1 /* 2 * FP/SIMD context switching and fault handling 3 * 4 * Copyright (C) 2012 ARM Ltd. 5 * Author: Catalin Marinas <catalin.marinas@arm.com> 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License version 2 as 9 * published by the Free Software Foundation. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program. If not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include <linux/cpu_pm.h> 21 #include <linux/kernel.h> 22 #include <linux/init.h> 23 #include <linux/sched.h> 24 #include <linux/signal.h> 25 #include <linux/hardirq.h> 26 27 #include <asm/fpsimd.h> 28 #include <asm/cputype.h> 29 30 #define FPEXC_IOF (1 << 0) 31 #define FPEXC_DZF (1 << 1) 32 #define FPEXC_OFF (1 << 2) 33 #define FPEXC_UFF (1 << 3) 34 #define FPEXC_IXF (1 << 4) 35 #define FPEXC_IDF (1 << 7) 36 37 /* 38 * In order to reduce the number of times the FPSIMD state is needlessly saved 39 * and restored, we need to keep track of two things: 40 * (a) for each task, we need to remember which CPU was the last one to have 41 * the task's FPSIMD state loaded into its FPSIMD registers; 42 * (b) for each CPU, we need to remember which task's userland FPSIMD state has 43 * been loaded into its FPSIMD registers most recently, or whether it has 44 * been used to perform kernel mode NEON in the meantime. 45 * 46 * For (a), we add a 'cpu' field to struct fpsimd_state, which gets updated to 47 * the id of the current CPU everytime the state is loaded onto a CPU. For (b), 48 * we add the per-cpu variable 'fpsimd_last_state' (below), which contains the 49 * address of the userland FPSIMD state of the task that was loaded onto the CPU 50 * the most recently, or NULL if kernel mode NEON has been performed after that. 51 * 52 * With this in place, we no longer have to restore the next FPSIMD state right 53 * when switching between tasks. Instead, we can defer this check to userland 54 * resume, at which time we verify whether the CPU's fpsimd_last_state and the 55 * task's fpsimd_state.cpu are still mutually in sync. If this is the case, we 56 * can omit the FPSIMD restore. 57 * 58 * As an optimization, we use the thread_info flag TIF_FOREIGN_FPSTATE to 59 * indicate whether or not the userland FPSIMD state of the current task is 60 * present in the registers. The flag is set unless the FPSIMD registers of this 61 * CPU currently contain the most recent userland FPSIMD state of the current 62 * task. 63 * 64 * For a certain task, the sequence may look something like this: 65 * - the task gets scheduled in; if both the task's fpsimd_state.cpu field 66 * contains the id of the current CPU, and the CPU's fpsimd_last_state per-cpu 67 * variable points to the task's fpsimd_state, the TIF_FOREIGN_FPSTATE flag is 68 * cleared, otherwise it is set; 69 * 70 * - the task returns to userland; if TIF_FOREIGN_FPSTATE is set, the task's 71 * userland FPSIMD state is copied from memory to the registers, the task's 72 * fpsimd_state.cpu field is set to the id of the current CPU, the current 73 * CPU's fpsimd_last_state pointer is set to this task's fpsimd_state and the 74 * TIF_FOREIGN_FPSTATE flag is cleared; 75 * 76 * - the task executes an ordinary syscall; upon return to userland, the 77 * TIF_FOREIGN_FPSTATE flag will still be cleared, so no FPSIMD state is 78 * restored; 79 * 80 * - the task executes a syscall which executes some NEON instructions; this is 81 * preceded by a call to kernel_neon_begin(), which copies the task's FPSIMD 82 * register contents to memory, clears the fpsimd_last_state per-cpu variable 83 * and sets the TIF_FOREIGN_FPSTATE flag; 84 * 85 * - the task gets preempted after kernel_neon_end() is called; as we have not 86 * returned from the 2nd syscall yet, TIF_FOREIGN_FPSTATE is still set so 87 * whatever is in the FPSIMD registers is not saved to memory, but discarded. 88 */ 89 static DEFINE_PER_CPU(struct fpsimd_state *, fpsimd_last_state); 90 91 /* 92 * Trapped FP/ASIMD access. 93 */ 94 void do_fpsimd_acc(unsigned int esr, struct pt_regs *regs) 95 { 96 /* TODO: implement lazy context saving/restoring */ 97 WARN_ON(1); 98 } 99 100 /* 101 * Raise a SIGFPE for the current process. 102 */ 103 void do_fpsimd_exc(unsigned int esr, struct pt_regs *regs) 104 { 105 siginfo_t info; 106 unsigned int si_code = 0; 107 108 if (esr & FPEXC_IOF) 109 si_code = FPE_FLTINV; 110 else if (esr & FPEXC_DZF) 111 si_code = FPE_FLTDIV; 112 else if (esr & FPEXC_OFF) 113 si_code = FPE_FLTOVF; 114 else if (esr & FPEXC_UFF) 115 si_code = FPE_FLTUND; 116 else if (esr & FPEXC_IXF) 117 si_code = FPE_FLTRES; 118 119 memset(&info, 0, sizeof(info)); 120 info.si_signo = SIGFPE; 121 info.si_code = si_code; 122 info.si_addr = (void __user *)instruction_pointer(regs); 123 124 send_sig_info(SIGFPE, &info, current); 125 } 126 127 void fpsimd_thread_switch(struct task_struct *next) 128 { 129 /* 130 * Save the current FPSIMD state to memory, but only if whatever is in 131 * the registers is in fact the most recent userland FPSIMD state of 132 * 'current'. 133 */ 134 if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE)) 135 fpsimd_save_state(¤t->thread.fpsimd_state); 136 137 if (next->mm) { 138 /* 139 * If we are switching to a task whose most recent userland 140 * FPSIMD state is already in the registers of *this* cpu, 141 * we can skip loading the state from memory. Otherwise, set 142 * the TIF_FOREIGN_FPSTATE flag so the state will be loaded 143 * upon the next return to userland. 144 */ 145 struct fpsimd_state *st = &next->thread.fpsimd_state; 146 147 if (__this_cpu_read(fpsimd_last_state) == st 148 && st->cpu == smp_processor_id()) 149 clear_ti_thread_flag(task_thread_info(next), 150 TIF_FOREIGN_FPSTATE); 151 else 152 set_ti_thread_flag(task_thread_info(next), 153 TIF_FOREIGN_FPSTATE); 154 } 155 } 156 157 void fpsimd_flush_thread(void) 158 { 159 memset(¤t->thread.fpsimd_state, 0, sizeof(struct fpsimd_state)); 160 set_thread_flag(TIF_FOREIGN_FPSTATE); 161 } 162 163 /* 164 * Save the userland FPSIMD state of 'current' to memory, but only if the state 165 * currently held in the registers does in fact belong to 'current' 166 */ 167 void fpsimd_preserve_current_state(void) 168 { 169 preempt_disable(); 170 if (!test_thread_flag(TIF_FOREIGN_FPSTATE)) 171 fpsimd_save_state(¤t->thread.fpsimd_state); 172 preempt_enable(); 173 } 174 175 /* 176 * Load the userland FPSIMD state of 'current' from memory, but only if the 177 * FPSIMD state already held in the registers is /not/ the most recent FPSIMD 178 * state of 'current' 179 */ 180 void fpsimd_restore_current_state(void) 181 { 182 preempt_disable(); 183 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) { 184 struct fpsimd_state *st = ¤t->thread.fpsimd_state; 185 186 fpsimd_load_state(st); 187 this_cpu_write(fpsimd_last_state, st); 188 st->cpu = smp_processor_id(); 189 } 190 preempt_enable(); 191 } 192 193 /* 194 * Load an updated userland FPSIMD state for 'current' from memory and set the 195 * flag that indicates that the FPSIMD register contents are the most recent 196 * FPSIMD state of 'current' 197 */ 198 void fpsimd_update_current_state(struct fpsimd_state *state) 199 { 200 preempt_disable(); 201 fpsimd_load_state(state); 202 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) { 203 struct fpsimd_state *st = ¤t->thread.fpsimd_state; 204 205 this_cpu_write(fpsimd_last_state, st); 206 st->cpu = smp_processor_id(); 207 } 208 preempt_enable(); 209 } 210 211 /* 212 * Invalidate live CPU copies of task t's FPSIMD state 213 */ 214 void fpsimd_flush_task_state(struct task_struct *t) 215 { 216 t->thread.fpsimd_state.cpu = NR_CPUS; 217 } 218 219 #ifdef CONFIG_KERNEL_MODE_NEON 220 221 static DEFINE_PER_CPU(struct fpsimd_partial_state, hardirq_fpsimdstate); 222 static DEFINE_PER_CPU(struct fpsimd_partial_state, softirq_fpsimdstate); 223 224 /* 225 * Kernel-side NEON support functions 226 */ 227 void kernel_neon_begin_partial(u32 num_regs) 228 { 229 if (in_interrupt()) { 230 struct fpsimd_partial_state *s = this_cpu_ptr( 231 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate); 232 233 BUG_ON(num_regs > 32); 234 fpsimd_save_partial_state(s, roundup(num_regs, 2)); 235 } else { 236 /* 237 * Save the userland FPSIMD state if we have one and if we 238 * haven't done so already. Clear fpsimd_last_state to indicate 239 * that there is no longer userland FPSIMD state in the 240 * registers. 241 */ 242 preempt_disable(); 243 if (current->mm && 244 !test_and_set_thread_flag(TIF_FOREIGN_FPSTATE)) 245 fpsimd_save_state(¤t->thread.fpsimd_state); 246 this_cpu_write(fpsimd_last_state, NULL); 247 } 248 } 249 EXPORT_SYMBOL(kernel_neon_begin_partial); 250 251 void kernel_neon_end(void) 252 { 253 if (in_interrupt()) { 254 struct fpsimd_partial_state *s = this_cpu_ptr( 255 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate); 256 fpsimd_load_partial_state(s); 257 } else { 258 preempt_enable(); 259 } 260 } 261 EXPORT_SYMBOL(kernel_neon_end); 262 263 #endif /* CONFIG_KERNEL_MODE_NEON */ 264 265 #ifdef CONFIG_CPU_PM 266 static int fpsimd_cpu_pm_notifier(struct notifier_block *self, 267 unsigned long cmd, void *v) 268 { 269 switch (cmd) { 270 case CPU_PM_ENTER: 271 if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE)) 272 fpsimd_save_state(¤t->thread.fpsimd_state); 273 break; 274 case CPU_PM_EXIT: 275 if (current->mm) 276 set_thread_flag(TIF_FOREIGN_FPSTATE); 277 break; 278 case CPU_PM_ENTER_FAILED: 279 default: 280 return NOTIFY_DONE; 281 } 282 return NOTIFY_OK; 283 } 284 285 static struct notifier_block fpsimd_cpu_pm_notifier_block = { 286 .notifier_call = fpsimd_cpu_pm_notifier, 287 }; 288 289 static void fpsimd_pm_init(void) 290 { 291 cpu_pm_register_notifier(&fpsimd_cpu_pm_notifier_block); 292 } 293 294 #else 295 static inline void fpsimd_pm_init(void) { } 296 #endif /* CONFIG_CPU_PM */ 297 298 /* 299 * FP/SIMD support code initialisation. 300 */ 301 static int __init fpsimd_init(void) 302 { 303 u64 pfr = read_cpuid(ID_AA64PFR0_EL1); 304 305 if (pfr & (0xf << 16)) { 306 pr_notice("Floating-point is not implemented\n"); 307 return 0; 308 } 309 elf_hwcap |= HWCAP_FP; 310 311 if (pfr & (0xf << 20)) 312 pr_notice("Advanced SIMD is not implemented\n"); 313 else 314 elf_hwcap |= HWCAP_ASIMD; 315 316 fpsimd_pm_init(); 317 318 return 0; 319 } 320 late_initcall(fpsimd_init); 321