xref: /illumos-gate/usr/src/uts/common/io/chxge/com/tp.c (revision 2d6eb4a5)
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 /*
23  * Copyright (C) 2003-2005 Chelsio Communications.  All rights reserved.
24  */
25 
26 #include "common.h"
27 #include "regs.h"
28 #include "tp.h"
29 #ifdef CONFIG_CHELSIO_T1_1G
30 #include "fpga_defs.h"
31 #endif
32 
33 struct petp {
34 	adapter_t *adapter;
35 };
36 
37 /* Pause deadlock avoidance parameters */
38 #define DROP_MSEC 16
39 #define DROP_PKTS_CNT  1
40 
41 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
42 
pm_num_pages(u32 size,u32 pg_size)43 static inline u32 pm_num_pages(u32 size, u32 pg_size)
44 {
45 	u32 num = size / pg_size;
46 	num -= num % 24;
47 	return num;
48 }
49 
tp_pm_configure(adapter_t * adapter,struct tp_params * p)50 static void tp_pm_configure(adapter_t *adapter, struct tp_params *p)
51 {
52 	u32 num;
53 
54 	num = pm_num_pages(p->pm_size - p->pm_rx_base, p->pm_rx_pg_size);
55 	if (p->pm_rx_num_pgs > num)
56 		p->pm_rx_num_pgs = num;
57 
58 	num = pm_num_pages(p->pm_rx_base - p->pm_tx_base, p->pm_tx_pg_size);
59 	if (p->pm_tx_num_pgs > num)
60 		p->pm_tx_num_pgs = num;
61 
62 	t1_write_reg_4(adapter, A_TP_PM_SIZE, p->pm_size);
63 	t1_write_reg_4(adapter, A_TP_PM_RX_BASE, p->pm_rx_base);
64 	t1_write_reg_4(adapter, A_TP_PM_TX_BASE, p->pm_tx_base);
65 	t1_write_reg_4(adapter, A_TP_PM_DEFRAG_BASE, p->pm_size);
66 	t1_write_reg_4(adapter, A_TP_PM_RX_PG_SIZE, p->pm_rx_pg_size);
67 	t1_write_reg_4(adapter, A_TP_PM_RX_MAX_PGS, p->pm_rx_num_pgs);
68 	t1_write_reg_4(adapter, A_TP_PM_TX_PG_SIZE, p->pm_tx_pg_size);
69 	t1_write_reg_4(adapter, A_TP_PM_TX_MAX_PGS, p->pm_tx_num_pgs);
70 }
71 
tp_cm_configure(adapter_t * adapter,u32 cm_size)72 static void tp_cm_configure(adapter_t *adapter, u32 cm_size)
73 {
74 	u32 mm_base = (cm_size >> 1);
75 	u32 mm_sub_size = (cm_size >> 5);
76 
77 	t1_write_reg_4(adapter, A_TP_CM_SIZE, cm_size);
78 	t1_write_reg_4(adapter, A_TP_CM_MM_BASE, mm_base);
79 	t1_write_reg_4(adapter, A_TP_CM_TIMER_BASE, (cm_size >> 2) * 3);
80 	t1_write_reg_4(adapter, A_TP_CM_MM_P_FLST_BASE,
81 		       mm_base + 5 * mm_sub_size);
82 	t1_write_reg_4(adapter, A_TP_CM_MM_TX_FLST_BASE,
83 		       mm_base + 6 * mm_sub_size);
84 	t1_write_reg_4(adapter, A_TP_CM_MM_RX_FLST_BASE,
85 		       mm_base + 7 * mm_sub_size);
86 	t1_write_reg_4(adapter, A_TP_CM_MM_MAX_P, 0x40000);
87 }
88 
tp_delayed_ack_ticks(adapter_t * adap,unsigned int tp_clk)89 static unsigned int tp_delayed_ack_ticks(adapter_t *adap, unsigned int tp_clk)
90 {
91 	u32 tr = t1_read_reg_4(adap, A_TP_TIMER_RESOLUTION);
92 
93 	return tp_clk /	(1 << G_DELAYED_ACK_TIMER_RESOLUTION(tr));
94 }
95 
t1_tp_ticks_per_sec(adapter_t * adap,unsigned int tp_clk)96 static unsigned int t1_tp_ticks_per_sec(adapter_t *adap, unsigned int tp_clk)
97 {
98 	u32 tr = t1_read_reg_4(adap, A_TP_TIMER_RESOLUTION);
99 
100 	return tp_clk /	(1 << G_GENERIC_TIMER_RESOLUTION(tr));
101 }
102 
tp_set_tcp_time_params(adapter_t * adapter,unsigned int tp_clk)103 static void tp_set_tcp_time_params(adapter_t *adapter, unsigned int tp_clk)
104 {
105 	u32 tps = t1_tp_ticks_per_sec(adapter, tp_clk);
106 	u32 tp_scnt;
107 
108 #define SECONDS * tps
109 	t1_write_reg_4(adapter, A_TP_2MSL, (1 SECONDS)/2);
110 	t1_write_reg_4(adapter, A_TP_RXT_MIN, (1 SECONDS)/4);
111 	t1_write_reg_4(adapter, A_TP_RXT_MAX, 64 SECONDS);
112 	t1_write_reg_4(adapter, A_TP_PERS_MIN, (1 SECONDS)/2);
113 	t1_write_reg_4(adapter, A_TP_PERS_MAX, 64 SECONDS);
114 	t1_write_reg_4(adapter, A_TP_KEEP_IDLE, 7200 SECONDS);
115 	t1_write_reg_4(adapter, A_TP_KEEP_INTVL, 75 SECONDS);
116 	t1_write_reg_4(adapter, A_TP_INIT_SRTT, 3 SECONDS);
117 	t1_write_reg_4(adapter, A_TP_FINWAIT2_TIME, 60 SECONDS);
118 	t1_write_reg_4(adapter, A_TP_FAST_FINWAIT2_TIME, 3 SECONDS);
119 #undef SECONDS
120 
121 	/* Set Retransmission shift max */
122 	tp_scnt = t1_read_reg_4(adapter, A_TP_SHIFT_CNT);
123 	tp_scnt &= (~V_RETRANSMISSION_MAX(0x3f));
124 	tp_scnt |= V_RETRANSMISSION_MAX(14);
125 	t1_write_reg_4(adapter, A_TP_SHIFT_CNT, tp_scnt);
126 
127 	/* Set DACK timer to 200ms */
128 	t1_write_reg_4(adapter, A_TP_DACK_TIME,
129 		       tp_delayed_ack_ticks(adapter, tp_clk) / 5);
130 }
131 
t1_tp_set_coalescing_size(struct petp * tp,unsigned int size)132 int t1_tp_set_coalescing_size(struct petp *tp, unsigned int size)
133 {
134 	u32 val;
135 
136 	if (size > TP_MAX_RX_COALESCING_SIZE)
137 		return -EINVAL;
138 
139 	val = t1_read_reg_4(tp->adapter, A_TP_PARA_REG3);
140 
141 	if (tp->adapter->params.nports > 1)
142 		size = 9904;
143 
144 	if (size) {
145 		u32 v = t1_is_T1B(tp->adapter) ? 0 : V_MAX_RX_SIZE(size);
146 
147 		/* Set coalescing size. */
148 		t1_write_reg_4(tp->adapter, A_TP_PARA_REG2,
149 			       V_RX_COALESCE_SIZE(size) | v);
150 
151 		val |= (F_RX_COALESCING_PSH_DELIVER | F_RX_COALESCING_ENABLE);
152 	} else
153 		val &= ~F_RX_COALESCING_ENABLE;
154 
155 	t1_write_reg_4(tp->adapter, A_TP_PARA_REG3, val);
156 	return 0;
157 }
158 
t1_tp_get_mib_statistics(adapter_t * adap,struct tp_mib_statistics * tps)159 void t1_tp_get_mib_statistics(adapter_t *adap, struct tp_mib_statistics *tps)
160 {
161 	u32 *data = (u32 *)tps;
162 	int i;
163 
164 	t1_write_reg_4(adap, A_TP_MIB_INDEX, 0);
165 
166 	for (i = 0; i < sizeof(*tps) / sizeof(u32); i++)
167 		*data++ = t1_read_reg_4(adap, A_TP_MIB_DATA);
168 }
169 #endif
170 
tp_init(adapter_t * ap,const struct tp_params * p,unsigned int tp_clk)171 static void tp_init(adapter_t *ap, const struct tp_params *p,
172 		    unsigned int tp_clk)
173 {
174 	if (t1_is_asic(ap)) {
175 		u32 val;
176 
177 		val = F_TP_IN_CSPI_CPL | F_TP_IN_CSPI_CHECK_IP_CSUM |
178 		      F_TP_IN_CSPI_CHECK_TCP_CSUM | F_TP_IN_ESPI_ETHERNET;
179 		if (!p->pm_size)
180 			val |= F_OFFLOAD_DISABLE;
181 		else
182 			val |= F_TP_IN_ESPI_CHECK_IP_CSUM |
183 				F_TP_IN_ESPI_CHECK_TCP_CSUM;
184 		t1_write_reg_4(ap, A_TP_IN_CONFIG, val);
185 		t1_write_reg_4(ap, A_TP_OUT_CONFIG, F_TP_OUT_CSPI_CPL |
186 			       F_TP_OUT_ESPI_ETHERNET |
187 			       F_TP_OUT_ESPI_GENERATE_IP_CSUM |
188 			       F_TP_OUT_ESPI_GENERATE_TCP_CSUM);
189 		t1_write_reg_4(ap, A_TP_GLOBAL_CONFIG, V_IP_TTL(64) |
190 			       F_PATH_MTU /* IP DF bit */ |
191 			       V_5TUPLE_LOOKUP(p->use_5tuple_mode) |
192 			       V_SYN_COOKIE_PARAMETER(29));
193 
194                 /*
195                  * Enable pause frame deadlock prevention.
196                  */
197                 if (is_T2(ap) && ap->params.nports > 1) {
198                         u32 drop_ticks = DROP_MSEC * (tp_clk / 1000);
199 
200                         t1_write_reg_4(ap, A_TP_TX_DROP_CONFIG,
201                                        F_ENABLE_TX_DROP | F_ENABLE_TX_ERROR |
202                                        V_DROP_TICKS_CNT(drop_ticks) |
203                                        V_NUM_PKTS_DROPPED(DROP_PKTS_CNT));
204                 }
205 
206 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
207 		t1_write_reg_4(ap, A_TP_GLOBAL_RX_CREDITS, 0xffffffff);
208 		val = V_WINDOW_SCALE(1) | F_MSS | V_DEFAULT_PEER_MSS(576);
209 
210 		/* We don't want timestamps for T204, otherwise we don't know
211 		 * the MSS.
212 		 */
213 		if (ap->params.nports == 1)
214 			val |= V_TIMESTAMP(1);
215 		t1_write_reg_4(ap, A_TP_TCP_OPTIONS, val);
216 		t1_write_reg_4(ap, A_TP_DACK_CONFIG, V_DACK_MSS_SELECTOR(1) |
217 			       F_DACK_AUTO_CAREFUL | V_DACK_MODE(1));
218 		t1_write_reg_4(ap, A_TP_BACKOFF0, 0x3020100);
219 		t1_write_reg_4(ap, A_TP_BACKOFF1, 0x7060504);
220 		t1_write_reg_4(ap, A_TP_BACKOFF2, 0xb0a0908);
221 		t1_write_reg_4(ap, A_TP_BACKOFF3, 0xf0e0d0c);
222 
223 		/* We do scheduling in software for T204, increase the cong.
224 		 * window to avoid TP holding on to payload longer than we
225 		 * expect.
226 		 */
227 		if (ap->params.nports == 1)
228 			t1_write_reg_4(ap, A_TP_PARA_REG0, 0xd1269324);
229 		else
230 			t1_write_reg_4(ap, A_TP_PARA_REG0, 0xd6269324);
231 		t1_write_reg_4(ap, A_TP_SYNC_TIME_HI, 0);
232 		t1_write_reg_4(ap, A_TP_SYNC_TIME_LO, 0);
233 		t1_write_reg_4(ap, A_TP_INT_ENABLE, 0);
234 		t1_write_reg_4(ap, A_TP_CM_FC_MODE, 0);   /* Enable CM cache */
235 		t1_write_reg_4(ap, A_TP_PC_CONGESTION_CNTL, 0x6186);
236 
237 		/*
238 		 * Calculate the time between modulation events, which affects
239 		 * both the Tx and Rx pipelines.  Larger values force the Tx
240 		 * pipeline to wait before processing modulation events, thus
241 		 * allowing Rx to use the pipeline.  A really small delay can
242 		 * starve the Rx side from accessing the pipeline.
243 		 *
244 		 * A balanced value is optimal.  This is roughly 9us per 1G.
245 		 * The Tx needs a low delay time for handling a lot of small
246 		 * packets. Too big of a delay could cause Tx not to achieve
247 		 * line rate.
248 		 */
249 		val = (9 * tp_clk) / 1000000;
250 		/* adjust for multiple ports */
251 		if (ap->params.nports > 1) {
252 			val = 0;
253 		}
254 		if (is_10G(ap))               /* adjust for 10G */
255 			val /= 10;
256 		/*
257 		 * Bit 0 must be 0 to keep the timer insertion property.
258 		 */
259 		t1_write_reg_4(ap, A_TP_TIMER_SEPARATOR, val & ~1);
260 
261 		t1_write_reg_4(ap, A_TP_TIMER_RESOLUTION, 0xF0011);
262 		tp_set_tcp_time_params(ap, tp_clk);
263 
264 		/* PR3229 */
265 		if (is_T2(ap)) {
266 			val = t1_read_reg_4(ap, A_TP_PC_CONFIG);
267 			val |= V_DIS_TX_FILL_WIN_PUSH(1);
268 			t1_write_reg_4(ap, A_TP_PC_CONFIG, val);
269 		}
270 
271 #ifdef CONFIG_CHELSIO_T1_1G
272 	} else {    /* FPGA */
273 		t1_write_reg_4(ap, A_TP_TIMER_RESOLUTION, 0xD000A);
274 #endif
275 #endif
276 	}
277 }
278 
t1_tp_destroy(struct petp * tp)279 void t1_tp_destroy(struct petp *tp)
280 {
281 	t1_os_free((void *)tp, sizeof(*tp));
282 }
283 
t1_tp_create(adapter_t * adapter,struct tp_params * p)284 struct petp * __devinit t1_tp_create(adapter_t *adapter, struct tp_params *p)
285 {
286 	struct petp *tp = t1_os_malloc_wait_zero(sizeof(*tp));
287 	if (!tp)
288 		return NULL;
289 
290 	tp->adapter = adapter;
291 
292 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
293 	if (p->pm_size) {                     /* Default PM partitioning */
294 		p->pm_rx_base = p->pm_size >> 1;
295 #ifdef TDI_SUPPORT
296 		p->pm_tx_base = 2048 * 1024;    /* reserve 2 MByte for REGION MAP */
297 #else
298 		p->pm_tx_base = 64 * 1024;    /* reserve 64 kbytes for REGION MAP */
299 #endif
300 		p->pm_rx_pg_size = 64 * 1024;
301 
302 		if (adapter->params.nports == 1)
303 			p->pm_tx_pg_size = 64 * 1024;
304 		else
305 			p->pm_tx_pg_size = 16 * 1024;
306 		p->pm_rx_num_pgs = pm_num_pages(p->pm_size - p->pm_rx_base,
307 						p->pm_rx_pg_size);
308 		p->pm_tx_num_pgs = pm_num_pages(p->pm_rx_base - p->pm_tx_base,
309 						p->pm_tx_pg_size);
310 	}
311 #endif
312 	return tp;
313 }
314 
t1_tp_intr_enable(struct petp * tp)315 void t1_tp_intr_enable(struct petp *tp)
316 {
317 	u32 tp_intr = t1_read_reg_4(tp->adapter, A_PL_ENABLE);
318 
319 #ifdef CONFIG_CHELSIO_T1_1G
320 	if (!t1_is_asic(tp->adapter)) {
321 		/* FPGA */
322 		t1_write_reg_4(tp->adapter, FPGA_TP_ADDR_INTERRUPT_ENABLE,
323 			       0xffffffff);
324 		t1_write_reg_4(tp->adapter, A_PL_ENABLE,
325 			       tp_intr | FPGA_PCIX_INTERRUPT_TP);
326 	} else
327 #endif
328 	{
329 		/* We don't use any TP interrupts */
330 		t1_write_reg_4(tp->adapter, A_TP_INT_ENABLE, 0);
331 		t1_write_reg_4(tp->adapter, A_PL_ENABLE,
332 			       tp_intr | F_PL_INTR_TP);
333 	}
334 }
335 
t1_tp_intr_disable(struct petp * tp)336 void t1_tp_intr_disable(struct petp *tp)
337 {
338 	u32 tp_intr = t1_read_reg_4(tp->adapter, A_PL_ENABLE);
339 
340 #ifdef CONFIG_CHELSIO_T1_1G
341 	if (!t1_is_asic(tp->adapter)) {
342 		/* FPGA */
343 		t1_write_reg_4(tp->adapter, FPGA_TP_ADDR_INTERRUPT_ENABLE, 0);
344 		t1_write_reg_4(tp->adapter, A_PL_ENABLE,
345 			       tp_intr & ~FPGA_PCIX_INTERRUPT_TP);
346 	} else
347 #endif
348 	{
349 		t1_write_reg_4(tp->adapter, A_TP_INT_ENABLE, 0);
350 		t1_write_reg_4(tp->adapter, A_PL_ENABLE,
351 			       tp_intr & ~F_PL_INTR_TP);
352 	}
353 }
354 
t1_tp_intr_clear(struct petp * tp)355 void t1_tp_intr_clear(struct petp *tp)
356 {
357 #ifdef CONFIG_CHELSIO_T1_1G
358 	if (!t1_is_asic(tp->adapter)) {
359 		t1_write_reg_4(tp->adapter, FPGA_TP_ADDR_INTERRUPT_CAUSE,
360 			       0xffffffff);
361 		t1_write_reg_4(tp->adapter, A_PL_CAUSE, FPGA_PCIX_INTERRUPT_TP);
362 		return;
363 	}
364 #endif
365 	t1_write_reg_4(tp->adapter, A_TP_INT_CAUSE, 0xffffffff);
366 	t1_write_reg_4(tp->adapter, A_PL_CAUSE, F_PL_INTR_TP);
367 }
368 
t1_tp_intr_handler(struct petp * tp)369 int t1_tp_intr_handler(struct petp *tp)
370 {
371 	u32 cause;
372 
373 #ifdef CONFIG_CHELSIO_T1_1G
374 	/* FPGA doesn't support TP interrupts. */
375 	if (!t1_is_asic(tp->adapter))
376 		return 1;
377 #endif
378 
379 	cause = t1_read_reg_4(tp->adapter, A_TP_INT_CAUSE);
380 	t1_write_reg_4(tp->adapter, A_TP_INT_CAUSE, cause);
381 	return 0;
382 }
383 
set_csum_offload(struct petp * tp,u32 csum_bit,int enable)384 static void set_csum_offload(struct petp *tp, u32 csum_bit, int enable)
385 {
386 	u32 val = t1_read_reg_4(tp->adapter, A_TP_GLOBAL_CONFIG);
387 
388 	if (enable)
389 		val |= csum_bit;
390 	else
391 		val &= ~csum_bit;
392 	t1_write_reg_4(tp->adapter, A_TP_GLOBAL_CONFIG, val);
393 }
394 
t1_tp_set_ip_checksum_offload(struct petp * tp,int enable)395 void t1_tp_set_ip_checksum_offload(struct petp *tp, int enable)
396 {
397 	set_csum_offload(tp, F_IP_CSUM, enable);
398 }
399 
t1_tp_set_udp_checksum_offload(struct petp * tp,int enable)400 void t1_tp_set_udp_checksum_offload(struct petp *tp, int enable)
401 {
402 	set_csum_offload(tp, F_UDP_CSUM, enable);
403 }
404 
t1_tp_set_tcp_checksum_offload(struct petp * tp,int enable)405 void t1_tp_set_tcp_checksum_offload(struct petp *tp, int enable)
406 {
407 	set_csum_offload(tp, F_TCP_CSUM, enable);
408 }
409 
410 /*
411  * Initialize TP state.  tp_params contains initial settings for some TP
412  * parameters, particularly the one-time PM and CM settings.
413  */
t1_tp_reset(struct petp * tp,struct tp_params * p,unsigned int tp_clk)414 int t1_tp_reset(struct petp *tp, struct tp_params *p, unsigned int tp_clk)
415 {
416 	int busy = 0;
417 	adapter_t *adapter = tp->adapter;
418 
419 	tp_init(adapter, p, tp_clk);
420 #ifdef CONFIG_CHELSIO_T1_OFFLOAD
421 	if (p->pm_size) {
422 		tp_pm_configure(adapter, p);
423 		tp_cm_configure(adapter, p->cm_size);
424 
425 		t1_write_reg_4(adapter, A_TP_RESET, F_CM_MEMMGR_INIT);
426 		busy = t1_wait_op_done(adapter, A_TP_RESET, F_CM_MEMMGR_INIT,
427 				0, 1000, 5);
428 	}
429 #endif
430 	if (!busy)
431 		t1_write_reg_4(adapter, A_TP_RESET, F_TP_RESET);
432 	else
433 		CH_ERR("%s: TP initialization timed out\n",
434 		       adapter_name(adapter));
435 	return busy;
436 }
437