1 | #
|
---|
2 | # Copyright (c) 2005 Jakub Jermar
|
---|
3 | # All rights reserved.
|
---|
4 | #
|
---|
5 | # Redistribution and use in source and binary forms, with or without
|
---|
6 | # modification, are permitted provided that the following conditions
|
---|
7 | # are met:
|
---|
8 | #
|
---|
9 | # - Redistributions of source code must retain the above copyright
|
---|
10 | # notice, this list of conditions and the following disclaimer.
|
---|
11 | # - Redistributions in binary form must reproduce the above copyright
|
---|
12 | # notice, this list of conditions and the following disclaimer in the
|
---|
13 | # documentation and/or other materials provided with the distribution.
|
---|
14 | # - The name of the author may not be used to endorse or promote products
|
---|
15 | # derived from this software without specific prior written permission.
|
---|
16 | #
|
---|
17 | # THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
|
---|
18 | # IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
|
---|
19 | # OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
---|
20 | # IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
|
---|
21 | # INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
---|
22 | # NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
---|
23 | # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
---|
24 | # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
---|
25 | # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
---|
26 | # THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
---|
27 | #
|
---|
28 |
|
---|
29 | #include <arch/arch.h>
|
---|
30 | #include <arch/cpu.h>
|
---|
31 | #include <arch/regdef.h>
|
---|
32 | #include <arch/boot/boot.h>
|
---|
33 | #include <arch/stack.h>
|
---|
34 |
|
---|
35 | #include <arch/mm/mmu.h>
|
---|
36 | #include <arch/mm/tlb.h>
|
---|
37 | #include <arch/mm/tte.h>
|
---|
38 |
|
---|
39 | #ifdef CONFIG_SMP
|
---|
40 | #include <arch/context_offset.h>
|
---|
41 | #endif
|
---|
42 |
|
---|
43 | .register %g2, #scratch
|
---|
44 | .register %g3, #scratch
|
---|
45 |
|
---|
46 | .section K_TEXT_START, "ax"
|
---|
47 |
|
---|
48 | #define BSP_FLAG 1
|
---|
49 |
|
---|
50 | /*
|
---|
51 | * 2^PHYSMEM_ADDR_SIZE is the size of the physical address space on
|
---|
52 | * a given processor.
|
---|
53 | */
|
---|
54 | #if defined (US)
|
---|
55 | #define PHYSMEM_ADDR_SIZE 41
|
---|
56 | #elif defined (US3)
|
---|
57 | #define PHYSMEM_ADDR_SIZE 43
|
---|
58 | #endif
|
---|
59 |
|
---|
60 | /*
|
---|
61 | * Here is where the kernel is passed control from the boot loader.
|
---|
62 | *
|
---|
63 | * The registers are expected to be in this state:
|
---|
64 | * - %o0 starting address of physical memory + bootstrap processor flag
|
---|
65 | * bits 63...1: physical memory starting address / 2
|
---|
66 | * bit 0: non-zero on BSP processor, zero on AP processors
|
---|
67 | * - %o1 bootinfo structure address (BSP only)
|
---|
68 | * - %o2 bootinfo structure size (BSP only)
|
---|
69 | *
|
---|
70 | * Moreover, we depend on boot having established the following environment:
|
---|
71 | * - TLBs are on
|
---|
72 | * - identity mapping for the kernel image
|
---|
73 | */
|
---|
74 |
|
---|
75 | .global kernel_image_start
|
---|
76 | kernel_image_start:
|
---|
77 | mov BSP_FLAG, %l0
|
---|
78 | and %o0, %l0, %l7 ! l7 <= bootstrap processor?
|
---|
79 | andn %o0, %l0, %l6 ! l6 <= start of physical memory
|
---|
80 |
|
---|
81 | ! Get bits (PHYSMEM_ADDR_SIZE - 1):13 of physmem_base.
|
---|
82 | srlx %l6, 13, %l5
|
---|
83 |
|
---|
84 | ! l5 <= physmem_base[(PHYSMEM_ADDR_SIZE - 1):13]
|
---|
85 | sllx %l5, 13 + (63 - (PHYSMEM_ADDR_SIZE - 1)), %l5
|
---|
86 | srlx %l5, 63 - (PHYSMEM_ADDR_SIZE - 1), %l5
|
---|
87 |
|
---|
88 | /*
|
---|
89 | * Setup basic runtime environment.
|
---|
90 | */
|
---|
91 |
|
---|
92 | wrpr %g0, NWINDOWS - 2, %cansave ! set maximum saveable windows
|
---|
93 | wrpr %g0, 0, %canrestore ! get rid of windows we will
|
---|
94 | ! never need again
|
---|
95 | wrpr %g0, 0, %otherwin ! make sure the window state is
|
---|
96 | ! consistent
|
---|
97 | wrpr %g0, NWINDOWS - 1, %cleanwin ! prevent needless clean_window
|
---|
98 | ! traps for kernel
|
---|
99 |
|
---|
100 | wrpr %g0, 0, %wstate ! use default spill/fill trap
|
---|
101 |
|
---|
102 | wrpr %g0, 0, %tl ! TL = 0, primary context
|
---|
103 | ! register is used
|
---|
104 |
|
---|
105 | wrpr %g0, PSTATE_PRIV_BIT, %pstate ! disable interrupts and disable
|
---|
106 | ! 32-bit address masking
|
---|
107 |
|
---|
108 | wrpr %g0, 0, %pil ! intialize %pil
|
---|
109 |
|
---|
110 | /*
|
---|
111 | * Switch to kernel trap table.
|
---|
112 | */
|
---|
113 | sethi %hi(trap_table), %g1
|
---|
114 | wrpr %g1, %lo(trap_table), %tba
|
---|
115 |
|
---|
116 | /*
|
---|
117 | * Take over the DMMU by installing locked TTE entry identically
|
---|
118 | * mapping the first 4M of memory.
|
---|
119 | *
|
---|
120 | * In case of DMMU, no FLUSH instructions need to be issued. Because of
|
---|
121 | * that, the old DTLB contents can be demapped pretty straightforwardly
|
---|
122 | * and without causing any traps.
|
---|
123 | */
|
---|
124 |
|
---|
125 | wr %g0, ASI_DMMU, %asi
|
---|
126 |
|
---|
127 | #define SET_TLB_DEMAP_CMD(r1, context_id) \
|
---|
128 | set (TLB_DEMAP_CONTEXT << TLB_DEMAP_TYPE_SHIFT) | (context_id << \
|
---|
129 | TLB_DEMAP_CONTEXT_SHIFT), %r1
|
---|
130 |
|
---|
131 | ! demap context 0
|
---|
132 | SET_TLB_DEMAP_CMD(g1, TLB_DEMAP_NUCLEUS)
|
---|
133 | stxa %g0, [%g1] ASI_DMMU_DEMAP
|
---|
134 | membar #Sync
|
---|
135 |
|
---|
136 | #define SET_TLB_TAG(r1, context) \
|
---|
137 | set VMA | (context << TLB_TAG_ACCESS_CONTEXT_SHIFT), %r1
|
---|
138 |
|
---|
139 | ! write DTLB tag
|
---|
140 | SET_TLB_TAG(g1, MEM_CONTEXT_KERNEL)
|
---|
141 | stxa %g1, [VA_DMMU_TAG_ACCESS] %asi
|
---|
142 | membar #Sync
|
---|
143 |
|
---|
144 | #ifdef CONFIG_VIRT_IDX_DCACHE
|
---|
145 | #define TTE_LOW_DATA(imm) (TTE_CP | TTE_CV | TTE_P | LMA | (imm))
|
---|
146 | #else /* CONFIG_VIRT_IDX_DCACHE */
|
---|
147 | #define TTE_LOW_DATA(imm) (TTE_CP | TTE_P | LMA | (imm))
|
---|
148 | #endif /* CONFIG_VIRT_IDX_DCACHE */
|
---|
149 |
|
---|
150 | #define SET_TLB_DATA(r1, r2, imm) \
|
---|
151 | set TTE_LOW_DATA(imm), %r1; \
|
---|
152 | or %r1, %l5, %r1; \
|
---|
153 | mov PAGESIZE_4M, %r2; \
|
---|
154 | sllx %r2, TTE_SIZE_SHIFT, %r2; \
|
---|
155 | or %r1, %r2, %r1; \
|
---|
156 | mov 1, %r2; \
|
---|
157 | sllx %r2, TTE_V_SHIFT, %r2; \
|
---|
158 | or %r1, %r2, %r1;
|
---|
159 |
|
---|
160 | ! write DTLB data and install the kernel mapping
|
---|
161 | SET_TLB_DATA(g1, g2, TTE_L | TTE_W) ! use non-global mapping
|
---|
162 | stxa %g1, [%g0] ASI_DTLB_DATA_IN_REG
|
---|
163 | membar #Sync
|
---|
164 |
|
---|
165 | /*
|
---|
166 | * Because we cannot use global mappings (because we want to have
|
---|
167 | * separate 64-bit address spaces for both the kernel and the
|
---|
168 | * userspace), we prepare the identity mapping also in context 1. This
|
---|
169 | * step is required by the code installing the ITLB mapping.
|
---|
170 | */
|
---|
171 | ! write DTLB tag of context 1 (i.e. MEM_CONTEXT_TEMP)
|
---|
172 | SET_TLB_TAG(g1, MEM_CONTEXT_TEMP)
|
---|
173 | stxa %g1, [VA_DMMU_TAG_ACCESS] %asi
|
---|
174 | membar #Sync
|
---|
175 |
|
---|
176 | ! write DTLB data and install the kernel mapping in context 1
|
---|
177 | SET_TLB_DATA(g1, g2, TTE_W) ! use non-global mapping
|
---|
178 | stxa %g1, [%g0] ASI_DTLB_DATA_IN_REG
|
---|
179 | membar #Sync
|
---|
180 |
|
---|
181 | /*
|
---|
182 | * Now is time to take over the IMMU. Unfortunatelly, it cannot be done
|
---|
183 | * as easily as the DMMU, because the IMMU is mapping the code it
|
---|
184 | * executes.
|
---|
185 | *
|
---|
186 | * [ Note that brave experiments with disabling the IMMU and using the
|
---|
187 | * DMMU approach failed after a dozen of desparate days with only little
|
---|
188 | * success. ]
|
---|
189 | *
|
---|
190 | * The approach used here is inspired from OpenBSD. First, the kernel
|
---|
191 | * creates IMMU mapping for itself in context 1 (MEM_CONTEXT_TEMP) and
|
---|
192 | * switches to it. Context 0 (MEM_CONTEXT_KERNEL) can be demapped
|
---|
193 | * afterwards and replaced with the kernel permanent mapping. Finally,
|
---|
194 | * the kernel switches back to context 0 and demaps context 1.
|
---|
195 | *
|
---|
196 | * Moreover, the IMMU requires use of the FLUSH instructions. But that
|
---|
197 | * is OK because we always use operands with addresses already mapped by
|
---|
198 | * the taken over DTLB.
|
---|
199 | */
|
---|
200 |
|
---|
201 | set kernel_image_start, %g5
|
---|
202 |
|
---|
203 | ! write ITLB tag of context 1
|
---|
204 | SET_TLB_TAG(g1, MEM_CONTEXT_TEMP)
|
---|
205 | mov VA_DMMU_TAG_ACCESS, %g2
|
---|
206 | stxa %g1, [%g2] ASI_IMMU
|
---|
207 | flush %g5
|
---|
208 |
|
---|
209 | ! write ITLB data and install the temporary mapping in context 1
|
---|
210 | SET_TLB_DATA(g1, g2, 0) ! use non-global mapping
|
---|
211 | stxa %g1, [%g0] ASI_ITLB_DATA_IN_REG
|
---|
212 | flush %g5
|
---|
213 |
|
---|
214 | ! switch to context 1
|
---|
215 | mov MEM_CONTEXT_TEMP, %g1
|
---|
216 | stxa %g1, [VA_PRIMARY_CONTEXT_REG] %asi ! ASI_DMMU is correct here !!!
|
---|
217 | flush %g5
|
---|
218 |
|
---|
219 | ! demap context 0
|
---|
220 | SET_TLB_DEMAP_CMD(g1, TLB_DEMAP_NUCLEUS)
|
---|
221 | stxa %g0, [%g1] ASI_IMMU_DEMAP
|
---|
222 | flush %g5
|
---|
223 |
|
---|
224 | ! write ITLB tag of context 0
|
---|
225 | SET_TLB_TAG(g1, MEM_CONTEXT_KERNEL)
|
---|
226 | mov VA_DMMU_TAG_ACCESS, %g2
|
---|
227 | stxa %g1, [%g2] ASI_IMMU
|
---|
228 | flush %g5
|
---|
229 |
|
---|
230 | ! write ITLB data and install the permanent kernel mapping in context 0
|
---|
231 | SET_TLB_DATA(g1, g2, TTE_L) ! use non-global mapping
|
---|
232 | stxa %g1, [%g0] ASI_ITLB_DATA_IN_REG
|
---|
233 | flush %g5
|
---|
234 |
|
---|
235 | ! enter nucleus - using context 0
|
---|
236 | wrpr %g0, 1, %tl
|
---|
237 |
|
---|
238 | ! demap context 1
|
---|
239 | SET_TLB_DEMAP_CMD(g1, TLB_DEMAP_PRIMARY)
|
---|
240 | stxa %g0, [%g1] ASI_IMMU_DEMAP
|
---|
241 | flush %g5
|
---|
242 |
|
---|
243 | ! set context 0 in the primary context register
|
---|
244 | stxa %g0, [VA_PRIMARY_CONTEXT_REG] %asi ! ASI_DMMU is correct here !!!
|
---|
245 | flush %g5
|
---|
246 |
|
---|
247 | ! leave nucleus - using primary context, i.e. context 0
|
---|
248 | wrpr %g0, 0, %tl
|
---|
249 |
|
---|
250 | brz %l7, 1f ! skip if you are not the bootstrap CPU
|
---|
251 | nop
|
---|
252 |
|
---|
253 | /*
|
---|
254 | * Save physmem_base for use by the mm subsystem.
|
---|
255 | * %l6 contains starting physical address
|
---|
256 | */
|
---|
257 | sethi %hi(physmem_base), %l4
|
---|
258 | stx %l6, [%l4 + %lo(physmem_base)]
|
---|
259 |
|
---|
260 | /*
|
---|
261 | * Precompute kernel 8K TLB data template.
|
---|
262 | * %l5 contains starting physical address
|
---|
263 | * bits [(PHYSMEM_ADDR_SIZE - 1):13]
|
---|
264 | */
|
---|
265 | sethi %hi(kernel_8k_tlb_data_template), %l4
|
---|
266 | ldx [%l4 + %lo(kernel_8k_tlb_data_template)], %l3
|
---|
267 | or %l3, %l5, %l3
|
---|
268 | stx %l3, [%l4 + %lo(kernel_8k_tlb_data_template)]
|
---|
269 |
|
---|
270 | /*
|
---|
271 | * Flush D-Cache.
|
---|
272 | */
|
---|
273 | call dcache_flush
|
---|
274 | nop
|
---|
275 |
|
---|
276 | /*
|
---|
277 | * So far, we have not touched the stack.
|
---|
278 | * It is a good idea to set the kernel stack to a known state now.
|
---|
279 | */
|
---|
280 | sethi %hi(temporary_boot_stack), %sp
|
---|
281 | or %sp, %lo(temporary_boot_stack), %sp
|
---|
282 | sub %sp, STACK_BIAS, %sp
|
---|
283 |
|
---|
284 | sethi %hi(bootinfo), %o0
|
---|
285 | call memcpy ! copy bootinfo
|
---|
286 | or %o0, %lo(bootinfo), %o0
|
---|
287 |
|
---|
288 | call arch_pre_main
|
---|
289 | nop
|
---|
290 |
|
---|
291 | call main_bsp
|
---|
292 | nop
|
---|
293 |
|
---|
294 | /* Not reached. */
|
---|
295 |
|
---|
296 | 0:
|
---|
297 | ba 0b
|
---|
298 | nop
|
---|
299 |
|
---|
300 |
|
---|
301 | 1:
|
---|
302 | #ifdef CONFIG_SMP
|
---|
303 | /*
|
---|
304 | * Determine the width of the MID and save its mask to %g3. The width
|
---|
305 | * is
|
---|
306 | * * 5 for US and US-IIIi,
|
---|
307 | * * 10 for US3 except US-IIIi.
|
---|
308 | */
|
---|
309 | #if defined(US)
|
---|
310 | mov 0x1f, %g3
|
---|
311 | #elif defined(US3)
|
---|
312 | mov 0x3ff, %g3
|
---|
313 | rdpr %ver, %g2
|
---|
314 | sllx %g2, 16, %g2
|
---|
315 | srlx %g2, 48, %g2
|
---|
316 | cmp %g2, IMPL_ULTRASPARCIII_I
|
---|
317 | move %xcc, 0x1f, %g3
|
---|
318 | #endif
|
---|
319 |
|
---|
320 | /*
|
---|
321 | * Read MID from the processor.
|
---|
322 | */
|
---|
323 | ldxa [%g0] ASI_ICBUS_CONFIG, %g1
|
---|
324 | srlx %g1, ICBUS_CONFIG_MID_SHIFT, %g1
|
---|
325 | and %g1, %g3, %g1
|
---|
326 |
|
---|
327 | /*
|
---|
328 | * Active loop for APs until the BSP picks them up. A processor cannot
|
---|
329 | * leave the loop until the global variable 'waking_up_mid' equals its
|
---|
330 | * MID.
|
---|
331 | */
|
---|
332 | set waking_up_mid, %g2
|
---|
333 | 2:
|
---|
334 | ldx [%g2], %g3
|
---|
335 | cmp %g3, %g1
|
---|
336 | bne 2b
|
---|
337 | nop
|
---|
338 |
|
---|
339 | /*
|
---|
340 | * Configure stack for the AP.
|
---|
341 | * The AP is expected to use the stack saved
|
---|
342 | * in the ctx global variable.
|
---|
343 | */
|
---|
344 | set ctx, %g1
|
---|
345 | add %g1, OFFSET_SP, %g1
|
---|
346 | ldx [%g1], %o6
|
---|
347 |
|
---|
348 | call main_ap
|
---|
349 | nop
|
---|
350 |
|
---|
351 | /* Not reached. */
|
---|
352 | #endif
|
---|
353 |
|
---|
354 | 0:
|
---|
355 | ba 0b
|
---|
356 | nop
|
---|
357 |
|
---|
358 |
|
---|
359 | .section K_DATA_START, "aw", @progbits
|
---|
360 |
|
---|
361 | /*
|
---|
362 | * Create small stack to be used by the bootstrap processor. It is going to be
|
---|
363 | * used only for a very limited period of time, but we switch to it anyway,
|
---|
364 | * just to be sure we are properly initialized.
|
---|
365 | */
|
---|
366 |
|
---|
367 | #define INITIAL_STACK_SIZE 1024
|
---|
368 |
|
---|
369 | .align STACK_ALIGNMENT
|
---|
370 | .space INITIAL_STACK_SIZE
|
---|
371 | .align STACK_ALIGNMENT
|
---|
372 | temporary_boot_stack:
|
---|
373 | .space STACK_WINDOW_SAVE_AREA_SIZE
|
---|
374 |
|
---|
375 |
|
---|
376 | .data
|
---|
377 |
|
---|
378 | .align 8
|
---|
379 | .global physmem_base ! copy of the physical memory base address
|
---|
380 | physmem_base:
|
---|
381 | .quad 0
|
---|
382 |
|
---|
383 | /*
|
---|
384 | * This variable is used by the fast_data_MMU_miss trap handler. In runtime, it
|
---|
385 | * is further modified to reflect the starting address of physical memory.
|
---|
386 | */
|
---|
387 | .global kernel_8k_tlb_data_template
|
---|
388 | kernel_8k_tlb_data_template:
|
---|
389 | #ifdef CONFIG_VIRT_IDX_DCACHE
|
---|
390 | .quad ((1 << TTE_V_SHIFT) | (PAGESIZE_8K << TTE_SIZE_SHIFT) | TTE_CP | \
|
---|
391 | TTE_CV | TTE_P | TTE_W)
|
---|
392 | #else /* CONFIG_VIRT_IDX_DCACHE */
|
---|
393 | .quad ((1 << TTE_V_SHIFT) | (PAGESIZE_8K << TTE_SIZE_SHIFT) | TTE_CP | \
|
---|
394 | TTE_P | TTE_W)
|
---|
395 | #endif /* CONFIG_VIRT_IDX_DCACHE */
|
---|
396 |
|
---|