xref: /titanic_41/usr/src/uts/sun4u/cpu/us3_cheetah_asm.s (revision 4c06356b0f0fffb4fc1b6eccc8e5d8e2254a84d6)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License, Version 1.0 only
6 * (the "License").  You may not use this file except in compliance
7 * with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or http://www.opensolaris.org/os/licensing.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22/*
23 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 *
26 * Assembly code support for the Cheetah module
27 */
28
29#pragma ident	"%Z%%M%	%I%	%E% SMI"
30
31#if !defined(lint)
32#include "assym.h"
33#endif	/* lint */
34
35#include <sys/asm_linkage.h>
36#include <sys/mmu.h>
37#include <vm/hat_sfmmu.h>
38#include <sys/machparam.h>
39#include <sys/machcpuvar.h>
40#include <sys/machthread.h>
41#include <sys/machtrap.h>
42#include <sys/privregs.h>
43#include <sys/asm_linkage.h>
44#include <sys/trap.h>
45#include <sys/cheetahregs.h>
46#include <sys/us3_module.h>
47#include <sys/xc_impl.h>
48#include <sys/intreg.h>
49#include <sys/async.h>
50#include <sys/clock.h>
51#include <sys/cheetahasm.h>
52
53#ifdef TRAPTRACE
54#include <sys/traptrace.h>
55#endif /* TRAPTRACE */
56
57#if !defined(lint)
58
59/* BEGIN CSTYLED */
60
61/*
62 * Cheetah version to flush an Ecache line by index (aliased address)
63 */
64#define	ECACHE_REFLUSH_LINE(ecache_size, alias_address, scr2)		\
65	ldxa	[alias_address]ASI_MEM, %g0
66
67#define	ECACHE_FLUSH_LINE(physaddr, ecache_size, scr1, scr2)		\
68	xor	physaddr, ecache_size, scr1;				\
69	add	ecache_size, ecache_size, scr2;				\
70	sub	scr2, 1, scr2;						\
71	and	scr1, scr2, scr1;					\
72	ASM_LDX(scr2, ecache_flushaddr);				\
73	add	scr1, scr2, scr1;					\
74	ECACHE_REFLUSH_LINE(ecache_size, scr1, scr2)
75
76/* END CSTYLED */
77
78#endif	/* !lint */
79
80
81/*
82 * Fast ECC error at TL>0 handler
83 * We get here via trap 70 at TL>0->Software trap 0 at TL>0.  We enter
84 * this routine with %g1 and %g2 already saved in %tpc, %tnpc and %tstate.
85 * For a complete description of the Fast ECC at TL>0 handling see the
86 * comment block "Cheetah/Cheetah+ Fast ECC at TL>0 trap strategy" in
87 * us3_common_asm.s
88 */
89#if defined(lint)
90
91void
92fast_ecc_tl1_err(void)
93{}
94
95#else	/* lint */
96
97	.section ".text"
98	.align	64
99	ENTRY_NP(fast_ecc_tl1_err)
100
101	/*
102	 * This macro turns off the D$/I$ if they are on and saves their
103	 * original state in ch_err_tl1_tmp, saves all the %g registers in the
104	 * ch_err_tl1_data structure, updates the ch_err_tl1_flags and saves
105	 * the %tpc in ch_err_tl1_tpc.  At the end of this macro, %g1 will
106	 * point to the ch_err_tl1_data structure and the original D$/I$ state
107	 * will be saved in ch_err_tl1_tmp.  All %g registers except for %g1
108	 * will be available.
109	 */
110	CH_ERR_TL1_FECC_ENTER;
111
112	/*
113	 * Get the diagnostic logout data.  %g4 must be initialized to
114	 * current CEEN state, %g5 must point to logout structure in
115	 * ch_err_tl1_data_t.  %g3 will contain the nesting count upon
116	 * return.
117	 */
118	ldxa	[%g0]ASI_ESTATE_ERR, %g4
119	and	%g4, EN_REG_CEEN, %g4
120	add	%g1, CH_ERR_TL1_LOGOUT, %g5
121	DO_TL1_CPU_LOGOUT(%g3, %g2, %g4, %g5, %g6, %g3, %g4)
122
123	/*
124	 * If the logout nesting count is exceeded, we're probably
125	 * not making any progress, try to panic instead.
126	 */
127	cmp	%g3, CLO_NESTING_MAX
128	bge	fecc_tl1_err
129	  nop
130
131	/*
132	 * Save the current CEEN and NCEEN state in %g7 and turn them off
133	 * before flushing the Ecache.
134	 */
135	ldxa	[%g0]ASI_ESTATE_ERR, %g7
136	andn	%g7, EN_REG_CEEN | EN_REG_NCEEN, %g5
137	stxa	%g5, [%g0]ASI_ESTATE_ERR
138	membar	#Sync
139
140	/*
141	 * Flush the Ecache, using the largest possible cache size with the
142	 * smallest possible line size since we can't get the actual sizes
143	 * from the cpu_node due to DTLB misses.
144	 */
145	set	CH_ECACHE_8M_SIZE, %g4
146	set	CH_ECACHE_MIN_LSIZE, %g5
147
148	/*
149	 * Use a different flush address to avoid recursion if the error
150	 * exists in ecache_flushaddr.
151	 */
152	ASM_LDX(%g6, ecache_tl1_flushaddr)
153	cmp	%g6, -1		! check if address is valid
154	be	%xcc, fecc_tl1_err
155	  nop
156	CH_ECACHE_FLUSHALL(%g4, %g5, %g6)
157
158	/*
159	 * Restore CEEN and NCEEN to the previous state.
160	 */
161	stxa	%g7, [%g0]ASI_ESTATE_ERR
162	membar	#Sync
163
164	/*
165	 * If we turned off the D$, then flush it and turn it back on.
166	 */
167	ldxa	[%g1 + CH_ERR_TL1_TMP]%asi, %g3
168	andcc	%g3, CH_ERR_TSTATE_DC_ON, %g0
169	bz	%xcc, 3f
170	  nop
171
172	/*
173	 * Flush the D$.
174	 */
175	ASM_LD(%g4, dcache_size)
176	ASM_LD(%g5, dcache_linesize)
177	CH_DCACHE_FLUSHALL(%g4, %g5, %g6)
178
179	/*
180	 * Turn the D$ back on.
181	 */
182	ldxa	[%g0]ASI_DCU, %g3
183	or	%g3, DCU_DC, %g3
184	stxa	%g3, [%g0]ASI_DCU
185	membar	#Sync
1863:
187	/*
188	 * If we turned off the I$, then flush it and turn it back on.
189	 */
190	ldxa	[%g1 + CH_ERR_TL1_TMP]%asi, %g3
191	andcc	%g3, CH_ERR_TSTATE_IC_ON, %g0
192	bz	%xcc, 4f
193	  nop
194
195	/*
196	 * Flush the I$.
197	 */
198	ASM_LD(%g4, icache_size)
199	ASM_LD(%g5, icache_linesize)
200	CH_ICACHE_FLUSHALL(%g4, %g5, %g6, %g3)
201
202	/*
203	 * Turn the I$ back on.  Changing DCU_IC requires flush.
204	 */
205	ldxa	[%g0]ASI_DCU, %g3
206	or	%g3, DCU_IC, %g3
207	stxa	%g3, [%g0]ASI_DCU
208	flush	%g0
2094:
210
211#ifdef TRAPTRACE
212	/*
213	 * Get current trap trace entry physical pointer.
214	 */
215	CPU_INDEX(%g6, %g5)
216	sll	%g6, TRAPTR_SIZE_SHIFT, %g6
217	set	trap_trace_ctl, %g5
218	add	%g6, %g5, %g6
219	ld	[%g6 + TRAPTR_LIMIT], %g5
220	tst	%g5
221	be	%icc, skip_traptrace
222	  nop
223	ldx	[%g6 + TRAPTR_PBASE], %g5
224	ld	[%g6 + TRAPTR_OFFSET], %g4
225	add	%g5, %g4, %g5
226
227	/*
228	 * Create trap trace entry.
229	 */
230	rd	%asi, %g7
231	wr	%g0, TRAPTR_ASI, %asi
232	rd	STICK, %g4
233	stxa	%g4, [%g5 + TRAP_ENT_TICK]%asi
234	rdpr	%tl, %g4
235	stha	%g4, [%g5 + TRAP_ENT_TL]%asi
236	rdpr	%tt, %g4
237	stha	%g4, [%g5 + TRAP_ENT_TT]%asi
238	rdpr	%tpc, %g4
239	stna	%g4, [%g5 + TRAP_ENT_TPC]%asi
240	rdpr	%tstate, %g4
241	stxa	%g4, [%g5 + TRAP_ENT_TSTATE]%asi
242	stna	%sp, [%g5 + TRAP_ENT_SP]%asi
243	stna	%g0, [%g5 + TRAP_ENT_TR]%asi
244	wr	%g0, %g7, %asi
245	ldxa	[%g1 + CH_ERR_TL1_SDW_AFAR]%asi, %g3
246	ldxa	[%g1 + CH_ERR_TL1_SDW_AFSR]%asi, %g4
247	wr	%g0, TRAPTR_ASI, %asi
248	stna	%g3, [%g5 + TRAP_ENT_F1]%asi
249	stna	%g4, [%g5 + TRAP_ENT_F2]%asi
250	wr	%g0, %g7, %asi
251	ldxa	[%g1 + CH_ERR_TL1_AFAR]%asi, %g3
252	ldxa	[%g1 + CH_ERR_TL1_AFSR]%asi, %g4
253	wr	%g0, TRAPTR_ASI, %asi
254	stna	%g3, [%g5 + TRAP_ENT_F3]%asi
255	stna	%g4, [%g5 + TRAP_ENT_F4]%asi
256	wr	%g0, %g7, %asi
257
258	/*
259	 * Advance trap trace pointer.
260	 */
261	ld	[%g6 + TRAPTR_OFFSET], %g5
262	ld	[%g6 + TRAPTR_LIMIT], %g4
263	st	%g5, [%g6 + TRAPTR_LAST_OFFSET]
264	add	%g5, TRAP_ENT_SIZE, %g5
265	sub	%g4, TRAP_ENT_SIZE, %g4
266	cmp	%g5, %g4
267	movge	%icc, 0, %g5
268	st	%g5, [%g6 + TRAPTR_OFFSET]
269skip_traptrace:
270#endif	/* TRAPTRACE */
271
272	/*
273	 * If nesting count is not zero, skip all the AFSR/AFAR
274	 * handling and just do the necessary cache-flushing.
275	 */
276	ldxa	[%g1 + CH_ERR_TL1_NEST_CNT]%asi, %g2
277	brnz	%g2, 6f
278	  nop
279
280	/*
281	 * If a UCU followed by a WDU has occurred go ahead and panic
282	 * since a UE will occur (on the retry) before the UCU and WDU
283	 * messages are enqueued.
284	 */
285	ldxa	[%g1 + CH_ERR_TL1_AFSR]%asi, %g3
286	set	1, %g4
287	sllx	%g4, C_AFSR_UCU_SHIFT, %g4
288	btst	%g4, %g3		! UCU in original AFSR?
289	bz	%xcc, 6f
290	  nop
291	ldxa	[%g0]ASI_AFSR, %g4	! current AFSR
292	or	%g3, %g4, %g3		! %g3 = original + current AFSR
293	set	1, %g4
294	sllx	%g4, C_AFSR_WDU_SHIFT, %g4
295	btst	%g4, %g3		! WDU in original or current AFSR?
296	bnz	%xcc, fecc_tl1_err
297	  nop
298
2996:
300	/*
301	 * We fall into this macro if we've successfully logged the error in
302	 * the ch_err_tl1_data structure and want the PIL15 softint to pick
303	 * it up and log it.  %g1 must point to the ch_err_tl1_data structure.
304	 * Restores the %g registers and issues retry.
305	 */
306	CH_ERR_TL1_EXIT;
307
308	/*
309	 * Establish panic exit label.
310	 */
311	CH_ERR_TL1_PANIC_EXIT(fecc_tl1_err);
312
313	SET_SIZE(fast_ecc_tl1_err)
314
315#endif	/* lint */
316
317
318#if defined(lint)
319/*
320 * scrubphys - Pass in the aligned physical memory address
321 * that you want to scrub, along with the ecache set size.
322 *
323 *	1) Displacement flush the E$ line corresponding to %addr.
324 *	   The first ldxa guarantees that the %addr is no longer in
325 *	   M, O, or E (goes to I or S (if instruction fetch also happens).
326 *	2) "Write" the data using a CAS %addr,%g0,%g0.
327 *	   The casxa guarantees a transition from I to M or S to M.
328 *	3) Displacement flush the E$ line corresponding to %addr.
329 *	   The second ldxa pushes the M line out of the ecache, into the
330 *	   writeback buffers, on the way to memory.
331 *	4) The "membar #Sync" pushes the cache line out of the writeback
332 *	   buffers onto the bus, on the way to dram finally.
333 *
334 * This is a modified version of the algorithm suggested by Gary Lauterbach.
335 * In theory the CAS %addr,%g0,%g0 is supposed to mark the addr's cache line
336 * as modified, but then we found out that for spitfire, if it misses in the
337 * E$ it will probably install as an M, but if it hits in the E$, then it
338 * will stay E, if the store doesn't happen. So the first displacement flush
339 * should ensure that the CAS will miss in the E$.  Arrgh.
340 */
341/* ARGSUSED */
342void
343scrubphys(uint64_t paddr, int ecache_set_size)
344{}
345
346#else	/* lint */
347	ENTRY(scrubphys)
348	rdpr	%pstate, %o4
349	andn	%o4, PSTATE_IE | PSTATE_AM, %o5
350	wrpr	%o5, %g0, %pstate	! clear IE, AM bits
351
352	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
353	casxa	[%o0]ASI_MEM, %g0, %g0
354	ECACHE_REFLUSH_LINE(%o1, %o2, %o3)
355
356	wrpr	%g0, %o4, %pstate	! restore earlier pstate register value
357
358	retl
359	membar	#Sync			! move the data out of the load buffer
360	SET_SIZE(scrubphys)
361
362#endif	/* lint */
363
364
365#if defined(lint)
366 /*
367 * clearphys - Pass in the physical memory address of the checkblock
368 * that you want to push out, cleared with a recognizable pattern,
369 * from the ecache.
370 *
371 * To ensure that the ecc gets recalculated after the bad data is cleared,
372 * we must write out enough data to fill the w$ line (64 bytes). So we read
373 * in an entire ecache subblock's worth of data, and write it back out.
374 * Then we overwrite the 16 bytes of bad data with the pattern.
375 */
376/* ARGSUSED */
377void
378clearphys(uint64_t paddr, int ecache_set_size, int ecache_linesize)
379{
380}
381
382#else	/* lint */
383	ENTRY(clearphys)
384	/* turn off IE, AM bits */
385	rdpr	%pstate, %o4
386	andn	%o4, PSTATE_IE | PSTATE_AM, %o5
387	wrpr	%o5, %g0, %pstate
388
389	/* turn off NCEEN */
390	ldxa	[%g0]ASI_ESTATE_ERR, %o5
391	andn	%o5, EN_REG_NCEEN, %o3
392	stxa	%o3, [%g0]ASI_ESTATE_ERR
393	membar	#Sync
394
395	/* align address passed with 64 bytes subblock size */
396	mov	CH_ECACHE_SUBBLK_SIZE, %o2
397	andn	%o0, (CH_ECACHE_SUBBLK_SIZE - 1), %g1
398
399	/* move the good data into the W$ */
4001:
401	subcc	%o2, 8, %o2
402	ldxa	[%g1 + %o2]ASI_MEM, %g2
403	bge	1b
404	  stxa	%g2, [%g1 + %o2]ASI_MEM
405
406	/* now overwrite the bad data */
407	setx	0xbadecc00badecc01, %g1, %g2
408	stxa	%g2, [%o0]ASI_MEM
409	mov	8, %g1
410	stxa	%g2, [%o0 + %g1]ASI_MEM
411
412	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
413	casxa	[%o0]ASI_MEM, %g0, %g0
414	ECACHE_REFLUSH_LINE(%o1, %o2, %o3)
415
416	/* clear the AFSR */
417	ldxa	[%g0]ASI_AFSR, %o1
418	stxa	%o1, [%g0]ASI_AFSR
419	membar	#Sync
420
421	/* turn NCEEN back on */
422	stxa	%o5, [%g0]ASI_ESTATE_ERR
423	membar	#Sync
424
425	/* return and re-enable IE and AM */
426	retl
427	  wrpr	%g0, %o4, %pstate
428	SET_SIZE(clearphys)
429
430#endif	/* lint */
431
432
433#if defined(lint)
434/*
435 * Cheetah Ecache displacement flush the specified line from the E$
436 *
437 * Register usage:
438 *	%o0 - 64 bit physical address for flushing
439 *	%o1 - Ecache set size
440 */
441/*ARGSUSED*/
442void
443ecache_flush_line(uint64_t flushaddr, int ec_set_size)
444{
445}
446#else	/* lint */
447	ENTRY(ecache_flush_line)
448
449	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
450
451	retl
452	  nop
453	SET_SIZE(ecache_flush_line)
454#endif	/* lint */
455
456
457#if defined(lint)
458/*
459 * This routine will not be called in Cheetah systems.
460 */
461void
462flush_ipb(void)
463{ return; }
464
465#else	/* lint */
466
467	ENTRY(flush_ipb)
468	retl
469	nop
470	SET_SIZE(flush_ipb)
471
472#endif	/* lint */
473