2 * FP/SIMD context switching and fault handling
4 * Copyright (C) 2012 ARM Ltd.
5 * Author: Catalin Marinas <catalin.marinas@arm.com>
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.
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.
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/>.
20 #include <linux/cpu.h>
21 #include <linux/cpu_pm.h>
22 #include <linux/kernel.h>
23 #include <linux/init.h>
24 #include <linux/sched.h>
25 #include <linux/signal.h>
26 #include <linux/hardirq.h>
28 #include <asm/fpsimd.h>
29 #include <asm/cpufeature.h>
30 #include <asm/cputype.h>
32 #define FPEXC_IOF (1 << 0)
33 #define FPEXC_DZF (1 << 1)
34 #define FPEXC_OFF (1 << 2)
35 #define FPEXC_UFF (1 << 3)
36 #define FPEXC_IXF (1 << 4)
37 #define FPEXC_IDF (1 << 7)
40 * In order to reduce the number of times the FPSIMD state is needlessly saved
41 * and restored, we need to keep track of two things:
42 * (a) for each task, we need to remember which CPU was the last one to have
43 * the task's FPSIMD state loaded into its FPSIMD registers;
44 * (b) for each CPU, we need to remember which task's userland FPSIMD state has
45 * been loaded into its FPSIMD registers most recently, or whether it has
46 * been used to perform kernel mode NEON in the meantime.
48 * For (a), we add a 'cpu' field to struct fpsimd_state, which gets updated to
49 * the id of the current CPU every time the state is loaded onto a CPU. For (b),
50 * we add the per-cpu variable 'fpsimd_last_state' (below), which contains the
51 * address of the userland FPSIMD state of the task that was loaded onto the CPU
52 * the most recently, or NULL if kernel mode NEON has been performed after that.
54 * With this in place, we no longer have to restore the next FPSIMD state right
55 * when switching between tasks. Instead, we can defer this check to userland
56 * resume, at which time we verify whether the CPU's fpsimd_last_state and the
57 * task's fpsimd_state.cpu are still mutually in sync. If this is the case, we
58 * can omit the FPSIMD restore.
60 * As an optimization, we use the thread_info flag TIF_FOREIGN_FPSTATE to
61 * indicate whether or not the userland FPSIMD state of the current task is
62 * present in the registers. The flag is set unless the FPSIMD registers of this
63 * CPU currently contain the most recent userland FPSIMD state of the current
66 * For a certain task, the sequence may look something like this:
67 * - the task gets scheduled in; if both the task's fpsimd_state.cpu field
68 * contains the id of the current CPU, and the CPU's fpsimd_last_state per-cpu
69 * variable points to the task's fpsimd_state, the TIF_FOREIGN_FPSTATE flag is
70 * cleared, otherwise it is set;
72 * - the task returns to userland; if TIF_FOREIGN_FPSTATE is set, the task's
73 * userland FPSIMD state is copied from memory to the registers, the task's
74 * fpsimd_state.cpu field is set to the id of the current CPU, the current
75 * CPU's fpsimd_last_state pointer is set to this task's fpsimd_state and the
76 * TIF_FOREIGN_FPSTATE flag is cleared;
78 * - the task executes an ordinary syscall; upon return to userland, the
79 * TIF_FOREIGN_FPSTATE flag will still be cleared, so no FPSIMD state is
82 * - the task executes a syscall which executes some NEON instructions; this is
83 * preceded by a call to kernel_neon_begin(), which copies the task's FPSIMD
84 * register contents to memory, clears the fpsimd_last_state per-cpu variable
85 * and sets the TIF_FOREIGN_FPSTATE flag;
87 * - the task gets preempted after kernel_neon_end() is called; as we have not
88 * returned from the 2nd syscall yet, TIF_FOREIGN_FPSTATE is still set so
89 * whatever is in the FPSIMD registers is not saved to memory, but discarded.
91 static DEFINE_PER_CPU(struct fpsimd_state *, fpsimd_last_state);
94 * Trapped FP/ASIMD access.
96 void do_fpsimd_acc(unsigned int esr, struct pt_regs *regs)
98 /* TODO: implement lazy context saving/restoring */
103 * Raise a SIGFPE for the current process.
105 void do_fpsimd_exc(unsigned int esr, struct pt_regs *regs)
108 unsigned int si_code = 0;
111 si_code = FPE_FLTINV;
112 else if (esr & FPEXC_DZF)
113 si_code = FPE_FLTDIV;
114 else if (esr & FPEXC_OFF)
115 si_code = FPE_FLTOVF;
116 else if (esr & FPEXC_UFF)
117 si_code = FPE_FLTUND;
118 else if (esr & FPEXC_IXF)
119 si_code = FPE_FLTRES;
121 memset(&info, 0, sizeof(info));
122 info.si_signo = SIGFPE;
123 info.si_code = si_code;
124 info.si_addr = (void __user *)instruction_pointer(regs);
126 send_sig_info(SIGFPE, &info, current);
129 void fpsimd_thread_switch(struct task_struct *next)
132 * Save the current FPSIMD state to memory, but only if whatever is in
133 * the registers is in fact the most recent userland FPSIMD state of
136 if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE))
137 fpsimd_save_state(¤t->thread.fpsimd_state);
141 * If we are switching to a task whose most recent userland
142 * FPSIMD state is already in the registers of *this* cpu,
143 * we can skip loading the state from memory. Otherwise, set
144 * the TIF_FOREIGN_FPSTATE flag so the state will be loaded
145 * upon the next return to userland.
147 struct fpsimd_state *st = &next->thread.fpsimd_state;
149 if (__this_cpu_read(fpsimd_last_state) == st
150 && st->cpu == smp_processor_id())
151 clear_ti_thread_flag(task_thread_info(next),
152 TIF_FOREIGN_FPSTATE);
154 set_ti_thread_flag(task_thread_info(next),
155 TIF_FOREIGN_FPSTATE);
159 void fpsimd_flush_thread(void)
162 memset(¤t->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
163 fpsimd_flush_task_state(current);
164 set_thread_flag(TIF_FOREIGN_FPSTATE);
169 * Save the userland FPSIMD state of 'current' to memory, but only if the state
170 * currently held in the registers does in fact belong to 'current'
172 void fpsimd_preserve_current_state(void)
175 if (!test_thread_flag(TIF_FOREIGN_FPSTATE))
176 fpsimd_save_state(¤t->thread.fpsimd_state);
181 * Load the userland FPSIMD state of 'current' from memory, but only if the
182 * FPSIMD state already held in the registers is /not/ the most recent FPSIMD
185 void fpsimd_restore_current_state(void)
188 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) {
189 struct fpsimd_state *st = ¤t->thread.fpsimd_state;
191 fpsimd_load_state(st);
192 this_cpu_write(fpsimd_last_state, st);
193 st->cpu = smp_processor_id();
199 * Load an updated userland FPSIMD state for 'current' from memory and set the
200 * flag that indicates that the FPSIMD register contents are the most recent
201 * FPSIMD state of 'current'
203 void fpsimd_update_current_state(struct fpsimd_state *state)
206 fpsimd_load_state(state);
207 if (test_and_clear_thread_flag(TIF_FOREIGN_FPSTATE)) {
208 struct fpsimd_state *st = ¤t->thread.fpsimd_state;
210 this_cpu_write(fpsimd_last_state, st);
211 st->cpu = smp_processor_id();
217 * Invalidate live CPU copies of task t's FPSIMD state
219 void fpsimd_flush_task_state(struct task_struct *t)
221 t->thread.fpsimd_state.cpu = NR_CPUS;
224 #ifdef CONFIG_KERNEL_MODE_NEON
226 static DEFINE_PER_CPU(struct fpsimd_partial_state, hardirq_fpsimdstate);
227 static DEFINE_PER_CPU(struct fpsimd_partial_state, softirq_fpsimdstate);
230 * Kernel-side NEON support functions
232 void kernel_neon_begin_partial(u32 num_regs)
234 if (in_interrupt()) {
235 struct fpsimd_partial_state *s = this_cpu_ptr(
236 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate);
238 BUG_ON(num_regs > 32);
239 fpsimd_save_partial_state(s, roundup(num_regs, 2));
242 * Save the userland FPSIMD state if we have one and if we
243 * haven't done so already. Clear fpsimd_last_state to indicate
244 * that there is no longer userland FPSIMD state in the
249 !test_and_set_thread_flag(TIF_FOREIGN_FPSTATE))
250 fpsimd_save_state(¤t->thread.fpsimd_state);
251 this_cpu_write(fpsimd_last_state, NULL);
254 EXPORT_SYMBOL(kernel_neon_begin_partial);
256 void kernel_neon_end(void)
258 if (in_interrupt()) {
259 struct fpsimd_partial_state *s = this_cpu_ptr(
260 in_irq() ? &hardirq_fpsimdstate : &softirq_fpsimdstate);
261 fpsimd_load_partial_state(s);
266 EXPORT_SYMBOL(kernel_neon_end);
268 #endif /* CONFIG_KERNEL_MODE_NEON */
271 static int fpsimd_cpu_pm_notifier(struct notifier_block *self,
272 unsigned long cmd, void *v)
276 if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE))
277 fpsimd_save_state(¤t->thread.fpsimd_state);
278 this_cpu_write(fpsimd_last_state, NULL);
282 set_thread_flag(TIF_FOREIGN_FPSTATE);
284 case CPU_PM_ENTER_FAILED:
291 static struct notifier_block fpsimd_cpu_pm_notifier_block = {
292 .notifier_call = fpsimd_cpu_pm_notifier,
295 static void __init fpsimd_pm_init(void)
297 cpu_pm_register_notifier(&fpsimd_cpu_pm_notifier_block);
301 static inline void fpsimd_pm_init(void) { }
302 #endif /* CONFIG_CPU_PM */
304 #ifdef CONFIG_HOTPLUG_CPU
305 static int fpsimd_cpu_dead(unsigned int cpu)
307 per_cpu(fpsimd_last_state, cpu) = NULL;
311 static inline void fpsimd_hotplug_init(void)
313 cpuhp_setup_state_nocalls(CPUHP_ARM64_FPSIMD_DEAD, "arm64/fpsimd:dead",
314 NULL, fpsimd_cpu_dead);
318 static inline void fpsimd_hotplug_init(void) { }
322 * FP/SIMD support code initialisation.
324 static int __init fpsimd_init(void)
326 if (elf_hwcap & HWCAP_FP) {
328 fpsimd_hotplug_init();
330 pr_notice("Floating-point is not implemented\n");
333 if (!(elf_hwcap & HWCAP_ASIMD))
334 pr_notice("Advanced SIMD is not implemented\n");
338 late_initcall(fpsimd_init);