1 /*
2  * Copyright (c) 2008-2015 Travis Geiselbrecht
3  *
4  * Use of this source code is governed by a MIT-style
5  * license that can be found in the LICENSE file or at
6  * https://opensource.org/licenses/MIT
7  */
8 #include <lk/debug.h>
9 #include <lk/trace.h>
10 #include <stdlib.h>
11 #include <lk/err.h>
12 #include <lk/trace.h>
13 #include <stdio.h>
14 #include <lk/reg.h>
15 #include <arch.h>
16 #include <arch/atomic.h>
17 #include <arch/ops.h>
18 #include <arch/mmu.h>
19 #include <arch/arm.h>
20 #include <arch/arm/mmu.h>
21 #include <arch/mp.h>
22 #include <kernel/spinlock.h>
23 #include <kernel/thread.h>
24 #include <lk/main.h>
25 #include <lk/init.h>
26 #include <platform.h>
27 #include <target.h>
28 #include <kernel/thread.h>
29 #if WITH_KERNEL_VM
30 #include <kernel/vm.h>
31 #endif
32 
33 #define LOCAL_TRACE 0
34 
35 #if WITH_DEV_TIMER_ARM_CORTEX_A9
36 #include <dev/timer/arm_cortex_a9.h>
37 #endif
38 #if WITH_DEV_INTERRUPT_ARM_GIC
39 #include <dev/interrupt/arm_gic.h>
40 #endif
41 #if WITH_DEV_CACHE_PL310
42 #include <dev/cache/pl310.h>
43 #endif
44 
45 /* initial and abort stacks */
46 uint8_t abort_stack[ARCH_DEFAULT_STACK_SIZE *SMP_MAX_CPUS] __CPU_ALIGN;
47 
48 static void arm_basic_setup(void);
49 static void spinlock_test(void);
50 static void spinlock_test_secondary(void);
51 
52 #if WITH_SMP
53 /* smp boot lock */
54 spin_lock_t arm_boot_cpu_lock = 1;
55 volatile int secondaries_to_init = 0;
56 #endif
57 
arch_early_init(void)58 void arch_early_init(void) {
59     /* turn off the cache */
60     arch_disable_cache(UCACHE);
61 #if WITH_DEV_CACHE_PL310
62     pl310_set_enable(false);
63 #endif
64 
65     arm_basic_setup();
66 
67 #if WITH_SMP && ARM_CPU_CORTEX_A9
68     /* enable snoop control */
69     addr_t scu_base = arm_read_cbar();
70     *REG32(scu_base) |= (1<<0); /* enable SCU */
71 #endif
72 
73 #if ARCH_HAS_MMU
74     arm_mmu_early_init();
75 
76     platform_init_mmu_mappings();
77 #endif
78 
79     /* turn the cache back on */
80 #if WITH_DEV_CACHE_PL310
81     pl310_set_enable(true);
82 #endif
83     arch_enable_cache(UCACHE);
84 }
85 
arch_init(void)86 void arch_init(void) {
87 #if WITH_SMP
88     arch_mp_init_percpu();
89 
90     LTRACEF("midr 0x%x\n", arm_read_midr());
91     LTRACEF("sctlr 0x%x\n", arm_read_sctlr());
92     LTRACEF("actlr 0x%x\n", arm_read_actlr());
93 #if ARM_CPU_CORTEX_A9
94     LTRACEF("cbar 0x%x\n", arm_read_cbar());
95 #endif
96     LTRACEF("mpidr 0x%x\n", arm_read_mpidr());
97     LTRACEF("ttbcr 0x%x\n", arm_read_ttbcr());
98     LTRACEF("ttbr0 0x%x\n", arm_read_ttbr0());
99     LTRACEF("dacr 0x%x\n", arm_read_dacr());
100 #if ARM_CPU_CORTEX_A7
101     LTRACEF("l2ctlr 0x%x\n", arm_read_l2ctlr());
102     LTRACEF("l2ectlr 0x%x\n", arm_read_l2ectlr());
103 #endif
104 
105 #if ARM_CPU_CORTEX_A9
106     addr_t scu_base = arm_read_cbar();
107     uint32_t scu_config = *REG32(scu_base + 4);
108     secondaries_to_init = scu_config & 0x3;
109 #elif ARM_CPU_CORTEX_A7 || ARM_CPU_CORTEX_A15
110     uint32_t l2ctlr = arm_read_l2ctlr();
111     secondaries_to_init = (l2ctlr >> 24);
112 #else
113     secondaries_to_init = SMP_MAX_CPUS - 1; /* TODO: get count from somewhere else, or add cpus as they boot */
114 #endif
115 
116     lk_init_secondary_cpus(secondaries_to_init);
117 
118     /* in platforms where the cpus have already been started, go ahead and wake up all the
119      * secondary cpus here.
120      */
121     dprintf(SPEW, "releasing %d secondary cpu%c\n", secondaries_to_init, secondaries_to_init != 1 ? 's' : ' ');
122 
123     /* release the secondary cpus */
124     spin_unlock(&arm_boot_cpu_lock);
125 
126     /* flush the release of the lock, since the secondary cpus are running without cache on */
127     arch_clean_cache_range((addr_t)&arm_boot_cpu_lock, sizeof(arm_boot_cpu_lock));
128 
129 #if ARM_ARCH_WAIT_FOR_SECONDARIES
130     /* wait for secondary cpus to boot before arm_mmu_init below, which will remove
131      * temporary boot mappings
132      * TODO: find a cleaner way to do this than this #define
133      */
134     while (secondaries_to_init > 0) {
135         __asm__ volatile("wfe");
136     }
137 #endif
138 #endif // WITH_SMP
139 
140     //spinlock_test();
141 
142 #if ARCH_HAS_MMU
143     /* finish initializing the mmu */
144     arm_mmu_init();
145 #endif
146 }
147 
148 #if WITH_SMP
149 void arm_secondary_entry(uint asm_cpu_num);
arm_secondary_entry(uint asm_cpu_num)150 void arm_secondary_entry(uint asm_cpu_num) {
151     uint cpu = arch_curr_cpu_num();
152     if (cpu != asm_cpu_num)
153         return;
154 
155     arm_basic_setup();
156 
157     /* enable the local L1 cache */
158     //arch_enable_cache(UCACHE);
159 
160     // XXX may not be safe, but just hard enable i and d cache here
161     // at the moment cannot rely on arch_enable_cache not dumping the L2
162     uint32_t sctlr = arm_read_sctlr();
163     sctlr |= (1<<12) | (1<<2); // enable i and dcache
164     arm_write_sctlr(sctlr);
165 
166     /* run early secondary cpu init routines up to the threading level */
167     lk_init_level(LK_INIT_FLAG_SECONDARY_CPUS, LK_INIT_LEVEL_EARLIEST, LK_INIT_LEVEL_THREADING - 1);
168 
169     arch_mp_init_percpu();
170 
171     LTRACEF("cpu num %d\n", cpu);
172     LTRACEF("sctlr 0x%x\n", arm_read_sctlr());
173     LTRACEF("actlr 0x%x\n", arm_read_actlr());
174 
175     /* we're done, tell the main cpu we're up */
176     atomic_add(&secondaries_to_init, -1);
177     smp_mb();
178     __asm__ volatile("sev");
179 
180     lk_secondary_cpu_entry();
181 }
182 #endif
183 
arm_basic_setup(void)184 static void arm_basic_setup(void) {
185     uint32_t sctlr = arm_read_sctlr();
186 
187     /* ARMV7 bits */
188     sctlr &= ~(1<<10); /* swp disable */
189     sctlr |=  (1<<11); /* enable program flow prediction */
190     sctlr &= ~(1<<14); /* random cache/tlb replacement */
191     sctlr &= ~(1<<25); /* E bit set to 0 on exception */
192     sctlr &= ~(1<<30); /* no thumb exceptions */
193     sctlr |=  (1<<22); /* enable unaligned access */
194     sctlr &= ~(1<<1);  /* disable alignment abort */
195 
196     arm_write_sctlr(sctlr);
197 
198     uint32_t actlr = arm_read_actlr();
199 #if ARM_CPU_CORTEX_A9
200     actlr |= (1<<2); /* enable dcache prefetch */
201 #if WITH_DEV_CACHE_PL310
202     actlr |= (1<<7); /* L2 exclusive cache */
203     actlr |= (1<<3); /* L2 write full line of zeroes */
204     actlr |= (1<<1); /* L2 prefetch hint enable */
205 #endif
206 #if WITH_SMP
207     /* enable smp mode, cache and tlb broadcast */
208     actlr |= (1<<6) | (1<<0);
209 #endif
210 #endif // ARM_CPU_CORTEX_A9
211 #if ARM_CPU_CORTEX_A7
212 #if WITH_SMP
213     /* enable smp mode */
214     actlr |= (1<<6);
215 #endif
216 #endif // ARM_CPU_CORTEX_A7
217 
218     arm_write_actlr(actlr);
219 
220 #if ENABLE_CYCLE_COUNTER && ARM_ISA_ARMV7
221     /* enable the cycle count register */
222     uint32_t en;
223     __asm__ volatile("mrc	p15, 0, %0, c9, c12, 0" : "=r" (en));
224     en &= ~(1<<3); /* cycle count every cycle */
225     en |= 1; /* enable all performance counters */
226     __asm__ volatile("mcr	p15, 0, %0, c9, c12, 0" :: "r" (en));
227 
228     /* enable cycle counter */
229     en = (1<<31);
230     __asm__ volatile("mcr	p15, 0, %0, c9, c12, 1" :: "r" (en));
231 #endif
232 
233 #if ARM_WITH_VFP
234     /* enable cp10 and cp11 */
235     uint32_t val = arm_read_cpacr();
236     val |= (3<<22)|(3<<20);
237     arm_write_cpacr(val);
238 
239     /* set enable bit in fpexc */
240     __asm__ volatile("mrc  p10, 7, %0, c8, c0, 0" : "=r" (val));
241     val |= (1<<30);
242     __asm__ volatile("mcr  p10, 7, %0, c8, c0, 0" :: "r" (val));
243 
244     /* make sure the fpu starts off disabled */
245     arm_fpu_set_enable(false);
246 #endif
247 
248     /* set the vector base to our exception vectors so we don't need to double map at 0 */
249 #if ARM_ISA_ARMV7
250     arm_write_vbar(KERNEL_BASE + KERNEL_LOAD_OFFSET);
251 #endif
252 }
253 
arch_quiesce(void)254 void arch_quiesce(void) {
255 #if ENABLE_CYCLE_COUNTER
256 #if ARM_ISA_ARMV7
257     /* disable the cycle count and performance counters */
258     uint32_t en;
259     __asm__ volatile("mrc	p15, 0, %0, c9, c12, 0" : "=r" (en));
260     en &= ~1; /* disable all performance counters */
261     __asm__ volatile("mcr	p15, 0, %0, c9, c12, 0" :: "r" (en));
262 
263     /* disable cycle counter */
264     en = 0;
265     __asm__ volatile("mcr	p15, 0, %0, c9, c12, 1" :: "r" (en));
266 #endif
267 #if ARM_CPU_ARM1136
268     /* disable the cycle count and performance counters */
269     uint32_t en;
270     __asm__ volatile("mrc	p15, 0, %0, c15, c12, 0" : "=r" (en));
271     en &= ~1; /* disable all performance counters */
272     __asm__ volatile("mcr	p15, 0, %0, c15, c12, 0" :: "r" (en));
273 #endif
274 #endif
275 
276     uint32_t actlr = arm_read_actlr();
277 #if ARM_CPU_CORTEX_A9
278     actlr = 0; /* put the aux control register back to default */
279 #endif // ARM_CPU_CORTEX_A9
280     arm_write_actlr(actlr);
281 }
282 
283 #if ARM_ISA_ARMV7
284 /* virtual to physical translation */
arm_vtop(addr_t va,addr_t * pa)285 status_t arm_vtop(addr_t va, addr_t *pa) {
286     spin_lock_saved_state_t irqstate;
287 
288     arch_interrupt_save(&irqstate, SPIN_LOCK_FLAG_INTERRUPTS);
289 
290     arm_write_ats1cpr(va & ~(PAGE_SIZE-1));
291     uint32_t par = arm_read_par();
292 
293     arch_interrupt_restore(irqstate, SPIN_LOCK_FLAG_INTERRUPTS);
294 
295     if (par & 1)
296         return ERR_NOT_FOUND;
297 
298     if (pa) {
299         *pa = (par & 0xfffff000) | (va & 0xfff);
300     }
301 
302     return NO_ERROR;
303 }
304 #endif
305 
arch_chain_load(void * entry,ulong arg0,ulong arg1,ulong arg2,ulong arg3)306 void arch_chain_load(void *entry, ulong arg0, ulong arg1, ulong arg2, ulong arg3) {
307     LTRACEF("entry %p, args 0x%lx 0x%lx 0x%lx 0x%lx\n", entry, arg0, arg1, arg2, arg3);
308 
309     /* we are going to shut down the system, start by disabling interrupts */
310     arch_disable_ints();
311 
312     /* give target and platform a chance to put hardware into a suitable
313      * state for chain loading.
314      */
315     target_quiesce();
316     platform_quiesce();
317 
318     paddr_t entry_pa;
319     paddr_t loader_pa;
320 
321 #if WITH_KERNEL_VM
322     /* get the physical address of the entry point we're going to branch to */
323     if (arm_vtop((addr_t)entry, &entry_pa) < 0) {
324         panic("error translating entry physical address\n");
325     }
326 
327     /* add the low bits of the virtual address back */
328     entry_pa |= ((addr_t)entry & 0xfff);
329 
330     LTRACEF("entry pa 0x%lx\n", entry_pa);
331 
332     /* figure out the mapping for the chain load routine */
333     if (arm_vtop((addr_t)&arm_chain_load, &loader_pa) < 0) {
334         panic("error translating loader physical address\n");
335     }
336 
337     /* add the low bits of the virtual address back */
338     loader_pa |= ((addr_t)&arm_chain_load & 0xfff);
339 
340     paddr_t loader_pa_section = ROUNDDOWN(loader_pa, SECTION_SIZE);
341 
342     LTRACEF("loader address %p, phys 0x%lx, surrounding large page 0x%lx\n",
343             &arm_chain_load, loader_pa, loader_pa_section);
344 
345     arch_aspace_t *aspace;
346     bool need_context_switch;
347     // if loader_pa is within the kernel aspace, we can simply use arch_mmu_map to identity map it
348     // if its outside, we need to create a new aspace and context switch to it
349     if (arch_mmu_is_valid_vaddr(&vmm_get_kernel_aspace()->arch_aspace, loader_pa)) {
350       aspace = &vmm_get_kernel_aspace()->arch_aspace;
351       need_context_switch = false;
352     } else {
353       aspace = malloc(sizeof(*aspace));
354       arch_mmu_init_aspace(aspace, loader_pa_section, SECTION_SIZE, 0);
355       need_context_switch = true;
356     }
357 
358     /* using large pages, map around the target location */
359     arch_mmu_map(aspace, loader_pa_section, loader_pa_section, (2 * SECTION_SIZE / PAGE_SIZE), 0);
360     if (need_context_switch) arch_mmu_context_switch(aspace);
361 #else
362     /* for non vm case, just branch directly into it */
363     entry_pa = (paddr_t)entry;
364     loader_pa = (paddr_t)&arm_chain_load;
365 #endif
366 
367     LTRACEF("disabling instruction/data cache\n");
368     arch_disable_cache(UCACHE);
369 #if WITH_DEV_CACHE_PL310
370     pl310_set_enable(false);
371 #endif
372 
373     /* put the booting cpu back into close to a default state */
374     arch_quiesce();
375 
376     // linux wont re-enable the FPU during boot, so it must be enabled when chainloading
377     arm_fpu_set_enable(true);
378 
379     LTRACEF("branching to physical address of loader\n");
380 
381     /* branch to the physical address version of the chain loader routine */
382     void (*loader)(paddr_t entry, ulong, ulong, ulong, ulong) __NO_RETURN = (void *)loader_pa;
383     loader(entry_pa, arg0, arg1, arg2, arg3);
384 }
385 
386 static spin_lock_t lock = 0;
387 
spinlock_test(void)388 static void spinlock_test(void) {
389     TRACE_ENTRY;
390 
391     spin_lock_saved_state_t state;
392     spin_lock_irqsave(&lock, state);
393 
394     TRACEF("cpu0: i have the lock\n");
395     spin(1000000);
396     TRACEF("cpu0: releasing it\n");
397 
398     spin_unlock_irqrestore(&lock, state);
399 
400     spin(1000000);
401 }
402 
spinlock_test_secondary(void)403 static void spinlock_test_secondary(void) {
404     TRACE_ENTRY;
405 
406     spin(500000);
407     spin_lock_saved_state_t state;
408     spin_lock_irqsave(&lock, state);
409 
410     TRACEF("cpu1: i have the lock\n");
411     spin(250000);
412     TRACEF("cpu1: releasing it\n");
413 
414     spin_unlock_irqrestore(&lock, state);
415 }
416 
417 /* switch to user mode, set the user stack pointer to user_stack_top, put the svc stack pointer to the top of the kernel stack */
arch_enter_uspace(vaddr_t entry_point,vaddr_t user_stack_top)418 void arch_enter_uspace(vaddr_t entry_point, vaddr_t user_stack_top) {
419     DEBUG_ASSERT(IS_ALIGNED(user_stack_top, 8));
420 
421     thread_t *ct = get_current_thread();
422 
423     vaddr_t kernel_stack_top = (uintptr_t)ct->stack + ct->stack_size;
424     kernel_stack_top = ROUNDDOWN(kernel_stack_top, 8);
425 
426     uint32_t spsr = CPSR_MODE_USR;
427     spsr |= (entry_point & 1) ? CPSR_THUMB : 0;
428 
429     arch_disable_ints();
430 
431     asm volatile(
432         "ldmia  %[ustack], { sp }^;"
433         "msr	spsr, %[spsr];"
434         "mov	sp, %[kstack];"
435         "movs	pc, %[entry];"
436         :
437         : [ustack]"r"(&user_stack_top),
438         [kstack]"r"(kernel_stack_top),
439         [entry]"r"(entry_point),
440         [spsr]"r"(spsr)
441         : "memory");
442     __UNREACHABLE;
443 }
444