add usermode helper for the ipvr kernel driver. test_ioctl: test kernel driver by directly ioctl

v2: take Emil's comments - correctly align ipvr_drm.h

v3: take Daniel Vetter and Daniel Stone's comments, and implement PRIME - correctly align ipvr_drm.h - use __u32 family in ipvr_drm.h - rip out explicit fence from libdrm_ipvr - implemented PRIME support - add relocation fixup implementation

Signed-off-by: Yao Cheng yao.cheng@intel.com --- Makefile.am | 6 +- Makefile.sources | 1 + configure.ac | 26 +- include/drm/ipvr_drm.h | 278 +++++++++++ ipvr/Makefile.am | 57 +++ ipvr/Makefile.sources | 5 + ipvr/ipvr_bufmgr.h | 149 ++++++ ipvr/ipvr_bufmgr_gem.c | 1200 ++++++++++++++++++++++++++++++++++++++++++++++++ ipvr/libdrm_ipvr.pc.in | 11 + ipvr/test_ioctl.c | 423 +++++++++++++++++ 10 files changed, 2154 insertions(+), 2 deletions(-) create mode 100644 include/drm/ipvr_drm.h create mode 100644 ipvr/Makefile.am create mode 100644 ipvr/Makefile.sources create mode 100644 ipvr/ipvr_bufmgr.h create mode 100644 ipvr/ipvr_bufmgr_gem.c create mode 100644 ipvr/libdrm_ipvr.pc.in create mode 100644 ipvr/test_ioctl.c

diff --git a/Makefile.am b/Makefile.am index 3952a88..2227add 100644 --- a/Makefile.am +++ b/Makefile.am @@ -33,6 +33,10 @@ if HAVE_INTEL INTEL_SUBDIR = intel endif

+if HAVE_IPVR +IPVR_SUBDIR = ipvr +endif + if HAVE_NOUVEAU NOUVEAU_SUBDIR = nouveau endif @@ -53,7 +57,7 @@ if HAVE_FREEDRENO FREEDRENO_SUBDIR = freedreno endif

-SUBDIRS = . $(LIBKMS_SUBDIR) $(INTEL_SUBDIR) $(NOUVEAU_SUBDIR) $(RADEON_SUBDIR) $(OMAP_SUBDIR) $(EXYNOS_SUBDIR) $(FREEDRENO_SUBDIR) tests man +SUBDIRS = . $(LIBKMS_SUBDIR) $(INTEL_SUBDIR) $(IPVR_SUBDIR) $(NOUVEAU_SUBDIR) $(RADEON_SUBDIR) $(OMAP_SUBDIR) $(EXYNOS_SUBDIR) $(FREEDRENO_SUBDIR) tests man

libdrm_la_LTLIBRARIES = libdrm.la libdrm_ladir = $(libdir) diff --git a/Makefile.sources b/Makefile.sources index d86fb2a..96f8c60 100644 --- a/Makefile.sources +++ b/Makefile.sources @@ -18,6 +18,7 @@ LIBDRM_INCLUDE_H_FILES := \ include/drm/drm_mode.h \ include/drm/drm_sarea.h \ include/drm/i915_drm.h \ + include/drm/ipvr_drm.h \ include/drm/mach64_drm.h \ include/drm/mga_drm.h \ include/drm/nouveau_drm.h \ diff --git a/configure.ac b/configure.ac index ee59b03..6dcf1b2 100644 --- a/configure.ac +++ b/configure.ac @@ -68,6 +68,11 @@ AC_ARG_ENABLE(intel, [Enable support for intel's KMS API (default: auto)]), [INTEL=$enableval], [INTEL=auto])

+AC_ARG_ENABLE(ipvr, + AS_HELP_STRING([--disable-ipvr], + [Enable support for baytrail's hardware VP8 decode (default: auto)]), + [IPVR=$enableval], [IPVR=auto]) + AC_ARG_ENABLE(radeon, AS_HELP_STRING([--disable-radeon], [Enable support for radeon's KMS API (default: auto)]), @@ -204,7 +209,7 @@ if test "x$drm_cv_atomic_primitives" = "xlibatomic-ops"; then AC_DEFINE(HAVE_LIB_ATOMIC_OPS, 1, [Enable if you have libatomic-ops-dev installed]) fi

-if test "x$INTEL" != "xno" -o "x$RADEON" != "xno" -o "x$NOUVEAU" != "xno"; then +if test "x$INTEL" != "xno" -o "x$IPVR" != "xno" -o "x$RADEON" != "xno" -o "x$NOUVEAU" != "xno"; then if test "x$drm_cv_atomic_primitives" = "xnone"; then if test "x$INTEL" != "xauto"; then if test "x$INTEL" != "xno"; then @@ -214,6 +219,14 @@ if test "x$INTEL" != "xno" -o "x$RADEON" != "xno" -o "x$NOUVEAU" != "xno"; then AC_MSG_WARN([Disabling libdrm_intel. It depends on atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package.]) INTEL=no fi + if test "x$IPVR" != "xauto"; then + if test "x$IPVR" != "xno"; then + AC_MSG_ERROR([libdrm_ipvr depends upon atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package, or, failing both of those, disable support for Baytrail VP8 by passing --disable-ipvr to ./configure]) + fi + else + AC_MSG_WARN([Disabling libdrm_ipvr. It depends on atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package.]) + INTEL=no + fi if test "x$RADEON" != "xauto"; then if test "x$RADEON" != "xno"; then AC_MSG_ERROR([libdrm_radeon depends upon atomic operations, which were not found for your compiler/cpu. Try compiling with -march=native, or install the libatomics-op-dev package, or, failing both of those, disable support for Radeon GPUs by passing --disable-radeon to ./configure]) @@ -237,6 +250,9 @@ if test "x$INTEL" != "xno" -o "x$RADEON" != "xno" -o "x$NOUVEAU" != "xno"; then *) INTEL=no ;; esac fi + if test "x$IPVR" != "xno"; then + IPVR=yes + fi if test "x$RADEON" != "xno"; then RADEON=yes fi @@ -274,6 +290,11 @@ if test "x$INTEL" = xyes; then AC_DEFINE(HAVE_INTEL, 1, [Have intel support]) fi

+AM_CONDITIONAL(HAVE_IPVR, [test "x$IPVR" = xyes]) +if test "x$IPVR" = xyes; then + AC_DEFINE(HAVE_PVR, 1, [Have ipvr support]) +fi + AM_CONDITIONAL(HAVE_VMWGFX, [test "x$VMWGFX" = xyes]) if test "x$VMWGFX" = xyes; then AC_DEFINE(HAVE_VMWGFX, 1, [Have vmwgfx kernel headers]) @@ -393,6 +414,8 @@ AC_CONFIG_FILES([ libkms/libkms.pc intel/Makefile intel/libdrm_intel.pc + ipvr/Makefile + ipvr/libdrm_ipvr.pc radeon/Makefile radeon/libdrm_radeon.pc nouveau/Makefile @@ -419,6 +442,7 @@ echo "$PACKAGE_STRING will be compiled with:" echo "" echo " libkms $LIBKMS" echo " Intel API $INTEL" +echo " Ipvr API $IPVR" echo " vmwgfx API $VMWGFX" echo " Radeon API $RADEON" echo " Nouveau API $NOUVEAU" diff --git a/include/drm/ipvr_drm.h b/include/drm/ipvr_drm.h new file mode 100644 index 0000000..d1916fa --- /dev/null +++ b/include/drm/ipvr_drm.h @@ -0,0 +1,278 @@ +/************************************************************************** + * ipvr_drm.h: IPVR header file exported to user space + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Fei Jiang fei.jiang@intel.com + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ + + +/* this file only define structs and macro which need export to user space */ +#ifndef _IPVR_DRM_H_ +#define _IPVR_DRM_H_ + +#include <drm/drm.h> +struct drm_ipvr_context_create { + /* passed ctx_info, including codec, profile info */ +#define IPVR_CONTEXT_TYPE_VED (0x1) + __u32 ctx_type; + /* returned back ctx_id */ + __u32 ctx_id; + /* + * following tiling strides for VED are supported: + * stride 0: 512 for scheme 0, 1024 for scheme 1 + * stride 1: 1024 for scheme 0, 2048 for scheme 1 + * stride 2: 2048 for scheme 0, 4096 for scheme 1 + * stride 3: 4096 for scheme 0 + */ + __u32 tiling_stride; + /* + * scheme 0: tile is 256x16, while minimal tile stride is 512 + * scheme 1: tile is 512x8, while minimal tile stride is 1024 + */ + __u32 tiling_scheme; +}; + +struct drm_ipvr_context_destroy { + __u32 ctx_id; + __u32 pad64; +}; + +enum drm_ipvr_misc_key { + IPVR_DEVICE_INFO, + IPVR_UPDATE_TILING, + IPVR_SET_DISPLAYING_FRAME, + IPVR_GET_DISPLAYING_FRAME, + IPVR_QUERY_ENTRY +}; + +/* + * different context maybe has different tiling stride, + * then tiling info need be bound with ctx + */ +struct drm_ipvr_update_tiling { + __u32 ctx_id; + __u32 tiling_stride; + __u32 tiling_scheme; + __u32 pad64; +}; + +/* Ioctl to set/get misc params: + */ +struct drm_ipvr_misc { + __u64 key; + __u64 arg; /* argument pointer */ + __u64 value; /* feed back pointer */ +}; + +struct drm_ipvr_gem_relocation_entry { + /** + * Handle of the buffer being pointed to by this relocation entry. + * + * It's appealing to make this be an index into the mm_validate_entry + * list to refer to the buffer, but this allows the driver to create + * a relocation list for state buffers and not re-write it per + * exec using the buffer. + */ + __u32 target_handle; + + /** + * Value to be added to the offset of the target buffer to make up + * the relocation entry. + */ + __u32 delta; + + /** Offset in the buffer the relocation entry will be written into */ + __u64 offset; + + /** + * Offset value of the target buffer that the relocation entry was last + * written as. + * + * If the buffer has the same offset as last time, we can skip syncing + * and writing the relocation. This value is written back out by + * the execbuffer ioctl when the relocation is written. + */ + __u64 presumed_offset; + + /** + * Target memory domains read by this operation. + */ + __u32 read_domains; + + /** + * Target memory domains written by this operation. + * + * Note that only one domain may be written by the whole + * execbuffer operation, so that where there are conflicts, + * the application will get -EINVAL back. + */ + __u32 write_domain; +}; + +struct drm_ipvr_gem_exec_object { + /** + * User's handle for a buffer to be bound into the MMU for this + * operation. + */ + __u32 handle; + + /** Number of relocations to be performed on this buffer */ + __u32 relocation_count; + /** + * Pointer to array of struct drm_i915_gem_relocation_entry containing + * the relocations to be performed in this buffer. + */ + __u64 relocs_ptr; + + /** Required alignment in graphics aperture */ + __u64 alignment; + + /** + * Returned value of the updated offset of the object, for future + * presumed_offset writes. + */ + __u64 offset; + +#define IPVR_EXEC_OBJECT_NEED_FENCE (1 << 0) +#define IPVR_EXEC_OBJECT_SUBMIT (1 << 1) + __u64 flags; + + __u64 rsvd1; + __u64 rsvd2; +}; + +struct drm_ipvr_gem_execbuffer { + /** + * List of gem_exec_object2 structs + */ + __u64 buffers_ptr; + __u32 buffer_count; + + /** Offset in the batchbuffer to start execution from. */ + __u32 exec_start_offset; + /** Bytes used in batchbuffer from batch_start_offset */ + __u32 exec_len; + + /** + * ID of hardware context. + */ + __u32 ctx_id; + + __u64 flags; + __u64 rsvd1; + __u64 rsvd2; +}; + +enum ipvr_cache_level +{ + IPVR_CACHE_NOACCESS, + IPVR_CACHE_UNCACHED, + IPVR_CACHE_WRITEBACK, + IPVR_CACHE_WRITECOMBINE, + IPVR_CACHE_MAX, +}; + +struct drm_ipvr_gem_create { + /* + * Requested size for the object. + * The (page-aligned) allocated size for the object will be returned. + */ + __u64 size; + __u64 rounded_size; + __u64 mmu_offset; + /* + * Returned handle for the object. + * Object handles are nonzero. + */ + __u32 handle; + __u32 tiling; + + __u32 cache_level; + __u32 pad64; + /* + * Handle used for user to mmap BO + */ + __u64 map_offset; +}; + +struct drm_ipvr_gem_busy { + /* Handle of the buffer to check for busy */ + __u32 handle; + + /* + * Return busy status (1 if busy, 0 if idle). + * The high word is used to indicate on which rings the object + * currently resides: + * 16:31 - busy (r or r/w) rings (16 render, 17 bsd, 18 blt, etc) + */ + __u32 busy; +}; + +struct drm_ipvr_gem_mmap_offset { + /** Handle for the object being mapped. */ + __u32 handle; + __u32 pad64; + /** + * Fake offset to use for subsequent mmap call + * + * This is a fixed-size type for 32/64 compatibility. + */ + __u64 offset; +}; + +struct drm_ipvr_gem_wait { + /* Handle of BO we shall wait on */ + __u32 handle; + __u32 flags; + /** Number of nanoseconds to wait, Returns time remaining. */ + __s64 timeout_ns; +}; + +/* + * IPVR GEM specific ioctls + */ +#define DRM_IPVR_CONTEXT_CREATE 0x00 +#define DRM_IPVR_CONTEXT_DESTROY 0x01 +#define DRM_IPVR_MISC 0x02 +#define DRM_IPVR_GEM_EXECBUFFER 0x03 +#define DRM_IPVR_GEM_BUSY 0x04 +#define DRM_IPVR_GEM_CREATE 0x05 +#define DRM_IPVR_GEM_WAIT 0x06 +#define DRM_IPVR_GEM_MMAP_OFFSET 0x07 + +#define DRM_IOCTL_IPVR_CONTEXT_CREATE \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_CONTEXT_CREATE, struct drm_ipvr_context_create) +#define DRM_IOCTL_IPVR_CONTEXT_DESTROY \ + DRM_IOW(DRM_COMMAND_BASE + DRM_IPVR_CONTEXT_DESTROY, struct drm_ipvr_context_destroy) +#define DRM_IOCTL_IPVR_MISC \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_MISC, struct drm_ipvr_misc) +#define DRM_IOCTL_IPVR_GEM_EXECBUFFER \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_EXECBUFFER, struct drm_ipvr_gem_execbuffer) +#define DRM_IOCTL_IPVR_GEM_BUSY \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_BUSY, struct drm_ipvr_gem_busy) +#define DRM_IOCTL_IPVR_GEM_CREATE \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_CREATE, struct drm_ipvr_gem_create) +#define DRM_IOCTL_IPVR_GEM_WAIT \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_WAIT, struct drm_ipvr_gem_wait) +#define DRM_IOCTL_IPVR_GEM_MMAP_OFFSET \ + DRM_IOWR(DRM_COMMAND_BASE + DRM_IPVR_GEM_MMAP_OFFSET, struct drm_ipvr_gem_mmap_offset) + +#endif diff --git a/ipvr/Makefile.am b/ipvr/Makefile.am new file mode 100644 index 0000000..60cbb36 --- /dev/null +++ b/ipvr/Makefile.am @@ -0,0 +1,57 @@ +# Copyright © 2014 Intel Corporation +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. +# +# Authors: +# Yao Cheng yao.cheng@intel.com + +include Makefile.sources + +AM_CFLAGS = \ + $(WARN_CFLAGS) \ + $(VISIBILITY_CFLAGS) \ + -I$(top_srcdir) \ + -I$(top_srcdir)/ipvr \ + $(PTHREADSTUBS_CFLAGS) \ + $(PCIACCESS_CFLAGS) \ + $(VALGRIND_CFLAGS) \ + -I$(top_srcdir)/include/drm + +libdrm_ipvr_la_LTLIBRARIES = libdrm_ipvr.la +libdrm_ipvr_ladir = $(libdir) +libdrm_ipvr_la_LDFLAGS = -version-number 0:1:0 -no-undefined +libdrm_ipvr_la_LIBADD = ../libdrm.la \ + @PTHREADSTUBS_LIBS@ \ + @PCIACCESS_LIBS@ \ + @CLOCK_LIB@ + +libdrm_ipvr_la_SOURCES = $(LIBDRM_IPVR_FILES) + +ipvr_bufmgr_gem_o_CFLAGS = $(AM_CFLAGS) -c99 + +libdrm_ipvrincludedir = ${includedir}/libdrm +libdrm_ipvrinclude_HEADERS = $(LIBDRM_IPVR_H_FILES) + +# This may be interesting even outside of "make check", due to the -dump option. +noinst_PROGRAMS = test_ioctl + +test_ioctl_LDADD = libdrm_ipvr.la ../libdrm.la ../intel/libdrm_intel.la + +pkgconfig_DATA = libdrm_ipvr.pc diff --git a/ipvr/Makefile.sources b/ipvr/Makefile.sources new file mode 100644 index 0000000..5103a02 --- /dev/null +++ b/ipvr/Makefile.sources @@ -0,0 +1,5 @@ +LIBDRM_IPVR_FILES := \ + ipvr_bufmgr_gem.c + +LIBDRM_IPVR_H_FILES := \ + ipvr_bufmgr.h diff --git a/ipvr/ipvr_bufmgr.h b/ipvr/ipvr_bufmgr.h new file mode 100644 index 0000000..31df82f --- /dev/null +++ b/ipvr/ipvr_bufmgr.h @@ -0,0 +1,149 @@ +/* + * Copyright 2014 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Yao Cheng yao.cheng@intel.com + * + */ +#ifndef IPVR_BUFMGR_H +#define IPVR_BUFMGR_H +#include <stdio.h> +#include <stdint.h> +#include <stdio.h> + +typedef struct _drm_ipvr_bo drm_ipvr_bo; +typedef struct _drm_ipvr_bufmgr drm_ipvr_bufmgr; +typedef struct _drm_ipvr_context drm_ipvr_context; + +struct _drm_ipvr_bo { + /** + * Size in bytes of the buffer object. + * + * The size may be larger than the size originally requested for the + * allocation, such as being aligned to page size. + */ + size_t size; + + /** rounded up size of actual allocation */ + size_t alloc_size; + + /** GEM handle of the BO */ + uint32_t handle; + + /** + * several buffers may share one BO (camera/RAR), and use offset to distinguish it + */ + uint64_t buffer_ofs; + + /** + * Last seen card virtual address (offset from the beginning of the + * aperture) for the object. This should be used to fill relocation + * entries when calling drm_ipvr_gem_bo_emit_reloc() + */ + uint64_t offset; + + /** + * Virtual address for accessing the buffer data. Only valid while + * mapped. + */ + void *virt; + + /** + * Generic external buffer info e.g. gfx handle on Android + */ + uint64_t ext_handle; + + /** Indicates the BO can be re-used or not after being recycled */ + int reusable; + + /** Buffer manager context associated with this buffer object */ + drm_ipvr_bufmgr *bufmgr; +}; + +struct _drm_ipvr_bufmgr { + int debug; +}; + +struct _drm_ipvr_context +{ + uint32_t ctx_id; + uint32_t ctx_type; + drm_ipvr_bufmgr *bufmgr; +}; + +drm_ipvr_bufmgr * +drm_ipvr_gem_bufmgr_init(int fd, int batch_size); + +void drm_ipvr_gem_bufmgr_destroy(drm_ipvr_bufmgr *bufmgr); + +int drm_ipvr_gem_bufmgr_get_device_info(drm_ipvr_bufmgr *bufmgr, uint16_t *dev_id, uint16_t *caps); + +/*void drm_ipvr_gem_bufmgr_trace_all_bos(drm_ipvr_bufmgr *bufmgr);*/ + + +drm_ipvr_bo * +drm_ipvr_gem_bo_alloc(drm_ipvr_bufmgr *bufmgr, drm_ipvr_context *ctx, + const char *name, size_t size, + uint8_t tiling, uint8_t cache_level, uint8_t reusable); + +void drm_ipvr_gem_bo_reference(drm_ipvr_bo *bo); + +void drm_ipvr_gem_bo_unreference(drm_ipvr_bo *bo); + +int drm_ipvr_gem_bo_map(drm_ipvr_bo *bo, unsigned long offset, size_t size, int write_enable); + +int drm_ipvr_gem_bo_unmap(drm_ipvr_bo *bo); + +void drm_ipvr_gem_bo_wait(drm_ipvr_bo *bo); + +int drm_ipvr_gem_bo_emit_reloc(drm_ipvr_bo *bo, unsigned long offset, + drm_ipvr_bo *target_bo, unsigned long target_offset, uint32_t skip_fence); + +int drm_ipvr_gem_bo_exec(drm_ipvr_bo *bo, unsigned long offset, size_t len, + int fence_in, int *fence_out); + +drm_ipvr_bo * +drm_ipvr_gem_bo_create_from_name(drm_ipvr_bufmgr *bufmgr, drm_ipvr_context *ctx, + const char *name, uint32_t handle); + +int drm_ipvr_gem_bo_flink(drm_ipvr_bo *bo, uint32_t * name); + +int drm_ipvr_gem_bo_export_to_prime(drm_ipvr_bo *bo, int *prime_fd); + +drm_ipvr_bo *drm_ipvr_gem_bo_create_from_prime(drm_ipvr_bufmgr *bufmgr, + int prime_fd, int size); + +int drm_ipvr_gem_bo_busy(drm_ipvr_bo *bo); + +int drm_ipvr_gem_bo_references(drm_ipvr_bo *bo, drm_ipvr_bo *target_bo); + +void drm_ipvr_gem_bo_remove_relocs(drm_ipvr_bo *bo, uint8_t recursive); + +/*int drm_ipvr_gem_bo_get_MB_error(drm_ipvr_bo *bo);*/ + +drm_ipvr_context * +drm_ipvr_gem_context_create(drm_ipvr_bufmgr *bufmgr, uint32_t ctx_type, uint32_t tiling_stride, uint32_t tiling_mode); + +void +drm_ipvr_gem_context_destroy(drm_ipvr_context *ctx); + +#endif diff --git a/ipvr/ipvr_bufmgr_gem.c b/ipvr/ipvr_bufmgr_gem.c new file mode 100644 index 0000000..1f6e7c5 --- /dev/null +++ b/ipvr/ipvr_bufmgr_gem.c @@ -0,0 +1,1200 @@ +/************************************************************************** + * + * Copyright 2014 Intel Corporation + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * + **************************************************************************/ +/* + * Authors: Yao Cheng yao.cheng@intel.com + * + */ + +#include <assert.h> +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#include <xf86drm.h> +#include <xf86atomic.h> +#include <fcntl.h> +#include <inttypes.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <assert.h> +#include <pthread.h> +#include <sys/ioctl.h> +#include <sys/mman.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <stdbool.h> +#include <sys/time.h> +#include "errno.h" +#include "libdrm.h" +#include "libdrm_lists.h" +#include "ipvr_bufmgr.h" +#include "string.h" +#include "ipvr_drm.h" +#ifdef ANDROID +#include <utils/Timers.h> +#endif + +static int debug_level = 3; + +typedef uint64_t nsecs_t; +#define VERB(fmt, ...) _DBGPRINT(0, fmt, ##__VA_ARGS__) +#define DBG(fmt, ...) _DBGPRINT(1, fmt, ##__VA_ARGS__) +#define INFO(fmt, ...) _DBGPRINT(2, fmt, ##__VA_ARGS__) +#define WARN(fmt, ...) _DBGPRINT(3, fmt, ##__VA_ARGS__) +#define ERR(fmt, ...) _DBGPRINT(4, fmt, ##__VA_ARGS__) +#define _DBGPRINT(level, fmt, ...) \ + do { \ + if (debug_level <= level) { \ + fprintf(stderr, fmt, ##__VA_ARGS__); \ + } \ + } while (false); + +#define MAX_SELECT_TIMEOUT (10 * 1000) // 10 msec for 1MB BO +#define COUNT_BEFORE_SELECT (5) + +#define IPVR_TIMEOUT_USEC 990000LL + +typedef struct _drm_ipvr_bo_gem drm_ipvr_bo_gem; + +typedef struct _drm_ipvr_reloc_target_info { + drm_ipvr_bo *bo; + uint64_t flags; +} drm_ipvr_reloc_target; + +struct drm_ipvr_cache_bucket { + unsigned long size; + drmMMListHead head; + size_t count; + size_t limit; +}; + +typedef struct _drm_ipvr_bufmgr_gem +{ + drm_ipvr_bufmgr base; + int fd; + pthread_mutex_t lock; + pthread_mutex_t list_lock; + + struct drm_ipvr_gem_exec_object *exec_objs; + drm_ipvr_bo **exec_bos; + int exec_size; + int exec_count; + time_t time; + int max_relocs; + + /* cache bucket of power-of-two */ + int cache_bucket_size; + + /* only cache linear BOs */ + struct drm_ipvr_cache_bucket *cache_buckets; + + /* seqno used to check BO's last operation oldness */ + int exec_seq; +} drm_ipvr_bufmgr_gem; + +typedef struct _drm_ipvr_bo_gem +{ + drm_ipvr_bo base; + const char *name; + /** + * Kenel-assigned global name for this object + */ + unsigned int global_name; + atomic_t refcount; + uint32_t tiling; + uint32_t caching; + unsigned long stride; + atomic_t mapcount; + struct drm_ipvr_gem_relocation_entry *relocs; + /* + * Handle for mmap + */ + uint64_t map_offset; + /** + * Array of info structs corresponding to relocs[i].target_handle etc + */ + drm_ipvr_reloc_target *reloc_target_info; + /** Number of entries in relocs */ + int reloc_count; + /** + * Index of the buffer within the validation list while preparing a + * batchbuffer execution. + */ + int validate_index; + /** + * Boolean of whether this buffer has been used as a relocation + * target and had its size accounted for, and thus can't have any + * further relocations added to it. + */ + bool used_as_reloc_target; + /*bool is_vmap;*/ + bool is_prime_imported; + /*void *user_virt;*/ + + /** BO cache list */ + drmMMListHead cache_head; + + drm_ipvr_context *ctx; + int exec_seq; +} drm_ipvr_bo_gem; + +static nsecs_t get_time_ns(void) +{ +#ifdef ANDROID + return systemTime(SYSTEM_TIME_MONOTONIC); +#else + struct timeval tv; + gettimeofday(&tv, NULL); + return tv.tv_sec * 1000000000LL + tv.tv_usec * 1000LL; +#endif +} + +static struct drm_ipvr_cache_bucket * +drm__ipvr_cache_bucket_for_size(drm_ipvr_bufmgr *bufmgr, + size_t size, uint8_t tiling) +{ + int i; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bufmgr; + + assert(tiling == 0); + for (i = 0; i < bufmgr_gem->cache_bucket_size; i++) { + struct drm_ipvr_cache_bucket *bucket = + &bufmgr_gem->cache_buckets[i]; + if (bucket->size >= size) { + return bucket; + } + } + return NULL; +} + +static int +drm__ipvr_gem_bo_free(drm_ipvr_bo *bo) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*) bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + struct drm_gem_close arg; + int ret; + bool fence_destroyed = false; + + /* Close this object */ + arg.handle = bo->handle; + + ret = drmIoctl(bufmgr_gem->fd, DRM_IOCTL_GEM_CLOSE, &arg); + if (ret != 0) { + ERR("%s: DRM_IOCTL_GEM_CLOSE "%s" handle 0x%x offset 0x%lx failed: %s\n", + __FUNCTION__, bo_gem->name, + bo->handle, bo->offset, strerror(errno)); + } + + if (bo_gem->relocs) + free(bo_gem->relocs); + if (bo_gem->reloc_target_info) + free(bo_gem->reloc_target_info); + + DBG("%s: freed buf khandle "%s" hnd %x, offset 0x%lx, %s\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, + (fence_destroyed? "fence is destroyed": "no fence to destroy")); + + memset(bo_gem, 0, sizeof(*bo_gem)); + free(bo_gem); + return 0; +} + +drm_public int +drm_ipvr_gem_bo_map(drm_ipvr_bo *bo, unsigned long offset, size_t size, int write_enable); +drm_public int +drm_ipvr_gem_bo_unmap(drm_ipvr_bo *bo); + +drm_public drm_ipvr_bo * +drm_ipvr_gem_bo_alloc(drm_ipvr_bufmgr *bufmgr, drm_ipvr_context *ctx, + const char *name, size_t size, + uint8_t tiling, uint8_t cache_level, uint8_t reusable) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*) bufmgr; + drm_ipvr_bo_gem *allocated_bo_gem = NULL; + int ret; + struct drm_ipvr_gem_create arg; + bool alloc_from_cache = false; + size_t alloc_size = size; + struct drm_ipvr_cache_bucket *bucket = NULL; + int bo_movings = 0; + int bucket_free_before, bucket_free_after; + nsecs_t begin_ns, end_ns; + nsecs_t time_wait, time_bo_moving, time_fence_destroy, time_kernel_alloc; + nsecs_t now; + time_wait = time_bo_moving = time_fence_destroy = time_kernel_alloc = 0; + bucket_free_before = bucket_free_after = 0; + + begin_ns = get_time_ns(); + if (reusable) + bucket = drm__ipvr_cache_bucket_for_size(&bufmgr_gem->base, size, tiling); + + if (reusable && !bucket) { + reusable = false; + WARN("%s size %zu cannot fit into any cache bucket, set as non-reusable\n", + __FUNCTION__, size); + } + if (reusable && bucket) { + /** + * only expand size when it's reusable + */ + alloc_size = bucket->size; + } + + if (reusable && bucket && !DRMLISTEMPTY(&bucket->head)) { + VERB("%s searching cached BOs in fast mode: free_list\n", __FUNCTION__); + now = get_time_ns(); + pthread_mutex_lock(&bufmgr_gem->list_lock); + allocated_bo_gem = DRMLISTENTRY(drm_ipvr_bo_gem, + bucket->head.next, cache_head); + if (bufmgr_gem->exec_seq - allocated_bo_gem->exec_seq < 5) { + alloc_from_cache = false; + allocated_bo_gem = NULL; + } + else if (!drm_ipvr_gem_bo_busy(&allocated_bo_gem->base)) { + alloc_from_cache = true; + DRMLISTDELINIT(&allocated_bo_gem->cache_head); + bucket->count--; + } + else { + drm_ipvr_gem_bo_wait(&allocated_bo_gem->base); + alloc_from_cache = true; + DRMLISTDELINIT(&allocated_bo_gem->cache_head); + bucket->count--; + } + pthread_mutex_unlock(&bufmgr_gem->list_lock); + time_wait += get_time_ns() - now; + VERB("%s got free BO from cache\n", __FUNCTION__); + } + else { + VERB("%s empty free_list, alloc_from kernel\n", __FUNCTION__); + alloc_from_cache = false; + } + + if (!alloc_from_cache) { + allocated_bo_gem = calloc(1, sizeof(*allocated_bo_gem)); + if (!allocated_bo_gem) { + ERR("%s: calloc failed: %s\n", __FUNCTION__, strerror(errno)); + return NULL; + } + + if (reusable) + arg.size = alloc_size; + else + arg.size = size; + arg.tiling = tiling; + arg.cache_level = cache_level; + + now = get_time_ns(); + ret = drmCommandWriteRead(bufmgr_gem->fd, + DRM_IPVR_GEM_CREATE, + &arg, sizeof(arg)); + + if (ret != 0) { + ERR("%s: IOCTL GEM_CREATE failed: %d\n", + __FUNCTION__, ret); + free(allocated_bo_gem); + allocated_bo_gem = NULL; + return NULL; + } + time_kernel_alloc = get_time_ns() - now; + + allocated_bo_gem->base.bufmgr = bufmgr; + allocated_bo_gem->base.handle = arg.handle; + allocated_bo_gem->tiling = arg.tiling; + allocated_bo_gem->caching = arg.cache_level; + allocated_bo_gem->stride = 0; /* FIXME: does ipvr_drv_video maintain stride/sliceHeight itself? */ + allocated_bo_gem->base.offset = arg.mmu_offset; + allocated_bo_gem->base.alloc_size = arg.rounded_size; + allocated_bo_gem->base.size = size; + allocated_bo_gem->name = name; + allocated_bo_gem->base.virt = NULL; + allocated_bo_gem->map_offset = arg.map_offset; + allocated_bo_gem->is_prime_imported = false; + atomic_set(&allocated_bo_gem->mapcount, 0); + + DBG("%s: allocated buf handle 0x%x from kernel, " + "offset 0x%lx, name "%s" size %lu (%lu), tiling %u, cache_level %u, bufmgr %p\n", + __FUNCTION__, + allocated_bo_gem->base.handle, allocated_bo_gem->base.offset, allocated_bo_gem->name, + allocated_bo_gem->base.size, allocated_bo_gem->base.alloc_size, allocated_bo_gem->tiling, allocated_bo_gem->caching, bufmgr); + } + else { + assert(allocated_bo_gem); + DBG("%s: allocated buf handle 0x%x from userspace cache, " + "offset 0x%lx, name "%s" size %lu (%lu), tiling %u, cache_level %u, bufmgr %p\n", + __FUNCTION__, + allocated_bo_gem->base.handle, allocated_bo_gem->base.offset, allocated_bo_gem->name, + allocated_bo_gem->base.size, allocated_bo_gem->base.alloc_size, allocated_bo_gem->tiling, allocated_bo_gem->caching, bufmgr); + } + + allocated_bo_gem->name = name; + allocated_bo_gem->base.size = size; + allocated_bo_gem->validate_index = -1; + allocated_bo_gem->ctx = ctx; + allocated_bo_gem->base.reusable = reusable; + allocated_bo_gem->exec_seq = -1; + + atomic_set(&allocated_bo_gem->refcount, 1); + end_ns = get_time_ns(); + if (bucket) { + bucket_free_after = bucket->count; + } + assert(atomic_read(&allocated_bo_gem->mapcount) == 0); + + if (end_ns - begin_ns > 2 * 1000 * 1000) { + DBG("%s got from %s: "%s" hnd %u offset 0x%lx size %lu took %.2f ms, " + "(wait %.2f, " + "bo_moving %.2f, kernel_alloc %.2f), " + "bucket (%d/%zu)=>(%d/%zu)\n", + __FUNCTION__, (alloc_from_cache? "CACHE": "KERNEL"), + allocated_bo_gem->name, allocated_bo_gem->base.handle, allocated_bo_gem->base.offset, + allocated_bo_gem->base.alloc_size, (end_ns - begin_ns)/1000000.0, + time_wait/1000000.0, + bo_movings/1000000.0, time_kernel_alloc/1000000.0, + bucket_free_before, (bucket? bucket->limit: 0), + bucket_free_after, (bucket? bucket->limit: 0)); + } + return &allocated_bo_gem->base; +} + +drm_public void drm_ipvr_gem_bo_unreference(drm_ipvr_bo *bo); + +static void +drm__ipvr_gem_bo_remove_relocs(drm_ipvr_bo *bo, int depth, bool recursive) +{ + int i; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem*)bo; + drm_ipvr_bo_gem *target_bo_gem; + for (i = 0; i < bo_gem->reloc_count; i++) { + if (bo_gem->reloc_target_info[i].bo != bo) { + target_bo_gem = (drm_ipvr_bo_gem*)bo_gem->reloc_target_info[i].bo; + VERB("%s (depth: %d) remove "%s" hnd %x off 0x%lx => "%s" hnd %x off 0x%lx\n", + __FUNCTION__, depth, bo_gem->name, bo_gem->base.handle, bo->offset, + target_bo_gem->name, target_bo_gem->base.handle, target_bo_gem->base.offset); + if (recursive) + drm__ipvr_gem_bo_remove_relocs(bo_gem->reloc_target_info[i].bo, depth + 1, true); + drm_ipvr_gem_bo_unreference(bo_gem->reloc_target_info[i].bo); + } + bo_gem->reloc_target_info[i].bo = NULL; + memset(&bo_gem->relocs[i], 0, sizeof(bo_gem->relocs[i])); + } + bo_gem->reloc_count = 0; +} + + +drm_public void +drm_ipvr_gem_bo_remove_relocs(drm_ipvr_bo *bo, uint8_t recursive) +{ + drm__ipvr_gem_bo_remove_relocs(bo, 0, recursive); +} + +drm_public void +drm_ipvr_gem_bo_reference(drm_ipvr_bo *bo) +{ + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *)bo; + atomic_inc(&bo_gem->refcount); + VERB("%s "%s" hnd 0x%x offset 0x%lx, refcount became %d\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, atomic_read(&bo_gem->refcount)); +} + +static void +drm__ipvr_gem_bo_finalize(drm_ipvr_bo *bo) +{ + struct drm_ipvr_cache_bucket *bucket; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem*)bo; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bo->bufmgr; + int mapcount; + /* add lock? */ + /* Automatically unreference all the target buffers */ + drm_ipvr_gem_bo_remove_relocs(bo, false); + mapcount = atomic_read(&bo_gem->mapcount); + if (mapcount != 0) + WARN("%s BO is finalized with mapcount %d\n", __FUNCTION__, mapcount); + + assert(atomic_read(&bo_gem->mapcount) == 0); + + bo_gem->validate_index = -1; + + bucket = drm__ipvr_cache_bucket_for_size(bo->bufmgr, bo->size, bo_gem->tiling); + + /* Put the buffer into our internal cache for reuse if we can. */ + if (bo->reusable && bucket != NULL) { + bo_gem->validate_index = -1; + DBG("%s adding bo "%s" hnd %x (offset 0x%lx) to internal cache\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset); + + pthread_mutex_lock(&bufmgr_gem->list_lock); + DRMLISTADDTAIL(&bo_gem->cache_head, &bucket->head); + bucket->count++; + + pthread_mutex_unlock(&bufmgr_gem->list_lock); + } else { + VERB("%s freeing bo "%s" hnd %x (offset 0x%lx)\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset); + drm__ipvr_gem_bo_free(bo); + } +} + +drm_public void +drm_ipvr_gem_bo_unreference(drm_ipvr_bo *bo) +{ + drm_ipvr_bo_gem * bo_gem = (drm_ipvr_bo_gem*)bo; + assert(atomic_read(&bo_gem->refcount) > 0); + if (atomic_dec_and_test(&bo_gem->refcount)) { + drm__ipvr_gem_bo_finalize(bo); + } else { + VERB("%s "%s" handle 0x%x offset 0x%lx, refcount became %d\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, atomic_read(&bo_gem->refcount)); + } +} + +drm_public int +drm_ipvr_gem_bo_map(drm_ipvr_bo *bo, unsigned long offset, + size_t size, int write_enable) +{ + int ret; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *)bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem*)bo; + struct drm_ipvr_gem_mmap_offset arg; + + assert(atomic_read(&bo_gem->mapcount) >= 0); + /* maintain userspace mapping cache, to avoid unnecessary IOCTLs */ + if (atomic_read(&bo_gem->mapcount) == 0 && !bo->virt) { + /* no need to call synccpu-grab, kernel do it automatically */ + VERB("Calling GEM_MMAP with handle %u offset 0x%"PRIx64", size %"PRIu64"\n", + bo->handle, offset, bo->size); + + if (!bo_gem->map_offset) { + memset(&arg, 0, sizeof(arg)); + arg.handle = bo->handle; + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_GEM_MMAP_OFFSET, + &arg, sizeof(arg)); + if (ret < 0) { + ERR("%s: failed to get mmap offset at bo "%s" 0x%x: %d (%s)\n", + __FUNCTION__, bo_gem->name, bo->handle, ret, strerror(ret)); + return ret; + } + } + bo->virt = drm_mmap(0, bo->alloc_size, PROT_READ | PROT_WRITE, + MAP_SHARED, bufmgr_gem->fd, bo_gem->map_offset); + if (bo->virt == MAP_FAILED) { + bo->virt = NULL; + return -errno; + } + + VERB("%s: map for first time, "%s" hnd 0x%x (offset 0x%lx) (%s) -> %p\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, bo_gem->name, bo->virt); + } + else { + VERB("%s: already mapped, "%s" hnd 0x%x (offset 0x%lx) (%s) -> %p\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, bo_gem->name, bo->virt); + } + atomic_inc(&bo_gem->mapcount); + assert(bo->virt); + VERB("%s: map_count became %d\n", + __FUNCTION__, atomic_read(&bo_gem->mapcount)); + + return 0; +} + +drm_public int +drm_ipvr_gem_bo_unmap(drm_ipvr_bo *bo) +{ + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem*)bo; + int mapcount; + + mapcount = atomic_read(&bo_gem->mapcount); + if (mapcount <= 0) { + VERB("%s: unexpected unmap() when mapcount is %d, ignore it\n", __FUNCTION__, mapcount); + } + else if (atomic_dec_and_test(&bo_gem->mapcount)) { + if (bo->virt) + drm_munmap(bo->virt, bo->size); + bo->virt = NULL; + } + VERB("%s: "%s" hnd 0x%x offset 0x%lx map_count became %d\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, atomic_read(&bo_gem->mapcount)); + return 0; +} + +drm_public void +drm_ipvr_gem_bo_wait(drm_ipvr_bo *bo) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*) bo->bufmgr; + drm_ipvr_bo_gem * bo_gem = (drm_ipvr_bo_gem*)bo; + struct drm_ipvr_gem_wait arg; + int ret; + VERB("%s: wait "%s" hnd 0x%x offset 0x%lx\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset); + + arg.handle = bo->handle; + arg.flags = 0; + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_GEM_WAIT, &arg, sizeof(arg)); + if (ret < 0) { + ERR("%s "%s" hnd %u offset 0x%lx failed, %d (%s)\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, ret, strerror(ret)); + } +} + +static unsigned +drm__ipvr_time_diff(struct timeval *now, struct timeval *then) +{ + long long val; + + val = now->tv_sec - then->tv_sec; + val *= 1000000LL; + val += now->tv_usec; + val -= then->tv_usec; + if (val < 1LL) + val = 1LL; + + return (unsigned) val; +} + +static int +drm__ipvr_setup_reloc_list(drm_ipvr_bo *bo) +{ + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bo->bufmgr; + unsigned int max_relocs = bufmgr_gem->max_relocs; + + if (bo->size / 4 < max_relocs) + max_relocs = bo->size / 4; + + bo_gem->relocs = malloc(max_relocs * + sizeof(struct drm_ipvr_gem_relocation_entry)); + bo_gem->reloc_target_info = malloc(max_relocs * + sizeof(drm_ipvr_reloc_target)); + if (bo_gem->relocs == NULL || bo_gem->reloc_target_info == NULL) { + ERR("%s: failed to alloc relocs and reloc_target_info\n", __FUNCTION__); + free (bo_gem->relocs); + bo_gem->relocs = NULL; + + free (bo_gem->reloc_target_info); + bo_gem->reloc_target_info = NULL; + + return 1; + } + + return 0; +} + +drm_public int +drm_ipvr_gem_bo_emit_reloc(drm_ipvr_bo *bo, unsigned long offset, + drm_ipvr_bo *target_bo, unsigned long target_offset, uint32_t skip_fence) +{ + int ret; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + drm_ipvr_bo_gem *target_bo_gem = (drm_ipvr_bo_gem *) target_bo; + VERB("%s::%d\n", __func__, __LINE__); + /* Create a new relocation list if needed */ + if (bo_gem->relocs == NULL) { + ret = drm__ipvr_setup_reloc_list(bo); + if (ret) { + ERR("%s failed: relocs = %p, setup_reloc_list returns %d (%s)\n", + __FUNCTION__, bo_gem->relocs, ret, strerror(ret)); + return -ENOMEM; + } + } + + /* Check overflow */ + assert(bo_gem->reloc_count < bufmgr_gem->max_relocs); + + /* Check args */ + assert(offset <= bo->size - 4); + + /* Make sure that we're not adding a reloc to something whose size has + * already been accounted for. + */ + if (target_bo_gem != bo_gem) { + /* todo: check it */ + } + /* An object needing a fence is a tiled buffer, so it won't have + * relocs to other buffers. + */ + bo_gem->relocs[bo_gem->reloc_count].offset = offset; + bo_gem->relocs[bo_gem->reloc_count].delta = target_offset; + bo_gem->relocs[bo_gem->reloc_count].target_handle = + target_bo_gem->base.handle; + bo_gem->relocs[bo_gem->reloc_count].presumed_offset = target_bo->offset; + bo_gem->reloc_target_info[bo_gem->reloc_count].bo = target_bo; + + if (target_bo != bo) + drm_ipvr_gem_bo_reference(target_bo); + bo_gem->reloc_target_info[bo_gem->reloc_count].flags = skip_fence? + 0: IPVR_EXEC_OBJECT_NEED_FENCE; + + bo_gem->reloc_count++; + + VERB("%s emitted reloc: "%s" hnd %x (0x%lx) => "%s" hnd %x (0x%lx = 0x%lx+0x%lx)\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, + target_bo_gem->name, target_bo->handle, + target_bo->offset + target_offset, target_bo->offset, target_offset); + return 0; +} + +drm_public drm_ipvr_bo * +drm_ipvr_gem_bo_create_from_name(drm_ipvr_bufmgr *bufmgr, drm_ipvr_context *ctx, + const char *name, uint32_t handle) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bufmgr; + drm_ipvr_bo_gem *bo_gem; + int ret; + struct drm_gem_open open_arg; + + open_arg.name = handle; + ret = drmIoctl(bufmgr_gem->fd, + DRM_IOCTL_GEM_OPEN, + &open_arg); + if (ret) { + ERR("Couldn't reference %s handle 0x%08x: %s\n", + name, handle, strerror(errno)); + return NULL; + } + bo_gem = calloc(1, sizeof(*bo_gem)); + if (!bo_gem) + return NULL; + + bo_gem->base.size = open_arg.size; + bo_gem->base.offset = 0; + bo_gem->base.bufmgr = bufmgr; + bo_gem->name = name; + atomic_set(&bo_gem->refcount, 1); + atomic_set(&bo_gem->mapcount, 0); + bo_gem->validate_index = -1; + bo_gem->base.handle = open_arg.handle; + bo_gem->global_name = handle; + bo_gem->base.reusable = false; + DBG("bo_create_from_handle: %d (%s)\n", handle, bo_gem->name); + + return &bo_gem->base; +} + +drm_public int +drm_ipvr_gem_bo_flink(drm_ipvr_bo *bo, uint32_t * name) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + int ret; + struct drm_gem_flink flink; + if (!bo_gem->global_name) { + flink.handle = bo_gem->base.handle; + + ret = drmIoctl(bufmgr_gem->fd, DRM_IOCTL_GEM_FLINK, &flink); + if (ret != 0) + return -errno; + + bo_gem->global_name = flink.name; + bo_gem->base.reusable = false; + } + + *name = bo_gem->global_name; + return 0; +} + +drm_public int +drm_ipvr_gem_bo_export_to_prime(drm_ipvr_bo *bo, int *prime_fd) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + int ret; + ret = drmPrimeHandleToFD(bufmgr_gem->fd, bo_gem->base.handle, + DRM_CLOEXEC, prime_fd); + if (ret) { + ERR("%s: failed calling drmPrimeHandleToFD: %d (%s)\n", + __FUNCTION__, ret, strerror(ret)) + return ret; + } + + bo_gem->base.reusable = false; + + return 0; +} + +drm_public drm_ipvr_bo* +drm_ipvr_gem_bo_create_from_prime(drm_ipvr_bufmgr *bufmgr, + int prime_fd, int size) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem *) bufmgr; + int ret; + uint32_t handle; + drm_ipvr_bo_gem *bo_gem; + + ret = drmPrimeFDToHandle(bufmgr_gem->fd, prime_fd, &handle); + if (ret) { + ERR("%s: failed calling drmPrimeFDToHandle: %d (%s)\n", + __FUNCTION__, ret, strerror(ret)) + return NULL; + } + + /* + * See if the kernel has already returned this buffer to us. Just as + * for named buffers, we must not create two bo's pointing at the same + * kernel object + */ + pthread_mutex_lock(&bufmgr_gem->lock); + + bo_gem = calloc(1, sizeof(*bo_gem)); + if (!bo_gem) { + pthread_mutex_unlock(&bufmgr_gem->lock); + return NULL; + } + /* Determine size of bo. The fd-to-handle ioctl really should + * return the size, but it doesn't. If we have kernel 3.12 or + * later, we can lseek on the prime fd to get the size. Older + * kernels will just fail, in which case we fall back to the + * provided (estimated or guess size). */ + ret = lseek(prime_fd, 0, SEEK_END); + if (ret != -1) { + bo_gem->base.size = ret; + bo_gem->base.alloc_size = ret; + } + else { + bo_gem->base.size = size; + bo_gem->base.alloc_size = size; + } + + bo_gem->base.bufmgr = bufmgr; + bo_gem->base.handle = handle; + bo_gem->tiling = 0; + bo_gem->caching = 0; + bo_gem->stride = 0; /* FIXME: does ipvr_drv_video maintain stride/sliceHeight itself? */ + bo_gem->name = "prime"; + bo_gem->base.virt = NULL; + bo_gem->map_offset = 0; + atomic_set(&bo_gem->mapcount, 0); + atomic_set(&bo_gem->refcount, 1); + bo_gem->validate_index = -1; + bo_gem->ctx = NULL; + bo_gem->base.reusable = false; + bo_gem->exec_seq = -1; + bo_gem->is_prime_imported = true; + + pthread_mutex_unlock(&bufmgr_gem->lock); + + return &bo_gem->base; +} + +static int +drm__ipvr_add_validate_buffer(drm_ipvr_bo *bo, uint64_t flags) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = NULL; + drm_ipvr_bo_gem *bo_gem = NULL; + int index; + + if (!bo) { + ERR("%s, invalid bo %p\n", __FUNCTION__, bo); + return -EINVAL; + } + bo_gem = (drm_ipvr_bo_gem *) bo; + if (!(bo->bufmgr)) { + ERR("%s, invalid bo bufmgr %p for bo "%s" hnd %x (off 0x%lx)\n", + __FUNCTION__, bo->bufmgr, bo_gem->name, bo->handle, bo->offset); + return -EINVAL; + } + + bufmgr_gem = (drm_ipvr_bufmgr_gem *)bo->bufmgr; + + if (bo_gem->validate_index != -1) { + if (!(bufmgr_gem->exec_objs[bo_gem->validate_index].flags & IPVR_EXEC_OBJECT_NEED_FENCE) + && (flags & IPVR_EXEC_OBJECT_NEED_FENCE)) { + VERB("%s: already has validate_index %d, update flags: 0x%llx->0x%llx\n", + __FUNCTION__, bo_gem->validate_index, + bufmgr_gem->exec_objs[bo_gem->validate_index].flags, + IPVR_EXEC_OBJECT_NEED_FENCE | bufmgr_gem->exec_objs[bo_gem->validate_index].flags); + bufmgr_gem->exec_objs[bo_gem->validate_index].flags |= IPVR_EXEC_OBJECT_NEED_FENCE; + } + return 0; + } + + /* Extend the array of validation entries as necessary. */ + if (bufmgr_gem->exec_count == bufmgr_gem->exec_size) { + int new_size = bufmgr_gem->exec_size * 2; + + if (new_size == 0) + new_size = 10; + + bufmgr_gem->exec_objs = + realloc(bufmgr_gem->exec_objs, + sizeof(*bufmgr_gem->exec_objs) * new_size); + bufmgr_gem->exec_bos = + realloc(bufmgr_gem->exec_bos, + sizeof(*bufmgr_gem->exec_bos) * new_size); + bufmgr_gem->exec_size = new_size; + } + + if (!bufmgr_gem->exec_objs || !bufmgr_gem->exec_bos) { + ERR("%s val_args = %p and exec_bos = %p\n", + __FUNCTION__, bufmgr_gem->exec_objs, bufmgr_gem->exec_bos); + return -ENOMEM; + } + + index = bufmgr_gem->exec_count; + bo_gem->validate_index = index; + /* Fill in array entry */ + if (bufmgr_gem->exec_objs && bufmgr_gem->exec_bos) { + bufmgr_gem->exec_objs[index].offset = bo->offset; + bufmgr_gem->exec_objs[index].handle = bo_gem->base.handle; + bufmgr_gem->exec_objs[index].flags = flags; + bufmgr_gem->exec_objs[index].relocs_ptr = (uintptr_t)bo_gem->relocs; + bufmgr_gem->exec_objs[index].relocation_count = bo_gem->reloc_count; + bufmgr_gem->exec_bos[index] = bo; + + VERB("%s added validate buffer "%s" hnd %x (off 0x%lx) at [%d]\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, index); + bufmgr_gem->exec_count++; + } + return 0; +} + +static int +drm__ipvr_bo_process_reloc(drm_ipvr_bo *bo, uint64_t flags) +{ + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *)bo; + int i; + int ret = 0; + + /* Add the target to the validate list */ + ret = drm__ipvr_add_validate_buffer(bo, flags); + if (ret != 0) + return ret; + + if (bo_gem->relocs == NULL) + return 0; + + for (i = 0; i < bo_gem->reloc_count; i++) { + drm_ipvr_bo *target_bo = bo_gem->reloc_target_info[i].bo; + uint64_t target_flags = bo_gem->reloc_target_info[i].flags; + if (target_bo == bo) + continue; + + /* Continue walking the tree depth-first. */ + ret = drm__ipvr_bo_process_reloc(target_bo, target_flags); + if (ret != 0) + return ret; + } + + return ret; +} + +drm_public int +drm_ipvr_gem_bo_exec(drm_ipvr_bo *bo, unsigned long offset, size_t len, + int fence_in, int *fence_out) +{ + int ret; + int i; + struct timeval then, now; + bool have_then = false; + struct drm_ipvr_gem_execbuffer exec_arg; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *)bo; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bo->bufmgr; + + pthread_mutex_lock(&bufmgr_gem->lock); + + ret = drm__ipvr_bo_process_reloc(bo, + IPVR_EXEC_OBJECT_NEED_FENCE | IPVR_EXEC_OBJECT_SUBMIT); + if (ret != 0) { + ERR("%s bo_process_reloc2 failed: %d (%s)\n", + __FUNCTION__, ret, strerror(ret)); + pthread_mutex_unlock(&bufmgr_gem->lock); + return ret; + } + + VERB("%s finished the reloc/validate processing. exec_count=%d\n", + __FUNCTION__, bufmgr_gem->exec_count); + exec_arg.exec_start_offset = offset; + exec_arg.exec_len = len; + exec_arg.buffers_ptr = (uintptr_t)bufmgr_gem->exec_objs; + exec_arg.buffer_count = bufmgr_gem->exec_count; + exec_arg.ctx_id = bo_gem->ctx->ctx_id; + VERB("%s sending EXEC IOCTL to kernel with: cmdbuf_handle %x (0x%lx), size %zu, " + "buffer_list_count %u, ctx_id 0x%08x\n", __FUNCTION__, bo->handle, + bo->offset, len, + exec_arg.buffer_count, exec_arg.ctx_id); + + bufmgr_gem->exec_seq ++; + do { + ret = drmCommandWriteRead(bufmgr_gem->fd, + DRM_IPVR_GEM_EXECBUFFER, + &exec_arg, sizeof(exec_arg)); + if (ret == EAGAIN) { + if (!have_then) { + if (gettimeofday(&then, NULL)) { + ERR("%s have no then, gettimeofday error.\n", __FUNCTION__); + break; + } + + have_then = true; + } + if (gettimeofday(&now, NULL)) { + ERR("%s: Gettimeofday error.\n", __FUNCTION__); + break; + } + + } + } while ((ret == EAGAIN) && (drm__ipvr_time_diff(&now, &then) < IPVR_TIMEOUT_USEC)); + + if (ret) { + WARN("%s: command write return is %d\n", __FUNCTION__, ret); + goto out; + } + + /* update the presumed_mmu_offsets + * in case we support eviction in IPVR driver later */ + for (i = 0; i < bufmgr_gem->exec_count; i++) { + drm_ipvr_bo *bo = bufmgr_gem->exec_bos[i]; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *)bo; + + /* Update the buffer offset to accelerate future exec */ + if (bufmgr_gem->exec_objs[i].offset != bo->offset) { + VERB("BO "%s" hnd %x migrated: 0x%"PRIx64" -> 0x%llx\n", + bo_gem->name, bo->handle, bo->offset, + bufmgr_gem->exec_objs[i].offset); + bo->offset = bufmgr_gem->exec_objs[i].offset; + } + bo_gem->exec_seq = bufmgr_gem->exec_seq; + } + + VERB("%s EXECBUFFER IOCTL succeeded\n", __FUNCTION__); +out: + for (i = 0; i < bufmgr_gem->exec_count; i++) { + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem*)bufmgr_gem->exec_bos[i]; + + /* Disconnect the buffer from the validate list */ + bo_gem->validate_index = -1; + bufmgr_gem->exec_bos[i] = NULL; + } + + /* autmatically decrease all bo's refcount in the reloc tree. + * VA created bo will be cached if unreferencing in vaRenderPicture. + * execbuf/mtxmsg/surface/colocated_buffer's refcount will become 1. + */ + drm_ipvr_gem_bo_remove_relocs(bo, 1); + + bufmgr_gem->exec_count = 0; + pthread_mutex_unlock(&bufmgr_gem->lock); + return 0; +} + +drm_public int +drm_ipvr_gem_bo_busy(drm_ipvr_bo *bo) +{ + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bo->bufmgr; + drm_ipvr_bo_gem *bo_gem = (drm_ipvr_bo_gem *) bo; + struct drm_ipvr_gem_busy arg; + int ret; + + arg.handle = bo->handle; + + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_GEM_BUSY, &arg, sizeof(arg)); + if (ret == -EBUSY || (ret == 0 && arg.busy)) { + VERB("%s returns %d: buffer "%s" hnd %x (offset 0x%lx) param is busy".\n", + __FUNCTION__, ret, bo_gem->name, bo->handle, bo->offset); + return 1; + } + else if (!ret && !arg.busy){ + VERB("%s: buffer "%s" hnd %x (offset 0x%lx) param is free".\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset); + return 0; + } + else { + WARN("%s: checking buffer busy "%s" hnd %u (offset 0x%lx) " + "got unexpected result: %d (%s), busy state=%d, view it as free.\n", + __FUNCTION__, bo_gem->name, bo->handle, bo->offset, ret, strerror(ret), arg.busy); + + return 0; + } +} + +drm_public void +drm_ipvr_gem_bufmgr_destroy(drm_ipvr_bufmgr *bufmgr) +{ + int j; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bufmgr; + + if (bufmgr_gem->exec_bos) + free(bufmgr_gem->exec_bos); + if (bufmgr_gem->exec_objs) + free(bufmgr_gem->exec_objs); + + if (bufmgr_gem->cache_buckets) { + for (j = 0; j < bufmgr_gem->cache_bucket_size; j++) { + struct drm_ipvr_cache_bucket *bucket = + &bufmgr_gem->cache_buckets[j]; + + while (!DRMLISTEMPTY(&bucket->head)) { + drm_ipvr_bo_gem *bo_gem; + + bo_gem = DRMLISTENTRY(drm_ipvr_bo_gem, + bucket->head.next, cache_head); + + DRMLISTDELINIT(&bo_gem->cache_head); + bucket->count--; + assert(bo_gem->reloc_count == 0); + drm__ipvr_gem_bo_free(&bo_gem->base); + } + } + + free(bufmgr_gem->cache_buckets); + } + + pthread_mutex_destroy(&bufmgr_gem->lock); + free(bufmgr_gem); +} + +drm_public int +drm_ipvr_gem_bo_references(drm_ipvr_bo *bo, drm_ipvr_bo *target_bo) +{ + return -1; +} + +drm_public void +drm_ipvr_gem_context_destroy(drm_ipvr_context *ctx) +{ + int ret; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)ctx->bufmgr; + struct drm_ipvr_context_destroy arg; + arg.ctx_id = ctx->ctx_id; + + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_CONTEXT_DESTROY, &arg, sizeof(arg)); + if (ret != 0) { + ERR("%s: Error destroying context %u: %s .\n", + __FUNCTION__, ctx->ctx_id, strerror(ret)); + } + + pthread_mutex_destroy(&bufmgr_gem->lock); + pthread_mutex_destroy(&bufmgr_gem->list_lock); + free(ctx); +} + +drm_public drm_ipvr_context* +drm_ipvr_gem_context_create(drm_ipvr_bufmgr *bufmgr, uint32_t ctx_type, + uint32_t tiling_stride, uint32_t tiling_mode) +{ + int ret; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bufmgr; + struct drm_ipvr_context_create arg; + drm_ipvr_context *ctx = calloc(1, sizeof(drm_ipvr_context)); + arg.ctx_type = ctx_type; + arg.tiling_stride = tiling_stride; + arg.tiling_scheme = tiling_mode; + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_CONTEXT_CREATE, &arg, sizeof(arg)); + if (ret != 0) { + ERR("%s: Error creating context %d: %s. tiling stride=%u, scheme=%u\n", + __FUNCTION__, ctx_type, strerror(ret), + arg.tiling_stride, arg.tiling_scheme); + free(ctx); + return NULL; + } + ctx->ctx_id = arg.ctx_id; + ctx->ctx_type = ctx_type; + ctx->bufmgr = bufmgr; + + DBG("%s: successfully create drm context: id=%x, tiling stride=%u, scheme=%u\n", + __FUNCTION__, ctx->ctx_id, arg.tiling_stride, arg.tiling_scheme); + return ctx; +} + +drm_public int +drm_ipvr_gem_bufmgr_get_device_info(drm_ipvr_bufmgr *bufmgr, uint16_t *dev_id, uint16_t *caps) +{ + struct drm_ipvr_misc arg; + int ret; + drm_ipvr_bufmgr_gem *bufmgr_gem = (drm_ipvr_bufmgr_gem*)bufmgr; + unsigned long info; + arg.key = IPVR_DEVICE_INFO; + arg.value = (uint64_t)((unsigned long)&info); + ret = drmCommandWriteRead(bufmgr_gem->fd, DRM_IPVR_MISC, &arg, sizeof(arg)); + if (ret != 0) { + ERR("%s: Error getting device info: %s .\n", + __FUNCTION__, strerror(ret)); + return ret; + } + *dev_id = (info >> 16) & 0xffff; + *caps = info & 0xffff; + return 0; +} + +drm_public drm_ipvr_bufmgr * +drm_ipvr_gem_bufmgr_init(int fd, int batch_size) +{ + drm_ipvr_bufmgr_gem * bufmgr_gem = NULL; + int j; + + bufmgr_gem = calloc(1, sizeof(*bufmgr_gem)); + if (bufmgr_gem == NULL) { + ERR("bufmgr init: calloc failed: %s\n", strerror(errno)); + return NULL; + } + + bufmgr_gem->fd = fd; + + if (pthread_mutex_init(&bufmgr_gem->lock, NULL) != 0) { + ERR("bufmgr init: mutex init failed: %s\n", strerror(errno)); + free(bufmgr_gem); + bufmgr_gem = NULL; + return NULL; + } + if (pthread_mutex_init(&bufmgr_gem->list_lock, NULL) != 0) { + ERR("bufmgr init: mutex init failed: %s\n", strerror(errno)); + free(bufmgr_gem); + bufmgr_gem = NULL; + return NULL; + } + + /* Hard code one. + */ + bufmgr_gem->max_relocs = 32; + + bufmgr_gem->exec_seq = -1; + + /* init cache buckets */ + /* 4KB, 8KB, 16KB, 32KB, 64KB, 128KB, 256KB, 512KB, 1MB, 2MB, 4MB, 8MB, 16MB, 32MB */ + bufmgr_gem->cache_bucket_size = 14; + bufmgr_gem->cache_buckets = calloc(bufmgr_gem->cache_bucket_size, + sizeof(struct drm_ipvr_cache_bucket)); + if (!bufmgr_gem->cache_buckets) { + ERR("%s failed to allocate cache buckets\n", __FUNCTION__); + return NULL; + } + + for (j = 0; j < bufmgr_gem->cache_bucket_size; ++j) { + DRMINITLISTHEAD(&bufmgr_gem->cache_buckets[j].head); + bufmgr_gem->cache_buckets[j].size = 0x1000 << j; + bufmgr_gem->cache_buckets[j].limit = 2 << ((14 - j)/2); + bufmgr_gem->cache_buckets[j].count = 0; + } + + DBG("%s created bufmgr %p with cache_bucket_size %u\n", + __FUNCTION__, &bufmgr_gem->base, bufmgr_gem->cache_bucket_size); + + return &bufmgr_gem->base; +} diff --git a/ipvr/libdrm_ipvr.pc.in b/ipvr/libdrm_ipvr.pc.in new file mode 100644 index 0000000..2fe9ab6 --- /dev/null +++ b/ipvr/libdrm_ipvr.pc.in @@ -0,0 +1,11 @@ +prefix=@prefix@ +exec_prefix=@exec_prefix@ +libdir=@libdir@ +includedir=@includedir@ + +Name: libdrm_ipvr +Description: Userspace interface to VP8 hardware decoder on baytrail +Version: @PACKAGE_VERSION@ +Requires: libdrm +Libs: -L${libdir} -ldrm_ipvr +Cflags: -I${includedir} -I${includedir}/libdrm diff --git a/ipvr/test_ioctl.c b/ipvr/test_ioctl.c new file mode 100644 index 0000000..976c244 --- /dev/null +++ b/ipvr/test_ioctl.c @@ -0,0 +1,423 @@ +/************************************************************************** + * test_ioctl.c: it is gem ioctl unit test + * + * Copyright (c) 2014 Intel Corporation, Hillsboro, OR, USA + * All Rights Reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License along with + * this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + * + * Authors: + * Yao Cheng yao.cheng@intel.com + * + **************************************************************************/ +#include <stdlib.h> +#include <stdio.h> +#include <unistd.h> +#include <string.h> +#include <assert.h> +#include <fcntl.h> +#include <inttypes.h> +#include <errno.h> +#include <sys/stat.h> +#include <sys/ioctl.h> +#include <sys/mman.h> +#include <stdint.h> +#include <sys/ioctl.h> +#include "ipvr_drm.h" +#include "i915_drm.h" +#include "../intel/intel_bufmgr.h" +#include <xf86drm.h> +#include <sys/time.h> +#include <time.h> +#include <errno.h> +#include <libdrm.h> + +#define exiterr(ret, fmt, ...) \ + do { \ + fprintf(stderr, fmt, ##__VA_ARGS__); \ + exit(ret); \ + } while (0); + +#define check_ret(ret) \ + do { \ + if (ret) { \ + exiterr(ret, "failed at %s::%d: %d (%s)\n", __FILE__, __LINE__, ret, strerror(ret)); \ + } \ + } while (0); +#define check_addr(addr) \ + do { \ + if (!addr) { \ + exiterr(-1, "failed at %s::%d: NULL address\n", __FILE__, __LINE__); \ + } \ + } while (0); + +static unsigned char msg_data[] = { + 0x14,0x81,0x80,0x01,0x11,0x49,0x85,0x00,0x00,0xd0,0x5b,0x00,0x95,0x70,0xee,0x31, + 0x00,0x10,0x21,0x04 +}; + +static void basic_test(int fd) +{ + struct drm_ipvr_gem_create create_arg; + struct drm_gem_close close_arg; + struct drm_gem_flink flink_arg; + struct drm_gem_open open_arg; + struct drm_ipvr_gem_mmap_offset map_offset_arg; + int i915_fd, export_fd, i915_export_fd; + dri_bufmgr *i915_bufmgr; + dri_bo *i915_bo; + uint32_t global_name; + uint32_t *vaddr; + uint32_t import_handle; + int ret; + + printf("\nTesting gem create.\n"); + + /* create bo iotcl*/ + memset(&create_arg, 0, sizeof(create_arg)); + create_arg.size = 16 * 1024; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_CREATE, &create_arg, sizeof(create_arg)); + check_ret(ret); + + printf("\nTest mmap.\n"); + + /* mmap bo ioctl*/ + if (!create_arg.map_offset) { + printf("didn't get valid map_offset during BO creation. call MMAP_OFFSET\n"); + map_offset_arg.handle = create_arg.handle; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_MMAP_OFFSET, &map_offset_arg, sizeof(map_offset_arg)); + check_ret(ret); + } + vaddr = drm_mmap(NULL, create_arg.rounded_size, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, create_arg.map_offset); + if (vaddr) { + printf("succeed to call drm_mmap, addr_ptr is %p, writting" + " 0xa5a5a5a5 0xb6b6b6b6 0xc7c7c7c7 0xd8d8d8d8 into it.\n", vaddr); + *vaddr++ = 0xa5a5a5a5; + *vaddr++ = 0xb6b6b6b6; + *vaddr++ = 0xc7c7c7c7; + *vaddr++ = 0xd8d8d8d8; + } else + exiterr(-1, "failed to call DRM_IOCTL_IPVR_GEM_MMAP ioctl: %d (%s).\n", ret, strerror(ret)); + drm_munmap(vaddr, create_arg.rounded_size); + + printf("\nTest opening flinked BOs.\n"); + + { + /* flink the bo */ + flink_arg.handle = create_arg.handle; + flink_arg.name = 0; + ret = drmIoctl(fd, DRM_IOCTL_GEM_FLINK, &flink_arg); + check_ret(ret); + global_name = flink_arg.name; + + /* open from flink name */ + open_arg.name = global_name; + ret = drmIoctl(fd, DRM_IOCTL_GEM_OPEN, &open_arg); + check_ret(ret); + + if (open_arg.size != create_arg.rounded_size) + exiterr(-1, "opened size %llu not consistent with created size %llu\n", + open_arg.size, create_arg.rounded_size); + + /* mmap and verify the bo content */ + map_offset_arg.handle = open_arg.handle; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_MMAP_OFFSET, &map_offset_arg, sizeof(map_offset_arg)); + check_ret(ret); + /* mmap bo ioctl*/ + vaddr = drm_mmap(NULL, open_arg.size, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, map_offset_arg.offset); + check_addr(vaddr); + printf("succeed to call drm_mmap, addr_ptr is %p, readding got " + "0x%08x 0x%08x 0x%08x 0x%08x.\n", vaddr, + *vaddr, *(vaddr+1), *(vaddr+2), *(vaddr+3)); + if (*vaddr == 0xa5a5a5a5 && + *(vaddr+1) == 0xb6b6b6b6 && + *(vaddr+2) == 0xc7c7c7c7 && + *(vaddr+3) == 0xd8d8d8d8) { + printf("succeed to verify opened BO content.\n"); + } + else { + drm_munmap(vaddr, create_arg.rounded_size); + exiterr(-1, "failed to call DRM_IOCTL_IPVR_GEM_MMAP ioctl: %d (%s).\n", ret, strerror(ret)); + } + drm_munmap(vaddr, create_arg.rounded_size); + + /* close opened bo ioctl*/ + close_arg.handle = open_arg.handle; + ret = drmIoctl(fd, DRM_IOCTL_GEM_CLOSE, &close_arg); + check_ret(ret); + } + + printf("\nTest importing i915 prime BOs.\n"); + + i915_fd = open("/dev/dri/card0", O_RDWR); + i915_bufmgr = drm_intel_bufmgr_gem_init(i915_fd, 4096); + + /* test prime import */ + { + i915_bo = dri_bo_alloc(i915_bufmgr, "to-be-exported", 4096, 0x1000); + check_addr(i915_bo); + ret = dri_bo_map(i915_bo, 1); + check_ret(ret); + check_addr(i915_bo->virtual); + vaddr = i915_bo->virtual; + printf("mapped i915 BO to %p and write" + " 0xd8d8d8d8 0xc7c7c7c7 0xb6b6b6b6 0xa5a5a5a5 into it.\n", vaddr); + *vaddr++ = 0xd8d8d8d8; + *vaddr++ = 0xc7c7c7c7; + *vaddr++ = 0xb6b6b6b6; + *vaddr++ = 0xa5a5a5a5; + ret = dri_bo_unmap(i915_bo); + check_ret(ret); + ret = drm_intel_bo_gem_export_to_prime(i915_bo, &i915_export_fd); + check_ret(ret); + ret = drmPrimeFDToHandle(fd, i915_export_fd, &import_handle); + map_offset_arg.handle = import_handle; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_MMAP_OFFSET, &map_offset_arg, sizeof(map_offset_arg)); + check_ret(ret); + vaddr = drm_mmap(NULL, 4096, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, map_offset_arg.offset); + check_addr(vaddr); + printf("succeed mapping the imported BO to %p, content is " + "0x%08x 0x%08x 0x%08x 0x%08x.\n", vaddr, + *vaddr, *(vaddr+1), *(vaddr+2), *(vaddr+3)); + if (*vaddr == 0xd8d8d8d8 && + *(vaddr+1) == 0xc7c7c7c7 && + *(vaddr+2) == 0xb6b6b6b6 && + *(vaddr+3) == 0xa5a5a5a5) { + printf("succeeded verifying imported i915 prime BO content.\n"); + } + else + exiterr(-1, "failed to verify imported i915 prime BO content\n"); + ret = drm_munmap(vaddr, 4096); + check_ret(ret); + close_arg.handle = import_handle; + ret = drmIoctl(fd, DRM_IOCTL_GEM_CLOSE, &close_arg); + check_ret(ret); + close(i915_export_fd); + dri_bo_unreference(i915_bo); + } + + printf("\nTest exporting prime BO and import with i915.\n"); + + /* test prime export */ + { + ret = drmPrimeHandleToFD(fd, create_arg.handle, 0, &export_fd); + check_ret(ret); + i915_bo = drm_intel_bo_gem_create_from_prime(i915_bufmgr, export_fd, create_arg.rounded_size); + check_addr(i915_bo); + ret = drm_intel_gem_bo_map_gtt(i915_bo); + check_ret(ret); + check_addr (i915_bo->virtual); + vaddr = i915_bo->virtual; + printf("succeed to import with i915 and mapping to %p, content is " + "0x%08x 0x%08x 0x%08x 0x%08x.\n", vaddr, + *vaddr, *(vaddr+1), *(vaddr+2), *(vaddr+3)); + if (*vaddr == 0xa5a5a5a5 && + *(vaddr+1) == 0xb6b6b6b6 && + *(vaddr+2) == 0xc7c7c7c7 && + *(vaddr+3) == 0xd8d8d8d8) { + printf("succeed to verify exported prime BO content.\n"); + } + else + exiterr(-1, "failed to verify exported prime BO content.\n"); + ret = dri_bo_unmap(i915_bo); + check_ret(ret); + dri_bo_unreference(i915_bo); + close(export_fd); + } + dri_bufmgr_destroy(i915_bufmgr); + close(i915_fd); + + /* close bo ioctl*/ + close_arg.handle = create_arg.handle; + ret = drmIoctl(fd, DRM_IOCTL_GEM_CLOSE, &close_arg); + check_ret(ret); + + printf("\nExit gem basic test.\n"); +} + +static void exec_test(int fd) +{ + struct drm_ipvr_context_create ctx_create_arg; + struct drm_ipvr_context_destroy ctx_destroy_arg; + struct drm_ipvr_gem_create create_arg; + struct drm_ipvr_gem_execbuffer exec_arg; + struct drm_ipvr_gem_wait wait_arg; + struct drm_ipvr_gem_busy busy_arg; + struct drm_gem_close close_arg; + struct drm_ipvr_gem_relocation_entry reloc_entries[3]; + struct drm_ipvr_gem_exec_object *exec_objs; + struct drm_ipvr_gem_exec_object *arg; + int i915_fd, i915_export_fd; + dri_bufmgr *i915_bufmgr; + dri_bo *i915_bo; + uint32_t import_handle; + uint32_t *dword; + uint32_t *vaddr; + int i, ret, buffer_count, cmdbuf_size, ctx_type; + + printf("\nTest simple exec ioctl.\n"); + i915_fd = open("/dev/dri/card0", O_RDWR); + i915_bufmgr = drm_intel_bufmgr_gem_init(i915_fd, 4096); + + /* create a ctx_context, which is needed for exec ioctl */ + memset(&ctx_create_arg, 0, sizeof(ctx_create_arg)); + ctx_type = IPVR_CONTEXT_TYPE_VED; + ctx_create_arg.ctx_type = ctx_type; + ctx_create_arg.tiling_scheme = 0; + ctx_create_arg.tiling_stride = 0; + ret = drmCommandWriteRead(fd, DRM_IPVR_CONTEXT_CREATE, &ctx_create_arg, sizeof(ctx_create_arg)); + check_ret(ret); + + buffer_count = 4; + dword = msg_data; + exec_objs = calloc(1, sizeof(struct drm_ipvr_gem_exec_object) * buffer_count); + + /* create 1 bo with correct mmu offset */ + arg = &exec_objs[0]; + memset(&create_arg, 0, sizeof(create_arg)); + create_arg.size = 16 * 1024; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_CREATE, &create_arg, sizeof(create_arg)); + check_ret(ret); + arg->handle = create_arg.handle; + arg->offset = create_arg.mmu_offset; + arg->flags |= IPVR_EXEC_OBJECT_NEED_FENCE; + arg->relocation_count = 0; + arg->relocs_ptr = (uintptr_t)NULL; + reloc_entries[0].delta = 0; + reloc_entries[0].offset = 8; + reloc_entries[0].presumed_offset = create_arg.mmu_offset; + reloc_entries[0].target_handle = create_arg.handle; + + /* create 1 bo with wrong mmu offset */ + arg = &exec_objs[1]; + memset(&create_arg, 0, sizeof(create_arg)); + create_arg.size = 16 * 1024; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_CREATE, &create_arg, sizeof(create_arg)); + check_ret(ret); + arg->handle = create_arg.handle; + arg->offset = create_arg.mmu_offset; + arg->flags |= IPVR_EXEC_OBJECT_NEED_FENCE; + arg->relocation_count = 0; + arg->relocs_ptr = (uintptr_t)NULL; + reloc_entries[1].delta = 0; + reloc_entries[1].offset = 12; + reloc_entries[1].presumed_offset = dword[reloc_entries[1].offset/4]; + arg->offset = reloc_entries[1].presumed_offset; + reloc_entries[1].target_handle = create_arg.handle; + + /* import a i915 bo */ + i915_bo = dri_bo_alloc(i915_bufmgr, "to-be-exported", 16 * 1024, 0x1000); + check_addr(i915_bo); + ret = drm_intel_bo_gem_export_to_prime(i915_bo, &i915_export_fd); + check_ret(ret); + ret = drmPrimeFDToHandle(fd, i915_export_fd, &import_handle); + check_ret(ret); + arg = &exec_objs[2]; + arg->handle = import_handle; + arg->offset = 0; /* we don't know the mmu offset */ + arg->flags |= IPVR_EXEC_OBJECT_NEED_FENCE; + arg->relocation_count = 0; + arg->relocs_ptr = (uintptr_t)NULL; + reloc_entries[2].delta = 0; + reloc_entries[2].offset = 16; + reloc_entries[2].presumed_offset = 0; + reloc_entries[2].target_handle = import_handle; + + /* create a cmd bo with 1 page */ + memset(&create_arg, 0, sizeof(create_arg)); + cmdbuf_size = 4 * 1024; + create_arg.size = cmdbuf_size; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_CREATE, &create_arg, sizeof(create_arg)); + check_ret(ret); + + vaddr = drm_mmap(NULL, create_arg.rounded_size, PROT_READ | PROT_WRITE, + MAP_SHARED, fd, create_arg.map_offset); + if (vaddr) { + printf("succeed to map CMD BO, addr_ptr is %p.\n", vaddr); + memcpy((void *)vaddr, msg_data, sizeof(msg_data)); + } else + exiterr(-1, "failed to map CMD BO.\n"); + drm_munmap(vaddr, create_arg.rounded_size); + arg = &exec_objs[3]; + arg->handle = create_arg.handle; + arg->offset = create_arg.mmu_offset; + arg->flags |= IPVR_EXEC_OBJECT_NEED_FENCE | IPVR_EXEC_OBJECT_SUBMIT; + arg->relocation_count = 3; + arg->relocs_ptr = (uintptr_t)reloc_entries; + + /* run execbuffer ioctl */ + memset(&exec_arg, 0, sizeof(exec_arg)); + exec_arg.buffer_count = buffer_count; + exec_arg.buffers_ptr = (uintptr_t)exec_objs; + exec_arg.exec_start_offset = 0; + exec_arg.exec_len = 20; + exec_arg.ctx_id = ctx_create_arg.ctx_id; + + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_EXECBUFFER, &exec_arg, sizeof(exec_arg)); + check_ret(ret); + + /*check busy ioctl*/ + memset(&busy_arg, 0, sizeof(busy_arg)); + busy_arg.handle = create_arg.handle; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_BUSY, &busy_arg, sizeof(busy_arg)); + if (ret != 0 && ret != -EBUSY) + exiterr(ret, "failed calling DRM_IPVR_GEM_BUSY: %d (%s)\n", ret, strerror(ret)); + + /*wait for finish*/ + memset(&wait_arg, 0, sizeof(wait_arg)); + wait_arg.handle = create_arg.handle; + ret = drmCommandWriteRead(fd, DRM_IPVR_GEM_WAIT, &wait_arg, sizeof(wait_arg)); + /* the random command causes PANIC */ + check_ret(!(ret == 0) && !(ret == -EDEADLK)); + + /* close val bo and cmd bo */ + memset(&close_arg, 0, sizeof(close_arg)); + for (i = 0; i < buffer_count; i++) { + close_arg.handle = exec_objs[i].handle; + ret = drmIoctl(fd, DRM_IOCTL_GEM_CLOSE, &close_arg); + if (ret == 0) + printf("succeed to call DRM_IOCTL_GEM_CLOSE ioctl.\n"); + else + exiterr(-1, "failed to call DRM_IOCTL_GEM_CLOSE ioctl.\n"); + } + + /* clean up i915 */ + close(i915_export_fd); + dri_bo_unreference(i915_bo); + dri_bufmgr_destroy(i915_bufmgr); + close(i915_fd); + + /* destroy the ctx_context */ + memset(&ctx_destroy_arg, 0, sizeof(ctx_destroy_arg)); + ctx_destroy_arg.ctx_id = ctx_create_arg.ctx_id; + ret = drmCommandWriteRead(fd, DRM_IPVR_CONTEXT_DESTROY, &ctx_destroy_arg, sizeof(ctx_destroy_arg)); + check_ret(ret); + return; +} + +int main(int argc, char **argv) +{ + int fd; + + fd = open("/dev/dri/card1", O_RDWR); + /* test bo create, mmap, close */ + basic_test(fd); + /* simulate exec ioctl */ + exec_test(fd); + close(fd); + return 0; +}