1 /*- 2 * Copyright (C) 2012 Intel Corporation 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 __FBSDID("$FreeBSD$"); 28 29 #ifndef __IOAT_H__ 30 #define __IOAT_H__ 31 32 #include <sys/param.h> 33 #include <machine/bus.h> 34 35 /* 36 * This file defines the public interface to the IOAT driver. 37 */ 38 39 /* 40 * Enables an interrupt for this operation. Typically, you would only enable 41 * this on the last operation in a group 42 */ 43 #define DMA_INT_EN 0x1 44 /* 45 * Like M_NOWAIT. Operations will return NULL if they cannot allocate a 46 * descriptor without blocking. 47 */ 48 #define DMA_NO_WAIT 0x2 49 /* 50 * Disallow prefetching the source of the following operation. Ordinarily, DMA 51 * operations can be pipelined on some hardware. E.g., operation 2's source 52 * may be prefetched before operation 1 completes. 53 */ 54 #define DMA_FENCE 0x4 55 #define _DMA_GENERIC_FLAGS (DMA_INT_EN | DMA_NO_WAIT | DMA_FENCE) 56 57 /* 58 * Emit a CRC32C as the result of a ioat_copy_crc() or ioat_crc(). 59 */ 60 #define DMA_CRC_STORE 0x8 61 62 /* 63 * Compare the CRC32C of a ioat_copy_crc() or ioat_crc() against an expeceted 64 * value. It is invalid to specify both TEST and STORE. 65 */ 66 #define DMA_CRC_TEST 0x10 67 #define _DMA_CRC_TESTSTORE (DMA_CRC_STORE | DMA_CRC_TEST) 68 69 /* 70 * Use an inline comparison CRC32C or emit an inline CRC32C result. Invalid 71 * without one of STORE or TEST. 72 */ 73 #define DMA_CRC_INLINE 0x20 74 #define _DMA_CRC_FLAGS (DMA_CRC_STORE | DMA_CRC_TEST | DMA_CRC_INLINE) 75 76 /* 77 * Hardware revision number. Different hardware revisions support different 78 * features. For example, 3.2 cannot read from MMIO space, while 3.3 can. 79 */ 80 #define IOAT_VER_3_0 0x30 81 #define IOAT_VER_3_2 0x32 82 #define IOAT_VER_3_3 0x33 83 84 /* 85 * Hardware capabilities. Different hardware revisions support different 86 * features. It is often useful to detect specific features than try to infer 87 * them from hardware version. 88 * 89 * Different channels may support different features too; for example, 'PQ' may 90 * only be supported on the first two channels of some hardware. 91 */ 92 #define IOAT_DMACAP_PB (1 << 0) 93 #define IOAT_DMACAP_CRC (1 << 1) 94 #define IOAT_DMACAP_MARKER_SKIP (1 << 2) 95 #define IOAT_DMACAP_OLD_XOR (1 << 3) 96 #define IOAT_DMACAP_DCA (1 << 4) 97 #define IOAT_DMACAP_MOVECRC (1 << 5) 98 #define IOAT_DMACAP_BFILL (1 << 6) 99 #define IOAT_DMACAP_EXT_APIC (1 << 7) 100 #define IOAT_DMACAP_XOR (1 << 8) 101 #define IOAT_DMACAP_PQ (1 << 9) 102 #define IOAT_DMACAP_DMA_DIF (1 << 10) 103 #define IOAT_DMACAP_DWBES (1 << 13) 104 #define IOAT_DMACAP_RAID16SS (1 << 17) 105 #define IOAT_DMACAP_DMAMC (1 << 18) 106 #define IOAT_DMACAP_CTOS (1 << 19) 107 108 #define IOAT_DMACAP_STR \ 109 "\20\24Completion_Timeout_Support\23DMA_with_Multicasting_Support" \ 110 "\22RAID_Super_descriptors\16Descriptor_Write_Back_Error_Support" \ 111 "\13DMA_with_DIF\12PQ\11XOR\10Extended_APIC_ID\07Block_Fill\06Move_CRC" \ 112 "\05DCA\04Old_XOR\03Marker_Skipping\02CRC\01Page_Break" 113 114 typedef void *bus_dmaengine_t; 115 struct bus_dmadesc; 116 typedef void (*bus_dmaengine_callback_t)(void *arg, int error); 117 118 unsigned ioat_get_nchannels(void); 119 120 /* 121 * Called first to acquire a reference to the DMA channel 122 * 123 * Flags may be M_WAITOK or M_NOWAIT. 124 */ 125 bus_dmaengine_t ioat_get_dmaengine(uint32_t channel_index, int flags); 126 127 /* Release the DMA channel */ 128 void ioat_put_dmaengine(bus_dmaengine_t dmaengine); 129 130 /* Check the DMA engine's HW version */ 131 int ioat_get_hwversion(bus_dmaengine_t dmaengine); 132 size_t ioat_get_max_io_size(bus_dmaengine_t dmaengine); 133 uint32_t ioat_get_capabilities(bus_dmaengine_t dmaengine); 134 135 /* 136 * Set interrupt coalescing on a DMA channel. 137 * 138 * The argument is in microseconds. A zero value disables coalescing. Any 139 * other value delays interrupt generation for N microseconds to provide 140 * opportunity to coalesce multiple operations into a single interrupt. 141 * 142 * Returns an error status, or zero on success. 143 * 144 * - ERANGE if the given value exceeds the delay supported by the hardware. 145 * (All current hardware supports a maximum of 0x3fff microseconds delay.) 146 * - ENODEV if the hardware does not support interrupt coalescing. 147 */ 148 int ioat_set_interrupt_coalesce(bus_dmaengine_t dmaengine, uint16_t delay); 149 150 /* 151 * Return the maximum supported coalescing period, for use in 152 * ioat_set_interrupt_coalesce(). If the hardware does not support coalescing, 153 * returns zero. 154 */ 155 uint16_t ioat_get_max_coalesce_period(bus_dmaengine_t dmaengine); 156 157 /* 158 * Acquire must be called before issuing an operation to perform. Release is 159 * called after. Multiple operations can be issued within the context of one 160 * acquire and release 161 */ 162 void ioat_acquire(bus_dmaengine_t dmaengine); 163 void ioat_release(bus_dmaengine_t dmaengine); 164 165 /* 166 * Acquire_reserve can be called to ensure there is room for N descriptors. If 167 * it succeeds, the next N valid operations will successfully enqueue. 168 * 169 * It may fail with: 170 * - ENXIO if the channel is in an errored state, or the driver is being 171 * unloaded 172 * - EAGAIN if mflags included M_NOWAIT 173 * 174 * On failure, the caller does not hold the dmaengine. 175 */ 176 int ioat_acquire_reserve(bus_dmaengine_t dmaengine, unsigned n, int mflags) 177 __result_use_check; 178 179 /* 180 * Issue a blockfill operation. The 64-bit pattern 'fillpattern' is written to 181 * 'len' physically contiguous bytes at 'dst'. 182 * 183 * Only supported on devices with the BFILL capability. 184 */ 185 struct bus_dmadesc *ioat_blockfill(bus_dmaengine_t dmaengine, bus_addr_t dst, 186 uint64_t fillpattern, bus_size_t len, bus_dmaengine_callback_t callback_fn, 187 void *callback_arg, uint32_t flags); 188 189 /* Issues the copy data operation */ 190 struct bus_dmadesc *ioat_copy(bus_dmaengine_t dmaengine, bus_addr_t dst, 191 bus_addr_t src, bus_size_t len, bus_dmaengine_callback_t callback_fn, 192 void *callback_arg, uint32_t flags); 193 194 /* 195 * Issue a copy data operation, with constraints: 196 * - src1, src2, dst1, dst2 are all page-aligned addresses 197 * - The quantity to copy is exactly 2 pages; 198 * - src1 -> dst1, src2 -> dst2 199 * 200 * Why use this instead of normal _copy()? You can copy two non-contiguous 201 * pages (src, dst, or both) with one descriptor. 202 */ 203 struct bus_dmadesc *ioat_copy_8k_aligned(bus_dmaengine_t dmaengine, 204 bus_addr_t dst1, bus_addr_t dst2, bus_addr_t src1, bus_addr_t src2, 205 bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags); 206 207 /* 208 * Copy len bytes from dst to src, like ioat_copy(). 209 * 210 * Additionally, accumulate a CRC32C of the data. 211 * 212 * If initialseed is not NULL, the value it points to is used to seed the 213 * initial value of the CRC32C. 214 * 215 * If flags include DMA_CRC_STORE and not DMA_CRC_INLINE, crcptr is written 216 * with the 32-bit CRC32C result (in wire format). 217 * 218 * If flags include DMA_CRC_TEST and not DMA_CRC_INLINE, the computed CRC32C is 219 * compared with the 32-bit CRC32C pointed to by crcptr. If they do not match, 220 * a channel error is raised. 221 * 222 * If the DMA_CRC_INLINE flag is set, crcptr is ignored and the DMA engine uses 223 * the 4 bytes trailing the source data (TEST) or the destination data (STORE). 224 */ 225 struct bus_dmadesc *ioat_copy_crc(bus_dmaengine_t dmaengine, bus_addr_t dst, 226 bus_addr_t src, bus_size_t len, uint32_t *initialseed, bus_addr_t crcptr, 227 bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags); 228 229 /* 230 * ioat_crc() is nearly identical to ioat_copy_crc(), but does not actually 231 * move data around. 232 * 233 * Like ioat_copy_crc, ioat_crc computes a CRC32C over len bytes pointed to by 234 * src. The flags affect its operation in the same way, with one exception: 235 * 236 * If flags includes both DMA_CRC_STORE and DMA_CRC_INLINE, the computed CRC32C 237 * is written to the 4 bytes trailing the *source* data. 238 */ 239 struct bus_dmadesc *ioat_crc(bus_dmaengine_t dmaengine, bus_addr_t src, 240 bus_size_t len, uint32_t *initialseed, bus_addr_t crcptr, 241 bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags); 242 243 /* 244 * Issues a null operation. This issues the operation to the hardware, but the 245 * hardware doesn't do anything with it. 246 */ 247 struct bus_dmadesc *ioat_null(bus_dmaengine_t dmaengine, 248 bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags); 249 250 251 #endif /* __IOAT_H__ */ 252 253