cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

vc4_drm.h (14472B)


      1/*
      2 * Copyright © 2014-2015 Broadcom
      3 *
      4 * Permission is hereby granted, free of charge, to any person obtaining a
      5 * copy of this software and associated documentation files (the "Software"),
      6 * to deal in the Software without restriction, including without limitation
      7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
      8 * and/or sell copies of the Software, and to permit persons to whom the
      9 * Software is furnished to do so, subject to the following conditions:
     10 *
     11 * The above copyright notice and this permission notice (including the next
     12 * paragraph) shall be included in all copies or substantial portions of the
     13 * Software.
     14 *
     15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
     18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
     19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
     20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
     21 * IN THE SOFTWARE.
     22 */
     23
     24#ifndef _UAPI_VC4_DRM_H_
     25#define _UAPI_VC4_DRM_H_
     26
     27#include "drm.h"
     28
     29#if defined(__cplusplus)
     30extern "C" {
     31#endif
     32
     33#define DRM_VC4_SUBMIT_CL                         0x00
     34#define DRM_VC4_WAIT_SEQNO                        0x01
     35#define DRM_VC4_WAIT_BO                           0x02
     36#define DRM_VC4_CREATE_BO                         0x03
     37#define DRM_VC4_MMAP_BO                           0x04
     38#define DRM_VC4_CREATE_SHADER_BO                  0x05
     39#define DRM_VC4_GET_HANG_STATE                    0x06
     40#define DRM_VC4_GET_PARAM                         0x07
     41#define DRM_VC4_SET_TILING                        0x08
     42#define DRM_VC4_GET_TILING                        0x09
     43#define DRM_VC4_LABEL_BO                          0x0a
     44#define DRM_VC4_GEM_MADVISE                       0x0b
     45#define DRM_VC4_PERFMON_CREATE                    0x0c
     46#define DRM_VC4_PERFMON_DESTROY                   0x0d
     47#define DRM_VC4_PERFMON_GET_VALUES                0x0e
     48
     49#define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
     50#define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
     51#define DRM_IOCTL_VC4_WAIT_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_WAIT_BO, struct drm_vc4_wait_bo)
     52#define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
     53#define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
     54#define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
     55#define DRM_IOCTL_VC4_GET_HANG_STATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state)
     56#define DRM_IOCTL_VC4_GET_PARAM           DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_PARAM, struct drm_vc4_get_param)
     57#define DRM_IOCTL_VC4_SET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_SET_TILING, struct drm_vc4_set_tiling)
     58#define DRM_IOCTL_VC4_GET_TILING          DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GET_TILING, struct drm_vc4_get_tiling)
     59#define DRM_IOCTL_VC4_LABEL_BO            DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_LABEL_BO, struct drm_vc4_label_bo)
     60#define DRM_IOCTL_VC4_GEM_MADVISE         DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_GEM_MADVISE, struct drm_vc4_gem_madvise)
     61#define DRM_IOCTL_VC4_PERFMON_CREATE      DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_CREATE, struct drm_vc4_perfmon_create)
     62#define DRM_IOCTL_VC4_PERFMON_DESTROY     DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_DESTROY, struct drm_vc4_perfmon_destroy)
     63#define DRM_IOCTL_VC4_PERFMON_GET_VALUES  DRM_IOWR(DRM_COMMAND_BASE + DRM_VC4_PERFMON_GET_VALUES, struct drm_vc4_perfmon_get_values)
     64
     65struct drm_vc4_submit_rcl_surface {
     66	__u32 hindex; /* Handle index, or ~0 if not present. */
     67	__u32 offset; /* Offset to start of buffer. */
     68	/*
     69	 * Bits for either render config (color_write) or load/store packet.
     70	 * Bits should all be 0 for MSAA load/stores.
     71	 */
     72	__u16 bits;
     73
     74#define VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES		(1 << 0)
     75	__u16 flags;
     76};
     77
     78/**
     79 * struct drm_vc4_submit_cl - ioctl argument for submitting commands to the 3D
     80 * engine.
     81 *
     82 * Drivers typically use GPU BOs to store batchbuffers / command lists and
     83 * their associated state.  However, because the VC4 lacks an MMU, we have to
     84 * do validation of memory accesses by the GPU commands.  If we were to store
     85 * our commands in BOs, we'd need to do uncached readback from them to do the
     86 * validation process, which is too expensive.  Instead, userspace accumulates
     87 * commands and associated state in plain memory, then the kernel copies the
     88 * data to its own address space, and then validates and stores it in a GPU
     89 * BO.
     90 */
     91struct drm_vc4_submit_cl {
     92	/* Pointer to the binner command list.
     93	 *
     94	 * This is the first set of commands executed, which runs the
     95	 * coordinate shader to determine where primitives land on the screen,
     96	 * then writes out the state updates and draw calls necessary per tile
     97	 * to the tile allocation BO.
     98	 */
     99	__u64 bin_cl;
    100
    101	/* Pointer to the shader records.
    102	 *
    103	 * Shader records are the structures read by the hardware that contain
    104	 * pointers to uniforms, shaders, and vertex attributes.  The
    105	 * reference to the shader record has enough information to determine
    106	 * how many pointers are necessary (fixed number for shaders/uniforms,
    107	 * and an attribute count), so those BO indices into bo_handles are
    108	 * just stored as __u32s before each shader record passed in.
    109	 */
    110	__u64 shader_rec;
    111
    112	/* Pointer to uniform data and texture handles for the textures
    113	 * referenced by the shader.
    114	 *
    115	 * For each shader state record, there is a set of uniform data in the
    116	 * order referenced by the record (FS, VS, then CS).  Each set of
    117	 * uniform data has a __u32 index into bo_handles per texture
    118	 * sample operation, in the order the QPU_W_TMUn_S writes appear in
    119	 * the program.  Following the texture BO handle indices is the actual
    120	 * uniform data.
    121	 *
    122	 * The individual uniform state blocks don't have sizes passed in,
    123	 * because the kernel has to determine the sizes anyway during shader
    124	 * code validation.
    125	 */
    126	__u64 uniforms;
    127	__u64 bo_handles;
    128
    129	/* Size in bytes of the binner command list. */
    130	__u32 bin_cl_size;
    131	/* Size in bytes of the set of shader records. */
    132	__u32 shader_rec_size;
    133	/* Number of shader records.
    134	 *
    135	 * This could just be computed from the contents of shader_records and
    136	 * the address bits of references to them from the bin CL, but it
    137	 * keeps the kernel from having to resize some allocations it makes.
    138	 */
    139	__u32 shader_rec_count;
    140	/* Size in bytes of the uniform state. */
    141	__u32 uniforms_size;
    142
    143	/* Number of BO handles passed in (size is that times 4). */
    144	__u32 bo_handle_count;
    145
    146	/* RCL setup: */
    147	__u16 width;
    148	__u16 height;
    149	__u8 min_x_tile;
    150	__u8 min_y_tile;
    151	__u8 max_x_tile;
    152	__u8 max_y_tile;
    153	struct drm_vc4_submit_rcl_surface color_read;
    154	struct drm_vc4_submit_rcl_surface color_write;
    155	struct drm_vc4_submit_rcl_surface zs_read;
    156	struct drm_vc4_submit_rcl_surface zs_write;
    157	struct drm_vc4_submit_rcl_surface msaa_color_write;
    158	struct drm_vc4_submit_rcl_surface msaa_zs_write;
    159	__u32 clear_color[2];
    160	__u32 clear_z;
    161	__u8 clear_s;
    162
    163	__u32 pad:24;
    164
    165#define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
    166/* By default, the kernel gets to choose the order that the tiles are
    167 * rendered in.  If this is set, then the tiles will be rendered in a
    168 * raster order, with the right-to-left vs left-to-right and
    169 * top-to-bottom vs bottom-to-top dictated by
    170 * VC4_SUBMIT_CL_RCL_ORDER_INCREASING_*.  This allows overlapping
    171 * blits to be implemented using the 3D engine.
    172 */
    173#define VC4_SUBMIT_CL_FIXED_RCL_ORDER			(1 << 1)
    174#define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X		(1 << 2)
    175#define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y		(1 << 3)
    176	__u32 flags;
    177
    178	/* Returned value of the seqno of this render job (for the
    179	 * wait ioctl).
    180	 */
    181	__u64 seqno;
    182
    183	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
    184	__u32 perfmonid;
    185
    186	/* Syncobj handle to wait on. If set, processing of this render job
    187	 * will not start until the syncobj is signaled. 0 means ignore.
    188	 */
    189	__u32 in_sync;
    190
    191	/* Syncobj handle to export fence to. If set, the fence in the syncobj
    192	 * will be replaced with a fence that signals upon completion of this
    193	 * render job. 0 means ignore.
    194	 */
    195	__u32 out_sync;
    196
    197	__u32 pad2;
    198};
    199
    200/**
    201 * struct drm_vc4_wait_seqno - ioctl argument for waiting for
    202 * DRM_VC4_SUBMIT_CL completion using its returned seqno.
    203 *
    204 * timeout_ns is the timeout in nanoseconds, where "0" means "don't
    205 * block, just return the status."
    206 */
    207struct drm_vc4_wait_seqno {
    208	__u64 seqno;
    209	__u64 timeout_ns;
    210};
    211
    212/**
    213 * struct drm_vc4_wait_bo - ioctl argument for waiting for
    214 * completion of the last DRM_VC4_SUBMIT_CL on a BO.
    215 *
    216 * This is useful for cases where multiple processes might be
    217 * rendering to a BO and you want to wait for all rendering to be
    218 * completed.
    219 */
    220struct drm_vc4_wait_bo {
    221	__u32 handle;
    222	__u32 pad;
    223	__u64 timeout_ns;
    224};
    225
    226/**
    227 * struct drm_vc4_create_bo - ioctl argument for creating VC4 BOs.
    228 *
    229 * There are currently no values for the flags argument, but it may be
    230 * used in a future extension.
    231 */
    232struct drm_vc4_create_bo {
    233	__u32 size;
    234	__u32 flags;
    235	/** Returned GEM handle for the BO. */
    236	__u32 handle;
    237	__u32 pad;
    238};
    239
    240/**
    241 * struct drm_vc4_mmap_bo - ioctl argument for mapping VC4 BOs.
    242 *
    243 * This doesn't actually perform an mmap.  Instead, it returns the
    244 * offset you need to use in an mmap on the DRM device node.  This
    245 * means that tools like valgrind end up knowing about the mapped
    246 * memory.
    247 *
    248 * There are currently no values for the flags argument, but it may be
    249 * used in a future extension.
    250 */
    251struct drm_vc4_mmap_bo {
    252	/** Handle for the object being mapped. */
    253	__u32 handle;
    254	__u32 flags;
    255	/** offset into the drm node to use for subsequent mmap call. */
    256	__u64 offset;
    257};
    258
    259/**
    260 * struct drm_vc4_create_shader_bo - ioctl argument for creating VC4
    261 * shader BOs.
    262 *
    263 * Since allowing a shader to be overwritten while it's also being
    264 * executed from would allow privlege escalation, shaders must be
    265 * created using this ioctl, and they can't be mmapped later.
    266 */
    267struct drm_vc4_create_shader_bo {
    268	/* Size of the data argument. */
    269	__u32 size;
    270	/* Flags, currently must be 0. */
    271	__u32 flags;
    272
    273	/* Pointer to the data. */
    274	__u64 data;
    275
    276	/** Returned GEM handle for the BO. */
    277	__u32 handle;
    278	/* Pad, must be 0. */
    279	__u32 pad;
    280};
    281
    282struct drm_vc4_get_hang_state_bo {
    283	__u32 handle;
    284	__u32 paddr;
    285	__u32 size;
    286	__u32 pad;
    287};
    288
    289/**
    290 * struct drm_vc4_hang_state - ioctl argument for collecting state
    291 * from a GPU hang for analysis.
    292*/
    293struct drm_vc4_get_hang_state {
    294	/** Pointer to array of struct drm_vc4_get_hang_state_bo. */
    295	__u64 bo;
    296	/**
    297	 * On input, the size of the bo array.  Output is the number
    298	 * of bos to be returned.
    299	 */
    300	__u32 bo_count;
    301
    302	__u32 start_bin, start_render;
    303
    304	__u32 ct0ca, ct0ea;
    305	__u32 ct1ca, ct1ea;
    306	__u32 ct0cs, ct1cs;
    307	__u32 ct0ra0, ct1ra0;
    308
    309	__u32 bpca, bpcs;
    310	__u32 bpoa, bpos;
    311
    312	__u32 vpmbase;
    313
    314	__u32 dbge;
    315	__u32 fdbgo;
    316	__u32 fdbgb;
    317	__u32 fdbgr;
    318	__u32 fdbgs;
    319	__u32 errstat;
    320
    321	/* Pad that we may save more registers into in the future. */
    322	__u32 pad[16];
    323};
    324
    325#define DRM_VC4_PARAM_V3D_IDENT0		0
    326#define DRM_VC4_PARAM_V3D_IDENT1		1
    327#define DRM_VC4_PARAM_V3D_IDENT2		2
    328#define DRM_VC4_PARAM_SUPPORTS_BRANCHES		3
    329#define DRM_VC4_PARAM_SUPPORTS_ETC1		4
    330#define DRM_VC4_PARAM_SUPPORTS_THREADED_FS	5
    331#define DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER	6
    332#define DRM_VC4_PARAM_SUPPORTS_MADVISE		7
    333#define DRM_VC4_PARAM_SUPPORTS_PERFMON		8
    334
    335struct drm_vc4_get_param {
    336	__u32 param;
    337	__u32 pad;
    338	__u64 value;
    339};
    340
    341struct drm_vc4_get_tiling {
    342	__u32 handle;
    343	__u32 flags;
    344	__u64 modifier;
    345};
    346
    347struct drm_vc4_set_tiling {
    348	__u32 handle;
    349	__u32 flags;
    350	__u64 modifier;
    351};
    352
    353/**
    354 * struct drm_vc4_label_bo - Attach a name to a BO for debug purposes.
    355 */
    356struct drm_vc4_label_bo {
    357	__u32 handle;
    358	__u32 len;
    359	__u64 name;
    360};
    361
    362/*
    363 * States prefixed with '__' are internal states and cannot be passed to the
    364 * DRM_IOCTL_VC4_GEM_MADVISE ioctl.
    365 */
    366#define VC4_MADV_WILLNEED			0
    367#define VC4_MADV_DONTNEED			1
    368#define __VC4_MADV_PURGED			2
    369#define __VC4_MADV_NOTSUPP			3
    370
    371struct drm_vc4_gem_madvise {
    372	__u32 handle;
    373	__u32 madv;
    374	__u32 retained;
    375	__u32 pad;
    376};
    377
    378enum {
    379	VC4_PERFCNT_FEP_VALID_PRIMS_NO_RENDER,
    380	VC4_PERFCNT_FEP_VALID_PRIMS_RENDER,
    381	VC4_PERFCNT_FEP_CLIPPED_QUADS,
    382	VC4_PERFCNT_FEP_VALID_QUADS,
    383	VC4_PERFCNT_TLB_QUADS_NOT_PASSING_STENCIL,
    384	VC4_PERFCNT_TLB_QUADS_NOT_PASSING_Z_AND_STENCIL,
    385	VC4_PERFCNT_TLB_QUADS_PASSING_Z_AND_STENCIL,
    386	VC4_PERFCNT_TLB_QUADS_ZERO_COVERAGE,
    387	VC4_PERFCNT_TLB_QUADS_NON_ZERO_COVERAGE,
    388	VC4_PERFCNT_TLB_QUADS_WRITTEN_TO_COLOR_BUF,
    389	VC4_PERFCNT_PLB_PRIMS_OUTSIDE_VIEWPORT,
    390	VC4_PERFCNT_PLB_PRIMS_NEED_CLIPPING,
    391	VC4_PERFCNT_PSE_PRIMS_REVERSED,
    392	VC4_PERFCNT_QPU_TOTAL_IDLE_CYCLES,
    393	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_VERTEX_COORD_SHADING,
    394	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_FRAGMENT_SHADING,
    395	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_EXEC_VALID_INST,
    396	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_TMUS,
    397	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_SCOREBOARD,
    398	VC4_PERFCNT_QPU_TOTAL_CLK_CYCLES_WAITING_VARYINGS,
    399	VC4_PERFCNT_QPU_TOTAL_INST_CACHE_HIT,
    400	VC4_PERFCNT_QPU_TOTAL_INST_CACHE_MISS,
    401	VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_HIT,
    402	VC4_PERFCNT_QPU_TOTAL_UNIFORM_CACHE_MISS,
    403	VC4_PERFCNT_TMU_TOTAL_TEXT_QUADS_PROCESSED,
    404	VC4_PERFCNT_TMU_TOTAL_TEXT_CACHE_MISS,
    405	VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VDW_STALLED,
    406	VC4_PERFCNT_VPM_TOTAL_CLK_CYCLES_VCD_STALLED,
    407	VC4_PERFCNT_L2C_TOTAL_L2_CACHE_HIT,
    408	VC4_PERFCNT_L2C_TOTAL_L2_CACHE_MISS,
    409	VC4_PERFCNT_NUM_EVENTS,
    410};
    411
    412#define DRM_VC4_MAX_PERF_COUNTERS	16
    413
    414struct drm_vc4_perfmon_create {
    415	__u32 id;
    416	__u32 ncounters;
    417	__u8 events[DRM_VC4_MAX_PERF_COUNTERS];
    418};
    419
    420struct drm_vc4_perfmon_destroy {
    421	__u32 id;
    422};
    423
    424/*
    425 * Returns the values of the performance counters tracked by this
    426 * perfmon (as an array of ncounters u64 values).
    427 *
    428 * No implicit synchronization is performed, so the user has to
    429 * guarantee that any jobs using this perfmon have already been
    430 * completed  (probably by blocking on the seqno returned by the
    431 * last exec that used the perfmon).
    432 */
    433struct drm_vc4_perfmon_get_values {
    434	__u32 id;
    435	__u64 values_ptr;
    436};
    437
    438#if defined(__cplusplus)
    439}
    440#endif
    441
    442#endif /* _UAPI_VC4_DRM_H_ */