xref: /illumos-gate/usr/src/uts/sun4/io/px/px_msiq.c (revision d17daf0b)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * px_msiq.c
28  */
29 
30 #include <sys/types.h>
31 #include <sys/kmem.h>
32 #include <sys/conf.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/sysmacros.h>
36 #include <sys/machsystm.h>	/* intr_dist_add */
37 #include <sys/modctl.h>
38 #include <sys/disp.h>
39 #include <sys/stat.h>
40 #include <sys/ddi_impldefs.h>
41 #include "px_obj.h"
42 
43 static int px_msiq_get_props(px_t *px_p);
44 
45 /*
46  * px_msiq_attach()
47  */
48 int
px_msiq_attach(px_t * px_p)49 px_msiq_attach(px_t *px_p)
50 {
51 	px_ib_t		*ib_p = px_p->px_ib_p;
52 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
53 	int		qcnt, i, ret = DDI_SUCCESS;
54 
55 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_attach\n");
56 
57 	/*
58 	 * Check for all MSIQ related properties and
59 	 * save all information.
60 	 *
61 	 * Avaialble MSIQs and its properties.
62 	 */
63 	if (px_msiq_get_props(px_p) != DDI_SUCCESS)
64 		return (DDI_FAILURE);
65 
66 	/*
67 	 * 10% of available MSIQs are reserved for the PCIe messages.
68 	 * Around 90% of available MSIQs are reserved for the MSI/Xs.
69 	 */
70 	msiq_state_p->msiq_msg_qcnt = howmany(msiq_state_p->msiq_cnt, 10);
71 
72 	qcnt = MIN(msiq_state_p->msiq_msg_qcnt, px_max_msiq_msgs);
73 	msiq_state_p->msiq_msg_qcnt = qcnt = MAX(qcnt, px_min_msiq_msgs);
74 	msiq_state_p->msiq_msi_qcnt = msiq_state_p->msiq_cnt - qcnt;
75 
76 	msiq_state_p->msiq_1st_msi_qid = msiq_state_p->msiq_1st_msiq_id;
77 	msiq_state_p->msiq_1st_msg_qid = msiq_state_p->msiq_1st_msiq_id +
78 	    msiq_state_p->msiq_msi_qcnt;
79 
80 	msiq_state_p->msiq_p = kmem_zalloc(msiq_state_p->msiq_cnt *
81 	    sizeof (px_msiq_t), KM_SLEEP);
82 
83 	mutex_enter(&ib_p->ib_ino_lst_mutex);
84 	for (i = 0; i < msiq_state_p->msiq_cnt; i++) {
85 		msiq_state_p->msiq_p[i].msiq_id =
86 		    msiq_state_p->msiq_1st_msiq_id + i;
87 		msiq_state_p->msiq_p[i].msiq_refcnt = 0;
88 		msiq_state_p->msiq_p[i].msiq_state = MSIQ_STATE_FREE;
89 		(void) px_ib_alloc_ino(ib_p, px_msiqid_to_devino(px_p,
90 		    msiq_state_p->msiq_p[i].msiq_id));
91 	}
92 
93 	msiq_state_p->msiq_redist_flag = B_TRUE;
94 	mutex_exit(&ib_p->ib_ino_lst_mutex);
95 
96 	if ((ret = px_lib_msiq_init(px_p->px_dip)) != DDI_SUCCESS)
97 		px_msiq_detach(px_p);
98 
99 	return (ret);
100 }
101 
102 /*
103  * px_msiq_detach()
104  */
105 void
px_msiq_detach(px_t * px_p)106 px_msiq_detach(px_t *px_p)
107 {
108 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
109 
110 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_detach\n");
111 
112 	if (px_lib_msiq_fini(px_p->px_dip) != DDI_SUCCESS) {
113 		DBG(DBG_MSIQ, px_p->px_dip,
114 		    "px_lib_msiq_fini: failed\n");
115 	}
116 
117 	kmem_free(msiq_state_p->msiq_p,
118 	    msiq_state_p->msiq_cnt * sizeof (px_msiq_t));
119 
120 	bzero(msiq_state_p, sizeof (px_msiq_state_t));
121 }
122 
123 /*
124  * px_msiq_resume()
125  */
126 void
px_msiq_resume(px_t * px_p)127 px_msiq_resume(px_t *px_p)
128 {
129 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
130 	int		i;
131 
132 	for (i = 0; i < msiq_state_p->msiq_cnt; i++) {
133 		(void) px_lib_msiq_gethead(px_p->px_dip,
134 		    msiq_state_p->msiq_p[i].msiq_id,
135 		    &msiq_state_p->msiq_p[i].msiq_curr_head_index);
136 		msiq_state_p->msiq_p[i].msiq_new_head_index = 0;
137 		msiq_state_p->msiq_p[i].msiq_recs2process = 0;
138 	}
139 }
140 
141 /*
142  * px_msiq_alloc()
143  */
144 int
px_msiq_alloc(px_t * px_p,msiq_rec_type_t rec_type,msgcode_t msg_code,msiqid_t * msiq_id_p)145 px_msiq_alloc(px_t *px_p, msiq_rec_type_t rec_type, msgcode_t msg_code,
146     msiqid_t *msiq_id_p)
147 {
148 	px_ib_t		*ib_p = px_p->px_ib_p;
149 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
150 	msiqid_t	first_msiq_id;
151 	uint_t		msiq_cnt;
152 	ushort_t	least_refcnt;
153 	int		i;
154 
155 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_alloc\n");
156 
157 	ASSERT(MUTEX_HELD(&ib_p->ib_ino_lst_mutex));
158 
159 	if (rec_type == MSG_REC) {
160 		/*
161 		 * The first MSG EQ is dedicated to PCIE_MSG_CODE_ERR_COR
162 		 * messages. All other messages will be spread across
163 		 * the remaining MSG EQs.
164 		 */
165 		first_msiq_id = msiq_state_p->msiq_1st_msg_qid;
166 
167 		if (msg_code == PCIE_MSG_CODE_ERR_COR) {
168 			msiq_state_p->msiq_p[first_msiq_id].msiq_state =
169 			    MSIQ_STATE_INUSE;
170 
171 			(void) px_lib_msiq_gethead(px_p->px_dip, first_msiq_id,
172 			    &msiq_state_p->msiq_p[first_msiq_id].
173 			    msiq_curr_head_index);
174 
175 			*msiq_id_p =
176 			    msiq_state_p->msiq_p[first_msiq_id].msiq_id;
177 
178 			msiq_state_p->msiq_p[first_msiq_id].msiq_refcnt++;
179 
180 			DBG(DBG_MSIQ, px_p->px_dip,
181 			    "px_msiq_alloc: msiq_id 0x%x\n", *msiq_id_p);
182 
183 			return (DDI_SUCCESS);
184 		}
185 
186 		/* Jump past the first/dedicated EQ */
187 		first_msiq_id++;
188 		msiq_cnt = msiq_state_p->msiq_msg_qcnt - 1;
189 	} else {
190 		msiq_cnt = msiq_state_p->msiq_msi_qcnt;
191 		first_msiq_id = msiq_state_p->msiq_1st_msi_qid;
192 	}
193 
194 	*msiq_id_p = first_msiq_id;
195 	least_refcnt = msiq_state_p->msiq_p[first_msiq_id].msiq_refcnt;
196 
197 	/* Allocate MSIQs */
198 	for (i = first_msiq_id; i < (first_msiq_id + msiq_cnt); i++) {
199 		if (msiq_state_p->msiq_p[i].msiq_state == MSIQ_STATE_FREE) {
200 			msiq_state_p->msiq_p[i].msiq_state = MSIQ_STATE_INUSE;
201 			(void) px_lib_msiq_gethead(px_p->px_dip, i,
202 			    &msiq_state_p->msiq_p[i].msiq_curr_head_index);
203 			*msiq_id_p = msiq_state_p->msiq_p[i].msiq_id;
204 			break;
205 		}
206 
207 		if (least_refcnt > msiq_state_p->msiq_p[i].msiq_refcnt) {
208 			*msiq_id_p = msiq_state_p->msiq_p[i].msiq_id;
209 			least_refcnt = msiq_state_p->msiq_p[i].msiq_refcnt;
210 		}
211 	}
212 
213 	msiq_state_p->msiq_p[*msiq_id_p].msiq_refcnt++;
214 
215 	DBG(DBG_MSIQ, px_p->px_dip,
216 	    "px_msiq_alloc: msiq_id 0x%x\n", *msiq_id_p);
217 
218 	return (DDI_SUCCESS);
219 }
220 
221 /*
222  * px_msiq_alloc_based_on_cpuid()
223  */
224 int
px_msiq_alloc_based_on_cpuid(px_t * px_p,msiq_rec_type_t rec_type,cpuid_t cpuid,msiqid_t * msiq_id_p)225 px_msiq_alloc_based_on_cpuid(px_t *px_p, msiq_rec_type_t rec_type,
226     cpuid_t cpuid, msiqid_t *msiq_id_p)
227 {
228 	px_ib_t		*ib_p = px_p->px_ib_p;
229 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
230 	msiqid_t	first_msiq_id, free_msiq_id;
231 	uint_t		msiq_cnt;
232 	ushort_t	least_refcnt;
233 	px_ino_t	*ino_p;
234 	int		i;
235 
236 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_alloc_based_on_cpuid: "
237 	    "cpuid 0x%x\n", cpuid);
238 
239 	ASSERT(MUTEX_HELD(&ib_p->ib_ino_lst_mutex));
240 
241 	if (rec_type == MSG_REC) {
242 		msiq_cnt = msiq_state_p->msiq_msg_qcnt;
243 		first_msiq_id = msiq_state_p->msiq_1st_msg_qid;
244 	} else {
245 		msiq_cnt = msiq_state_p->msiq_msi_qcnt;
246 		first_msiq_id = msiq_state_p->msiq_1st_msi_qid;
247 	}
248 
249 	*msiq_id_p = free_msiq_id = (msiqid_t)-1;
250 	least_refcnt = (ushort_t)-1;
251 
252 	/* Allocate MSIQs */
253 	for (i = first_msiq_id; i < (first_msiq_id + msiq_cnt); i++) {
254 		ino_p = px_ib_locate_ino(ib_p, px_msiqid_to_devino(px_p, i));
255 
256 		if ((ino_p->ino_cpuid == cpuid) &&
257 		    (least_refcnt > msiq_state_p->msiq_p[i].msiq_refcnt)) {
258 			*msiq_id_p = msiq_state_p->msiq_p[i].msiq_id;
259 			least_refcnt = msiq_state_p->msiq_p[i].msiq_refcnt;
260 		}
261 
262 		if ((*msiq_id_p == -1) && (free_msiq_id == -1) &&
263 		    (msiq_state_p->msiq_p[i].msiq_state == MSIQ_STATE_FREE))
264 			free_msiq_id = msiq_state_p->msiq_p[i].msiq_id;
265 	}
266 
267 	if (*msiq_id_p == -1) {
268 		if (free_msiq_id == -1) {
269 			DBG(DBG_MSIQ, px_p->px_dip,
270 			    "px_msiq_alloc_based_on_cpuid: No EQ is available "
271 			    "for CPU 0x%x\n", cpuid);
272 			return (DDI_EINVAL);
273 		}
274 
275 		*msiq_id_p = free_msiq_id;
276 		ino_p = px_ib_locate_ino(ib_p,
277 		    px_msiqid_to_devino(px_p, *msiq_id_p));
278 		ino_p->ino_cpuid = ino_p->ino_default_cpuid = cpuid;
279 	}
280 
281 	if (msiq_state_p->msiq_p[*msiq_id_p].msiq_state == MSIQ_STATE_FREE) {
282 		msiq_state_p->msiq_p[*msiq_id_p].msiq_state = MSIQ_STATE_INUSE;
283 		(void) px_lib_msiq_gethead(px_p->px_dip, *msiq_id_p,
284 		    &msiq_state_p->msiq_p[*msiq_id_p].msiq_curr_head_index);
285 	}
286 
287 	msiq_state_p->msiq_p[*msiq_id_p].msiq_refcnt++;
288 
289 	DBG(DBG_MSIQ, px_p->px_dip,
290 	    "px_msiq_alloc_based_on_cpuid: msiq_id 0x%x\n", *msiq_id_p);
291 
292 	return (DDI_SUCCESS);
293 }
294 
295 /*
296  * px_msiq_free()
297  */
298 int
px_msiq_free(px_t * px_p,msiqid_t msiq_id)299 px_msiq_free(px_t *px_p, msiqid_t msiq_id)
300 {
301 	px_ib_t		*ib_p = px_p->px_ib_p;
302 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
303 
304 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_free: msiq_id 0x%x", msiq_id);
305 
306 	ASSERT(MUTEX_HELD(&ib_p->ib_ino_lst_mutex));
307 
308 	if ((msiq_id < msiq_state_p->msiq_1st_msiq_id) || (msiq_id >=
309 	    (msiq_state_p->msiq_1st_msiq_id + msiq_state_p->msiq_cnt))) {
310 		DBG(DBG_MSIQ, px_p->px_dip,
311 		    "px_msiq_free: Invalid msiq_id 0x%x", msiq_id);
312 		return (DDI_FAILURE);
313 	}
314 
315 	if (--msiq_state_p->msiq_p[msiq_id].msiq_refcnt == 0)
316 		msiq_state_p->msiq_p[msiq_id].msiq_state = MSIQ_STATE_FREE;
317 
318 	return (DDI_SUCCESS);
319 }
320 
321 /*
322  * px_msiq_redist()
323  */
324 void
px_msiq_redist(px_t * px_p)325 px_msiq_redist(px_t *px_p)
326 {
327 	px_ib_t		*ib_p = px_p->px_ib_p;
328 	px_msiq_state_t	*msiq_state_p = &ib_p->ib_msiq_state;
329 	px_ino_t	*ino_p;
330 	int		i;
331 
332 	ASSERT(MUTEX_HELD(&ib_p->ib_ino_lst_mutex));
333 
334 	if (msiq_state_p->msiq_redist_flag == B_FALSE)
335 		return;
336 
337 	for (i = 0; i < msiq_state_p->msiq_cnt; i++) {
338 		ino_p = px_ib_locate_ino(ib_p,
339 		    px_msiqid_to_devino(px_p, msiq_state_p->msiq_p[i].msiq_id));
340 
341 		if (ino_p) {
342 			ino_p->ino_cpuid = ino_p->ino_default_cpuid =
343 			    intr_dist_cpuid();
344 
345 			DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_redist: "
346 			    "sysino 0x%llx current cpuid 0x%x "
347 			    "default cpuid 0x%x\n", ino_p->ino_sysino,
348 			    ino_p->ino_cpuid, ino_p->ino_default_cpuid);
349 		}
350 	}
351 
352 	msiq_state_p->msiq_redist_flag = B_FALSE;
353 }
354 
355 /*
356  * px_msiqid_to_devino()
357  */
358 devino_t
px_msiqid_to_devino(px_t * px_p,msiqid_t msiq_id)359 px_msiqid_to_devino(px_t *px_p, msiqid_t msiq_id)
360 {
361 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
362 	devino_t	devino;
363 
364 	devino = msiq_state_p->msiq_1st_devino +
365 	    msiq_id - msiq_state_p->msiq_1st_msiq_id;
366 
367 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiqid_to_devino: "
368 	    "msiq_id 0x%x devino 0x%x\n", msiq_id, devino);
369 
370 	return (devino);
371 }
372 
373 /*
374  * px_devino_to_msiqid()
375  */
376 msiqid_t
px_devino_to_msiqid(px_t * px_p,devino_t devino)377 px_devino_to_msiqid(px_t *px_p, devino_t devino)
378 {
379 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
380 	msiqid_t	msiq_id;
381 
382 	msiq_id = msiq_state_p->msiq_1st_msiq_id +
383 	    devino - msiq_state_p->msiq_1st_devino;
384 
385 	DBG(DBG_MSIQ, px_p->px_dip, "px_devino_to_msiq: "
386 	    "devino 0x%x msiq_id 0x%x\n", devino, msiq_id);
387 
388 	return (msiq_id);
389 }
390 
391 /*
392  * px_msiq_get_props()
393  */
394 static int
px_msiq_get_props(px_t * px_p)395 px_msiq_get_props(px_t *px_p)
396 {
397 	px_msiq_state_t	*msiq_state_p = &px_p->px_ib_p->ib_msiq_state;
398 	int		length = sizeof (int);
399 	char		*valuep = NULL;
400 	int		ret;
401 
402 	DBG(DBG_MSIQ, px_p->px_dip, "px_msiq_get_props\n");
403 
404 	/* #msi-eqs */
405 	msiq_state_p->msiq_cnt = ddi_getprop(DDI_DEV_T_ANY, px_p->px_dip,
406 	    DDI_PROP_DONTPASS, "#msi-eqs", 0);
407 
408 	DBG(DBG_MSIQ, px_p->px_dip, "msiq_cnt=%d\n", msiq_state_p->msiq_cnt);
409 
410 	/* msi-eq-size */
411 	msiq_state_p->msiq_rec_cnt = ddi_getprop(DDI_DEV_T_ANY, px_p->px_dip,
412 	    DDI_PROP_DONTPASS, "msi-eq-size", 0);
413 
414 	DBG(DBG_MSIQ, px_p->px_dip, "msiq_rec_cnt=%d\n",
415 	    msiq_state_p->msiq_rec_cnt);
416 
417 	if ((msiq_state_p->msiq_cnt == 0) || (msiq_state_p->msiq_rec_cnt == 0))
418 		return (DDI_FAILURE);
419 
420 	/* msi-eq-to-devino: msi-eq#, devino# fields */
421 	ret = ddi_prop_op(DDI_DEV_T_ANY, px_p->px_dip, PROP_LEN_AND_VAL_ALLOC,
422 	    DDI_PROP_DONTPASS, "msi-eq-to-devino", (caddr_t)&valuep, &length);
423 
424 	/*
425 	 * NOTE:
426 	 * On sun4u PCIe systems, the msi-eq-to-devino property is broken and
427 	 * these systems defines this property as msi-eq-devino.
428 	 */
429 	if (ret == DDI_PROP_NOT_FOUND) {
430 		DBG(DBG_MSIQ, px_p->px_dip, "msi-eq-to-devino is not found\n");
431 		ret = ddi_prop_op(DDI_DEV_T_ANY, px_p->px_dip,
432 		    PROP_LEN_AND_VAL_ALLOC, DDI_PROP_DONTPASS, "msi-eq-devino",
433 		    (caddr_t)&valuep, &length);
434 	}
435 
436 	if (ret != DDI_PROP_SUCCESS) {
437 		return (DDI_FAILURE);
438 	}
439 
440 	msiq_state_p->msiq_1st_msiq_id =
441 	    ((px_msi_eq_to_devino_t *)valuep)->msi_eq_no;
442 	msiq_state_p->msiq_1st_devino =
443 	    ((px_msi_eq_to_devino_t *)valuep)->devino_no;
444 
445 	DBG(DBG_MSIQ, px_p->px_dip, "msiq_1st_msiq_id=%d\n",
446 	    msiq_state_p->msiq_1st_msiq_id);
447 
448 	DBG(DBG_MSIQ, px_p->px_dip, "msiq_1st_devino=%d\n",
449 	    msiq_state_p->msiq_1st_devino);
450 
451 	kmem_free(valuep, (size_t)length);
452 	return (DDI_SUCCESS);
453 }
454