xref: /illumos-gate/usr/src/uts/sun4u/cpu/us3_cheetah_asm.S (revision 20a7641f9918de8574b8b3b47dbe35c4bfc78df1)
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#include "assym.h"
30
31#include <sys/asm_linkage.h>
32#include <sys/mmu.h>
33#include <vm/hat_sfmmu.h>
34#include <sys/machparam.h>
35#include <sys/machcpuvar.h>
36#include <sys/machthread.h>
37#include <sys/machtrap.h>
38#include <sys/privregs.h>
39#include <sys/asm_linkage.h>
40#include <sys/trap.h>
41#include <sys/cheetahregs.h>
42#include <sys/us3_module.h>
43#include <sys/xc_impl.h>
44#include <sys/intreg.h>
45#include <sys/async.h>
46#include <sys/clock.h>
47#include <sys/cheetahasm.h>
48
49#ifdef TRAPTRACE
50#include <sys/traptrace.h>
51#endif /* TRAPTRACE */
52
53/* BEGIN CSTYLED */
54
55/*
56 * Cheetah version to flush an Ecache line by index (aliased address)
57 */
58#define	ECACHE_REFLUSH_LINE(ecache_size, alias_address, scr2)		\
59	ldxa	[alias_address]ASI_MEM, %g0
60
61#define	ECACHE_FLUSH_LINE(physaddr, ecache_size, scr1, scr2)		\
62	xor	physaddr, ecache_size, scr1;				\
63	add	ecache_size, ecache_size, scr2;				\
64	sub	scr2, 1, scr2;						\
65	and	scr1, scr2, scr1;					\
66	ASM_LDX(scr2, ecache_flushaddr);				\
67	add	scr1, scr2, scr1;					\
68	ECACHE_REFLUSH_LINE(ecache_size, scr1, scr2)
69
70/* END CSTYLED */
71
72
73/*
74 * Fast ECC error at TL>0 handler
75 * We get here via trap 70 at TL>0->Software trap 0 at TL>0.  We enter
76 * this routine with %g1 and %g2 already saved in %tpc, %tnpc and %tstate.
77 * For a complete description of the Fast ECC at TL>0 handling see the
78 * comment block "Cheetah/Cheetah+ Fast ECC at TL>0 trap strategy" in
79 * us3_common_asm.s
80 */
81
82	.section ".text"
83	.align	64
84	ENTRY_NP(fast_ecc_tl1_err)
85
86	/*
87	 * This macro turns off the D$/I$ if they are on and saves their
88	 * original state in ch_err_tl1_tmp, saves all the %g registers in the
89	 * ch_err_tl1_data structure, updates the ch_err_tl1_flags and saves
90	 * the %tpc in ch_err_tl1_tpc.  At the end of this macro, %g1 will
91	 * point to the ch_err_tl1_data structure and the original D$/I$ state
92	 * will be saved in ch_err_tl1_tmp.  All %g registers except for %g1
93	 * will be available.
94	 */
95	CH_ERR_TL1_FECC_ENTER;
96
97	/*
98	 * Get the diagnostic logout data.  %g4 must be initialized to
99	 * current CEEN state, %g5 must point to logout structure in
100	 * ch_err_tl1_data_t.  %g3 will contain the nesting count upon
101	 * return.
102	 */
103	ldxa	[%g0]ASI_ESTATE_ERR, %g4
104	and	%g4, EN_REG_CEEN, %g4
105	add	%g1, CH_ERR_TL1_LOGOUT, %g5
106	DO_TL1_CPU_LOGOUT(%g3, %g2, %g4, %g5, %g6, %g3, %g4)
107
108	/*
109	 * If the logout nesting count is exceeded, we're probably
110	 * not making any progress, try to panic instead.
111	 */
112	cmp	%g3, CLO_NESTING_MAX
113	bge	fecc_tl1_err
114	  nop
115
116	/*
117	 * Save the current CEEN and NCEEN state in %g7 and turn them off
118	 * before flushing the Ecache.
119	 */
120	ldxa	[%g0]ASI_ESTATE_ERR, %g7
121	andn	%g7, EN_REG_CEEN | EN_REG_NCEEN, %g5
122	stxa	%g5, [%g0]ASI_ESTATE_ERR
123	membar	#Sync
124
125	/*
126	 * Flush the Ecache, using the largest possible cache size with the
127	 * smallest possible line size since we can't get the actual sizes
128	 * from the cpu_node due to DTLB misses.
129	 */
130	set	CH_ECACHE_8M_SIZE, %g4
131	set	CH_ECACHE_MIN_LSIZE, %g5
132
133	/*
134	 * Use a different flush address to avoid recursion if the error
135	 * exists in ecache_flushaddr.
136	 */
137	ASM_LDX(%g6, ecache_tl1_flushaddr)
138	cmp	%g6, -1		! check if address is valid
139	be	%xcc, fecc_tl1_err
140	  nop
141	CH_ECACHE_FLUSHALL(%g4, %g5, %g6)
142
143	/*
144	 * Restore CEEN and NCEEN to the previous state.
145	 */
146	stxa	%g7, [%g0]ASI_ESTATE_ERR
147	membar	#Sync
148
149	/*
150	 * If we turned off the D$, then flush it and turn it back on.
151	 */
152	ldxa	[%g1 + CH_ERR_TL1_TMP]%asi, %g3
153	andcc	%g3, CH_ERR_TSTATE_DC_ON, %g0
154	bz	%xcc, 3f
155	  nop
156
157	/*
158	 * Flush the D$.
159	 */
160	ASM_LD(%g4, dcache_size)
161	ASM_LD(%g5, dcache_linesize)
162	CH_DCACHE_FLUSHALL(%g4, %g5, %g6)
163
164	/*
165	 * Turn the D$ back on.
166	 */
167	ldxa	[%g0]ASI_DCU, %g3
168	or	%g3, DCU_DC, %g3
169	stxa	%g3, [%g0]ASI_DCU
170	membar	#Sync
1713:
172	/*
173	 * If we turned off the I$, then flush it and turn it back on.
174	 */
175	ldxa	[%g1 + CH_ERR_TL1_TMP]%asi, %g3
176	andcc	%g3, CH_ERR_TSTATE_IC_ON, %g0
177	bz	%xcc, 4f
178	  nop
179
180	/*
181	 * Flush the I$.
182	 */
183	ASM_LD(%g4, icache_size)
184	ASM_LD(%g5, icache_linesize)
185	CH_ICACHE_FLUSHALL(%g4, %g5, %g6, %g3)
186
187	/*
188	 * Turn the I$ back on.  Changing DCU_IC requires flush.
189	 */
190	ldxa	[%g0]ASI_DCU, %g3
191	or	%g3, DCU_IC, %g3
192	stxa	%g3, [%g0]ASI_DCU
193	flush	%g0
1944:
195
196#ifdef TRAPTRACE
197	/*
198	 * Get current trap trace entry physical pointer.
199	 */
200	CPU_INDEX(%g6, %g5)
201	sll	%g6, TRAPTR_SIZE_SHIFT, %g6
202	set	trap_trace_ctl, %g5
203	add	%g6, %g5, %g6
204	ld	[%g6 + TRAPTR_LIMIT], %g5
205	tst	%g5
206	be	%icc, skip_traptrace
207	  nop
208	ldx	[%g6 + TRAPTR_PBASE], %g5
209	ld	[%g6 + TRAPTR_OFFSET], %g4
210	add	%g5, %g4, %g5
211
212	/*
213	 * Create trap trace entry.
214	 */
215	rd	%asi, %g7
216	wr	%g0, TRAPTR_ASI, %asi
217	rd	STICK, %g4
218	stxa	%g4, [%g5 + TRAP_ENT_TICK]%asi
219	rdpr	%tl, %g4
220	stha	%g4, [%g5 + TRAP_ENT_TL]%asi
221	rdpr	%tt, %g4
222	stha	%g4, [%g5 + TRAP_ENT_TT]%asi
223	rdpr	%tpc, %g4
224	stna	%g4, [%g5 + TRAP_ENT_TPC]%asi
225	rdpr	%tstate, %g4
226	stxa	%g4, [%g5 + TRAP_ENT_TSTATE]%asi
227	stna	%sp, [%g5 + TRAP_ENT_SP]%asi
228	stna	%g0, [%g5 + TRAP_ENT_TR]%asi
229	wr	%g0, %g7, %asi
230	ldxa	[%g1 + CH_ERR_TL1_SDW_AFAR]%asi, %g3
231	ldxa	[%g1 + CH_ERR_TL1_SDW_AFSR]%asi, %g4
232	wr	%g0, TRAPTR_ASI, %asi
233	stna	%g3, [%g5 + TRAP_ENT_F1]%asi
234	stna	%g4, [%g5 + TRAP_ENT_F2]%asi
235	wr	%g0, %g7, %asi
236	ldxa	[%g1 + CH_ERR_TL1_AFAR]%asi, %g3
237	ldxa	[%g1 + CH_ERR_TL1_AFSR]%asi, %g4
238	wr	%g0, TRAPTR_ASI, %asi
239	stna	%g3, [%g5 + TRAP_ENT_F3]%asi
240	stna	%g4, [%g5 + TRAP_ENT_F4]%asi
241	wr	%g0, %g7, %asi
242
243	/*
244	 * Advance trap trace pointer.
245	 */
246	ld	[%g6 + TRAPTR_OFFSET], %g5
247	ld	[%g6 + TRAPTR_LIMIT], %g4
248	st	%g5, [%g6 + TRAPTR_LAST_OFFSET]
249	add	%g5, TRAP_ENT_SIZE, %g5
250	sub	%g4, TRAP_ENT_SIZE, %g4
251	cmp	%g5, %g4
252	movge	%icc, 0, %g5
253	st	%g5, [%g6 + TRAPTR_OFFSET]
254skip_traptrace:
255#endif	/* TRAPTRACE */
256
257	/*
258	 * If nesting count is not zero, skip all the AFSR/AFAR
259	 * handling and just do the necessary cache-flushing.
260	 */
261	ldxa	[%g1 + CH_ERR_TL1_NEST_CNT]%asi, %g2
262	brnz	%g2, 6f
263	  nop
264
265	/*
266	 * If a UCU followed by a WDU has occurred go ahead and panic
267	 * since a UE will occur (on the retry) before the UCU and WDU
268	 * messages are enqueued.
269	 */
270	ldxa	[%g1 + CH_ERR_TL1_AFSR]%asi, %g3
271	set	1, %g4
272	sllx	%g4, C_AFSR_UCU_SHIFT, %g4
273	btst	%g4, %g3		! UCU in original AFSR?
274	bz	%xcc, 6f
275	  nop
276	ldxa	[%g0]ASI_AFSR, %g4	! current AFSR
277	or	%g3, %g4, %g3		! %g3 = original + current AFSR
278	set	1, %g4
279	sllx	%g4, C_AFSR_WDU_SHIFT, %g4
280	btst	%g4, %g3		! WDU in original or current AFSR?
281	bnz	%xcc, fecc_tl1_err
282	  nop
283
2846:
285	/*
286	 * We fall into this macro if we've successfully logged the error in
287	 * the ch_err_tl1_data structure and want the PIL15 softint to pick
288	 * it up and log it.  %g1 must point to the ch_err_tl1_data structure.
289	 * Restores the %g registers and issues retry.
290	 */
291	CH_ERR_TL1_EXIT;
292
293	/*
294	 * Establish panic exit label.
295	 */
296	CH_ERR_TL1_PANIC_EXIT(fecc_tl1_err);
297
298	SET_SIZE(fast_ecc_tl1_err)
299
300
301/*
302 * scrubphys - Pass in the aligned physical memory address
303 * that you want to scrub, along with the ecache set size.
304 *
305 *	1) Displacement flush the E$ line corresponding to %addr.
306 *	   The first ldxa guarantees that the %addr is no longer in
307 *	   M, O, or E (goes to I or S (if instruction fetch also happens).
308 *	2) "Write" the data using a CAS %addr,%g0,%g0.
309 *	   The casxa guarantees a transition from I to M or S to M.
310 *	3) Displacement flush the E$ line corresponding to %addr.
311 *	   The second ldxa pushes the M line out of the ecache, into the
312 *	   writeback buffers, on the way to memory.
313 *	4) The "membar #Sync" pushes the cache line out of the writeback
314 *	   buffers onto the bus, on the way to dram finally.
315 *
316 * This is a modified version of the algorithm suggested by Gary Lauterbach.
317 * In theory the CAS %addr,%g0,%g0 is supposed to mark the addr's cache line
318 * as modified, but then we found out that for spitfire, if it misses in the
319 * E$ it will probably install as an M, but if it hits in the E$, then it
320 * will stay E, if the store doesn't happen. So the first displacement flush
321 * should ensure that the CAS will miss in the E$.  Arrgh.
322 */
323	ENTRY(scrubphys)
324	rdpr	%pstate, %o4
325	andn	%o4, PSTATE_IE | PSTATE_AM, %o5
326	wrpr	%o5, %g0, %pstate	! clear IE, AM bits
327
328	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
329	casxa	[%o0]ASI_MEM, %g0, %g0
330	ECACHE_REFLUSH_LINE(%o1, %o2, %o3)
331
332	wrpr	%g0, %o4, %pstate	! restore earlier pstate register value
333
334	retl
335	membar	#Sync			! move the data out of the load buffer
336	SET_SIZE(scrubphys)
337
338
339/*
340 * clearphys - Pass in the physical memory address of the checkblock
341 * that you want to push out, cleared with a recognizable pattern,
342 * from the ecache.
343 *
344 * To ensure that the ecc gets recalculated after the bad data is cleared,
345 * we must write out enough data to fill the w$ line (64 bytes). So we read
346 * in an entire ecache subblock's worth of data, and write it back out.
347 * Then we overwrite the 16 bytes of bad data with the pattern.
348 */
349	ENTRY(clearphys)
350	/* turn off IE, AM bits */
351	rdpr	%pstate, %o4
352	andn	%o4, PSTATE_IE | PSTATE_AM, %o5
353	wrpr	%o5, %g0, %pstate
354
355	/* turn off NCEEN */
356	ldxa	[%g0]ASI_ESTATE_ERR, %o5
357	andn	%o5, EN_REG_NCEEN, %o3
358	stxa	%o3, [%g0]ASI_ESTATE_ERR
359	membar	#Sync
360
361	/* align address passed with 64 bytes subblock size */
362	mov	CH_ECACHE_SUBBLK_SIZE, %o2
363	andn	%o0, (CH_ECACHE_SUBBLK_SIZE - 1), %g1
364
365	/* move the good data into the W$ */
3661:
367	subcc	%o2, 8, %o2
368	ldxa	[%g1 + %o2]ASI_MEM, %g2
369	bge	1b
370	  stxa	%g2, [%g1 + %o2]ASI_MEM
371
372	/* now overwrite the bad data */
373	setx	0xbadecc00badecc01, %g1, %g2
374	stxa	%g2, [%o0]ASI_MEM
375	mov	8, %g1
376	stxa	%g2, [%o0 + %g1]ASI_MEM
377
378	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
379	casxa	[%o0]ASI_MEM, %g0, %g0
380	ECACHE_REFLUSH_LINE(%o1, %o2, %o3)
381
382	/* clear the AFSR */
383	ldxa	[%g0]ASI_AFSR, %o1
384	stxa	%o1, [%g0]ASI_AFSR
385	membar	#Sync
386
387	/* turn NCEEN back on */
388	stxa	%o5, [%g0]ASI_ESTATE_ERR
389	membar	#Sync
390
391	/* return and re-enable IE and AM */
392	retl
393	  wrpr	%g0, %o4, %pstate
394	SET_SIZE(clearphys)
395
396
397/*
398 * Cheetah Ecache displacement flush the specified line from the E$
399 *
400 * Register usage:
401 *	%o0 - 64 bit physical address for flushing
402 *	%o1 - Ecache set size
403 */
404	ENTRY(ecache_flush_line)
405
406	ECACHE_FLUSH_LINE(%o0, %o1, %o2, %o3)
407
408	retl
409	  nop
410	SET_SIZE(ecache_flush_line)
411
412/*
413 * This routine will not be called in Cheetah systems.
414 */
415	ENTRY(flush_ipb)
416	retl
417	nop
418	SET_SIZE(flush_ipb)
419
420