Merge tag 'lsk-v3.10-android-15.02'

[firefly-linux-kernel-4.4.55.git] / drivers / gpu / arm / midgard / mali_base_kernel.h

/*
 *
 * (C) COPYRIGHT ARM Limited. All rights reserved.
 *
 * This program is free software and is provided to you under the terms of the
 * GNU General Public License version 2 as published by the Free Software
 * Foundation, and any use by you of this program is subject to the terms
 * of such GNU licence.
 *
 * A copy of the licence is included with the program, and can also be obtained
 * from Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
 * Boston, MA  02110-1301, USA.
 *
 */





/**
 * @file
 * Base structures shared with the kernel.
 */

#ifndef _BASE_KERNEL_H_
#define _BASE_KERNEL_H_

#ifndef __user
#define __user
#endif

/* Support UK6 IOCTLS */
#define BASE_LEGACY_UK6_SUPPORT 1

/* Support UK7 IOCTLS */
/* NB: To support UK6 we also need to support UK7 */
#define BASE_LEGACY_UK7_SUPPORT 1

typedef mali_addr64 base_mem_handle;

#include "mali_base_mem_priv.h"
#include "mali_kbase_profiling_gator_api.h"

/*
 * Dependency stuff, keep it private for now. May want to expose it if
 * we decide to make the number of semaphores a configurable
 * option.
 */
#define BASE_JD_ATOM_COUNT              256

#define BASEP_JD_SEM_PER_WORD_LOG2      5
#define BASEP_JD_SEM_PER_WORD           (1 << BASEP_JD_SEM_PER_WORD_LOG2)
#define BASEP_JD_SEM_WORD_NR(x)         ((x) >> BASEP_JD_SEM_PER_WORD_LOG2)
#define BASEP_JD_SEM_MASK_IN_WORD(x)    (1 << ((x) & (BASEP_JD_SEM_PER_WORD - 1)))
#define BASEP_JD_SEM_ARRAY_SIZE         BASEP_JD_SEM_WORD_NR(BASE_JD_ATOM_COUNT)

#define BASE_GPU_NUM_TEXTURE_FEATURES_REGISTERS 3

#define BASE_MAX_COHERENT_GROUPS 16

#if defined CDBG_ASSERT
#define LOCAL_ASSERT CDBG_ASSERT
#elif defined KBASE_DEBUG_ASSERT
#define LOCAL_ASSERT KBASE_DEBUG_ASSERT
#else
#error assert macro not defined!
#endif

#if defined PAGE_MASK
#define LOCAL_PAGE_LSB ~PAGE_MASK
#else
#include <osu/mali_osu.h>

#if defined OSU_CONFIG_CPU_PAGE_SIZE_LOG2
#define LOCAL_PAGE_LSB ((1ul << OSU_CONFIG_CPU_PAGE_SIZE_LOG2) - 1)
#else
#error Failed to find page size
#endif
#endif

/** 32/64-bit neutral way to represent pointers */
typedef union kbase_pointer {
        void __user *value;       /**< client should store their pointers here */
        u32 compat_value; /**< 64-bit kernels should fetch value here when handling 32-bit clients */
        u64 sizer;        /**< Force 64-bit storage for all clients regardless */
} kbase_pointer;

/**
 * @addtogroup base_user_api User-side Base APIs
 * @{
 */

/**
 * @addtogroup base_user_api_memory User-side Base Memory APIs
 * @{
 */

/**
 * @brief Memory allocation, access/hint flags
 *
 * A combination of MEM_PROT/MEM_HINT flags must be passed to each allocator
 * in order to determine the best cache policy. Some combinations are
 * of course invalid (eg @c MEM_PROT_CPU_WR | @c MEM_HINT_CPU_RD),
 * which defines a @a write-only region on the CPU side, which is
 * heavily read by the CPU...
 * Other flags are only meaningful to a particular allocator.
 * More flags can be added to this list, as long as they don't clash
 * (see ::BASE_MEM_FLAGS_NR_TOTAL_BITS for the number of the first free bit).
 */
typedef u32 base_mem_alloc_flags;

/**
 * @brief Memory allocation, access/hint flags
 *
 * See ::base_mem_alloc_flags.
 *
 */
enum {
/* IN */
        BASE_MEM_PROT_CPU_RD = (1U << 0),      /**< Read access CPU side */
        BASE_MEM_PROT_CPU_WR = (1U << 1),      /**< Write access CPU side */
        BASE_MEM_PROT_GPU_RD = (1U << 2),      /**< Read access GPU side */
        BASE_MEM_PROT_GPU_WR = (1U << 3),      /**< Write access GPU side */
        BASE_MEM_PROT_GPU_EX = (1U << 4),      /**< Execute allowed on the GPU
                                                    side */

        /* Note that the HINT flags are obsolete now. If you want the memory
         * to be cached on the CPU please use the BASE_MEM_CACHED_CPU flag
         */
        BASE_MEM_HINT_CPU_RD = (1U << 5),      /**< Heavily read CPU side
                                                    - OBSOLETE */
        BASE_MEM_HINT_CPU_WR = (1U << 6),      /**< Heavily written CPU side
                                                    - OBSOLETE */
        BASE_MEM_HINT_GPU_RD = (1U << 7),      /**< Heavily read GPU side
                                                    - OBSOLETE */
        BASE_MEM_HINT_GPU_WR = (1U << 8),      /**< Heavily written GPU side
                                                    - OBSOLETE */

        BASE_MEM_GROW_ON_GPF = (1U << 9),      /**< Grow backing store on GPU
                                                    Page Fault */

        BASE_MEM_COHERENT_SYSTEM = (1U << 10), /**< Page coherence Outer
                                                    shareable */
        BASE_MEM_COHERENT_LOCAL = (1U << 11),  /**< Page coherence Inner
                                                    shareable */
        BASE_MEM_CACHED_CPU = (1U << 12),      /**< Should be cached on the
                                                    CPU */

/* IN/OUT */
        BASE_MEM_SAME_VA = (1U << 13), /**< Must have same VA on both the GPU
                                            and the CPU */
/* OUT */
        BASE_MEM_NEED_MMAP = (1U << 14) /**< Must call mmap to aquire a GPU
                                             address for the alloc */
};

/**
 * @brief Number of bits used as flags for base memory management
 *
 * Must be kept in sync with the ::base_mem_alloc_flags flags
 */
#define BASE_MEM_FLAGS_NR_INPUT_BITS  14
#define BASE_MEM_FLAGS_NR_OUTPUT_BITS 1
#define BASE_MEM_FLAGS_NR_TOTAL_BITS  ((BASE_MEM_FLAGS_NR_INPUT_BITS) + (BASE_MEM_FLAGS_NR_OUTPUT_BITS))
#define BASE_MEM_FLAGS_NR_BITS 15

#if BASE_MEM_FLAGS_NR_TOTAL_BITS > BASE_MEM_FLAGS_NR_BITS
#error "Too many flag bits, will require change in cmem"
#endif

/**
 * @brief Memory types supported by @a base_mem_import
 *
 * Each type defines what the supported handle type is.
 *
 * If any new type is added here ARM must be contacted
 * to allocate a numeric value for it.
 * Do not just add a new type without synchronizing with ARM
 * as future releases from ARM might include other new types
 * which could clash with your custom types.
 */
typedef enum base_mem_import_type {
        BASE_MEM_IMPORT_TYPE_INVALID = 0,
        /** UMP import. Handle type is ump_secure_id. */
        BASE_MEM_IMPORT_TYPE_UMP = 1,
        /** UMM import. Handle type is a file descriptor (int) */
        BASE_MEM_IMPORT_TYPE_UMM = 2
} base_mem_import_type;

/* legacy API wrappers */
#define base_tmem_import_type          base_mem_import_type
#define BASE_TMEM_IMPORT_TYPE_INVALID  BASE_MEM_IMPORT_TYPE_INVALID
#define BASE_TMEM_IMPORT_TYPE_UMP      BASE_MEM_IMPORT_TYPE_UMP
#define BASE_TMEM_IMPORT_TYPE_UMM      BASE_MEM_IMPORT_TYPE_UMM

/**
 * @brief Invalid memory handle type.
 * Return value from functions returning @a base_mem_handle on error.
 */
#define BASE_MEM_INVALID_HANDLE                (0ull  << 12)
#define BASE_MEM_MMU_DUMP_HANDLE               (1ull  << 12)
#define BASE_MEM_TRACE_BUFFER_HANDLE           (2ull  << 12)
#define BASE_MEM_MAP_TRACKING_HANDLE           (3ull  << 12)
#define BASE_MEM_WRITE_ALLOC_PAGES_HANDLE      (4ull  << 12)
/* reserved handles ..-64<<PAGE_SHIFT> for future special handles */
#define BASE_MEM_COOKIE_BASE                   (64ul  << 12)
#define BASE_MEM_FIRST_FREE_ADDRESS            ((BITS_PER_LONG << 12) + \
                                                BASE_MEM_COOKIE_BASE)

/* Bit mask of cookies used for for memory allocation setup */
#define KBASE_COOKIE_MASK  ~1UL /* bit 0 is reserved */


/**
 * @brief Result codes of changing the size of the backing store allocated to a tmem region
 */
typedef enum base_backing_threshold_status {
        BASE_BACKING_THRESHOLD_OK = 0,                      /**< Resize successful */
        BASE_BACKING_THRESHOLD_ERROR_NOT_GROWABLE = -1,     /**< Not a growable tmem object */
        BASE_BACKING_THRESHOLD_ERROR_OOM = -2,              /**< Increase failed due to an out-of-memory condition */
        BASE_BACKING_THRESHOLD_ERROR_MAPPED = -3,           /**< Resize attempted on buffer while it was mapped, which is not permitted */
        BASE_BACKING_THRESHOLD_ERROR_INVALID_ARGUMENTS = -4 /**< Invalid arguments (not tmem, illegal size request, etc.) */
} base_backing_threshold_status;

/**
 * @addtogroup base_user_api_memory_defered User-side Base Defered Memory Coherency APIs
 * @{
 */

/**
 * @brief a basic memory operation (sync-set).
 *
 * The content of this structure is private, and should only be used
 * by the accessors.
 */
typedef struct base_syncset {
        struct basep_syncset basep_sset;
} base_syncset;

/** @} end group base_user_api_memory_defered */

/**
 * Handle to represent imported memory object.
 * Simple opague handle to imported memory, can't be used
 * with anything but base_external_resource_init to bind to an atom.
 */
typedef struct base_import_handle {
        struct {
                mali_addr64 handle;
        } basep;
} base_import_handle;

/** @} end group base_user_api_memory */

/**
 * @addtogroup base_user_api_job_dispatch User-side Base Job Dispatcher APIs
 * @{
 */

typedef int platform_fence_type;
#define INVALID_PLATFORM_FENCE ((platform_fence_type)-1)

/**
 * Base stream handle.
 *
 * References an underlying base stream object.
 */
typedef struct base_stream {
        struct {
                int fd;
        } basep;
} base_stream;

/**
 * Base fence handle.
 *
 * References an underlying base fence object.
 */
typedef struct base_fence {
        struct {
                int fd;
                int stream_fd;
        } basep;
} base_fence;

/**
 * @brief Per-job data
 *
 * This structure is used to store per-job data, and is completly unused
 * by the Base driver. It can be used to store things such as callback
 * function pointer, data to handle job completion. It is guaranteed to be
 * untouched by the Base driver.
 */
typedef struct base_jd_udata {
        u64 blob[2];     /**< per-job data array */
} base_jd_udata;

/**
 * @brief Memory aliasing info
 *
 * Describes a memory handle to be aliased.
 * A subset of the handle can be chosen for aliasing, given an offset and a
 * length.
 * A special handle BASE_MEM_WRITE_ALLOC_PAGES_HANDLE is used to represent a
 * region where a special page is mapped with a write-alloc cache setup,
 * typically used when the write result of the GPU isn't needed, but the GPU
 * must write anyway.
 *
 * Offset and length are specified in pages.
 * Offset must be within the size of the handle.
 * Offset+length must not overrun the size of the handle.
 *
 * @handle Handle to alias, can be BASE_MEM_WRITE_ALLOC_PAGES_HANDLE
 * @offset Offset within the handle to start aliasing from, in pages.
 *         Not used with BASE_MEM_WRITE_ALLOC_PAGES_HANDLE.
 * @length Length to alias, in pages. For BASE_MEM_WRITE_ALLOC_PAGES_HANDLE
 *         specifies the number of times the special page is needed.
 */
struct base_mem_aliasing_info {
        base_mem_handle handle;
        u64 offset;
        u64 length;
};

/**
 * @brief Job dependency type.
 *
 * A flags field will be inserted into the atom structure to specify whether a dependency is a data or 
 * ordering dependency (by putting it before/after 'core_req' in the structure it should be possible to add without 
 * changing the structure size).
 * When the flag is set for a particular dependency to signal that it is an ordering only dependency then 
 * errors will not be propagated.
 */
typedef u8 base_jd_dep_type;


#define BASE_JD_DEP_TYPE_INVALID  (0)   /**< Invalid dependency */
#define BASE_JD_DEP_TYPE_DATA     (1U << 0)     /**< Data dependency */
#define BASE_JD_DEP_TYPE_ORDER    (1U << 1)     /**< Order dependency */

/**
 * @brief Job chain hardware requirements.
 *
 * A job chain must specify what GPU features it needs to allow the
 * driver to schedule the job correctly.  By not specifying the
 * correct settings can/will cause an early job termination.  Multiple
 * values can be ORed together to specify multiple requirements.
 * Special case is ::BASE_JD_REQ_DEP, which is used to express complex
 * dependencies, and that doesn't execute anything on the hardware.
 */
typedef u16 base_jd_core_req;

/* Requirements that come from the HW */
#define BASE_JD_REQ_DEP 0           /**< No requirement, dependency only */
#define BASE_JD_REQ_FS  (1U << 0)   /**< Requires fragment shaders */
/**
 * Requires compute shaders
 * This covers any of the following Midgard Job types:
 * - Vertex Shader Job
 * - Geometry Shader Job
 * - An actual Compute Shader Job
 *
 * Compare this with @ref BASE_JD_REQ_ONLY_COMPUTE, which specifies that the
 * job is specifically just the "Compute Shader" job type, and not the "Vertex
 * Shader" nor the "Geometry Shader" job type.
 */
#define BASE_JD_REQ_CS  (1U << 1)
#define BASE_JD_REQ_T   (1U << 2)   /**< Requires tiling */
#define BASE_JD_REQ_CF  (1U << 3)   /**< Requires cache flushes */
#define BASE_JD_REQ_V   (1U << 4)   /**< Requires value writeback */

/* SW-only requirements - the HW does not expose these as part of the job slot capabilities */

/* Requires fragment job with AFBC encoding */
#define BASE_JD_REQ_FS_AFBC  (1U << 13)

/**
 * SW Only requirement: the job chain requires a coherent core group. We don't
 * mind which coherent core group is used.
 */
#define BASE_JD_REQ_COHERENT_GROUP  (1U << 6)

/**
 * SW Only requirement: The performance counters should be enabled only when
 * they are needed, to reduce power consumption.
 */

#define BASE_JD_REQ_PERMON               (1U << 7)

/**
 * SW Only requirement: External resources are referenced by this atom.
 * When external resources are referenced no syncsets can be bundled with the atom
 * but should instead be part of a NULL jobs inserted into the dependency tree.
 * The first pre_dep object must be configured for the external resouces to use,
 * the second pre_dep object can be used to create other dependencies.
 */
#define BASE_JD_REQ_EXTERNAL_RESOURCES   (1U << 8)

/**
 * SW Only requirement: Software defined job. Jobs with this bit set will not be submitted
 * to the hardware but will cause some action to happen within the driver
 */
#define BASE_JD_REQ_SOFT_JOB        (1U << 9)

#define BASE_JD_REQ_SOFT_DUMP_CPU_GPU_TIME      (BASE_JD_REQ_SOFT_JOB | 0x1)
#define BASE_JD_REQ_SOFT_FENCE_TRIGGER          (BASE_JD_REQ_SOFT_JOB | 0x2)
#define BASE_JD_REQ_SOFT_FENCE_WAIT             (BASE_JD_REQ_SOFT_JOB | 0x3)

/**
 * SW Only requirement : Replay job.
 *
 * If the preceeding job fails, the replay job will cause the jobs specified in
 * the list of base_jd_replay_payload pointed to by the jc pointer to be
 * replayed.
 *
 * A replay job will only cause jobs to be replayed up to BASEP_JD_REPLAY_LIMIT
 * times. If a job fails more than BASEP_JD_REPLAY_LIMIT times then the replay
 * job is failed, as well as any following dependencies.
 *
 * The replayed jobs will require a number of atom IDs. If there are not enough
 * free atom IDs then the replay job will fail.
 *
 * If the preceeding job does not fail, then the replay job is returned as
 * completed.
 *
 * The replayed jobs will never be returned to userspace. The preceeding failed
 * job will be returned to userspace as failed; the status of this job should
 * be ignored. Completion should be determined by the status of the replay soft
 * job.
 *
 * In order for the jobs to be replayed, the job headers will have to be
 * modified. The Status field will be reset to NOT_STARTED. If the Job Type
 * field indicates a Vertex Shader Job then it will be changed to Null Job.
 *
 * The replayed jobs have the following assumptions :
 *
 * - No external resources. Any required external resources will be held by the
 *   replay atom.
 * - Pre-dependencies are created based on job order.
 * - Atom numbers are automatically assigned.
 * - device_nr is set to 0. This is not relevant as
 *   BASE_JD_REQ_SPECIFIC_COHERENT_GROUP should not be set.
 * - Priority is inherited from the replay job.
 */
#define BASE_JD_REQ_SOFT_REPLAY                 (BASE_JD_REQ_SOFT_JOB | 0x4)

/**
 * HW Requirement: Requires Compute shaders (but not Vertex or Geometry Shaders)
 *
 * This indicates that the Job Chain contains Midgard Jobs of the 'Compute Shaders' type.
 *
 * In contrast to @ref BASE_JD_REQ_CS, this does \b not indicate that the Job
 * Chain contains 'Geometry Shader' or 'Vertex Shader' jobs.
 *
 * @note This is a more flexible variant of the @ref BASE_CONTEXT_HINT_ONLY_COMPUTE flag,
 * allowing specific jobs to be marked as 'Only Compute' instead of the entire context
 */
#define BASE_JD_REQ_ONLY_COMPUTE    (1U << 10)

/**
 * HW Requirement: Use the base_jd_atom::device_nr field to specify a
 * particular core group
 *
 * If both BASE_JD_REQ_COHERENT_GROUP and this flag are set, this flag takes priority
 *
 * This is only guaranteed to work for BASE_JD_REQ_ONLY_COMPUTE atoms.
 *
 * If the core availability policy is keeping the required core group turned off, then 
 * the job will fail with a BASE_JD_EVENT_PM_EVENT error code.
 */
#define BASE_JD_REQ_SPECIFIC_COHERENT_GROUP (1U << 11)

/**
 * SW Flag: If this bit is set then the successful completion of this atom
 * will not cause an event to be sent to userspace
 */
#define BASE_JD_REQ_EVENT_ONLY_ON_FAILURE   (1U << 12)

/**
 * SW Flag: If this bit is set then completion of this atom will not cause an
 * event to be sent to userspace, whether successful or not.
 */
#define BASEP_JD_REQ_EVENT_NEVER (1U << 14)

/**
* These requirement bits are currently unused in base_jd_core_req (currently a u16)
*/

#define BASEP_JD_REQ_RESERVED_BIT5 (1U << 5)
#define BASEP_JD_REQ_RESERVED_BIT15 (1U << 15)

/**
* Mask of all the currently unused requirement bits in base_jd_core_req.
*/

#define BASEP_JD_REQ_RESERVED (BASEP_JD_REQ_RESERVED_BIT5 | \
                                BASEP_JD_REQ_RESERVED_BIT15)

/**
 * Mask of all bits in base_jd_core_req that control the type of the atom.
 *
 * This allows dependency only atoms to have flags set
 */
#define BASEP_JD_REQ_ATOM_TYPE (~(BASEP_JD_REQ_RESERVED | BASE_JD_REQ_EVENT_ONLY_ON_FAILURE |\
                                BASE_JD_REQ_EXTERNAL_RESOURCES | BASEP_JD_REQ_EVENT_NEVER))

/**
 * @brief States to model state machine processed by kbasep_js_job_check_ref_cores(), which
 * handles retaining cores for power management and affinity management.
 *
 * The state @ref KBASE_ATOM_COREREF_STATE_RECHECK_AFFINITY prevents an attack
 * where lots of atoms could be submitted before powerup, and each has an
 * affinity chosen that causes other atoms to have an affinity
 * violation. Whilst the affinity was not causing violations at the time it
 * was chosen, it could cause violations thereafter. For example, 1000 jobs
 * could have had their affinity chosen during the powerup time, so any of
 * those 1000 jobs could cause an affinity violation later on.
 *
 * The attack would otherwise occur because other atoms/contexts have to wait for:
 * -# the currently running atoms (which are causing the violation) to
 * finish
 * -# and, the atoms that had their affinity chosen during powerup to
 * finish. These are run preferrentially because they don't cause a
 * violation, but instead continue to cause the violation in others.
 * -# or, the attacker is scheduled out (which might not happen for just 2
 * contexts)
 *
 * By re-choosing the affinity (which is designed to avoid violations at the
 * time it's chosen), we break condition (2) of the wait, which minimizes the
 * problem to just waiting for current jobs to finish (which can be bounded if
 * the Job Scheduling Policy has a timer).
 */
enum kbase_atom_coreref_state {
        /** Starting state: No affinity chosen, and cores must be requested. kbase_jd_atom::affinity==0 */
        KBASE_ATOM_COREREF_STATE_NO_CORES_REQUESTED,
        /** Cores requested, but waiting for them to be powered. Requested cores given by kbase_jd_atom::affinity */
        KBASE_ATOM_COREREF_STATE_WAITING_FOR_REQUESTED_CORES,
        /** Cores given by kbase_jd_atom::affinity are powered, but affinity might be out-of-date, so must recheck */
        KBASE_ATOM_COREREF_STATE_RECHECK_AFFINITY,
        /** Cores given by kbase_jd_atom::affinity are powered, and affinity is up-to-date, but must check for violations */
        KBASE_ATOM_COREREF_STATE_CHECK_AFFINITY_VIOLATIONS,
        /** Cores are powered, kbase_jd_atom::affinity up-to-date, no affinity violations: atom can be submitted to HW */
        KBASE_ATOM_COREREF_STATE_READY
};

enum kbase_jd_atom_state {
        /** Atom is not used */
        KBASE_JD_ATOM_STATE_UNUSED,
        /** Atom is queued in JD */
        KBASE_JD_ATOM_STATE_QUEUED,
        /** Atom has been given to JS (is runnable/running) */
        KBASE_JD_ATOM_STATE_IN_JS,
        /** Atom has been completed, but not yet handed back to userspace */
        KBASE_JD_ATOM_STATE_COMPLETED
};

typedef u8 base_atom_id; /**< Type big enough to store an atom number in */

struct base_dependency {
        base_atom_id  atom_id;               /**< An atom number */
        base_jd_dep_type dependency_type;    /**< Dependency type */
}; 

typedef struct base_jd_atom_v2 {
        mali_addr64 jc;                     /**< job-chain GPU address */
        struct base_jd_udata udata;                 /**< user data */
        kbase_pointer extres_list;          /**< list of external resources */
        u16 nr_extres;                      /**< nr of external resources */
        base_jd_core_req core_req;          /**< core requirements */
        const struct base_dependency pre_dep[2]; /**< pre-dependencies, one need to use SETTER function to assign this field,
        this is done in order to reduce possibility of improper assigment of a dependency field */
        base_atom_id atom_number;           /**< unique number to identify the atom */
        s8 prio;                            /**< priority - smaller is higher priority */
        u8 device_nr;                       /**< coregroup when BASE_JD_REQ_SPECIFIC_COHERENT_GROUP specified */
        u8 padding[5];
} base_jd_atom_v2;

#ifdef BASE_LEGACY_UK6_SUPPORT
struct base_jd_atom_v2_uk6 {
        mali_addr64 jc;                     /**< job-chain GPU address */
        struct base_jd_udata udata;                 /**< user data */
        kbase_pointer extres_list;          /**< list of external resources */
        u16 nr_extres;                      /**< nr of external resources */
        base_jd_core_req core_req;          /**< core requirements */
        base_atom_id pre_dep[2]; /**< pre-dependencies */
        base_atom_id atom_number;           /**< unique number to identify the atom */
        s8 prio;                            /**< priority - smaller is higher priority */
        u8 device_nr;                       /**< coregroup when BASE_JD_REQ_SPECIFIC_COHERENT_GROUP specified */
        u8 padding[7];
};
#endif

typedef enum base_external_resource_access {
        BASE_EXT_RES_ACCESS_SHARED,
        BASE_EXT_RES_ACCESS_EXCLUSIVE
} base_external_resource_access;

typedef struct base_external_resource {
        u64 ext_resource;
} base_external_resource;

/**
 * @brief Setter for a dependency structure
 *
 * @param[in] dep          The kbase jd atom dependency to be initialized.
 * @param     id           The atom_id to be assigned.
 * @param     dep_type     The dep_type to be assigned.
 *
 */
static INLINE void base_jd_atom_dep_set(const struct base_dependency* const_dep, base_atom_id id, base_jd_dep_type dep_type)
{
        struct base_dependency* dep;
        
        LOCAL_ASSERT(const_dep != NULL);
        /* make sure we don't set not allowed combinations of atom_id/dependency_type */
        LOCAL_ASSERT( ( id == 0 && dep_type == BASE_JD_DEP_TYPE_INVALID) || 
                                (id > 0 && dep_type != BASE_JD_DEP_TYPE_INVALID) );

        dep = REINTERPRET_CAST(struct base_dependency*)const_dep;

        dep->atom_id = id;
        dep->dependency_type = dep_type;
}

/**
 * @brief Make a copy of a dependency structure
 *
 * @param[in,out] dep          The kbase jd atom dependency to be written.
 * @param[in]     from         The dependency to make a copy from.
 *
 */
static INLINE void base_jd_atom_dep_copy(const struct base_dependency* const_dep, const struct base_dependency* from)
{
        LOCAL_ASSERT(const_dep != NULL);

        base_jd_atom_dep_set(const_dep, from->atom_id, from->dependency_type);
}

/**
 * @brief Soft-atom fence trigger setup.
 *
 * Sets up an atom to be a SW-only atom signaling a fence
 * when it reaches the run state.
 *
 * Using the existing base dependency system the fence can
 * be set to trigger when a GPU job has finished.
 *
 * The base fence object must not be terminated until the atom
 * has been submitted to @a base_jd_submit_bag and @a base_jd_submit_bag has returned.
 *
 * @a fence must be a valid fence set up with @a base_fence_init.
 * Calling this function with a uninitialized fence results in undefined behavior.
 *
 * @param[out] atom A pre-allocated atom to configure as a fence trigger SW atom
 * @param[in] fence The base fence object to trigger.
 */
static INLINE void base_jd_fence_trigger_setup_v2(struct base_jd_atom_v2 *atom, struct base_fence *fence)
{
        LOCAL_ASSERT(atom);
        LOCAL_ASSERT(fence);
        LOCAL_ASSERT(fence->basep.fd == INVALID_PLATFORM_FENCE);
        LOCAL_ASSERT(fence->basep.stream_fd >= 0);
        atom->jc = (uintptr_t) fence;
        atom->core_req = BASE_JD_REQ_SOFT_FENCE_TRIGGER;
}

/**
 * @brief Soft-atom fence wait setup.
 *
 * Sets up an atom to be a SW-only atom waiting on a fence.
 * When the fence becomes triggered the atom becomes runnable
 * and completes immediately.
 *
 * Using the existing base dependency system the fence can
 * be set to block a GPU job until it has been triggered.
 *
 * The base fence object must not be terminated until the atom
 * has been submitted to @a base_jd_submit_bag and @a base_jd_submit_bag has returned.
 *
 * @a fence must be a valid fence set up with @a base_fence_init or @a base_fence_import.
 * Calling this function with a uninitialized fence results in undefined behavior.
 *
 * @param[out] atom A pre-allocated atom to configure as a fence wait SW atom
 * @param[in] fence The base fence object to wait on
 */
static INLINE void base_jd_fence_wait_setup_v2(struct base_jd_atom_v2 *atom, struct base_fence *fence)
{
        LOCAL_ASSERT(atom);
        LOCAL_ASSERT(fence);
        LOCAL_ASSERT(fence->basep.fd >= 0);
        atom->jc = (uintptr_t) fence;
        atom->core_req = BASE_JD_REQ_SOFT_FENCE_WAIT;
}

/**
 * @brief External resource info initialization.
 *
 * Sets up a external resource object to reference
 * a memory allocation and the type of access requested.
 *
 * @param[in] res     The resource object to initialize
 * @param     handle  The handle to the imported memory object
 * @param     access  The type of access requested
 */
static INLINE void base_external_resource_init(struct base_external_resource * res, struct base_import_handle handle, base_external_resource_access access)
{
        mali_addr64 address;
        address = handle.basep.handle;

        LOCAL_ASSERT(res != NULL);
        LOCAL_ASSERT(0 == (address & LOCAL_PAGE_LSB));
        LOCAL_ASSERT(access == BASE_EXT_RES_ACCESS_SHARED || access == BASE_EXT_RES_ACCESS_EXCLUSIVE);

        res->ext_resource = address | (access & LOCAL_PAGE_LSB);
}

/**
 * @brief Job chain event code bits
 * Defines the bits used to create ::base_jd_event_code
 */
enum {
        BASE_JD_SW_EVENT_KERNEL = (1u << 15), /**< Kernel side event */
        BASE_JD_SW_EVENT = (1u << 14), /**< SW defined event */
        BASE_JD_SW_EVENT_SUCCESS = (1u << 13), /**< Event idicates success (SW events only) */
        BASE_JD_SW_EVENT_JOB = (0u << 11), /**< Job related event */
        BASE_JD_SW_EVENT_BAG = (1u << 11), /**< Bag related event */
        BASE_JD_SW_EVENT_INFO = (2u << 11), /**< Misc/info event */
        BASE_JD_SW_EVENT_RESERVED = (3u << 11), /**< Reserved event type */
        BASE_JD_SW_EVENT_TYPE_MASK = (3u << 11)     /**< Mask to extract the type from an event code */
};

/**
 * @brief Job chain event codes
 *
 * HW and low-level SW events are represented by event codes.
 * The status of jobs which succeeded are also represented by
 * an event code (see ::BASE_JD_EVENT_DONE).
 * Events are usually reported as part of a ::base_jd_event.
 *
 * The event codes are encoded in the following way:
 * @li 10:0  - subtype
 * @li 12:11 - type
 * @li 13    - SW success (only valid if the SW bit is set)
 * @li 14    - SW event (HW event if not set)
 * @li 15    - Kernel event (should never be seen in userspace)
 *
 * Events are split up into ranges as follows:
 * - BASE_JD_EVENT_RANGE_\<description\>_START
 * - BASE_JD_EVENT_RANGE_\<description\>_END
 *
 * \a code is in \<description\>'s range when:
 * - <tt>BASE_JD_EVENT_RANGE_\<description\>_START <= code < BASE_JD_EVENT_RANGE_\<description\>_END </tt>
 *
 * Ranges can be asserted for adjacency by testing that the END of the previous
 * is equal to the START of the next. This is useful for optimizing some tests
 * for range.
 *
 * A limitation is that the last member of this enum must explicitly be handled
 * (with an assert-unreachable statement) in switch statements that use
 * variables of this type. Otherwise, the compiler warns that we have not
 * handled that enum value.
 */
typedef enum base_jd_event_code {
        /* HW defined exceptions */

        /** Start of HW Non-fault status codes
         *
         * @note Obscurely, BASE_JD_EVENT_TERMINATED indicates a real fault,
         * because the job was hard-stopped
         */
        BASE_JD_EVENT_RANGE_HW_NONFAULT_START = 0,

        /* non-fatal exceptions */
        BASE_JD_EVENT_NOT_STARTED = 0x00, /**< Can't be seen by userspace, treated as 'previous job done' */
        BASE_JD_EVENT_DONE = 0x01,
        BASE_JD_EVENT_STOPPED = 0x03,     /**< Can't be seen by userspace, becomes TERMINATED, DONE or JOB_CANCELLED */
        BASE_JD_EVENT_TERMINATED = 0x04,  /**< This is actually a fault status code - the job was hard stopped */
        BASE_JD_EVENT_ACTIVE = 0x08,      /**< Can't be seen by userspace, jobs only returned on complete/fail/cancel */

        /** End of HW Non-fault status codes
         *
         * @note Obscurely, BASE_JD_EVENT_TERMINATED indicates a real fault,
         * because the job was hard-stopped
         */
        BASE_JD_EVENT_RANGE_HW_NONFAULT_END = 0x40,

        /** Start of HW fault and SW Error status codes */
        BASE_JD_EVENT_RANGE_HW_FAULT_OR_SW_ERROR_START = 0x40,

        /* job exceptions */
        BASE_JD_EVENT_JOB_CONFIG_FAULT = 0x40,
        BASE_JD_EVENT_JOB_POWER_FAULT = 0x41,
        BASE_JD_EVENT_JOB_READ_FAULT = 0x42,
        BASE_JD_EVENT_JOB_WRITE_FAULT = 0x43,
        BASE_JD_EVENT_JOB_AFFINITY_FAULT = 0x44,
        BASE_JD_EVENT_JOB_BUS_FAULT = 0x48,
        BASE_JD_EVENT_INSTR_INVALID_PC = 0x50,
        BASE_JD_EVENT_INSTR_INVALID_ENC = 0x51,
        BASE_JD_EVENT_INSTR_TYPE_MISMATCH = 0x52,
        BASE_JD_EVENT_INSTR_OPERAND_FAULT = 0x53,
        BASE_JD_EVENT_INSTR_TLS_FAULT = 0x54,
        BASE_JD_EVENT_INSTR_BARRIER_FAULT = 0x55,
        BASE_JD_EVENT_INSTR_ALIGN_FAULT = 0x56,
        BASE_JD_EVENT_DATA_INVALID_FAULT = 0x58,
        BASE_JD_EVENT_TILE_RANGE_FAULT = 0x59,
        BASE_JD_EVENT_STATE_FAULT = 0x5A,
        BASE_JD_EVENT_OUT_OF_MEMORY = 0x60,
        BASE_JD_EVENT_UNKNOWN = 0x7F,

        /* GPU exceptions */
        BASE_JD_EVENT_DELAYED_BUS_FAULT = 0x80,
        BASE_JD_EVENT_SHAREABILITY_FAULT = 0x88,

        /* MMU exceptions */
        BASE_JD_EVENT_TRANSLATION_FAULT_LEVEL1 = 0xC1,
        BASE_JD_EVENT_TRANSLATION_FAULT_LEVEL2 = 0xC2,
        BASE_JD_EVENT_TRANSLATION_FAULT_LEVEL3 = 0xC3,
        BASE_JD_EVENT_TRANSLATION_FAULT_LEVEL4 = 0xC4,
        BASE_JD_EVENT_PERMISSION_FAULT = 0xC8,
        BASE_JD_EVENT_TRANSTAB_BUS_FAULT_LEVEL1 = 0xD1,
        BASE_JD_EVENT_TRANSTAB_BUS_FAULT_LEVEL2 = 0xD2,
        BASE_JD_EVENT_TRANSTAB_BUS_FAULT_LEVEL3 = 0xD3,
        BASE_JD_EVENT_TRANSTAB_BUS_FAULT_LEVEL4 = 0xD4,
        BASE_JD_EVENT_ACCESS_FLAG = 0xD8,

        /* SW defined exceptions */
        BASE_JD_EVENT_MEM_GROWTH_FAILED = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x000,
        BASE_JD_EVENT_TIMED_OUT         = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x001,
        BASE_JD_EVENT_JOB_CANCELLED     = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x002,
        BASE_JD_EVENT_JOB_INVALID       = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x003,
        BASE_JD_EVENT_PM_EVENT          = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x004,
        BASE_JD_EVENT_FORCE_REPLAY      = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_JOB | 0x005,

        BASE_JD_EVENT_BAG_INVALID       = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_BAG | 0x003,

        /** End of HW fault and SW Error status codes */
        BASE_JD_EVENT_RANGE_HW_FAULT_OR_SW_ERROR_END = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_RESERVED | 0x3FF,

        /** Start of SW Success status codes */
        BASE_JD_EVENT_RANGE_SW_SUCCESS_START = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_SUCCESS | 0x000,

        BASE_JD_EVENT_PROGRESS_REPORT = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_SUCCESS | BASE_JD_SW_EVENT_JOB | 0x000,
        BASE_JD_EVENT_BAG_DONE = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_SUCCESS | BASE_JD_SW_EVENT_BAG | 0x000,
        BASE_JD_EVENT_DRV_TERMINATED = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_SUCCESS | BASE_JD_SW_EVENT_INFO | 0x000,

        /** End of SW Success status codes */
        BASE_JD_EVENT_RANGE_SW_SUCCESS_END = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_SUCCESS | BASE_JD_SW_EVENT_RESERVED | 0x3FF,

        /** Start of Kernel-only status codes. Such codes are never returned to user-space */
        BASE_JD_EVENT_RANGE_KERNEL_ONLY_START = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_KERNEL | 0x000,
        BASE_JD_EVENT_REMOVED_FROM_NEXT = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_KERNEL | BASE_JD_SW_EVENT_JOB | 0x000,

        /** End of Kernel-only status codes. */
        BASE_JD_EVENT_RANGE_KERNEL_ONLY_END = BASE_JD_SW_EVENT | BASE_JD_SW_EVENT_KERNEL | BASE_JD_SW_EVENT_RESERVED | 0x3FF
} base_jd_event_code;

/**
 * @brief Event reporting structure
 *
 * This structure is used by the kernel driver to report information
 * about GPU events. The can either be HW-specific events or low-level
 * SW events, such as job-chain completion.
 *
 * The event code contains an event type field which can be extracted
 * by ANDing with ::BASE_JD_SW_EVENT_TYPE_MASK.
 *
 * Based on the event type base_jd_event::data holds:
 * @li ::BASE_JD_SW_EVENT_JOB : the offset in the ring-buffer for the completed
 * job-chain
 * @li ::BASE_JD_SW_EVENT_BAG : The address of the ::base_jd_bag that has
 * been completed (ie all contained job-chains have been completed).
 * @li ::BASE_JD_SW_EVENT_INFO : base_jd_event::data not used
 */
typedef struct base_jd_event_v2 {
        base_jd_event_code event_code;  /**< event code */
        base_atom_id atom_number;       /**< the atom number that has completed */
        struct base_jd_udata udata;     /**< user data */
} base_jd_event_v2;

/**
 * Padding required to ensure that the @ref struct base_dump_cpu_gpu_counters structure fills
 * a full cache line.
 */

#define BASE_CPU_GPU_CACHE_LINE_PADDING (36)


/**
 * @brief Structure for BASE_JD_REQ_SOFT_DUMP_CPU_GPU_COUNTERS jobs.
 *
 * This structure is stored into the memory pointed to by the @c jc field of @ref base_jd_atom.
 *
 * This structure must be padded to ensure that it will occupy whole cache lines. This is to avoid
 * cases where access to pages containing the structure is shared between cached and un-cached
 * memory regions, which would cause memory corruption.  Here we set the structure size to be 64 bytes
 * which is the cache line for ARM A15 processors.
 */

typedef struct base_dump_cpu_gpu_counters {
        u64 system_time;
        u64 cycle_counter;
        u64 sec;
        u32 usec;
        u8 padding[BASE_CPU_GPU_CACHE_LINE_PADDING];
} base_dump_cpu_gpu_counters;



/** @} end group base_user_api_job_dispatch */

#ifdef __KERNEL__
/*
 * The following typedefs should be removed when a midg types header is added.
 * See MIDCOM-1657 for details.
 */
typedef u32 midg_product_id;
typedef u32 midg_cache_features;
typedef u32 midg_tiler_features;
typedef u32 midg_mem_features;
typedef u32 midg_mmu_features;
typedef u32 midg_js_features;
typedef u32 midg_as_present;
typedef u32 midg_js_present;

#define MIDG_MAX_JOB_SLOTS 16

#else
#include <midg/mali_midg.h>
#endif

/**
 * @page page_base_user_api_gpuprops User-side Base GPU Property Query API
 *
 * The User-side Base GPU Property Query API encapsulates two
 * sub-modules:
 *
 * - @ref base_user_api_gpuprops_dyn "Dynamic GPU Properties"
 * - @ref base_plat_config_gpuprops "Base Platform Config GPU Properties"
 *
 * There is a related third module outside of Base, which is owned by the MIDG
 * module:
 * - @ref midg_gpuprops_static "Midgard Compile-time GPU Properties"
 *
 * Base only deals with properties that vary between different Midgard
 * implementations - the Dynamic GPU properties and the Platform Config
 * properties.
 *
 * For properties that are constant for the Midgard Architecture, refer to the
 * MIDG module. However, we will discuss their relevance here <b>just to
 * provide background information.</b>
 *
 * @section sec_base_user_api_gpuprops_about About the GPU Properties in Base and MIDG modules
 *
 * The compile-time properties (Platform Config, Midgard Compile-time
 * properties) are exposed as pre-processor macros.
 *
 * Complementing the compile-time properties are the Dynamic GPU
 * Properties, which act as a conduit for the Midgard Configuration
 * Discovery.
 *
 * In general, the dynamic properties are present to verify that the platform
 * has been configured correctly with the right set of Platform Config
 * Compile-time Properties.
 *
 * As a consistant guide across the entire DDK, the choice for dynamic or
 * compile-time should consider the following, in order:
 * -# Can the code be written so that it doesn't need to know the
 * implementation limits at all?
 * -# If you need the limits, get the information from the Dynamic Property
 * lookup. This should be done once as you fetch the context, and then cached
 * as part of the context data structure, so it's cheap to access.
 * -# If there's a clear and arguable inefficiency in using Dynamic Properties,
 * then use a Compile-Time Property (Platform Config, or Midgard Compile-time
 * property). Examples of where this might be sensible follow:
 *  - Part of a critical inner-loop
 *  - Frequent re-use throughout the driver, causing significant extra load
 * instructions or control flow that would be worthwhile optimizing out.
 *
 * We cannot provide an exhaustive set of examples, neither can we provide a
 * rule for every possible situation. Use common sense, and think about: what
 * the rest of the driver will be doing; how the compiler might represent the
 * value if it is a compile-time constant; whether an OEM shipping multiple
 * devices would benefit much more from a single DDK binary, instead of
 * insignificant micro-optimizations.
 *
 * @section sec_base_user_api_gpuprops_dyn Dynamic GPU Properties
 *
 * Dynamic GPU properties are presented in two sets:
 * -# the commonly used properties in @ref base_gpu_props, which have been
 * unpacked from GPU register bitfields.
 * -# The full set of raw, unprocessed properties in @ref midg_raw_gpu_props
 * (also a member of @ref base_gpu_props). All of these are presented in
 * the packed form, as presented by the GPU  registers themselves.
 *
 * @usecase The raw properties in @ref midg_raw_gpu_props are necessary to
 * allow a user of the Mali Tools (e.g. PAT) to determine "Why is this device
 * behaving differently?". In this case, all information about the
 * configuration is potentially useful, but it <b>does not need to be processed
 * by the driver</b>. Instead, the raw registers can be processed by the Mali
 * Tools software on the host PC.
 *
 * The properties returned extend the Midgard Configuration Discovery
 * registers. For example, GPU clock speed is not specified in the Midgard
 * Architecture, but is <b>necessary for OpenCL's clGetDeviceInfo() function</b>.
 *
 * The GPU properties are obtained by a call to
 * _mali_base_get_gpu_props(). This simply returns a pointer to a const
 * base_gpu_props structure. It is constant for the life of a base
 * context. Multiple calls to _mali_base_get_gpu_props() to a base context
 * return the same pointer to a constant structure. This avoids cache pollution
 * of the common data.
 *
 * This pointer must not be freed, because it does not point to the start of a
 * region allocated by the memory allocator; instead, just close the @ref
 * base_context.
 *
 *
 * @section sec_base_user_api_gpuprops_config Platform Config Compile-time Properties
 *
 * The Platform Config File sets up gpu properties that are specific to a
 * certain platform. Properties that are 'Implementation Defined' in the
 * Midgard Architecture spec are placed here.
 *
 * @note Reference configurations are provided for Midgard Implementations, such as
 * the Mali-T600 family. The customer need not repeat this information, and can select one of
 * these reference configurations. For example, VA_BITS, PA_BITS and the
 * maximum number of samples per pixel might vary between Midgard Implementations, but
 * \b not for platforms using the Mali-T604. This information is placed in
 * the reference configuration files.
 *
 * The System Integrator creates the following structure:
 * - platform_XYZ
 * - platform_XYZ/plat
 * - platform_XYZ/plat/plat_config.h
 *
 * They then edit plat_config.h, using the example plat_config.h files as a
 * guide.
 *
 * At the very least, the customer must set @ref CONFIG_GPU_CORE_TYPE, and will
 * receive a helpful \#error message if they do not do this correctly. This
 * selects the Reference Configuration for the Midgard Implementation. The rationale
 * behind this decision (against asking the customer to write \#include
 * <gpus/mali_t600.h> in their plat_config.h) is as follows:
 * - This mechanism 'looks' like a regular config file (such as Linux's
 * .config)
 * - It is difficult to get wrong in a way that will produce strange build
 * errors:
 *  - They need not know where the mali_t600.h, other_midg_gpu.h etc. files are stored - and
 *  so they won't accidentally pick another file with 'mali_t600' in its name
 *  - When the build doesn't work, the System Integrator may think the DDK is
 *  doesn't work, and attempt to fix it themselves:
 *   - For the @ref CONFIG_GPU_CORE_TYPE mechanism, the only way to get past the
 *   error is to set @ref CONFIG_GPU_CORE_TYPE, and this is what the \#error tells
 *   you.
 *   - For a \#include mechanism, checks must still be made elsewhere, which the
 *   System Integrator may try working around by setting \#defines (such as
 *   VA_BITS) themselves in their plat_config.h. In the  worst case, they may
 *   set the prevention-mechanism \#define of
 *   "A_CORRECT_MIDGARD_CORE_WAS_CHOSEN".
 *   - In this case, they would believe they are on the right track, because
 *   the build progresses with their fix, but with errors elsewhere.
 *
 * However, there is nothing to prevent the customer using \#include to organize
 * their own configurations files hierarchically.
 *
 * The mechanism for the header file processing is as follows:
 *
 * @dot
   digraph plat_config_mechanism {
           rankdir=BT
           size="6,6"

       "mali_base.h";
           "midg/midg.h";

           node [ shape=box ];
           {
               rank = same; ordering = out;

                   "midg/midg_gpu_props.h";
                   "base/midg_gpus/mali_t600.h";
                   "base/midg_gpus/other_midg_gpu.h";
           }
           { rank = same; "plat/plat_config.h"; }
           {
               rank = same;
                   "midg/midg.h" [ shape=box ];
                   gpu_chooser [ label="" style="invisible" width=0 height=0 fixedsize=true ];
                   select_gpu [ label="Mali-T600 | Other\n(select_gpu.h)" shape=polygon,sides=4,distortion=0.25 width=3.3 height=0.99 fixedsize=true ] ;
           }
           node [ shape=box ];
           { rank = same; "plat/plat_config.h"; }
           { rank = same; "mali_base.h"; }

           "mali_base.h" -> "midg/midg.h" -> "midg/midg_gpu_props.h";
           "mali_base.h" -> "plat/plat_config.h" ;
           "mali_base.h" -> select_gpu ;

           "plat/plat_config.h" -> gpu_chooser [style="dotted,bold" dir=none weight=4] ;
           gpu_chooser -> select_gpu [style="dotted,bold"] ;

           select_gpu -> "base/midg_gpus/mali_t600.h" ;
           select_gpu -> "base/midg_gpus/other_midg_gpu.h" ;
   }
   @enddot
 *
 *
 * @section sec_base_user_api_gpuprops_kernel Kernel Operation
 *
 * During Base Context Create time, user-side makes a single kernel call:
 * - A call to fill user memory with GPU information structures
 *
 * The kernel-side will fill the provided the entire processed @ref base_gpu_props
 * structure, because this information is required in both
 * user and kernel side; it does not make sense to decode it twice.
 *
 * Coherency groups must be derived from the bitmasks, but this can be done
 * kernel side, and just once at kernel startup: Coherency groups must already
 * be known kernel-side, to support chains that specify a 'Only Coherent Group'
 * SW requirement, or 'Only Coherent Group with Tiler' SW requirement.
 *
 * @section sec_base_user_api_gpuprops_cocalc Coherency Group calculation
 * Creation of the coherent group data is done at device-driver startup, and so
 * is one-time. This will most likely involve a loop with CLZ, shifting, and
 * bit clearing on the L2_PRESENT or L3_PRESENT masks, depending on whether the
 * system is L2 or L2+L3 Coherent. The number of shader cores is done by a
 * population count, since faulty cores may be disabled during production,
 * producing a non-contiguous mask.
 *
 * The memory requirements for this algoirthm can be determined either by a u64
 * population count on the L2/L3_PRESENT masks (a LUT helper already is
 * requried for the above), or simple assumption that there can be no more than
 * 16 coherent groups, since core groups are typically 4 cores.
 */

/**
 * @addtogroup base_user_api_gpuprops User-side Base GPU Property Query APIs
 * @{
 */

/**
 * @addtogroup base_user_api_gpuprops_dyn Dynamic HW Properties
 * @{
 */

#define BASE_GPU_NUM_TEXTURE_FEATURES_REGISTERS 3

#define BASE_MAX_COHERENT_GROUPS 16

struct mali_base_gpu_core_props {
        /**
         * Product specific value.
         */
        midg_product_id product_id;

        /**
         * Status of the GPU release.
     * No defined values, but starts at 0 and increases by one for each release
     * status (alpha, beta, EAC, etc.).
     * 4 bit values (0-15).
         */
        u16 version_status;

        /**
         * Minor release number of the GPU. "P" part of an "RnPn" release number.
     * 8 bit values (0-255).
         */
        u16 minor_revision;

        /**
         * Major release number of the GPU. "R" part of an "RnPn" release number.
     * 4 bit values (0-15).
         */
        u16 major_revision;

        u16 padding;

        /**
         * @usecase GPU clock speed is not specified in the Midgard Architecture, but is
         * <b>necessary for OpenCL's clGetDeviceInfo() function</b>.
         */
        u32 gpu_speed_mhz;

        /**
         * @usecase GPU clock max/min speed is required for computing best/worst case
         * in tasks as job scheduling ant irq_throttling. (It is not specified in the
         *  Midgard Architecture).
         */
        u32 gpu_freq_khz_max;
        u32 gpu_freq_khz_min;

        /**
         * Size of the shader program counter, in bits.
         */
        u32 log2_program_counter_size;

        /**
         * TEXTURE_FEATURES_x registers, as exposed by the GPU. This is a
         * bitpattern where a set bit indicates that the format is supported.
         *
         * Before using a texture format, it is recommended that the corresponding
         * bit be checked.
         */
        u32 texture_features[BASE_GPU_NUM_TEXTURE_FEATURES_REGISTERS];

        /**
         * Theoretical maximum memory available to the GPU. It is unlikely that a
         * client will be able to allocate all of this memory for their own
         * purposes, but this at least provides an upper bound on the memory
         * available to the GPU.
         *
         * This is required for OpenCL's clGetDeviceInfo() call when
         * CL_DEVICE_GLOBAL_MEM_SIZE is requested, for OpenCL GPU devices. The
         * client will not be expecting to allocate anywhere near this value.
         */
        u64 gpu_available_memory_size;
};

/**
 *
 * More information is possible - but associativity and bus width are not
 * required by upper-level apis.
 */
struct mali_base_gpu_l2_cache_props {
        u8 log2_line_size;
        u8 log2_cache_size;
        u8 num_l2_slices; /* Number of L2C slices. 1 or higher */
        u8 padding[5];
};

struct mali_base_gpu_l3_cache_props {
        u8 log2_line_size;
        u8 log2_cache_size;
        u8 padding[6];
};

struct mali_base_gpu_tiler_props {
        u32 bin_size_bytes;     /* Max is 4*2^15 */
        u32 max_active_levels;  /* Max is 2^15 */
};

/**
 * GPU threading system details.  
 */
struct mali_base_gpu_thread_props {
        u32 max_threads;            /* Max. number of threads per core */ 
        u32 max_workgroup_size;     /* Max. number of threads per workgroup */
        u32 max_barrier_size;       /* Max. number of threads that can synchronize on a simple barrier */
        u16 max_registers;                      /* Total size [1..65535] of the register file available per core. */
        u8  max_task_queue;                     /* Max. tasks [1..255] which may be sent to a core before it becomes blocked. */
        u8  max_thread_group_split;     /* Max. allowed value [1..15] of the Thread Group Split field. */
        u8  impl_tech;                  /* 0 = Not specified, 1 = Silicon, 2 = FPGA, 3 = SW Model/Emulation */
        u8  padding[7];
};

/**
 * @brief descriptor for a coherent group
 *
 * \c core_mask exposes all cores in that coherent group, and \c num_cores
 * provides a cached population-count for that mask.
 *
 * @note Whilst all cores are exposed in the mask, not all may be available to
 * the application, depending on the Kernel Job Scheduler policy. Therefore,
 * the application should not further restrict the core mask itself, as it may
 * result in an empty core mask. However, it can guarentee that there will be
 * at least one core available for each core group exposed .
 *
 * @usecase Chains marked at certain user-side priorities (e.g. the Long-running
 * (batch) priority ) can be prevented from running on entire core groups by the
 * Kernel Chain Scheduler policy.
 *
 * @note if u64s must be 8-byte aligned, then this structure has 32-bits of wastage.
 */
struct mali_base_gpu_coherent_group {
        u64 core_mask;         /**< Core restriction mask required for the group */
        u16 num_cores;         /**< Number of cores in the group */
        u16 padding[3];
};

/**
 * @brief Coherency group information
 *
 * Note that the sizes of the members could be reduced. However, the \c group
 * member might be 8-byte aligned to ensure the u64 core_mask is 8-byte
 * aligned, thus leading to wastage if the other members sizes were reduced.
 *
 * The groups are sorted by core mask. The core masks are non-repeating and do
 * not intersect.
 */
struct mali_base_gpu_coherent_group_info {
        u32 num_groups;

        /**
         * Number of core groups (coherent or not) in the GPU. Equivalent to the number of L2 Caches.
         *
         * The GPU Counter dumping writes 2048 bytes per core group, regardless of
         * whether the core groups are coherent or not. Hence this member is needed
         * to calculate how much memory is required for dumping.
         *
         * @note Do not use it to work out how many valid elements are in the
         * group[] member. Use num_groups instead.
         */
        u32 num_core_groups;

        /**
         * Coherency features of the memory, accessed by @ref midg_mem_features
         * methods
         */
        midg_mem_features coherency;

        u32 padding;

        /**
         * Descriptors of coherent groups
         */
        struct mali_base_gpu_coherent_group group[BASE_MAX_COHERENT_GROUPS];
};

/**
 * A complete description of the GPU's Hardware Configuration Discovery
 * registers.
 *
 * The information is presented inefficiently for access. For frequent access,
 * the values should be better expressed in an unpacked form in the
 * base_gpu_props structure.
 *
 * @usecase The raw properties in @ref midg_raw_gpu_props are necessary to
 * allow a user of the Mali Tools (e.g. PAT) to determine "Why is this device
 * behaving differently?". In this case, all information about the
 * configuration is potentially useful, but it <b>does not need to be processed
 * by the driver</b>. Instead, the raw registers can be processed by the Mali
 * Tools software on the host PC.
 *
 */
struct midg_raw_gpu_props {
        u64 shader_present;
        u64 tiler_present;
        u64 l2_present;
        u64 l3_present;

        midg_cache_features l2_features;
        midg_cache_features l3_features;
        midg_mem_features mem_features;
        midg_mmu_features mmu_features;

        midg_as_present as_present;

        u32 js_present;
        midg_js_features js_features[MIDG_MAX_JOB_SLOTS];
        midg_tiler_features tiler_features;
        u32 texture_features[3];

        u32 gpu_id;
        
        u32 thread_max_threads;
        u32 thread_max_workgroup_size;
        u32 thread_max_barrier_size;
        u32 thread_features;

        u32 padding;
};

/**
 * Return structure for _mali_base_get_gpu_props().
 *
 * NOTE: the raw_props member in this datastructure contains the register
 * values from which the value of the other members are derived. The derived
 * members exist to allow for efficient access and/or shielding the details
 * of the layout of the registers.
 *
 */
typedef struct mali_base_gpu_props {
        struct mali_base_gpu_core_props core_props;
        struct mali_base_gpu_l2_cache_props l2_props;
        struct mali_base_gpu_l3_cache_props l3_props;
        struct mali_base_gpu_tiler_props tiler_props;
        struct mali_base_gpu_thread_props thread_props;

        /** This member is large, likely to be 128 bytes */
        struct midg_raw_gpu_props raw_props;

        /** This must be last member of the structure */
        struct mali_base_gpu_coherent_group_info coherency_info;
} base_gpu_props;

/** @} end group base_user_api_gpuprops_dyn */

/** @} end group base_user_api_gpuprops */

/**
 * @addtogroup base_user_api_core User-side Base core APIs
 * @{
 */

/**
 * \enum base_context_create_flags
 *
 * Flags to pass to ::base_context_init.
 * Flags can be ORed together to enable multiple things.
 *
 * These share the same space as @ref basep_context_private_flags, and so must
 * not collide with them.
 */
enum base_context_create_flags {
        /** No flags set */
        BASE_CONTEXT_CREATE_FLAG_NONE = 0,

        /** Base context is embedded in a cctx object (flag used for CINSTR software counter macros) */
        BASE_CONTEXT_CCTX_EMBEDDED = (1u << 0),

        /** Base context is a 'System Monitor' context for Hardware counters.
         *
         * One important side effect of this is that job submission is disabled. */
        BASE_CONTEXT_SYSTEM_MONITOR_SUBMIT_DISABLED = (1u << 1),

        /** Base context flag indicating a 'hint' that this context uses Compute
         * Jobs only.
         *
         * Specifially, this means that it only sends atoms that <b>do not</b>
         * contain the following @ref base_jd_core_req :
         * - BASE_JD_REQ_FS
         * - BASE_JD_REQ_T
         *
         * Violation of these requirements will cause the Job-Chains to be rejected.
         *
         * In addition, it is inadvisable for the atom's Job-Chains to contain Jobs
         * of the following @ref midg_job_type (whilst it may work now, it may not
         * work in future) :
         * - @ref MIDG_JOB_VERTEX
         * - @ref MIDG_JOB_GEOMETRY
         *
         * @note An alternative to using this is to specify the BASE_JD_REQ_ONLY_COMPUTE
         * requirement in atoms.
         */
        BASE_CONTEXT_HINT_ONLY_COMPUTE = (1u << 2)
};

/**
 * Bitpattern describing the ::base_context_create_flags that can be passed to base_context_init()
 */
#define BASE_CONTEXT_CREATE_ALLOWED_FLAGS \
        (((u32)BASE_CONTEXT_CCTX_EMBEDDED) | \
          ((u32)BASE_CONTEXT_SYSTEM_MONITOR_SUBMIT_DISABLED) | \
          ((u32)BASE_CONTEXT_HINT_ONLY_COMPUTE))

/**
 * Bitpattern describing the ::base_context_create_flags that can be passed to the kernel
 */
#define BASE_CONTEXT_CREATE_KERNEL_FLAGS \
        (((u32)BASE_CONTEXT_SYSTEM_MONITOR_SUBMIT_DISABLED) | \
          ((u32)BASE_CONTEXT_HINT_ONLY_COMPUTE))

/**
 * Private flags used on the base context
 *
 * These start at bit 31, and run down to zero.
 *
 * They share the same space as @ref base_context_create_flags, and so must
 * not collide with them.
 */
enum basep_context_private_flags {
        /** Private flag tracking whether job descriptor dumping is disabled */
        BASEP_CONTEXT_FLAG_JOB_DUMP_DISABLED = (1 << 31)
};

/** @} end group base_user_api_core */

/** @} end group base_user_api */

/**
 * @addtogroup base_plat_config_gpuprops Base Platform Config GPU Properties
 * @{
 *
 * C Pre-processor macros are exposed here to do with Platform
 * Config.
 *
 * These include:
 * - GPU Properties that are constant on a particular Midgard Family
 * Implementation e.g. Maximum samples per pixel on Mali-T600.
 * - General platform config for the GPU, such as the GPU major and minor
 * revison.
 */

/** @} end group base_plat_config_gpuprops */

/**
 * @addtogroup base_api Base APIs
 * @{
 */

/**
 * @brief The payload for a replay job. This must be in GPU memory.
 */
typedef struct base_jd_replay_payload {
        /**
         * Pointer to the first entry in the base_jd_replay_jc list.  These
         * will be replayed in @b reverse order (so that extra ones can be added
         * to the head in future soft jobs without affecting this soft job)
         */
        mali_addr64 tiler_jc_list;

        /**
         * Pointer to the fragment job chain.
         */
        mali_addr64 fragment_jc;

        /**
         * Pointer to the tiler heap free FBD field to be modified.
         */
        mali_addr64 tiler_heap_free;

        /**
         * Hierarchy mask for the replayed fragment jobs. May be zero.
         */
        u16 fragment_hierarchy_mask;

        /**
         * Hierarchy mask for the replayed tiler jobs. May be zero.
         */
        u16 tiler_hierarchy_mask;

        /**
         * Default weight to be used for hierarchy levels not in the original
         * mask.
         */
        u32 hierarchy_default_weight;

        /**
         * Core requirements for the tiler job chain
         */
        base_jd_core_req tiler_core_req;

        /**
         * Core requirements for the fragment job chain
         */
        base_jd_core_req fragment_core_req;

        u8 padding[4];
} base_jd_replay_payload;

/**
 * @brief An entry in the linked list of job chains to be replayed. This must
 *        be in GPU memory.
 */
typedef struct base_jd_replay_jc {
        /**
         * Pointer to next entry in the list. A setting of NULL indicates the
         * end of the list.
         */
        mali_addr64 next;

        /**
         * Pointer to the job chain.
         */
        mali_addr64 jc;

} base_jd_replay_jc;

/* Maximum number of jobs allowed in a fragment chain in the payload of a
 * replay job */
#define BASE_JD_REPLAY_F_CHAIN_JOB_LIMIT 256

/** @} end group base_api */

typedef struct base_profiling_controls {
        u32 profiling_controls[FBDUMP_CONTROL_MAX];
} base_profiling_controls;

#endif                          /* _BASE_KERNEL_H_ */