resume_all_sms_args {
	__u32 mode;
} __packed;

#define NVGPU_DBG_GPU_IOCTL_SUSPEND_RESUME_ALL_SMS			\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 6, struct nvgpu_dbg_gpu_suspend_resume_all_sms_args)

struct nvgpu_dbg_gpu_perfbuf_map_args {
	__u32 dmabuf_fd;	/* in */
	__u32 reserved;
	__u64 mapping_size;	/* in, size of mapped buffer region */
	__u64 offset;		/* out, virtual address of the mapping */
};

struct nvgpu_dbg_gpu_perfbuf_unmap_args {
	__u64 offset;
};

#define NVGPU_DBG_GPU_IOCTL_PERFBUF_MAP \
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 7, struct nvgpu_dbg_gpu_perfbuf_map_args)
#define NVGPU_DBG_GPU_IOCTL_PERFBUF_UNMAP \
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 8, struct nvgpu_dbg_gpu_perfbuf_unmap_args)

/* Enable/disable PC Sampling */
struct nvgpu_dbg_gpu_pc_sampling_args {
	__u32 enable;
	__u32 _pad0[1];
};

#define NVGPU_DBG_GPU_IOCTL_PC_SAMPLING_DISABLE	0
#define NVGPU_DBG_GPU_IOCTL_PC_SAMPLING_ENABLE	1

#define NVGPU_DBG_GPU_IOCTL_PC_SAMPLING \
	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC,  9, struct nvgpu_dbg_gpu_pc_sampling_args)

/* Enable/Disable timeouts */
#define NVGPU_DBG_GPU_IOCTL_TIMEOUT_ENABLE                                   1
#define NVGPU_DBG_GPU_IOCTL_TIMEOUT_DISABLE                                  0

struct nvgpu_dbg_gpu_timeout_args {
	__u32 enable;
	__u32 padding;
};

#define NVGPU_DBG_GPU_IOCTL_TIMEOUT \
	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC, 10, struct nvgpu_dbg_gpu_timeout_args)

#define NVGPU_DBG_GPU_IOCTL_GET_TIMEOUT \
	_IOR(NVGPU_DBG_GPU_IOCTL_MAGIC, 11, struct nvgpu_dbg_gpu_timeout_args)


struct nvgpu_dbg_gpu_set_next_stop_trigger_type_args {
	__u32 broadcast;
	__u32 reserved;
};

#define NVGPU_DBG_GPU_IOCTL_SET_NEXT_STOP_TRIGGER_TYPE			\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 12, struct nvgpu_dbg_gpu_set_next_stop_trigger_type_args)


/* PM Context Switch Mode */
#define NVGPU_DBG_GPU_HWPM_CTXSW_MODE_NO_CTXSW               (0x00000000)
#define NVGPU_DBG_GPU_HWPM_CTXSW_MODE_CTXSW                  (0x00000001)

struct nvgpu_dbg_gpu_hwpm_ctxsw_mode_args {
	__u32 mode;
	__u32 reserved;
};

#define NVGPU_DBG_GPU_IOCTL_HWPM_CTXSW_MODE \
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 13, struct nvgpu_dbg_gpu_hwpm_ctxsw_mode_args)


struct nvgpu_dbg_gpu_sm_error_state_record {
	__u32 hww_global_esr;
	__u32 hww_warp_esr;
	__u64 hww_warp_esr_pc;
	__u32 hww_global_esr_report_mask;
	__u32 hww_warp_esr_report_mask;

	/*
	 * Notes
	 * - This struct can be safely appended with new fields. However, always
	 *   keep the structure size multiple of 8 and make sure that the binary
	 *   layout does not change between 32-bit and 64-bit architectures.
	 */
};

struct nvgpu_dbg_gpu_read_single_sm_error_state_args {
	__u32 sm_id;
	__u32 padding;
	__u64 sm_error_state_record_mem;
	__u64 sm_error_state_record_size;
};

#define NVGPU_DBG_GPU_IOCTL_READ_SINGLE_SM_ERROR_STATE			\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 14, struct nvgpu_dbg_gpu_read_single_sm_error_state_args)


struct nvgpu_dbg_gpu_clear_single_sm_error_state_args {
	__u32 sm_id;
	__u32 padding;
};

#define NVGPU_DBG_GPU_IOCTL_CLEAR_SINGLE_SM_ERROR_STATE			\
	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC, 15, struct nvgpu_dbg_gpu_clear_single_sm_error_state_args)


struct nvgpu_dbg_gpu_write_single_sm_error_state_args {
	__u32 sm_id;
	__u32 padding;
	__u64 sm_error_state_record_mem;
	__u64 sm_error_state_record_size;
};

#define NVGPU_DBG_GPU_IOCTL_WRITE_SINGLE_SM_ERROR_STATE			\
	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC, 16, struct nvgpu_dbg_gpu_write_single_sm_error_state_args)

/*
 * Unbinding/detaching a debugger session from a nvgpu channel
 *
 * The 'channel_fd' given here is the fd used to allocate the
 * gpu channel context.
 */
struct nvgpu_dbg_gpu_unbind_channel_args {
	__u32 channel_fd; /* in */
	__u32 _pad0[1];
};

#define NVGPU_DBG_GPU_IOCTL_UNBIND_CHANNEL				\
	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC, 17, struct nvgpu_dbg_gpu_unbind_channel_args)


#define NVGPU_DBG_GPU_SUSPEND_ALL_CONTEXTS	1
#define NVGPU_DBG_GPU_RESUME_ALL_CONTEXTS	2

struct nvgpu_dbg_gpu_suspend_resume_contexts_args {
	__u32 action;
	__u32 is_resident_context;
	__s32 resident_context_fd;
	__u32 padding;
};

#define NVGPU_DBG_GPU_IOCTL_SUSPEND_RESUME_CONTEXTS			\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 18, struct nvgpu_dbg_gpu_suspend_resume_contexts_args)


#define NVGPU_DBG_GPU_IOCTL_ACCESS_FB_MEMORY_CMD_READ	1
#define NVGPU_DBG_GPU_IOCTL_ACCESS_FB_MEMORY_CMD_WRITE	2

struct nvgpu_dbg_gpu_access_fb_memory_args {
	__u32 cmd;       /* in: either read or write */

	__s32 dmabuf_fd; /* in: dmabuf fd of the buffer in FB */
	__u64 offset;    /* in: offset within buffer in FB, should be 4B aligned */

	__u64 buffer;    /* in/out: temp buffer to read/write from */
	__u64 size;      /* in: size of the buffer i.e. size of read/write in bytes, should be 4B aligned */
};

#define NVGPU_DBG_GPU_IOCTL_ACCESS_FB_MEMORY	\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 19, struct nvgpu_dbg_gpu_access_fb_memory_args)

struct nvgpu_dbg_gpu_profiler_obj_mgt_args {
	__u32 profiler_handle;
	__u32 reserved;
};

#define NVGPU_DBG_GPU_IOCTL_PROFILER_ALLOCATE	\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 20, struct nvgpu_dbg_gpu_profiler_obj_mgt_args)

#define NVGPU_DBG_GPU_IOCTL_PROFILER_FREE	\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 21, struct nvgpu_dbg_gpu_profiler_obj_mgt_args)

struct nvgpu_dbg_gpu_profiler_reserve_args {
	__u32 profiler_handle;
	__u32 acquire;
};

#define NVGPU_DBG_GPU_IOCTL_PROFILER_RESERVE			\
	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 22, struct nvgpu_dbg_gpu_profiler_reserve_args)

#define NVGPU_DBG_GPU_IOCTL_LAST		\
	_IOC_NR(NVGPU_DBG_GPU_IOCTL_PROFILER_RESERVE)

#define NVGPU_DBG_GPU_IOCTL_MAX_ARG_SIZE		\
	sizeof(struct nvgpu_dbg_gpu_access_fb_memory_args)

/*
 * /dev/nvhost-gpu device
 */

#define NVGPU_IOCTL_MAGIC 'H'
#define NVGPU_NO_TIMEOUT ((u32)~0)