| 1 | /* Shared HTM header. Work around false transactional execution facility |
|---|
| 2 | intrinsics. |
|---|
| 3 | |
|---|
| 4 | Copyright (C) 2016-2022 Free Software Foundation, Inc. |
|---|
| 5 | This file is part of the GNU C Library. |
|---|
| 6 | |
|---|
| 7 | The GNU C Library is free software; you can redistribute it and/or |
|---|
| 8 | modify it under the terms of the GNU Lesser General Public |
|---|
| 9 | License as published by the Free Software Foundation; either |
|---|
| 10 | version 2.1 of the License, or (at your option) any later version. |
|---|
| 11 | |
|---|
| 12 | The GNU C Library is distributed in the hope that it will be useful, |
|---|
| 13 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
|---|
| 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|---|
| 15 | Lesser General Public License for more details. |
|---|
| 16 | |
|---|
| 17 | You should have received a copy of the GNU Lesser General Public |
|---|
| 18 | License along with the GNU C Library; if not, see |
|---|
| 19 | <https://www.gnu.org/licenses/>. */ |
|---|
| 20 | |
|---|
| 21 | #ifndef _HTM_H |
|---|
| 22 | #define _HTM_H 1 |
|---|
| 23 | |
|---|
| 24 | #include <htmintrin.h> |
|---|
| 25 | |
|---|
| 26 | #ifdef __s390x__ |
|---|
| 27 | # define TX_FPRS_BYTES 64 |
|---|
| 28 | # define TX_SAVE_FPRS \ |
|---|
| 29 | " std %%f8, 0(%[R_FPRS])\n\t" \ |
|---|
| 30 | " std %%f9, 8(%[R_FPRS])\n\t" \ |
|---|
| 31 | " std %%f10, 16(%[R_FPRS])\n\t" \ |
|---|
| 32 | " std %%f11, 24(%[R_FPRS])\n\t" \ |
|---|
| 33 | " std %%f12, 32(%[R_FPRS])\n\t" \ |
|---|
| 34 | " std %%f13, 40(%[R_FPRS])\n\t" \ |
|---|
| 35 | " std %%f14, 48(%[R_FPRS])\n\t" \ |
|---|
| 36 | " std %%f15, 56(%[R_FPRS])\n\t" |
|---|
| 37 | |
|---|
| 38 | # define TX_RESTORE_FPRS \ |
|---|
| 39 | " ld %%f8, 0(%[R_FPRS])\n\t" \ |
|---|
| 40 | " ld %%f9, 8(%[R_FPRS])\n\t" \ |
|---|
| 41 | " ld %%f10, 16(%[R_FPRS])\n\t" \ |
|---|
| 42 | " ld %%f11, 24(%[R_FPRS])\n\t" \ |
|---|
| 43 | " ld %%f12, 32(%[R_FPRS])\n\t" \ |
|---|
| 44 | " ld %%f13, 40(%[R_FPRS])\n\t" \ |
|---|
| 45 | " ld %%f14, 48(%[R_FPRS])\n\t" \ |
|---|
| 46 | " ld %%f15, 56(%[R_FPRS])\n\t" |
|---|
| 47 | |
|---|
| 48 | #else |
|---|
| 49 | |
|---|
| 50 | # define TX_FPRS_BYTES 16 |
|---|
| 51 | # define TX_SAVE_FPRS \ |
|---|
| 52 | " std %%f4, 0(%[R_FPRS])\n\t" \ |
|---|
| 53 | " std %%f6, 8(%[R_FPRS])\n\t" |
|---|
| 54 | |
|---|
| 55 | # define TX_RESTORE_FPRS \ |
|---|
| 56 | " ld %%f4, 0(%[R_FPRS])\n\t" \ |
|---|
| 57 | " ld %%f6, 8(%[R_FPRS])\n\t" |
|---|
| 58 | |
|---|
| 59 | #endif /* ! __s390x__ */ |
|---|
| 60 | |
|---|
| 61 | /* Use own inline assembly instead of __builtin_tbegin, as tbegin |
|---|
| 62 | has to filter program interruptions which can't be done with the builtin. |
|---|
| 63 | Now the fprs have to be saved / restored here, too. |
|---|
| 64 | The fpc is also not saved / restored with the builtin. |
|---|
| 65 | The used inline assembly does not clobber the volatile fprs / vrs! |
|---|
| 66 | Clobbering the latter ones would force the compiler to save / restore |
|---|
| 67 | the call saved fprs as those overlap with the vrs, but they only need to be |
|---|
| 68 | restored if the transaction fails but not if the transaction is successfully |
|---|
| 69 | started. Thus the user of the tbegin macros in this header file has to |
|---|
| 70 | compile the file / function with -msoft-float. It prevents gcc from using |
|---|
| 71 | fprs / vrs. */ |
|---|
| 72 | #define __libc_tbegin(tdb) __libc_tbegin_base(tdb,,,) |
|---|
| 73 | |
|---|
| 74 | #define __libc_tbegin_retry_output_regs , [R_TX_CNT] "+&d" (__tx_cnt) |
|---|
| 75 | #define __libc_tbegin_retry_input_regs(retry_cnt) , [R_RETRY] "d" (retry_cnt) |
|---|
| 76 | #define __libc_tbegin_retry_abort_path_insn \ |
|---|
| 77 | /* If tbegin returned _HTM_TBEGIN_TRANSIENT, retry immediately so \ |
|---|
| 78 | that max tbegin_cnt transactions are tried. Otherwise return and \ |
|---|
| 79 | let the caller of this macro do the fallback path. */ \ |
|---|
| 80 | " jnh 1f\n\t" /* cc 1/3: jump to fallback path. */ \ |
|---|
| 81 | /* tbegin returned _HTM_TBEGIN_TRANSIENT: retry with transaction. */ \ |
|---|
| 82 | " crje %[R_TX_CNT], %[R_RETRY], 1f\n\t" /* Reached max retries? */ \ |
|---|
| 83 | " ahi %[R_TX_CNT], 1\n\t" \ |
|---|
| 84 | " ppa %[R_TX_CNT], 0, 1\n\t" /* Transaction-Abort Assist. */ \ |
|---|
| 85 | " j 2b\n\t" /* Loop to tbegin. */ |
|---|
| 86 | |
|---|
| 87 | /* Same as __libc_tbegin except if tbegin aborts with _HTM_TBEGIN_TRANSIENT. |
|---|
| 88 | Then this macros restores the fpc, fprs and automatically retries up to |
|---|
| 89 | retry_cnt tbegins. Further saving of the state is omitted as it is already |
|---|
| 90 | saved. This macro calls tbegin at most as retry_cnt + 1 times. */ |
|---|
| 91 | #define __libc_tbegin_retry(tdb, retry_cnt) \ |
|---|
| 92 | ({ int __ret; \ |
|---|
| 93 | int __tx_cnt = 0; \ |
|---|
| 94 | __ret = __libc_tbegin_base(tdb, \ |
|---|
| 95 | __libc_tbegin_retry_abort_path_insn, \ |
|---|
| 96 | __libc_tbegin_retry_output_regs, \ |
|---|
| 97 | __libc_tbegin_retry_input_regs(retry_cnt)); \ |
|---|
| 98 | __ret; \ |
|---|
| 99 | }) |
|---|
| 100 | |
|---|
| 101 | #define __libc_tbegin_base(tdb, abort_path_insn, output_regs, input_regs) \ |
|---|
| 102 | ({ int __ret; \ |
|---|
| 103 | int __fpc; \ |
|---|
| 104 | char __fprs[TX_FPRS_BYTES]; \ |
|---|
| 105 | __asm__ __volatile__ (".machine push\n\t" \ |
|---|
| 106 | ".machinemode \"zarch_nohighgprs\"\n\t" \ |
|---|
| 107 | ".machine \"all\"\n\t" \ |
|---|
| 108 | /* Save state at the outermost transaction. \ |
|---|
| 109 | As extracting nesting depth is expensive \ |
|---|
| 110 | on at least zEC12, save fprs at inner \ |
|---|
| 111 | transactions, too. \ |
|---|
| 112 | The fpc and fprs are saved here as they \ |
|---|
| 113 | are not saved by tbegin. There exist no \ |
|---|
| 114 | call-saved vrs, thus they are not saved \ |
|---|
| 115 | here. */ \ |
|---|
| 116 | " efpc %[R_FPC]\n\t" \ |
|---|
| 117 | TX_SAVE_FPRS \ |
|---|
| 118 | /* Begin transaction: save all gprs, allow \ |
|---|
| 119 | ar modification and fp operations. Some \ |
|---|
| 120 | program-interruptions (e.g. a null \ |
|---|
| 121 | pointer access) are filtered and the \ |
|---|
| 122 | transaction will abort. In this case \ |
|---|
| 123 | the normal lock path will execute it \ |
|---|
| 124 | again and result in a core dump wich does \ |
|---|
| 125 | now show at tbegin but the real executed \ |
|---|
| 126 | instruction. \ |
|---|
| 127 | However it is not guaranteed that this \ |
|---|
| 128 | retry operate on the same data and thus \ |
|---|
| 129 | may not end in an program-interruption. \ |
|---|
| 130 | Note: This could also be used to probe \ |
|---|
| 131 | memory for being accessible! */ \ |
|---|
| 132 | "2: tbegin 0, 0xFF0E\n\t" \ |
|---|
| 133 | /* Branch away in abort case (this is the \ |
|---|
| 134 | prefered sequence. See PoP in chapter 5 \ |
|---|
| 135 | Transactional-Execution Facility \ |
|---|
| 136 | Operation). */ \ |
|---|
| 137 | " jnz 0f\n\t" \ |
|---|
| 138 | /* Transaction has successfully started. */ \ |
|---|
| 139 | " lhi %[R_RET], 0\n\t" \ |
|---|
| 140 | " j 1f\n\t" \ |
|---|
| 141 | /* Transaction has aborted. Now we are at \ |
|---|
| 142 | the outermost transaction. Restore fprs \ |
|---|
| 143 | and fpc. */ \ |
|---|
| 144 | "0: ipm %[R_RET]\n\t" \ |
|---|
| 145 | " srl %[R_RET], 28\n\t" \ |
|---|
| 146 | " sfpc %[R_FPC]\n\t" \ |
|---|
| 147 | TX_RESTORE_FPRS \ |
|---|
| 148 | abort_path_insn \ |
|---|
| 149 | "1:\n\t" \ |
|---|
| 150 | ".machine pop\n" \ |
|---|
| 151 | : [R_RET] "=&d" (__ret), \ |
|---|
| 152 | [R_FPC] "=&d" (__fpc) \ |
|---|
| 153 | output_regs \ |
|---|
| 154 | : [R_FPRS] "a" (__fprs) \ |
|---|
| 155 | input_regs \ |
|---|
| 156 | : "cc", "memory"); \ |
|---|
| 157 | __ret; \ |
|---|
| 158 | }) |
|---|
| 159 | |
|---|
| 160 | /* These builtins are usable in context of glibc lock elision code without any |
|---|
| 161 | changes. Use them. */ |
|---|
| 162 | #define __libc_tend() \ |
|---|
| 163 | ({ __asm__ __volatile__ (".machine push\n\t" \ |
|---|
| 164 | ".machinemode \"zarch_nohighgprs\"\n\t" \ |
|---|
| 165 | ".machine \"all\"\n\t"); \ |
|---|
| 166 | int __ret = __builtin_tend (); \ |
|---|
| 167 | __asm__ __volatile__ (".machine pop"); \ |
|---|
| 168 | __ret; \ |
|---|
| 169 | }) |
|---|
| 170 | |
|---|
| 171 | #define __libc_tabort(abortcode) \ |
|---|
| 172 | __asm__ __volatile__ (".machine push\n\t" \ |
|---|
| 173 | ".machinemode \"zarch_nohighgprs\"\n\t" \ |
|---|
| 174 | ".machine \"all\"\n\t"); \ |
|---|
| 175 | __builtin_tabort (abortcode); \ |
|---|
| 176 | __asm__ __volatile__ (".machine pop") |
|---|
| 177 | |
|---|
| 178 | #define __libc_tx_nesting_depth() \ |
|---|
| 179 | ({ __asm__ __volatile__ (".machine push\n\t" \ |
|---|
| 180 | ".machinemode \"zarch_nohighgprs\"\n\t" \ |
|---|
| 181 | ".machine \"all\"\n\t"); \ |
|---|
| 182 | int __ret = __builtin_tx_nesting_depth (); \ |
|---|
| 183 | __asm__ __volatile__ (".machine pop"); \ |
|---|
| 184 | __ret; \ |
|---|
| 185 | }) |
|---|
| 186 | |
|---|
| 187 | #endif |
|---|
| 188 | |
|---|
