xref: /openbmc/linux/arch/x86/kernel/process_32.c (revision e4781421e883340b796da5a724bda7226817990b)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *
4  *  Pentium III FXSR, SSE support
5  *	Gareth Hughes <gareth@valinux.com>, May 2000
6  */
7 
8 /*
9  * This file handles the architecture-dependent parts of process handling..
10  */
11 
12 #include <linux/cpu.h>
13 #include <linux/errno.h>
14 #include <linux/sched.h>
15 #include <linux/fs.h>
16 #include <linux/kernel.h>
17 #include <linux/mm.h>
18 #include <linux/elfcore.h>
19 #include <linux/smp.h>
20 #include <linux/stddef.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/user.h>
24 #include <linux/interrupt.h>
25 #include <linux/delay.h>
26 #include <linux/reboot.h>
27 #include <linux/mc146818rtc.h>
28 #include <linux/export.h>
29 #include <linux/kallsyms.h>
30 #include <linux/ptrace.h>
31 #include <linux/personality.h>
32 #include <linux/percpu.h>
33 #include <linux/prctl.h>
34 #include <linux/ftrace.h>
35 #include <linux/uaccess.h>
36 #include <linux/io.h>
37 #include <linux/kdebug.h>
38 
39 #include <asm/pgtable.h>
40 #include <asm/ldt.h>
41 #include <asm/processor.h>
42 #include <asm/fpu/internal.h>
43 #include <asm/desc.h>
44 #ifdef CONFIG_MATH_EMULATION
45 #include <asm/math_emu.h>
46 #endif
47 
48 #include <linux/err.h>
49 
50 #include <asm/tlbflush.h>
51 #include <asm/cpu.h>
52 #include <asm/syscalls.h>
53 #include <asm/debugreg.h>
54 #include <asm/switch_to.h>
55 #include <asm/vm86.h>
56 #include <asm/intel_rdt.h>
57 
58 void __show_regs(struct pt_regs *regs, int all)
59 {
60 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
61 	unsigned long d0, d1, d2, d3, d6, d7;
62 	unsigned long sp;
63 	unsigned short ss, gs;
64 
65 	if (user_mode(regs)) {
66 		sp = regs->sp;
67 		ss = regs->ss & 0xffff;
68 		gs = get_user_gs(regs);
69 	} else {
70 		sp = kernel_stack_pointer(regs);
71 		savesegment(ss, ss);
72 		savesegment(gs, gs);
73 	}
74 
75 	printk(KERN_DEFAULT "EIP: %pS\n", (void *)regs->ip);
76 	printk(KERN_DEFAULT "EFLAGS: %08lx CPU: %d\n", regs->flags,
77 		smp_processor_id());
78 
79 	printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
80 		regs->ax, regs->bx, regs->cx, regs->dx);
81 	printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
82 		regs->si, regs->di, regs->bp, sp);
83 	printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
84 	       (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
85 
86 	if (!all)
87 		return;
88 
89 	cr0 = read_cr0();
90 	cr2 = read_cr2();
91 	cr3 = read_cr3();
92 	cr4 = __read_cr4();
93 	printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
94 			cr0, cr2, cr3, cr4);
95 
96 	get_debugreg(d0, 0);
97 	get_debugreg(d1, 1);
98 	get_debugreg(d2, 2);
99 	get_debugreg(d3, 3);
100 	get_debugreg(d6, 6);
101 	get_debugreg(d7, 7);
102 
103 	/* Only print out debug registers if they are in their non-default state. */
104 	if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
105 	    (d6 == DR6_RESERVED) && (d7 == 0x400))
106 		return;
107 
108 	printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
109 			d0, d1, d2, d3);
110 	printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
111 			d6, d7);
112 }
113 
114 void release_thread(struct task_struct *dead_task)
115 {
116 	BUG_ON(dead_task->mm);
117 	release_vm86_irqs(dead_task);
118 }
119 
120 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
121 	unsigned long arg, struct task_struct *p, unsigned long tls)
122 {
123 	struct pt_regs *childregs = task_pt_regs(p);
124 	struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
125 	struct inactive_task_frame *frame = &fork_frame->frame;
126 	struct task_struct *tsk;
127 	int err;
128 
129 	frame->bp = 0;
130 	frame->ret_addr = (unsigned long) ret_from_fork;
131 	p->thread.sp = (unsigned long) fork_frame;
132 	p->thread.sp0 = (unsigned long) (childregs+1);
133 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
134 
135 	if (unlikely(p->flags & PF_KTHREAD)) {
136 		/* kernel thread */
137 		memset(childregs, 0, sizeof(struct pt_regs));
138 		frame->bx = sp;		/* function */
139 		frame->di = arg;
140 		p->thread.io_bitmap_ptr = NULL;
141 		return 0;
142 	}
143 	frame->bx = 0;
144 	*childregs = *current_pt_regs();
145 	childregs->ax = 0;
146 	if (sp)
147 		childregs->sp = sp;
148 
149 	task_user_gs(p) = get_user_gs(current_pt_regs());
150 
151 	p->thread.io_bitmap_ptr = NULL;
152 	tsk = current;
153 	err = -ENOMEM;
154 
155 	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
156 		p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
157 						IO_BITMAP_BYTES, GFP_KERNEL);
158 		if (!p->thread.io_bitmap_ptr) {
159 			p->thread.io_bitmap_max = 0;
160 			return -ENOMEM;
161 		}
162 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
163 	}
164 
165 	err = 0;
166 
167 	/*
168 	 * Set a new TLS for the child thread?
169 	 */
170 	if (clone_flags & CLONE_SETTLS)
171 		err = do_set_thread_area(p, -1,
172 			(struct user_desc __user *)tls, 0);
173 
174 	if (err && p->thread.io_bitmap_ptr) {
175 		kfree(p->thread.io_bitmap_ptr);
176 		p->thread.io_bitmap_max = 0;
177 	}
178 	return err;
179 }
180 
181 void
182 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
183 {
184 	set_user_gs(regs, 0);
185 	regs->fs		= 0;
186 	regs->ds		= __USER_DS;
187 	regs->es		= __USER_DS;
188 	regs->ss		= __USER_DS;
189 	regs->cs		= __USER_CS;
190 	regs->ip		= new_ip;
191 	regs->sp		= new_sp;
192 	regs->flags		= X86_EFLAGS_IF;
193 	force_iret();
194 }
195 EXPORT_SYMBOL_GPL(start_thread);
196 
197 
198 /*
199  *	switch_to(x,y) should switch tasks from x to y.
200  *
201  * We fsave/fwait so that an exception goes off at the right time
202  * (as a call from the fsave or fwait in effect) rather than to
203  * the wrong process. Lazy FP saving no longer makes any sense
204  * with modern CPU's, and this simplifies a lot of things (SMP
205  * and UP become the same).
206  *
207  * NOTE! We used to use the x86 hardware context switching. The
208  * reason for not using it any more becomes apparent when you
209  * try to recover gracefully from saved state that is no longer
210  * valid (stale segment register values in particular). With the
211  * hardware task-switch, there is no way to fix up bad state in
212  * a reasonable manner.
213  *
214  * The fact that Intel documents the hardware task-switching to
215  * be slow is a fairly red herring - this code is not noticeably
216  * faster. However, there _is_ some room for improvement here,
217  * so the performance issues may eventually be a valid point.
218  * More important, however, is the fact that this allows us much
219  * more flexibility.
220  *
221  * The return value (in %ax) will be the "prev" task after
222  * the task-switch, and shows up in ret_from_fork in entry.S,
223  * for example.
224  */
225 __visible __notrace_funcgraph struct task_struct *
226 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
227 {
228 	struct thread_struct *prev = &prev_p->thread,
229 			     *next = &next_p->thread;
230 	struct fpu *prev_fpu = &prev->fpu;
231 	struct fpu *next_fpu = &next->fpu;
232 	int cpu = smp_processor_id();
233 	struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
234 
235 	/* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
236 
237 	switch_fpu_prepare(prev_fpu, cpu);
238 
239 	/*
240 	 * Save away %gs. No need to save %fs, as it was saved on the
241 	 * stack on entry.  No need to save %es and %ds, as those are
242 	 * always kernel segments while inside the kernel.  Doing this
243 	 * before setting the new TLS descriptors avoids the situation
244 	 * where we temporarily have non-reloadable segments in %fs
245 	 * and %gs.  This could be an issue if the NMI handler ever
246 	 * used %fs or %gs (it does not today), or if the kernel is
247 	 * running inside of a hypervisor layer.
248 	 */
249 	lazy_save_gs(prev->gs);
250 
251 	/*
252 	 * Load the per-thread Thread-Local Storage descriptor.
253 	 */
254 	load_TLS(next, cpu);
255 
256 	/*
257 	 * Restore IOPL if needed.  In normal use, the flags restore
258 	 * in the switch assembly will handle this.  But if the kernel
259 	 * is running virtualized at a non-zero CPL, the popf will
260 	 * not restore flags, so it must be done in a separate step.
261 	 */
262 	if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
263 		set_iopl_mask(next->iopl);
264 
265 	/*
266 	 * Now maybe handle debug registers and/or IO bitmaps
267 	 */
268 	if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
269 		     task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
270 		__switch_to_xtra(prev_p, next_p, tss);
271 
272 	/*
273 	 * Leave lazy mode, flushing any hypercalls made here.
274 	 * This must be done before restoring TLS segments so
275 	 * the GDT and LDT are properly updated, and must be
276 	 * done before fpu__restore(), so the TS bit is up
277 	 * to date.
278 	 */
279 	arch_end_context_switch(next_p);
280 
281 	/*
282 	 * Reload esp0 and cpu_current_top_of_stack.  This changes
283 	 * current_thread_info().
284 	 */
285 	load_sp0(tss, next);
286 	this_cpu_write(cpu_current_top_of_stack,
287 		       (unsigned long)task_stack_page(next_p) +
288 		       THREAD_SIZE);
289 
290 	/*
291 	 * Restore %gs if needed (which is common)
292 	 */
293 	if (prev->gs | next->gs)
294 		lazy_load_gs(next->gs);
295 
296 	switch_fpu_finish(next_fpu, cpu);
297 
298 	this_cpu_write(current_task, next_p);
299 
300 	/* Load the Intel cache allocation PQR MSR. */
301 	intel_rdt_sched_in();
302 
303 	return prev_p;
304 }
305