xref: /openbmc/linux/arch/x86/kernel/fpu/init.c (revision 174cd4b1)
1 /*
2  * x86 FPU boot time init code:
3  */
4 #include <asm/fpu/internal.h>
5 #include <asm/tlbflush.h>
6 #include <asm/setup.h>
7 #include <asm/cmdline.h>
8 
9 #include <linux/sched.h>
10 #include <linux/init.h>
11 
12 /*
13  * Initialize the registers found in all CPUs, CR0 and CR4:
14  */
15 static void fpu__init_cpu_generic(void)
16 {
17 	unsigned long cr0;
18 	unsigned long cr4_mask = 0;
19 
20 	if (boot_cpu_has(X86_FEATURE_FXSR))
21 		cr4_mask |= X86_CR4_OSFXSR;
22 	if (boot_cpu_has(X86_FEATURE_XMM))
23 		cr4_mask |= X86_CR4_OSXMMEXCPT;
24 	if (cr4_mask)
25 		cr4_set_bits(cr4_mask);
26 
27 	cr0 = read_cr0();
28 	cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */
29 	if (!boot_cpu_has(X86_FEATURE_FPU))
30 		cr0 |= X86_CR0_EM;
31 	write_cr0(cr0);
32 
33 	/* Flush out any pending x87 state: */
34 #ifdef CONFIG_MATH_EMULATION
35 	if (!boot_cpu_has(X86_FEATURE_FPU))
36 		fpstate_init_soft(&current->thread.fpu.state.soft);
37 	else
38 #endif
39 		asm volatile ("fninit");
40 }
41 
42 /*
43  * Enable all supported FPU features. Called when a CPU is brought online:
44  */
45 void fpu__init_cpu(void)
46 {
47 	fpu__init_cpu_generic();
48 	fpu__init_cpu_xstate();
49 }
50 
51 static bool fpu__probe_without_cpuid(void)
52 {
53 	unsigned long cr0;
54 	u16 fsw, fcw;
55 
56 	fsw = fcw = 0xffff;
57 
58 	cr0 = read_cr0();
59 	cr0 &= ~(X86_CR0_TS | X86_CR0_EM);
60 	write_cr0(cr0);
61 
62 	asm volatile("fninit ; fnstsw %0 ; fnstcw %1" : "+m" (fsw), "+m" (fcw));
63 
64 	pr_info("x86/fpu: Probing for FPU: FSW=0x%04hx FCW=0x%04hx\n", fsw, fcw);
65 
66 	return fsw == 0 && (fcw & 0x103f) == 0x003f;
67 }
68 
69 static void fpu__init_system_early_generic(struct cpuinfo_x86 *c)
70 {
71 	if (!boot_cpu_has(X86_FEATURE_CPUID) &&
72 	    !test_bit(X86_FEATURE_FPU, (unsigned long *)cpu_caps_cleared)) {
73 		if (fpu__probe_without_cpuid())
74 			setup_force_cpu_cap(X86_FEATURE_FPU);
75 		else
76 			setup_clear_cpu_cap(X86_FEATURE_FPU);
77 	}
78 
79 #ifndef CONFIG_MATH_EMULATION
80 	if (!test_cpu_cap(&boot_cpu_data, X86_FEATURE_FPU)) {
81 		pr_emerg("x86/fpu: Giving up, no FPU found and no math emulation present\n");
82 		for (;;)
83 			asm volatile("hlt");
84 	}
85 #endif
86 }
87 
88 /*
89  * Boot time FPU feature detection code:
90  */
91 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
92 
93 static void __init fpu__init_system_mxcsr(void)
94 {
95 	unsigned int mask = 0;
96 
97 	if (boot_cpu_has(X86_FEATURE_FXSR)) {
98 		/* Static because GCC does not get 16-byte stack alignment right: */
99 		static struct fxregs_state fxregs __initdata;
100 
101 		asm volatile("fxsave %0" : "+m" (fxregs));
102 
103 		mask = fxregs.mxcsr_mask;
104 
105 		/*
106 		 * If zero then use the default features mask,
107 		 * which has all features set, except the
108 		 * denormals-are-zero feature bit:
109 		 */
110 		if (mask == 0)
111 			mask = 0x0000ffbf;
112 	}
113 	mxcsr_feature_mask &= mask;
114 }
115 
116 /*
117  * Once per bootup FPU initialization sequences that will run on most x86 CPUs:
118  */
119 static void __init fpu__init_system_generic(void)
120 {
121 	/*
122 	 * Set up the legacy init FPU context. (xstate init might overwrite this
123 	 * with a more modern format, if the CPU supports it.)
124 	 */
125 	fpstate_init(&init_fpstate);
126 
127 	fpu__init_system_mxcsr();
128 }
129 
130 /*
131  * Size of the FPU context state. All tasks in the system use the
132  * same context size, regardless of what portion they use.
133  * This is inherent to the XSAVE architecture which puts all state
134  * components into a single, continuous memory block:
135  */
136 unsigned int fpu_kernel_xstate_size;
137 EXPORT_SYMBOL_GPL(fpu_kernel_xstate_size);
138 
139 /* Get alignment of the TYPE. */
140 #define TYPE_ALIGN(TYPE) offsetof(struct { char x; TYPE test; }, test)
141 
142 /*
143  * Enforce that 'MEMBER' is the last field of 'TYPE'.
144  *
145  * Align the computed size with alignment of the TYPE,
146  * because that's how C aligns structs.
147  */
148 #define CHECK_MEMBER_AT_END_OF(TYPE, MEMBER) \
149 	BUILD_BUG_ON(sizeof(TYPE) != ALIGN(offsetofend(TYPE, MEMBER), \
150 					   TYPE_ALIGN(TYPE)))
151 
152 /*
153  * We append the 'struct fpu' to the task_struct:
154  */
155 static void __init fpu__init_task_struct_size(void)
156 {
157 	int task_size = sizeof(struct task_struct);
158 
159 	/*
160 	 * Subtract off the static size of the register state.
161 	 * It potentially has a bunch of padding.
162 	 */
163 	task_size -= sizeof(((struct task_struct *)0)->thread.fpu.state);
164 
165 	/*
166 	 * Add back the dynamically-calculated register state
167 	 * size.
168 	 */
169 	task_size += fpu_kernel_xstate_size;
170 
171 	/*
172 	 * We dynamically size 'struct fpu', so we require that
173 	 * it be at the end of 'thread_struct' and that
174 	 * 'thread_struct' be at the end of 'task_struct'.  If
175 	 * you hit a compile error here, check the structure to
176 	 * see if something got added to the end.
177 	 */
178 	CHECK_MEMBER_AT_END_OF(struct fpu, state);
179 	CHECK_MEMBER_AT_END_OF(struct thread_struct, fpu);
180 	CHECK_MEMBER_AT_END_OF(struct task_struct, thread);
181 
182 	arch_task_struct_size = task_size;
183 }
184 
185 /*
186  * Set up the user and kernel xstate sizes based on the legacy FPU context size.
187  *
188  * We set this up first, and later it will be overwritten by
189  * fpu__init_system_xstate() if the CPU knows about xstates.
190  */
191 static void __init fpu__init_system_xstate_size_legacy(void)
192 {
193 	static int on_boot_cpu __initdata = 1;
194 
195 	WARN_ON_FPU(!on_boot_cpu);
196 	on_boot_cpu = 0;
197 
198 	/*
199 	 * Note that xstate sizes might be overwritten later during
200 	 * fpu__init_system_xstate().
201 	 */
202 
203 	if (!boot_cpu_has(X86_FEATURE_FPU)) {
204 		/*
205 		 * Disable xsave as we do not support it if i387
206 		 * emulation is enabled.
207 		 */
208 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
209 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
210 		fpu_kernel_xstate_size = sizeof(struct swregs_state);
211 	} else {
212 		if (boot_cpu_has(X86_FEATURE_FXSR))
213 			fpu_kernel_xstate_size =
214 				sizeof(struct fxregs_state);
215 		else
216 			fpu_kernel_xstate_size =
217 				sizeof(struct fregs_state);
218 	}
219 
220 	fpu_user_xstate_size = fpu_kernel_xstate_size;
221 }
222 
223 /*
224  * Find supported xfeatures based on cpu features and command-line input.
225  * This must be called after fpu__init_parse_early_param() is called and
226  * xfeatures_mask is enumerated.
227  */
228 u64 __init fpu__get_supported_xfeatures_mask(void)
229 {
230 	return XCNTXT_MASK;
231 }
232 
233 /* Legacy code to initialize eager fpu mode. */
234 static void __init fpu__init_system_ctx_switch(void)
235 {
236 	static bool on_boot_cpu __initdata = 1;
237 
238 	WARN_ON_FPU(!on_boot_cpu);
239 	on_boot_cpu = 0;
240 
241 	WARN_ON_FPU(current->thread.fpu.fpstate_active);
242 }
243 
244 /*
245  * We parse fpu parameters early because fpu__init_system() is executed
246  * before parse_early_param().
247  */
248 static void __init fpu__init_parse_early_param(void)
249 {
250 	if (cmdline_find_option_bool(boot_command_line, "no387"))
251 		setup_clear_cpu_cap(X86_FEATURE_FPU);
252 
253 	if (cmdline_find_option_bool(boot_command_line, "nofxsr")) {
254 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
255 		setup_clear_cpu_cap(X86_FEATURE_FXSR_OPT);
256 		setup_clear_cpu_cap(X86_FEATURE_XMM);
257 	}
258 
259 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
260 		fpu__xstate_clear_all_cpu_caps();
261 
262 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
263 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
264 
265 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
266 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
267 }
268 
269 /*
270  * Called on the boot CPU once per system bootup, to set up the initial
271  * FPU state that is later cloned into all processes:
272  */
273 void __init fpu__init_system(struct cpuinfo_x86 *c)
274 {
275 	fpu__init_parse_early_param();
276 	fpu__init_system_early_generic(c);
277 
278 	/*
279 	 * The FPU has to be operational for some of the
280 	 * later FPU init activities:
281 	 */
282 	fpu__init_cpu();
283 
284 	fpu__init_system_generic();
285 	fpu__init_system_xstate_size_legacy();
286 	fpu__init_system_xstate();
287 	fpu__init_task_struct_size();
288 
289 	fpu__init_system_ctx_switch();
290 }
291