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 */ 89static DEFINE_PER_CPU(struct fpsimd_state *, fpsimd_last_state); 90 91/* 92 * Trapped FP/ASIMD access. 93 */ 94void 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 */ 103void 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 127void 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 157void fpsimd_flush_thread(void) 158{ 159 memset(¤t->thread.fpsimd_state, 0, sizeof(struct fpsimd_state)); 160 fpsimd_flush_task_state(current); 161 set_thread_flag(TIF_FOREIGN_FPSTATE); 162} 163 164/* 165 * Save the userland FPSIMD state of 'current' to memory, but only if the state 166 * currently held in the registers does in fact belong to 'current' 167 */ 168void fpsimd_preserve_current_state(void) 169{ 170 preempt_disable(); 171 if (!test_thread_flag(TIF_FOREIGN_FPSTATE)) 172 fpsimd_save_state(¤t->thread.fpsimd_state); 173 preempt_enable(); 174} 175 176/* 177 * Load the userland FPSIMD state of 'current' from memory, but only if the 178 * FPSIMD state already held in the registers is /not/ the most recent FPSIMD 179 * state of 'current' 180 */ 181void fpsimd_restore_current_state(void) 182{ 183 preempt_disable(); 184 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) { 185 struct fpsimd_state *st = ¤t->thread.fpsimd_state; 186 187 fpsimd_load_state(st); 188 this_cpu_write(fpsimd_last_state, st); 189 st->cpu = smp_processor_id(); 190 } 191 preempt_enable(); 192} 193 194/* 195 * Load an updated userland FPSIMD state for 'current' from memory and set the 196 * flag that indicates that the FPSIMD register contents are the most recent 197 * FPSIMD state of 'current' 198 */ 199void fpsimd_update_current_state(struct fpsimd_state *state) 200{ 201 preempt_disable(); 202 fpsimd_load_state(state); 203 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) { 204 struct fpsimd_state *st = ¤t->thread.fpsimd_state; 205 206 this_cpu_write(fpsimd_last_state, st); 207 st->cpu = smp_processor_id(); 208 } 209 preempt_enable(); 210} 211 212/* 213 * Invalidate live CPU copies of task t's FPSIMD state 214 */ 215void fpsimd_flush_task_state(struct task_struct *t) 216{ 217 t->thread.fpsimd_state.cpu = NR_CPUS; 218} 219 220#ifdef CONFIG_KERNEL_MODE_NEON 221 222static DEFINE_PER_CPU(struct fpsimd_partial_state, hardirq_fpsimdstate); 223static DEFINE_PER_CPU(struct fpsimd_partial_state, softirq_fpsimdstate); 224 225/* 226 * Kernel-side NEON support functions 227 */ 228void kernel_neon_begin_partial(u32 num_regs) 229{ 230 if (in_interrupt()) { 231 struct fpsimd_partial_state *s = this_cpu_ptr( 232 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate); 233 234 BUG_ON(num_regs > 32); 235 fpsimd_save_partial_state(s, roundup(num_regs, 2)); 236 } else { 237 /* 238 * Save the userland FPSIMD state if we have one and if we 239 * haven't done so already. Clear fpsimd_last_state to indicate 240 * that there is no longer userland FPSIMD state in the 241 * registers. 242 */ 243 preempt_disable(); 244 if (current->mm && 245 !test_and_set_thread_flag(TIF_FOREIGN_FPSTATE)) 246 fpsimd_save_state(¤t->thread.fpsimd_state); 247 this_cpu_write(fpsimd_last_state, NULL); 248 } 249} 250EXPORT_SYMBOL(kernel_neon_begin_partial); 251 252void kernel_neon_end(void) 253{ 254 if (in_interrupt()) { 255 struct fpsimd_partial_state *s = this_cpu_ptr( 256 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate); 257 fpsimd_load_partial_state(s); 258 } else { 259 preempt_enable(); 260 } 261} 262EXPORT_SYMBOL(kernel_neon_end); 263 264#endif /* CONFIG_KERNEL_MODE_NEON */ 265 266#ifdef CONFIG_CPU_PM 267static int fpsimd_cpu_pm_notifier(struct notifier_block *self, 268 unsigned long cmd, void *v) 269{ 270 switch (cmd) { 271 case CPU_PM_ENTER: 272 if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE)) 273 fpsimd_save_state(¤t->thread.fpsimd_state); 274 this_cpu_write(fpsimd_last_state, NULL); 275 break; 276 case CPU_PM_EXIT: 277 if (current->mm) 278 set_thread_flag(TIF_FOREIGN_FPSTATE); 279 break; 280 case CPU_PM_ENTER_FAILED: 281 default: 282 return NOTIFY_DONE; 283 } 284 return NOTIFY_OK; 285} 286 287static struct notifier_block fpsimd_cpu_pm_notifier_block = { 288 .notifier_call = fpsimd_cpu_pm_notifier, 289}; 290 291static void fpsimd_pm_init(void) 292{ 293 cpu_pm_register_notifier(&fpsimd_cpu_pm_notifier_block); 294} 295 296#else 297static inline void fpsimd_pm_init(void) { } 298#endif /* CONFIG_CPU_PM */ 299 300/* 301 * FP/SIMD support code initialisation. 302 */ 303static int __init fpsimd_init(void) 304{ 305 u64 pfr = read_cpuid(ID_AA64PFR0_EL1); 306 307 if (pfr & (0xf << 16)) { 308 pr_notice("Floating-point is not implemented\n"); 309 return 0; 310 } 311 elf_hwcap |= HWCAP_FP; 312 313 if (pfr & (0xf << 20)) 314 pr_notice("Advanced SIMD is not implemented\n"); 315 else 316 elf_hwcap |= HWCAP_ASIMD; 317 318 fpsimd_pm_init(); 319 320 return 0; 321} 322late_initcall(fpsimd_init); 323