xref: /minix3/minix/kernel/arch/i386/protect.c (revision 35b65c5af1d53112a49b08e87ddc00d7e24625e9)
1 /* This file contains code for initialization of protected mode, to initialize
2  * code and data segment descriptors, and to initialize global descriptors
3  * for local descriptors in the process table.
4  */
5 
6 #include <assert.h>
7 #include <string.h>
8 
9 #include <minix/cpufeature.h>
10 #include <sys/types.h>
11 #include "kernel/kernel.h"
12 
13 #include "arch_proto.h"
14 
15 #include <sys/exec.h>
16 #include <libexec.h>
17 
18 #define INT_GATE_TYPE	(INT_286_GATE | DESC_386_BIT)
19 #define TSS_TYPE	(AVL_286_TSS  | DESC_386_BIT)
20 
21 /* This is OK initially, when the 1:1 mapping is still there. */
22 char *video_mem = (char *) MULTIBOOT_VIDEO_BUFFER;
23 
24 /* Storage for gdt, idt and tss. */
25 struct segdesc_s gdt[GDT_SIZE] __aligned(DESC_SIZE);
26 struct gatedesc_s idt[IDT_SIZE] __aligned(DESC_SIZE);
27 struct tss_s tss[CONFIG_MAX_CPUS];
28 
29 u32_t k_percpu_stacks[CONFIG_MAX_CPUS];
30 
31 int prot_init_done = 0;
32 
33 phys_bytes vir2phys(void *vir)
34 {
35 	extern char _kern_vir_base, _kern_phys_base;	/* in kernel.lds */
36 	u32_t offset = (vir_bytes) &_kern_vir_base -
37 		(vir_bytes) &_kern_phys_base;
38 	return (phys_bytes)vir - offset;
39 }
40 
41 /*===========================================================================*
42  *				enable_iop				     *
43  *===========================================================================*/
44 void enable_iop(struct proc *pp)
45 {
46 /* Allow a user process to use I/O instructions.  Change the I/O Permission
47  * Level bits in the psw. These specify least-privileged Current Permission
48  * Level allowed to execute I/O instructions. Users and servers have CPL 3.
49  * You can't have less privilege than that. Kernel has CPL 0, tasks CPL 1.
50  */
51   pp->p_reg.psw |= 0x3000;
52 }
53 
54 
55 /*===========================================================================*
56  *				sdesc					     *
57  *===========================================================================*/
58  void sdesc(struct segdesc_s *segdp, phys_bytes base, vir_bytes size)
59 {
60 /* Fill in the size fields (base, limit and granularity) of a descriptor. */
61   segdp->base_low = base;
62   segdp->base_middle = base >> BASE_MIDDLE_SHIFT;
63   segdp->base_high = base >> BASE_HIGH_SHIFT;
64 
65   --size;			/* convert to a limit, 0 size means 4G */
66   if (size > BYTE_GRAN_MAX) {
67 	segdp->limit_low = size >> PAGE_GRAN_SHIFT;
68 	segdp->granularity = GRANULAR | (size >>
69 			     (PAGE_GRAN_SHIFT + GRANULARITY_SHIFT));
70   } else {
71 	segdp->limit_low = size;
72 	segdp->granularity = size >> GRANULARITY_SHIFT;
73   }
74   segdp->granularity |= DEFAULT;	/* means BIG for data seg */
75 }
76 
77 /*===========================================================================*
78  *				init_dataseg				     *
79  *===========================================================================*/
80 void init_param_dataseg(register struct segdesc_s *segdp,
81 	phys_bytes base, vir_bytes size, const int privilege)
82 {
83 	/* Build descriptor for a data segment. */
84 	sdesc(segdp, base, size);
85 	segdp->access = (privilege << DPL_SHIFT) | (PRESENT | SEGMENT |
86 		WRITEABLE | ACCESSED);
87 		/* EXECUTABLE = 0, EXPAND_DOWN = 0, ACCESSED = 0 */
88 }
89 
90 void init_dataseg(int index, const int privilege)
91 {
92 	init_param_dataseg(&gdt[index], 0, 0xFFFFFFFF, privilege);
93 }
94 
95 /*===========================================================================*
96  *				init_codeseg				     *
97  *===========================================================================*/
98 static void init_codeseg(int index, int privilege)
99 {
100 	/* Build descriptor for a code segment. */
101 	sdesc(&gdt[index], 0, 0xFFFFFFFF);
102 	gdt[index].access = (privilege << DPL_SHIFT)
103 	        | (PRESENT | SEGMENT | EXECUTABLE | READABLE);
104 		/* CONFORMING = 0, ACCESSED = 0 */
105 }
106 
107 static struct gate_table_s gate_table_pic[] = {
108 	{ hwint00, VECTOR( 0), INTR_PRIVILEGE },
109 	{ hwint01, VECTOR( 1), INTR_PRIVILEGE },
110 	{ hwint02, VECTOR( 2), INTR_PRIVILEGE },
111 	{ hwint03, VECTOR( 3), INTR_PRIVILEGE },
112 	{ hwint04, VECTOR( 4), INTR_PRIVILEGE },
113 	{ hwint05, VECTOR( 5), INTR_PRIVILEGE },
114 	{ hwint06, VECTOR( 6), INTR_PRIVILEGE },
115 	{ hwint07, VECTOR( 7), INTR_PRIVILEGE },
116 	{ hwint08, VECTOR( 8), INTR_PRIVILEGE },
117 	{ hwint09, VECTOR( 9), INTR_PRIVILEGE },
118 	{ hwint10, VECTOR(10), INTR_PRIVILEGE },
119 	{ hwint11, VECTOR(11), INTR_PRIVILEGE },
120 	{ hwint12, VECTOR(12), INTR_PRIVILEGE },
121 	{ hwint13, VECTOR(13), INTR_PRIVILEGE },
122 	{ hwint14, VECTOR(14), INTR_PRIVILEGE },
123 	{ hwint15, VECTOR(15), INTR_PRIVILEGE },
124 	{ NULL, 0, 0}
125 };
126 
127 static struct gate_table_s gate_table_exceptions[] = {
128 	{ divide_error, DIVIDE_VECTOR, INTR_PRIVILEGE },
129 	{ single_step_exception, DEBUG_VECTOR, INTR_PRIVILEGE },
130 	{ nmi, NMI_VECTOR, INTR_PRIVILEGE },
131 	{ breakpoint_exception, BREAKPOINT_VECTOR, USER_PRIVILEGE },
132 	{ overflow, OVERFLOW_VECTOR, USER_PRIVILEGE },
133 	{ bounds_check, BOUNDS_VECTOR, INTR_PRIVILEGE },
134 	{ inval_opcode, INVAL_OP_VECTOR, INTR_PRIVILEGE },
135 	{ copr_not_available, COPROC_NOT_VECTOR, INTR_PRIVILEGE },
136 	{ double_fault, DOUBLE_FAULT_VECTOR, INTR_PRIVILEGE },
137 	{ copr_seg_overrun, COPROC_SEG_VECTOR, INTR_PRIVILEGE },
138 	{ inval_tss, INVAL_TSS_VECTOR, INTR_PRIVILEGE },
139 	{ segment_not_present, SEG_NOT_VECTOR, INTR_PRIVILEGE },
140 	{ stack_exception, STACK_FAULT_VECTOR, INTR_PRIVILEGE },
141 	{ general_protection, PROTECTION_VECTOR, INTR_PRIVILEGE },
142 	{ page_fault, PAGE_FAULT_VECTOR, INTR_PRIVILEGE },
143 	{ copr_error, COPROC_ERR_VECTOR, INTR_PRIVILEGE },
144 	{ alignment_check, ALIGNMENT_CHECK_VECTOR, INTR_PRIVILEGE },
145 	{ machine_check, MACHINE_CHECK_VECTOR, INTR_PRIVILEGE },
146 	{ simd_exception, SIMD_EXCEPTION_VECTOR, INTR_PRIVILEGE },
147 	{ ipc_entry_softint_orig, IPC_VECTOR_ORIG, USER_PRIVILEGE },
148 	{ kernel_call_entry_orig, KERN_CALL_VECTOR_ORIG, USER_PRIVILEGE },
149 	{ ipc_entry_softint_um, IPC_VECTOR_UM, USER_PRIVILEGE },
150 	{ kernel_call_entry_um, KERN_CALL_VECTOR_UM, USER_PRIVILEGE },
151 	{ NULL, 0, 0}
152 };
153 
154 int tss_init(unsigned cpu, void * kernel_stack)
155 {
156 	struct tss_s * t = &tss[cpu];
157 	int index = TSS_INDEX(cpu);
158 	struct segdesc_s *tssgdt;
159 
160 	tssgdt = &gdt[index];
161 
162 	init_param_dataseg(tssgdt, (phys_bytes) t,
163 			sizeof(struct tss_s), INTR_PRIVILEGE);
164 	tssgdt->access = PRESENT | (INTR_PRIVILEGE << DPL_SHIFT) | TSS_TYPE;
165 
166 	/* Build TSS. */
167 	memset(t, 0, sizeof(*t));
168 	t->ds = t->es = t->fs = t->gs = t->ss0 = KERN_DS_SELECTOR;
169 	t->cs = KERN_CS_SELECTOR;
170 	t->iobase = sizeof(struct tss_s);	/* empty i/o permissions map */
171 
172 	/*
173 	 * make space for process pointer and cpu id and point to the first
174 	 * usable word
175 	 */
176 	k_percpu_stacks[cpu] = t->sp0 = ((unsigned) kernel_stack) - X86_STACK_TOP_RESERVED;
177 	/*
178 	 * set the cpu id at the top of the stack so we know on which cpu is
179 	 * this stack in use when we trap to kernel
180 	 */
181 	*((reg_t *)(t->sp0 + 1 * sizeof(reg_t))) = cpu;
182 
183 	/* Set up Intel SYSENTER support if available. */
184 	if(minix_feature_flags & MKF_I386_INTEL_SYSENTER) {
185 	  ia32_msr_write(INTEL_MSR_SYSENTER_CS, 0, KERN_CS_SELECTOR);
186   	  ia32_msr_write(INTEL_MSR_SYSENTER_ESP, 0, t->sp0);
187   	  ia32_msr_write(INTEL_MSR_SYSENTER_EIP, 0, (u32_t) ipc_entry_sysenter);
188   	}
189 
190 	/* Set up AMD SYSCALL support if available. */
191 	if(minix_feature_flags & MKF_I386_AMD_SYSCALL) {
192 		u32_t msr_lo, msr_hi;
193 
194 		/* set SYSCALL ENABLE bit in EFER MSR */
195 		ia32_msr_read(AMD_MSR_EFER, &msr_hi, &msr_lo);
196 		msr_lo |= AMD_EFER_SCE;
197 		ia32_msr_write(AMD_MSR_EFER, msr_hi, msr_lo);
198 
199 		/* set STAR register value */
200 #define set_star_cpu(forcpu) if(cpu == forcpu) {				\
201 		ia32_msr_write(AMD_MSR_STAR,					\
202 		  ((u32_t)USER_CS_SELECTOR << 16) | (u32_t)KERN_CS_SELECTOR,	\
203 		  (u32_t) ipc_entry_syscall_cpu ## forcpu); }
204 		set_star_cpu(0);
205 		set_star_cpu(1);
206 		set_star_cpu(2);
207 		set_star_cpu(3);
208 		set_star_cpu(4);
209 		set_star_cpu(5);
210 		set_star_cpu(6);
211 		set_star_cpu(7);
212 		assert(CONFIG_MAX_CPUS <= 8);
213   	}
214 
215 	return SEG_SELECTOR(index);
216 }
217 
218 phys_bytes init_segdesc(int gdt_index, void *base, int size)
219 {
220 	struct desctableptr_s *dtp = (struct desctableptr_s *) &gdt[gdt_index];
221 	dtp->limit = size - 1;
222 	dtp->base = (phys_bytes) base;
223 
224 	return (phys_bytes) dtp;
225 }
226 
227 void int_gate(struct gatedesc_s *tab,
228 	unsigned vec_nr, vir_bytes offset, unsigned dpl_type)
229 {
230 /* Build descriptor for an interrupt gate. */
231   register struct gatedesc_s *idp;
232 
233   idp = &tab[vec_nr];
234   idp->offset_low = offset;
235   idp->selector = KERN_CS_SELECTOR;
236   idp->p_dpl_type = dpl_type;
237   idp->offset_high = offset >> OFFSET_HIGH_SHIFT;
238 }
239 
240 void int_gate_idt(unsigned vec_nr, vir_bytes offset, unsigned dpl_type)
241 {
242 	int_gate(idt, vec_nr, offset, dpl_type);
243 }
244 
245 void idt_copy_vectors(struct gate_table_s * first)
246 {
247 	struct gate_table_s *gtp;
248 	for (gtp = first; gtp->gate; gtp++) {
249 		int_gate(idt, gtp->vec_nr, (vir_bytes) gtp->gate,
250 				PRESENT | INT_GATE_TYPE |
251 				(gtp->privilege << DPL_SHIFT));
252 	}
253 }
254 
255 void idt_copy_vectors_pic(void)
256 {
257 	idt_copy_vectors(gate_table_pic);
258 }
259 
260 void idt_init(void)
261 {
262 	idt_copy_vectors_pic();
263 	idt_copy_vectors(gate_table_exceptions);
264 }
265 
266 struct desctableptr_s gdt_desc, idt_desc;
267 
268 void idt_reload(void)
269 {
270 	x86_lidt(&idt_desc);
271 }
272 
273 multiboot_module_t *bootmod(int pnr)
274 {
275 	int i;
276 
277 	assert(pnr >= 0);
278 
279 	/* Search for desired process in boot process
280 	 * list. The first NR_TASKS ones do not correspond
281 	 * to a module, however, so we don't search those.
282 	 */
283 	for(i = NR_TASKS; i < NR_BOOT_PROCS; i++) {
284 		int p;
285 		p = i - NR_TASKS;
286 		if(image[i].proc_nr == pnr) {
287 			assert(p < MULTIBOOT_MAX_MODS);
288 			assert(p < kinfo.mbi.mi_mods_count);
289 			return &kinfo.module_list[p];
290 		}
291 	}
292 
293 	panic("boot module %d not found", pnr);
294 }
295 
296 int booting_cpu = 0;
297 
298 void prot_load_selectors(void)
299 {
300   /* this function is called by both prot_init by the BSP and
301    * the early AP booting code in mpx.S by secondary CPU's.
302    * everything is set up the same except for the TSS that is per-CPU.
303    */
304   x86_lgdt(&gdt_desc);	/* Load gdt */
305   idt_init();
306   idt_reload();
307   x86_lldt(LDT_SELECTOR); 	/* Load bogus ldt */
308   x86_ltr(TSS_SELECTOR(booting_cpu));
309 
310   x86_load_kerncs();
311   x86_load_ds(KERN_DS_SELECTOR);
312   x86_load_es(KERN_DS_SELECTOR);
313   x86_load_fs(KERN_DS_SELECTOR);
314   x86_load_gs(KERN_DS_SELECTOR);
315   x86_load_ss(KERN_DS_SELECTOR);
316 }
317 
318 /*===========================================================================*
319  *				prot_init				     *
320  *===========================================================================*/
321 void prot_init(void)
322 {
323   extern char k_boot_stktop;
324 
325   if(_cpufeature(_CPUF_I386_SYSENTER))
326 	minix_feature_flags |= MKF_I386_INTEL_SYSENTER;
327   if(_cpufeature(_CPUF_I386_SYSCALL))
328 	minix_feature_flags |= MKF_I386_AMD_SYSCALL;
329 
330   memset(gdt, 0, sizeof(gdt));
331   memset(idt, 0, sizeof(idt));
332 
333   /* Build GDT, IDT, IDT descriptors. */
334   gdt_desc.base = (u32_t) gdt;
335   gdt_desc.limit = sizeof(gdt)-1;
336   idt_desc.base = (u32_t) idt;
337   idt_desc.limit = sizeof(idt)-1;
338   tss_init(0, &k_boot_stktop);
339 
340   /* Build GDT */
341   init_param_dataseg(&gdt[LDT_INDEX],
342     (phys_bytes) 0, 0, INTR_PRIVILEGE); /* unusable LDT */
343   gdt[LDT_INDEX].access = PRESENT | LDT;
344   init_codeseg(KERN_CS_INDEX, INTR_PRIVILEGE);
345   init_dataseg(KERN_DS_INDEX, INTR_PRIVILEGE);
346   init_codeseg(USER_CS_INDEX, USER_PRIVILEGE);
347   init_dataseg(USER_DS_INDEX, USER_PRIVILEGE);
348 
349   /* Currently the multiboot segments are loaded; which is fine, but
350    * let's replace them with the ones from our own GDT so we test
351    * right away whether they work as expected.
352    */
353   prot_load_selectors();
354 
355   /* Set up a new post-relocate bootstrap pagetable so that
356    * we can map in VM, and we no longer rely on pre-relocated
357    * data.
358    */
359 
360   pg_clear();
361   pg_identity(&kinfo); /* Still need 1:1 for lapic and video mem and such. */
362   pg_mapkernel();
363   pg_load();
364 
365   prot_init_done = 1;
366 }
367 
368 static int alloc_for_vm = 0;
369 
370 void arch_post_init(void)
371 {
372   /* Let memory mapping code know what's going on at bootstrap time */
373   struct proc *vm;
374   vm = proc_addr(VM_PROC_NR);
375   get_cpulocal_var(ptproc) = vm;
376   pg_info(&vm->p_seg.p_cr3, &vm->p_seg.p_cr3_v);
377 }
378 
379 static int libexec_pg_alloc(struct exec_info *execi, vir_bytes vaddr, size_t len)
380 {
381         pg_map(PG_ALLOCATEME, vaddr, vaddr+len, &kinfo);
382   	pg_load();
383         memset((char *) vaddr, 0, len);
384 	alloc_for_vm += len;
385         return OK;
386 }
387 
388 void arch_boot_proc(struct boot_image *ip, struct proc *rp)
389 {
390 	multiboot_module_t *mod;
391 	struct ps_strings *psp;
392 	char *sp;
393 
394 	if(rp->p_nr < 0) return;
395 
396 	mod = bootmod(rp->p_nr);
397 
398 	/* Important special case: we put VM in the bootstrap pagetable
399 	 * so it can run.
400 	 */
401 
402 	if(rp->p_nr == VM_PROC_NR) {
403 		struct exec_info execi;
404 
405 		memset(&execi, 0, sizeof(execi));
406 
407 		/* exec parameters */
408 		execi.stack_high = kinfo.user_sp;
409 		execi.stack_size = 64 * 1024;	/* not too crazy as it must be preallocated */
410 		execi.proc_e = ip->endpoint;
411 		execi.hdr = (char *) mod->mod_start; /* phys mem direct */
412 		execi.filesize = execi.hdr_len = mod->mod_end - mod->mod_start;
413 		strlcpy(execi.progname, ip->proc_name, sizeof(execi.progname));
414 		execi.frame_len = 0;
415 
416 		/* callbacks for use in the kernel */
417 		execi.copymem = libexec_copy_memcpy;
418 		execi.clearmem = libexec_clear_memset;
419 		execi.allocmem_prealloc_junk = libexec_pg_alloc;
420 		execi.allocmem_prealloc_cleared = libexec_pg_alloc;
421 		execi.allocmem_ondemand = libexec_pg_alloc;
422 		execi.clearproc = NULL;
423 
424 		/* parse VM ELF binary and alloc/map it into bootstrap pagetable */
425 		if(libexec_load_elf(&execi) != OK)
426 			panic("VM loading failed");
427 
428 		/* Setup a ps_strings struct on the stack, pointing to the
429 		 * following argv, envp. */
430 		sp = (char *)execi.stack_high;
431 		sp -= sizeof(struct ps_strings);
432 		psp = (struct ps_strings *) sp;
433 
434 		/* Take the stack pointer down three words to give startup code
435 		 * something to use as "argc", "argv" and "envp".
436 		 */
437 		sp -= (sizeof(void *) + sizeof(void *) + sizeof(int));
438 
439 		// linear address space, so it is available.
440 		psp->ps_argvstr = (char **)(sp + sizeof(int));
441 		psp->ps_nargvstr = 0;
442 		psp->ps_envstr = psp->ps_argvstr + sizeof(void *);
443 		psp->ps_nenvstr = 0;
444 
445 		arch_proc_init(rp, execi.pc, (vir_bytes)sp,
446 			execi.stack_high - sizeof(struct ps_strings),
447 			ip->proc_name);
448 
449 		/* Free VM blob that was just copied into existence. */
450 		add_memmap(&kinfo, mod->mod_start, mod->mod_end-mod->mod_start);
451 		mod->mod_end = mod->mod_start = 0;
452 
453 		/* Remember them */
454 		kinfo.vm_allocated_bytes = alloc_for_vm;
455 	}
456 }
457