]> git.karo-electronics.de Git - karo-tx-linux.git/blob - drivers/net/ethernet/sfc/ef10.c
Merge remote-tracking branch 'slave-dma/next'
[karo-tx-linux.git] / drivers / net / ethernet / sfc / ef10.c
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9
10 #include "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include <linux/in.h>
18 #include <linux/jhash.h>
19 #include <linux/wait.h>
20 #include <linux/workqueue.h>
21
22 /* Hardware control for EF10 architecture including 'Huntington'. */
23
24 #define EFX_EF10_DRVGEN_EV              7
25 enum {
26         EFX_EF10_TEST = 1,
27         EFX_EF10_REFILL,
28 };
29
30 /* The reserved RSS context value */
31 #define EFX_EF10_RSS_CONTEXT_INVALID    0xffffffff
32
33 /* The filter table(s) are managed by firmware and we have write-only
34  * access.  When removing filters we must identify them to the
35  * firmware by a 64-bit handle, but this is too wide for Linux kernel
36  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
37  * be able to tell in advance whether a requested insertion will
38  * replace an existing filter.  Therefore we maintain a software hash
39  * table, which should be at least as large as the hardware hash
40  * table.
41  *
42  * Huntington has a single 8K filter table shared between all filter
43  * types and both ports.
44  */
45 #define HUNT_FILTER_TBL_ROWS 8192
46
47 struct efx_ef10_filter_table {
48 /* The RX match field masks supported by this fw & hw, in order of priority */
49         enum efx_filter_match_flags rx_match_flags[
50                 MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
51         unsigned int rx_match_count;
52
53         struct {
54                 unsigned long spec;     /* pointer to spec plus flag bits */
55 /* BUSY flag indicates that an update is in progress.  STACK_OLD is
56  * used to mark and sweep stack-owned MAC filters.
57  */
58 #define EFX_EF10_FILTER_FLAG_BUSY       1UL
59 #define EFX_EF10_FILTER_FLAG_STACK_OLD  2UL
60 #define EFX_EF10_FILTER_FLAGS           3UL
61                 u64 handle;             /* firmware handle */
62         } *entry;
63         wait_queue_head_t waitq;
64 /* Shadow of net_device address lists, guarded by mac_lock */
65 #define EFX_EF10_FILTER_STACK_UC_MAX    32
66 #define EFX_EF10_FILTER_STACK_MC_MAX    256
67         struct {
68                 u8 addr[ETH_ALEN];
69                 u16 id;
70         } stack_uc_list[EFX_EF10_FILTER_STACK_UC_MAX],
71           stack_mc_list[EFX_EF10_FILTER_STACK_MC_MAX];
72         int stack_uc_count;             /* negative for PROMISC */
73         int stack_mc_count;             /* negative for PROMISC/ALLMULTI */
74 };
75
76 /* An arbitrary search limit for the software hash table */
77 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
78
79 static void efx_ef10_rx_push_indir_table(struct efx_nic *efx);
80 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
81 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
82
83 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
84 {
85         efx_dword_t reg;
86
87         efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
88         return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
89                 EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
90 }
91
92 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
93 {
94         return resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]);
95 }
96
97 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
98 {
99         MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
100         struct efx_ef10_nic_data *nic_data = efx->nic_data;
101         size_t outlen;
102         int rc;
103
104         BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
105
106         rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
107                           outbuf, sizeof(outbuf), &outlen);
108         if (rc)
109                 return rc;
110         if (outlen < sizeof(outbuf)) {
111                 netif_err(efx, drv, efx->net_dev,
112                           "unable to read datapath firmware capabilities\n");
113                 return -EIO;
114         }
115
116         nic_data->datapath_caps =
117                 MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
118
119         if (!(nic_data->datapath_caps &
120               (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
121                 netif_err(efx, drv, efx->net_dev,
122                           "current firmware does not support TSO\n");
123                 return -ENODEV;
124         }
125
126         if (!(nic_data->datapath_caps &
127               (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
128                 netif_err(efx, probe, efx->net_dev,
129                           "current firmware does not support an RX prefix\n");
130                 return -ENODEV;
131         }
132
133         return 0;
134 }
135
136 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
137 {
138         MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
139         int rc;
140
141         rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
142                           outbuf, sizeof(outbuf), NULL);
143         if (rc)
144                 return rc;
145         rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
146         return rc > 0 ? rc : -ERANGE;
147 }
148
149 static int efx_ef10_get_mac_address(struct efx_nic *efx, u8 *mac_address)
150 {
151         MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
152         size_t outlen;
153         int rc;
154
155         BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
156
157         rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
158                           outbuf, sizeof(outbuf), &outlen);
159         if (rc)
160                 return rc;
161         if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
162                 return -EIO;
163
164         memcpy(mac_address,
165                MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE), ETH_ALEN);
166         return 0;
167 }
168
169 static int efx_ef10_probe(struct efx_nic *efx)
170 {
171         struct efx_ef10_nic_data *nic_data;
172         int i, rc;
173
174         /* We can have one VI for each 8K region.  However we need
175          * multiple TX queues per channel.
176          */
177         efx->max_channels =
178                 min_t(unsigned int,
179                       EFX_MAX_CHANNELS,
180                       resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]) /
181                       (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
182         BUG_ON(efx->max_channels == 0);
183
184         nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
185         if (!nic_data)
186                 return -ENOMEM;
187         efx->nic_data = nic_data;
188
189         rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
190                                   8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
191         if (rc)
192                 goto fail1;
193
194         /* Get the MC's warm boot count.  In case it's rebooting right
195          * now, be prepared to retry.
196          */
197         i = 0;
198         for (;;) {
199                 rc = efx_ef10_get_warm_boot_count(efx);
200                 if (rc >= 0)
201                         break;
202                 if (++i == 5)
203                         goto fail2;
204                 ssleep(1);
205         }
206         nic_data->warm_boot_count = rc;
207
208         nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
209
210         /* In case we're recovering from a crash (kexec), we want to
211          * cancel any outstanding request by the previous user of this
212          * function.  We send a special message using the least
213          * significant bits of the 'high' (doorbell) register.
214          */
215         _efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
216
217         rc = efx_mcdi_init(efx);
218         if (rc)
219                 goto fail2;
220
221         /* Reset (most) configuration for this function */
222         rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
223         if (rc)
224                 goto fail3;
225
226         /* Enable event logging */
227         rc = efx_mcdi_log_ctrl(efx, true, false, 0);
228         if (rc)
229                 goto fail3;
230
231         rc = efx_ef10_init_datapath_caps(efx);
232         if (rc < 0)
233                 goto fail3;
234
235         efx->rx_packet_len_offset =
236                 ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
237
238         rc = efx_mcdi_port_get_number(efx);
239         if (rc < 0)
240                 goto fail3;
241         efx->port_num = rc;
242
243         rc = efx_ef10_get_mac_address(efx, efx->net_dev->perm_addr);
244         if (rc)
245                 goto fail3;
246
247         rc = efx_ef10_get_sysclk_freq(efx);
248         if (rc < 0)
249                 goto fail3;
250         efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
251
252         /* Check whether firmware supports bug 35388 workaround */
253         rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true);
254         if (rc == 0)
255                 nic_data->workaround_35388 = true;
256         else if (rc != -ENOSYS && rc != -ENOENT)
257                 goto fail3;
258         netif_dbg(efx, probe, efx->net_dev,
259                   "workaround for bug 35388 is %sabled\n",
260                   nic_data->workaround_35388 ? "en" : "dis");
261
262         rc = efx_mcdi_mon_probe(efx);
263         if (rc)
264                 goto fail3;
265
266         return 0;
267
268 fail3:
269         efx_mcdi_fini(efx);
270 fail2:
271         efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
272 fail1:
273         kfree(nic_data);
274         efx->nic_data = NULL;
275         return rc;
276 }
277
278 static int efx_ef10_free_vis(struct efx_nic *efx)
279 {
280         int rc = efx_mcdi_rpc(efx, MC_CMD_FREE_VIS, NULL, 0, NULL, 0, NULL);
281
282         /* -EALREADY means nothing to free, so ignore */
283         if (rc == -EALREADY)
284                 rc = 0;
285         return rc;
286 }
287
288 static void efx_ef10_remove(struct efx_nic *efx)
289 {
290         struct efx_ef10_nic_data *nic_data = efx->nic_data;
291         int rc;
292
293         efx_mcdi_mon_remove(efx);
294
295         /* This needs to be after efx_ptp_remove_channel() with no filters */
296         efx_ef10_rx_free_indir_table(efx);
297
298         rc = efx_ef10_free_vis(efx);
299         WARN_ON(rc != 0);
300
301         efx_mcdi_fini(efx);
302         efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
303         kfree(nic_data);
304 }
305
306 static int efx_ef10_alloc_vis(struct efx_nic *efx,
307                               unsigned int min_vis, unsigned int max_vis)
308 {
309         MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
310         MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
311         struct efx_ef10_nic_data *nic_data = efx->nic_data;
312         size_t outlen;
313         int rc;
314
315         MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
316         MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
317         rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
318                           outbuf, sizeof(outbuf), &outlen);
319         if (rc != 0)
320                 return rc;
321
322         if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
323                 return -EIO;
324
325         netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
326                   MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
327
328         nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
329         nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
330         return 0;
331 }
332
333 static int efx_ef10_dimension_resources(struct efx_nic *efx)
334 {
335         unsigned int n_vis =
336                 max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
337
338         return efx_ef10_alloc_vis(efx, n_vis, n_vis);
339 }
340
341 static int efx_ef10_init_nic(struct efx_nic *efx)
342 {
343         struct efx_ef10_nic_data *nic_data = efx->nic_data;
344         int rc;
345
346         if (nic_data->must_check_datapath_caps) {
347                 rc = efx_ef10_init_datapath_caps(efx);
348                 if (rc)
349                         return rc;
350                 nic_data->must_check_datapath_caps = false;
351         }
352
353         if (nic_data->must_realloc_vis) {
354                 /* We cannot let the number of VIs change now */
355                 rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
356                                         nic_data->n_allocated_vis);
357                 if (rc)
358                         return rc;
359                 nic_data->must_realloc_vis = false;
360         }
361
362         efx_ef10_rx_push_indir_table(efx);
363         return 0;
364 }
365
366 static int efx_ef10_map_reset_flags(u32 *flags)
367 {
368         enum {
369                 EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
370                                    ETH_RESET_SHARED_SHIFT),
371                 EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
372                                   ETH_RESET_OFFLOAD | ETH_RESET_MAC |
373                                   ETH_RESET_PHY | ETH_RESET_MGMT) <<
374                                  ETH_RESET_SHARED_SHIFT)
375         };
376
377         /* We assume for now that our PCI function is permitted to
378          * reset everything.
379          */
380
381         if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
382                 *flags &= ~EF10_RESET_MC;
383                 return RESET_TYPE_WORLD;
384         }
385
386         if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
387                 *flags &= ~EF10_RESET_PORT;
388                 return RESET_TYPE_ALL;
389         }
390
391         /* no invisible reset implemented */
392
393         return -EINVAL;
394 }
395
396 #define EF10_DMA_STAT(ext_name, mcdi_name)                      \
397         [EF10_STAT_ ## ext_name] =                              \
398         { #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
399 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)                \
400         [EF10_STAT_ ## int_name] =                              \
401         { NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
402 #define EF10_OTHER_STAT(ext_name)                               \
403         [EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
404
405 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
406         EF10_DMA_STAT(tx_bytes, TX_BYTES),
407         EF10_DMA_STAT(tx_packets, TX_PKTS),
408         EF10_DMA_STAT(tx_pause, TX_PAUSE_PKTS),
409         EF10_DMA_STAT(tx_control, TX_CONTROL_PKTS),
410         EF10_DMA_STAT(tx_unicast, TX_UNICAST_PKTS),
411         EF10_DMA_STAT(tx_multicast, TX_MULTICAST_PKTS),
412         EF10_DMA_STAT(tx_broadcast, TX_BROADCAST_PKTS),
413         EF10_DMA_STAT(tx_lt64, TX_LT64_PKTS),
414         EF10_DMA_STAT(tx_64, TX_64_PKTS),
415         EF10_DMA_STAT(tx_65_to_127, TX_65_TO_127_PKTS),
416         EF10_DMA_STAT(tx_128_to_255, TX_128_TO_255_PKTS),
417         EF10_DMA_STAT(tx_256_to_511, TX_256_TO_511_PKTS),
418         EF10_DMA_STAT(tx_512_to_1023, TX_512_TO_1023_PKTS),
419         EF10_DMA_STAT(tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
420         EF10_DMA_STAT(tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
421         EF10_DMA_STAT(rx_bytes, RX_BYTES),
422         EF10_DMA_INVIS_STAT(rx_bytes_minus_good_bytes, RX_BAD_BYTES),
423         EF10_OTHER_STAT(rx_good_bytes),
424         EF10_OTHER_STAT(rx_bad_bytes),
425         EF10_DMA_STAT(rx_packets, RX_PKTS),
426         EF10_DMA_STAT(rx_good, RX_GOOD_PKTS),
427         EF10_DMA_STAT(rx_bad, RX_BAD_FCS_PKTS),
428         EF10_DMA_STAT(rx_pause, RX_PAUSE_PKTS),
429         EF10_DMA_STAT(rx_control, RX_CONTROL_PKTS),
430         EF10_DMA_STAT(rx_unicast, RX_UNICAST_PKTS),
431         EF10_DMA_STAT(rx_multicast, RX_MULTICAST_PKTS),
432         EF10_DMA_STAT(rx_broadcast, RX_BROADCAST_PKTS),
433         EF10_DMA_STAT(rx_lt64, RX_UNDERSIZE_PKTS),
434         EF10_DMA_STAT(rx_64, RX_64_PKTS),
435         EF10_DMA_STAT(rx_65_to_127, RX_65_TO_127_PKTS),
436         EF10_DMA_STAT(rx_128_to_255, RX_128_TO_255_PKTS),
437         EF10_DMA_STAT(rx_256_to_511, RX_256_TO_511_PKTS),
438         EF10_DMA_STAT(rx_512_to_1023, RX_512_TO_1023_PKTS),
439         EF10_DMA_STAT(rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
440         EF10_DMA_STAT(rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
441         EF10_DMA_STAT(rx_gtjumbo, RX_GTJUMBO_PKTS),
442         EF10_DMA_STAT(rx_bad_gtjumbo, RX_JABBER_PKTS),
443         EF10_DMA_STAT(rx_overflow, RX_OVERFLOW_PKTS),
444         EF10_DMA_STAT(rx_align_error, RX_ALIGN_ERROR_PKTS),
445         EF10_DMA_STAT(rx_length_error, RX_LENGTH_ERROR_PKTS),
446         EF10_DMA_STAT(rx_nodesc_drops, RX_NODESC_DROPS),
447         EF10_DMA_STAT(rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
448         EF10_DMA_STAT(rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
449         EF10_DMA_STAT(rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
450         EF10_DMA_STAT(rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
451         EF10_DMA_STAT(rx_pm_trunc_qbb, PM_TRUNC_QBB),
452         EF10_DMA_STAT(rx_pm_discard_qbb, PM_DISCARD_QBB),
453         EF10_DMA_STAT(rx_pm_discard_mapping, PM_DISCARD_MAPPING),
454         EF10_DMA_STAT(rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
455         EF10_DMA_STAT(rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
456         EF10_DMA_STAT(rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
457         EF10_DMA_STAT(rx_dp_emerg_fetch, RXDP_EMERGENCY_FETCH_CONDITIONS),
458         EF10_DMA_STAT(rx_dp_emerg_wait, RXDP_EMERGENCY_WAIT_CONDITIONS),
459 };
460
461 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_tx_bytes) |           \
462                                (1ULL << EF10_STAT_tx_packets) |         \
463                                (1ULL << EF10_STAT_tx_pause) |           \
464                                (1ULL << EF10_STAT_tx_unicast) |         \
465                                (1ULL << EF10_STAT_tx_multicast) |       \
466                                (1ULL << EF10_STAT_tx_broadcast) |       \
467                                (1ULL << EF10_STAT_rx_bytes) |           \
468                                (1ULL << EF10_STAT_rx_bytes_minus_good_bytes) | \
469                                (1ULL << EF10_STAT_rx_good_bytes) |      \
470                                (1ULL << EF10_STAT_rx_bad_bytes) |       \
471                                (1ULL << EF10_STAT_rx_packets) |         \
472                                (1ULL << EF10_STAT_rx_good) |            \
473                                (1ULL << EF10_STAT_rx_bad) |             \
474                                (1ULL << EF10_STAT_rx_pause) |           \
475                                (1ULL << EF10_STAT_rx_control) |         \
476                                (1ULL << EF10_STAT_rx_unicast) |         \
477                                (1ULL << EF10_STAT_rx_multicast) |       \
478                                (1ULL << EF10_STAT_rx_broadcast) |       \
479                                (1ULL << EF10_STAT_rx_lt64) |            \
480                                (1ULL << EF10_STAT_rx_64) |              \
481                                (1ULL << EF10_STAT_rx_65_to_127) |       \
482                                (1ULL << EF10_STAT_rx_128_to_255) |      \
483                                (1ULL << EF10_STAT_rx_256_to_511) |      \
484                                (1ULL << EF10_STAT_rx_512_to_1023) |     \
485                                (1ULL << EF10_STAT_rx_1024_to_15xx) |    \
486                                (1ULL << EF10_STAT_rx_15xx_to_jumbo) |   \
487                                (1ULL << EF10_STAT_rx_gtjumbo) |         \
488                                (1ULL << EF10_STAT_rx_bad_gtjumbo) |     \
489                                (1ULL << EF10_STAT_rx_overflow) |        \
490                                (1ULL << EF10_STAT_rx_nodesc_drops))
491
492 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
493  * switchable port we do not expose these because they might not
494  * include all the packets they should.
495  */
496 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_tx_control) |       \
497                                  (1ULL << EF10_STAT_tx_lt64) |          \
498                                  (1ULL << EF10_STAT_tx_64) |            \
499                                  (1ULL << EF10_STAT_tx_65_to_127) |     \
500                                  (1ULL << EF10_STAT_tx_128_to_255) |    \
501                                  (1ULL << EF10_STAT_tx_256_to_511) |    \
502                                  (1ULL << EF10_STAT_tx_512_to_1023) |   \
503                                  (1ULL << EF10_STAT_tx_1024_to_15xx) |  \
504                                  (1ULL << EF10_STAT_tx_15xx_to_jumbo))
505
506 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
507  * switchable port we do expose these because the errors will otherwise
508  * be silent.
509  */
510 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_rx_align_error) |  \
511                                   (1ULL << EF10_STAT_rx_length_error))
512
513 /* These statistics are only provided if the firmware supports the
514  * capability PM_AND_RXDP_COUNTERS.
515  */
516 #define HUNT_PM_AND_RXDP_STAT_MASK (                                    \
517         (1ULL << EF10_STAT_rx_pm_trunc_bb_overflow) |                   \
518         (1ULL << EF10_STAT_rx_pm_discard_bb_overflow) |                 \
519         (1ULL << EF10_STAT_rx_pm_trunc_vfifo_full) |                    \
520         (1ULL << EF10_STAT_rx_pm_discard_vfifo_full) |                  \
521         (1ULL << EF10_STAT_rx_pm_trunc_qbb) |                           \
522         (1ULL << EF10_STAT_rx_pm_discard_qbb) |                         \
523         (1ULL << EF10_STAT_rx_pm_discard_mapping) |                     \
524         (1ULL << EF10_STAT_rx_dp_q_disabled_packets) |                  \
525         (1ULL << EF10_STAT_rx_dp_di_dropped_packets) |                  \
526         (1ULL << EF10_STAT_rx_dp_streaming_packets) |                   \
527         (1ULL << EF10_STAT_rx_dp_emerg_fetch) |                         \
528         (1ULL << EF10_STAT_rx_dp_emerg_wait))
529
530 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
531 {
532         u64 raw_mask = HUNT_COMMON_STAT_MASK;
533         u32 port_caps = efx_mcdi_phy_get_caps(efx);
534         struct efx_ef10_nic_data *nic_data = efx->nic_data;
535
536         if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
537                 raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
538         else
539                 raw_mask |= HUNT_10G_ONLY_STAT_MASK;
540
541         if (nic_data->datapath_caps &
542             (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
543                 raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
544
545         return raw_mask;
546 }
547
548 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
549 {
550         u64 raw_mask = efx_ef10_raw_stat_mask(efx);
551
552 #if BITS_PER_LONG == 64
553         mask[0] = raw_mask;
554 #else
555         mask[0] = raw_mask & 0xffffffff;
556         mask[1] = raw_mask >> 32;
557 #endif
558 }
559
560 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
561 {
562         DECLARE_BITMAP(mask, EF10_STAT_COUNT);
563
564         efx_ef10_get_stat_mask(efx, mask);
565         return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
566                                       mask, names);
567 }
568
569 static int efx_ef10_try_update_nic_stats(struct efx_nic *efx)
570 {
571         struct efx_ef10_nic_data *nic_data = efx->nic_data;
572         DECLARE_BITMAP(mask, EF10_STAT_COUNT);
573         __le64 generation_start, generation_end;
574         u64 *stats = nic_data->stats;
575         __le64 *dma_stats;
576
577         efx_ef10_get_stat_mask(efx, mask);
578
579         dma_stats = efx->stats_buffer.addr;
580         nic_data = efx->nic_data;
581
582         generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
583         if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
584                 return 0;
585         rmb();
586         efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
587                              stats, efx->stats_buffer.addr, false);
588         rmb();
589         generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
590         if (generation_end != generation_start)
591                 return -EAGAIN;
592
593         /* Update derived statistics */
594         stats[EF10_STAT_rx_good_bytes] =
595                 stats[EF10_STAT_rx_bytes] -
596                 stats[EF10_STAT_rx_bytes_minus_good_bytes];
597         efx_update_diff_stat(&stats[EF10_STAT_rx_bad_bytes],
598                              stats[EF10_STAT_rx_bytes_minus_good_bytes]);
599
600         return 0;
601 }
602
603
604 static size_t efx_ef10_update_stats(struct efx_nic *efx, u64 *full_stats,
605                                     struct rtnl_link_stats64 *core_stats)
606 {
607         DECLARE_BITMAP(mask, EF10_STAT_COUNT);
608         struct efx_ef10_nic_data *nic_data = efx->nic_data;
609         u64 *stats = nic_data->stats;
610         size_t stats_count = 0, index;
611         int retry;
612
613         efx_ef10_get_stat_mask(efx, mask);
614
615         /* If we're unlucky enough to read statistics during the DMA, wait
616          * up to 10ms for it to finish (typically takes <500us)
617          */
618         for (retry = 0; retry < 100; ++retry) {
619                 if (efx_ef10_try_update_nic_stats(efx) == 0)
620                         break;
621                 udelay(100);
622         }
623
624         if (full_stats) {
625                 for_each_set_bit(index, mask, EF10_STAT_COUNT) {
626                         if (efx_ef10_stat_desc[index].name) {
627                                 *full_stats++ = stats[index];
628                                 ++stats_count;
629                         }
630                 }
631         }
632
633         if (core_stats) {
634                 core_stats->rx_packets = stats[EF10_STAT_rx_packets];
635                 core_stats->tx_packets = stats[EF10_STAT_tx_packets];
636                 core_stats->rx_bytes = stats[EF10_STAT_rx_bytes];
637                 core_stats->tx_bytes = stats[EF10_STAT_tx_bytes];
638                 core_stats->rx_dropped = stats[EF10_STAT_rx_nodesc_drops];
639                 core_stats->multicast = stats[EF10_STAT_rx_multicast];
640                 core_stats->rx_length_errors =
641                         stats[EF10_STAT_rx_gtjumbo] +
642                         stats[EF10_STAT_rx_length_error];
643                 core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
644                 core_stats->rx_frame_errors = stats[EF10_STAT_rx_align_error];
645                 core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
646                 core_stats->rx_errors = (core_stats->rx_length_errors +
647                                          core_stats->rx_crc_errors +
648                                          core_stats->rx_frame_errors);
649         }
650
651         return stats_count;
652 }
653
654 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
655 {
656         struct efx_nic *efx = channel->efx;
657         unsigned int mode, value;
658         efx_dword_t timer_cmd;
659
660         if (channel->irq_moderation) {
661                 mode = 3;
662                 value = channel->irq_moderation - 1;
663         } else {
664                 mode = 0;
665                 value = 0;
666         }
667
668         if (EFX_EF10_WORKAROUND_35388(efx)) {
669                 EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
670                                      EFE_DD_EVQ_IND_TIMER_FLAGS,
671                                      ERF_DD_EVQ_IND_TIMER_MODE, mode,
672                                      ERF_DD_EVQ_IND_TIMER_VAL, value);
673                 efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
674                                 channel->channel);
675         } else {
676                 EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
677                                      ERF_DZ_TC_TIMER_VAL, value);
678                 efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
679                                 channel->channel);
680         }
681 }
682
683 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
684 {
685         wol->supported = 0;
686         wol->wolopts = 0;
687         memset(&wol->sopass, 0, sizeof(wol->sopass));
688 }
689
690 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
691 {
692         if (type != 0)
693                 return -EINVAL;
694         return 0;
695 }
696
697 static void efx_ef10_mcdi_request(struct efx_nic *efx,
698                                   const efx_dword_t *hdr, size_t hdr_len,
699                                   const efx_dword_t *sdu, size_t sdu_len)
700 {
701         struct efx_ef10_nic_data *nic_data = efx->nic_data;
702         u8 *pdu = nic_data->mcdi_buf.addr;
703
704         memcpy(pdu, hdr, hdr_len);
705         memcpy(pdu + hdr_len, sdu, sdu_len);
706         wmb();
707
708         /* The hardware provides 'low' and 'high' (doorbell) registers
709          * for passing the 64-bit address of an MCDI request to
710          * firmware.  However the dwords are swapped by firmware.  The
711          * least significant bits of the doorbell are then 0 for all
712          * MCDI requests due to alignment.
713          */
714         _efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
715                     ER_DZ_MC_DB_LWRD);
716         _efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
717                     ER_DZ_MC_DB_HWRD);
718 }
719
720 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
721 {
722         struct efx_ef10_nic_data *nic_data = efx->nic_data;
723         const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
724
725         rmb();
726         return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
727 }
728
729 static void
730 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
731                             size_t offset, size_t outlen)
732 {
733         struct efx_ef10_nic_data *nic_data = efx->nic_data;
734         const u8 *pdu = nic_data->mcdi_buf.addr;
735
736         memcpy(outbuf, pdu + offset, outlen);
737 }
738
739 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
740 {
741         struct efx_ef10_nic_data *nic_data = efx->nic_data;
742         int rc;
743
744         rc = efx_ef10_get_warm_boot_count(efx);
745         if (rc < 0) {
746                 /* The firmware is presumably in the process of
747                  * rebooting.  However, we are supposed to report each
748                  * reboot just once, so we must only do that once we
749                  * can read and store the updated warm boot count.
750                  */
751                 return 0;
752         }
753
754         if (rc == nic_data->warm_boot_count)
755                 return 0;
756
757         nic_data->warm_boot_count = rc;
758
759         /* All our allocations have been reset */
760         nic_data->must_realloc_vis = true;
761         nic_data->must_restore_filters = true;
762         nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
763
764         /* The datapath firmware might have been changed */
765         nic_data->must_check_datapath_caps = true;
766
767         /* MAC statistics have been cleared on the NIC; clear the local
768          * statistic that we update with efx_update_diff_stat().
769          */
770         nic_data->stats[EF10_STAT_rx_bad_bytes] = 0;
771
772         return -EIO;
773 }
774
775 /* Handle an MSI interrupt
776  *
777  * Handle an MSI hardware interrupt.  This routine schedules event
778  * queue processing.  No interrupt acknowledgement cycle is necessary.
779  * Also, we never need to check that the interrupt is for us, since
780  * MSI interrupts cannot be shared.
781  */
782 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
783 {
784         struct efx_msi_context *context = dev_id;
785         struct efx_nic *efx = context->efx;
786
787         netif_vdbg(efx, intr, efx->net_dev,
788                    "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
789
790         if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
791                 /* Note test interrupts */
792                 if (context->index == efx->irq_level)
793                         efx->last_irq_cpu = raw_smp_processor_id();
794
795                 /* Schedule processing of the channel */
796                 efx_schedule_channel_irq(efx->channel[context->index]);
797         }
798
799         return IRQ_HANDLED;
800 }
801
802 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
803 {
804         struct efx_nic *efx = dev_id;
805         bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
806         struct efx_channel *channel;
807         efx_dword_t reg;
808         u32 queues;
809
810         /* Read the ISR which also ACKs the interrupts */
811         efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
812         queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
813
814         if (queues == 0)
815                 return IRQ_NONE;
816
817         if (likely(soft_enabled)) {
818                 /* Note test interrupts */
819                 if (queues & (1U << efx->irq_level))
820                         efx->last_irq_cpu = raw_smp_processor_id();
821
822                 efx_for_each_channel(channel, efx) {
823                         if (queues & 1)
824                                 efx_schedule_channel_irq(channel);
825                         queues >>= 1;
826                 }
827         }
828
829         netif_vdbg(efx, intr, efx->net_dev,
830                    "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
831                    irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
832
833         return IRQ_HANDLED;
834 }
835
836 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
837 {
838         MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
839
840         BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
841
842         MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
843         (void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
844                             inbuf, sizeof(inbuf), NULL, 0, NULL);
845 }
846
847 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
848 {
849         return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
850                                     (tx_queue->ptr_mask + 1) *
851                                     sizeof(efx_qword_t),
852                                     GFP_KERNEL);
853 }
854
855 /* This writes to the TX_DESC_WPTR and also pushes data */
856 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
857                                          const efx_qword_t *txd)
858 {
859         unsigned int write_ptr;
860         efx_oword_t reg;
861
862         write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
863         EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
864         reg.qword[0] = *txd;
865         efx_writeo_page(tx_queue->efx, &reg,
866                         ER_DZ_TX_DESC_UPD, tx_queue->queue);
867 }
868
869 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
870 {
871         MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
872                                                        EFX_BUF_SIZE));
873         MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_TXQ_OUT_LEN);
874         bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
875         size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
876         struct efx_channel *channel = tx_queue->channel;
877         struct efx_nic *efx = tx_queue->efx;
878         size_t inlen, outlen;
879         dma_addr_t dma_addr;
880         efx_qword_t *txd;
881         int rc;
882         int i;
883
884         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
885         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
886         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
887         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
888         MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
889                               INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
890                               INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
891         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
892         MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
893
894         dma_addr = tx_queue->txd.buf.dma_addr;
895
896         netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
897                   tx_queue->queue, entries, (u64)dma_addr);
898
899         for (i = 0; i < entries; ++i) {
900                 MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
901                 dma_addr += EFX_BUF_SIZE;
902         }
903
904         inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
905
906         rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
907                           outbuf, sizeof(outbuf), &outlen);
908         if (rc)
909                 goto fail;
910
911         /* A previous user of this TX queue might have set us up the
912          * bomb by writing a descriptor to the TX push collector but
913          * not the doorbell.  (Each collector belongs to a port, not a
914          * queue or function, so cannot easily be reset.)  We must
915          * attempt to push a no-op descriptor in its place.
916          */
917         tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
918         tx_queue->insert_count = 1;
919         txd = efx_tx_desc(tx_queue, 0);
920         EFX_POPULATE_QWORD_4(*txd,
921                              ESF_DZ_TX_DESC_IS_OPT, true,
922                              ESF_DZ_TX_OPTION_TYPE,
923                              ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
924                              ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
925                              ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
926         tx_queue->write_count = 1;
927         wmb();
928         efx_ef10_push_tx_desc(tx_queue, txd);
929
930         return;
931
932 fail:
933         WARN_ON(true);
934         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
935 }
936
937 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
938 {
939         MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
940         MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_TXQ_OUT_LEN);
941         struct efx_nic *efx = tx_queue->efx;
942         size_t outlen;
943         int rc;
944
945         MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
946                        tx_queue->queue);
947
948         rc = efx_mcdi_rpc(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
949                           outbuf, sizeof(outbuf), &outlen);
950
951         if (rc && rc != -EALREADY)
952                 goto fail;
953
954         return;
955
956 fail:
957         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
958 }
959
960 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
961 {
962         efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
963 }
964
965 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
966 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
967 {
968         unsigned int write_ptr;
969         efx_dword_t reg;
970
971         write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
972         EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
973         efx_writed_page(tx_queue->efx, &reg,
974                         ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
975 }
976
977 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
978 {
979         unsigned int old_write_count = tx_queue->write_count;
980         struct efx_tx_buffer *buffer;
981         unsigned int write_ptr;
982         efx_qword_t *txd;
983
984         BUG_ON(tx_queue->write_count == tx_queue->insert_count);
985
986         do {
987                 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
988                 buffer = &tx_queue->buffer[write_ptr];
989                 txd = efx_tx_desc(tx_queue, write_ptr);
990                 ++tx_queue->write_count;
991
992                 /* Create TX descriptor ring entry */
993                 if (buffer->flags & EFX_TX_BUF_OPTION) {
994                         *txd = buffer->option;
995                 } else {
996                         BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
997                         EFX_POPULATE_QWORD_3(
998                                 *txd,
999                                 ESF_DZ_TX_KER_CONT,
1000                                 buffer->flags & EFX_TX_BUF_CONT,
1001                                 ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1002                                 ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1003                 }
1004         } while (tx_queue->write_count != tx_queue->insert_count);
1005
1006         wmb(); /* Ensure descriptors are written before they are fetched */
1007
1008         if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1009                 txd = efx_tx_desc(tx_queue,
1010                                   old_write_count & tx_queue->ptr_mask);
1011                 efx_ef10_push_tx_desc(tx_queue, txd);
1012                 ++tx_queue->pushes;
1013         } else {
1014                 efx_ef10_notify_tx_desc(tx_queue);
1015         }
1016 }
1017
1018 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context)
1019 {
1020         MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1021         MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1022         size_t outlen;
1023         int rc;
1024
1025         MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1026                        EVB_PORT_ID_ASSIGNED);
1027         MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE,
1028                        MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE);
1029         MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES,
1030                        EFX_MAX_CHANNELS);
1031
1032         rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1033                 outbuf, sizeof(outbuf), &outlen);
1034         if (rc != 0)
1035                 return rc;
1036
1037         if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1038                 return -EIO;
1039
1040         *context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1041
1042         return 0;
1043 }
1044
1045 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1046 {
1047         MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1048         int rc;
1049
1050         MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1051                        context);
1052
1053         rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1054                             NULL, 0, NULL);
1055         WARN_ON(rc != 0);
1056 }
1057
1058 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context)
1059 {
1060         MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1061         MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1062         int i, rc;
1063
1064         MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1065                        context);
1066         BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1067                      MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1068
1069         for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1070                 MCDI_PTR(tablebuf,
1071                          RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1072                                 (u8) efx->rx_indir_table[i];
1073
1074         rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1075                           sizeof(tablebuf), NULL, 0, NULL);
1076         if (rc != 0)
1077                 return rc;
1078
1079         MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1080                        context);
1081         BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1082                      MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1083         for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1084                 MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1085                         efx->rx_hash_key[i];
1086
1087         return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1088                             sizeof(keybuf), NULL, 0, NULL);
1089 }
1090
1091 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1092 {
1093         struct efx_ef10_nic_data *nic_data = efx->nic_data;
1094
1095         if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1096                 efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1097         nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1098 }
1099
1100 static void efx_ef10_rx_push_indir_table(struct efx_nic *efx)
1101 {
1102         struct efx_ef10_nic_data *nic_data = efx->nic_data;
1103         int rc;
1104
1105         netif_dbg(efx, drv, efx->net_dev, "pushing RX indirection table\n");
1106
1107         if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID) {
1108                 rc = efx_ef10_alloc_rss_context(efx, &nic_data->rx_rss_context);
1109                 if (rc != 0)
1110                         goto fail;
1111         }
1112
1113         rc = efx_ef10_populate_rss_table(efx, nic_data->rx_rss_context);
1114         if (rc != 0)
1115                 goto fail;
1116
1117         return;
1118
1119 fail:
1120         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1121 }
1122
1123 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
1124 {
1125         return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
1126                                     (rx_queue->ptr_mask + 1) *
1127                                     sizeof(efx_qword_t),
1128                                     GFP_KERNEL);
1129 }
1130
1131 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
1132 {
1133         MCDI_DECLARE_BUF(inbuf,
1134                          MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1135                                                 EFX_BUF_SIZE));
1136         MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_RXQ_OUT_LEN);
1137         struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1138         size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
1139         struct efx_nic *efx = rx_queue->efx;
1140         size_t inlen, outlen;
1141         dma_addr_t dma_addr;
1142         int rc;
1143         int i;
1144
1145         rx_queue->scatter_n = 0;
1146         rx_queue->scatter_len = 0;
1147
1148         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
1149         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
1150         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
1151         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
1152                        efx_rx_queue_index(rx_queue));
1153         MCDI_POPULATE_DWORD_1(inbuf, INIT_RXQ_IN_FLAGS,
1154                               INIT_RXQ_IN_FLAG_PREFIX, 1);
1155         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
1156         MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1157
1158         dma_addr = rx_queue->rxd.buf.dma_addr;
1159
1160         netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
1161                   efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
1162
1163         for (i = 0; i < entries; ++i) {
1164                 MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
1165                 dma_addr += EFX_BUF_SIZE;
1166         }
1167
1168         inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
1169
1170         rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
1171                           outbuf, sizeof(outbuf), &outlen);
1172         if (rc)
1173                 goto fail;
1174
1175         return;
1176
1177 fail:
1178         WARN_ON(true);
1179         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1180 }
1181
1182 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
1183 {
1184         MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
1185         MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_RXQ_OUT_LEN);
1186         struct efx_nic *efx = rx_queue->efx;
1187         size_t outlen;
1188         int rc;
1189
1190         MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
1191                        efx_rx_queue_index(rx_queue));
1192
1193         rc = efx_mcdi_rpc(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
1194                           outbuf, sizeof(outbuf), &outlen);
1195
1196         if (rc && rc != -EALREADY)
1197                 goto fail;
1198
1199         return;
1200
1201 fail:
1202         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1203 }
1204
1205 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
1206 {
1207         efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
1208 }
1209
1210 /* This creates an entry in the RX descriptor queue */
1211 static inline void
1212 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
1213 {
1214         struct efx_rx_buffer *rx_buf;
1215         efx_qword_t *rxd;
1216
1217         rxd = efx_rx_desc(rx_queue, index);
1218         rx_buf = efx_rx_buffer(rx_queue, index);
1219         EFX_POPULATE_QWORD_2(*rxd,
1220                              ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
1221                              ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
1222 }
1223
1224 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
1225 {
1226         struct efx_nic *efx = rx_queue->efx;
1227         unsigned int write_count;
1228         efx_dword_t reg;
1229
1230         /* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
1231         write_count = rx_queue->added_count & ~7;
1232         if (rx_queue->notified_count == write_count)
1233                 return;
1234
1235         do
1236                 efx_ef10_build_rx_desc(
1237                         rx_queue,
1238                         rx_queue->notified_count & rx_queue->ptr_mask);
1239         while (++rx_queue->notified_count != write_count);
1240
1241         wmb();
1242         EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
1243                              write_count & rx_queue->ptr_mask);
1244         efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
1245                         efx_rx_queue_index(rx_queue));
1246 }
1247
1248 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
1249
1250 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
1251 {
1252         struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1253         MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1254         efx_qword_t event;
1255
1256         EFX_POPULATE_QWORD_2(event,
1257                              ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1258                              ESF_DZ_EV_DATA, EFX_EF10_REFILL);
1259
1260         MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1261
1262         /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1263          * already swapped the data to little-endian order.
1264          */
1265         memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1266                sizeof(efx_qword_t));
1267
1268         efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
1269                            inbuf, sizeof(inbuf), 0,
1270                            efx_ef10_rx_defer_refill_complete, 0);
1271 }
1272
1273 static void
1274 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
1275                                   int rc, efx_dword_t *outbuf,
1276                                   size_t outlen_actual)
1277 {
1278         /* nothing to do */
1279 }
1280
1281 static int efx_ef10_ev_probe(struct efx_channel *channel)
1282 {
1283         return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
1284                                     (channel->eventq_mask + 1) *
1285                                     sizeof(efx_qword_t),
1286                                     GFP_KERNEL);
1287 }
1288
1289 static int efx_ef10_ev_init(struct efx_channel *channel)
1290 {
1291         MCDI_DECLARE_BUF(inbuf,
1292                          MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
1293                                                 EFX_BUF_SIZE));
1294         MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
1295         size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
1296         struct efx_nic *efx = channel->efx;
1297         struct efx_ef10_nic_data *nic_data;
1298         bool supports_rx_merge;
1299         size_t inlen, outlen;
1300         dma_addr_t dma_addr;
1301         int rc;
1302         int i;
1303
1304         nic_data = efx->nic_data;
1305         supports_rx_merge =
1306                 !!(nic_data->datapath_caps &
1307                    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
1308
1309         /* Fill event queue with all ones (i.e. empty events) */
1310         memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1311
1312         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
1313         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
1314         /* INIT_EVQ expects index in vector table, not absolute */
1315         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
1316         MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
1317                               INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
1318                               INIT_EVQ_IN_FLAG_RX_MERGE, 1,
1319                               INIT_EVQ_IN_FLAG_TX_MERGE, 1,
1320                               INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
1321         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
1322                        MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
1323         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
1324         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
1325         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
1326                        MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
1327         MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
1328
1329         dma_addr = channel->eventq.buf.dma_addr;
1330         for (i = 0; i < entries; ++i) {
1331                 MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
1332                 dma_addr += EFX_BUF_SIZE;
1333         }
1334
1335         inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
1336
1337         rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
1338                           outbuf, sizeof(outbuf), &outlen);
1339         if (rc)
1340                 goto fail;
1341
1342         /* IRQ return is ignored */
1343
1344         return 0;
1345
1346 fail:
1347         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1348         return rc;
1349 }
1350
1351 static void efx_ef10_ev_fini(struct efx_channel *channel)
1352 {
1353         MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
1354         MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_EVQ_OUT_LEN);
1355         struct efx_nic *efx = channel->efx;
1356         size_t outlen;
1357         int rc;
1358
1359         MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
1360
1361         rc = efx_mcdi_rpc(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
1362                           outbuf, sizeof(outbuf), &outlen);
1363
1364         if (rc && rc != -EALREADY)
1365                 goto fail;
1366
1367         return;
1368
1369 fail:
1370         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1371 }
1372
1373 static void efx_ef10_ev_remove(struct efx_channel *channel)
1374 {
1375         efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
1376 }
1377
1378 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
1379                                            unsigned int rx_queue_label)
1380 {
1381         struct efx_nic *efx = rx_queue->efx;
1382
1383         netif_info(efx, hw, efx->net_dev,
1384                    "rx event arrived on queue %d labeled as queue %u\n",
1385                    efx_rx_queue_index(rx_queue), rx_queue_label);
1386
1387         efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1388 }
1389
1390 static void
1391 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
1392                              unsigned int actual, unsigned int expected)
1393 {
1394         unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
1395         struct efx_nic *efx = rx_queue->efx;
1396
1397         netif_info(efx, hw, efx->net_dev,
1398                    "dropped %d events (index=%d expected=%d)\n",
1399                    dropped, actual, expected);
1400
1401         efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1402 }
1403
1404 /* partially received RX was aborted. clean up. */
1405 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
1406 {
1407         unsigned int rx_desc_ptr;
1408
1409         WARN_ON(rx_queue->scatter_n == 0);
1410
1411         netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
1412                   "scattered RX aborted (dropping %u buffers)\n",
1413                   rx_queue->scatter_n);
1414
1415         rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
1416
1417         efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
1418                       0, EFX_RX_PKT_DISCARD);
1419
1420         rx_queue->removed_count += rx_queue->scatter_n;
1421         rx_queue->scatter_n = 0;
1422         rx_queue->scatter_len = 0;
1423         ++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
1424 }
1425
1426 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
1427                                     const efx_qword_t *event)
1428 {
1429         unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
1430         unsigned int n_descs, n_packets, i;
1431         struct efx_nic *efx = channel->efx;
1432         struct efx_rx_queue *rx_queue;
1433         bool rx_cont;
1434         u16 flags = 0;
1435
1436         if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1437                 return 0;
1438
1439         /* Basic packet information */
1440         rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
1441         next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
1442         rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
1443         rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
1444         rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
1445
1446         WARN_ON(EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT));
1447
1448         rx_queue = efx_channel_get_rx_queue(channel);
1449
1450         if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
1451                 efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
1452
1453         n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
1454                    ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1455
1456         if (n_descs != rx_queue->scatter_n + 1) {
1457                 /* detect rx abort */
1458                 if (unlikely(n_descs == rx_queue->scatter_n)) {
1459                         WARN_ON(rx_bytes != 0);
1460                         efx_ef10_handle_rx_abort(rx_queue);
1461                         return 0;
1462                 }
1463
1464                 if (unlikely(rx_queue->scatter_n != 0)) {
1465                         /* Scattered packet completions cannot be
1466                          * merged, so something has gone wrong.
1467                          */
1468                         efx_ef10_handle_rx_bad_lbits(
1469                                 rx_queue, next_ptr_lbits,
1470                                 (rx_queue->removed_count +
1471                                  rx_queue->scatter_n + 1) &
1472                                 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1473                         return 0;
1474                 }
1475
1476                 /* Merged completion for multiple non-scattered packets */
1477                 rx_queue->scatter_n = 1;
1478                 rx_queue->scatter_len = 0;
1479                 n_packets = n_descs;
1480                 ++channel->n_rx_merge_events;
1481                 channel->n_rx_merge_packets += n_packets;
1482                 flags |= EFX_RX_PKT_PREFIX_LEN;
1483         } else {
1484                 ++rx_queue->scatter_n;
1485                 rx_queue->scatter_len += rx_bytes;
1486                 if (rx_cont)
1487                         return 0;
1488                 n_packets = 1;
1489         }
1490
1491         if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
1492                 flags |= EFX_RX_PKT_DISCARD;
1493
1494         if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
1495                 channel->n_rx_ip_hdr_chksum_err += n_packets;
1496         } else if (unlikely(EFX_QWORD_FIELD(*event,
1497                                             ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
1498                 channel->n_rx_tcp_udp_chksum_err += n_packets;
1499         } else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
1500                    rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
1501                 flags |= EFX_RX_PKT_CSUMMED;
1502         }
1503
1504         if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
1505                 flags |= EFX_RX_PKT_TCP;
1506
1507         channel->irq_mod_score += 2 * n_packets;
1508
1509         /* Handle received packet(s) */
1510         for (i = 0; i < n_packets; i++) {
1511                 efx_rx_packet(rx_queue,
1512                               rx_queue->removed_count & rx_queue->ptr_mask,
1513                               rx_queue->scatter_n, rx_queue->scatter_len,
1514                               flags);
1515                 rx_queue->removed_count += rx_queue->scatter_n;
1516         }
1517
1518         rx_queue->scatter_n = 0;
1519         rx_queue->scatter_len = 0;
1520
1521         return n_packets;
1522 }
1523
1524 static int
1525 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
1526 {
1527         struct efx_nic *efx = channel->efx;
1528         struct efx_tx_queue *tx_queue;
1529         unsigned int tx_ev_desc_ptr;
1530         unsigned int tx_ev_q_label;
1531         int tx_descs = 0;
1532
1533         if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1534                 return 0;
1535
1536         if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
1537                 return 0;
1538
1539         /* Transmit completion */
1540         tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
1541         tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
1542         tx_queue = efx_channel_get_tx_queue(channel,
1543                                             tx_ev_q_label % EFX_TXQ_TYPES);
1544         tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
1545                     tx_queue->ptr_mask);
1546         efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
1547
1548         return tx_descs;
1549 }
1550
1551 static void
1552 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1553 {
1554         struct efx_nic *efx = channel->efx;
1555         int subcode;
1556
1557         subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
1558
1559         switch (subcode) {
1560         case ESE_DZ_DRV_TIMER_EV:
1561         case ESE_DZ_DRV_WAKE_UP_EV:
1562                 break;
1563         case ESE_DZ_DRV_START_UP_EV:
1564                 /* event queue init complete. ok. */
1565                 break;
1566         default:
1567                 netif_err(efx, hw, efx->net_dev,
1568                           "channel %d unknown driver event type %d"
1569                           " (data " EFX_QWORD_FMT ")\n",
1570                           channel->channel, subcode,
1571                           EFX_QWORD_VAL(*event));
1572
1573         }
1574 }
1575
1576 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
1577                                                    efx_qword_t *event)
1578 {
1579         struct efx_nic *efx = channel->efx;
1580         u32 subcode;
1581
1582         subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
1583
1584         switch (subcode) {
1585         case EFX_EF10_TEST:
1586                 channel->event_test_cpu = raw_smp_processor_id();
1587                 break;
1588         case EFX_EF10_REFILL:
1589                 /* The queue must be empty, so we won't receive any rx
1590                  * events, so efx_process_channel() won't refill the
1591                  * queue. Refill it here
1592                  */
1593                 efx_fast_push_rx_descriptors(&channel->rx_queue);
1594                 break;
1595         default:
1596                 netif_err(efx, hw, efx->net_dev,
1597                           "channel %d unknown driver event type %u"
1598                           " (data " EFX_QWORD_FMT ")\n",
1599                           channel->channel, (unsigned) subcode,
1600                           EFX_QWORD_VAL(*event));
1601         }
1602 }
1603
1604 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
1605 {
1606         struct efx_nic *efx = channel->efx;
1607         efx_qword_t event, *p_event;
1608         unsigned int read_ptr;
1609         int ev_code;
1610         int tx_descs = 0;
1611         int spent = 0;
1612
1613         read_ptr = channel->eventq_read_ptr;
1614
1615         for (;;) {
1616                 p_event = efx_event(channel, read_ptr);
1617                 event = *p_event;
1618
1619                 if (!efx_event_present(&event))
1620                         break;
1621
1622                 EFX_SET_QWORD(*p_event);
1623
1624                 ++read_ptr;
1625
1626                 ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
1627
1628                 netif_vdbg(efx, drv, efx->net_dev,
1629                            "processing event on %d " EFX_QWORD_FMT "\n",
1630                            channel->channel, EFX_QWORD_VAL(event));
1631
1632                 switch (ev_code) {
1633                 case ESE_DZ_EV_CODE_MCDI_EV:
1634                         efx_mcdi_process_event(channel, &event);
1635                         break;
1636                 case ESE_DZ_EV_CODE_RX_EV:
1637                         spent += efx_ef10_handle_rx_event(channel, &event);
1638                         if (spent >= quota) {
1639                                 /* XXX can we split a merged event to
1640                                  * avoid going over-quota?
1641                                  */
1642                                 spent = quota;
1643                                 goto out;
1644                         }
1645                         break;
1646                 case ESE_DZ_EV_CODE_TX_EV:
1647                         tx_descs += efx_ef10_handle_tx_event(channel, &event);
1648                         if (tx_descs > efx->txq_entries) {
1649                                 spent = quota;
1650                                 goto out;
1651                         } else if (++spent == quota) {
1652                                 goto out;
1653                         }
1654                         break;
1655                 case ESE_DZ_EV_CODE_DRIVER_EV:
1656                         efx_ef10_handle_driver_event(channel, &event);
1657                         if (++spent == quota)
1658                                 goto out;
1659                         break;
1660                 case EFX_EF10_DRVGEN_EV:
1661                         efx_ef10_handle_driver_generated_event(channel, &event);
1662                         break;
1663                 default:
1664                         netif_err(efx, hw, efx->net_dev,
1665                                   "channel %d unknown event type %d"
1666                                   " (data " EFX_QWORD_FMT ")\n",
1667                                   channel->channel, ev_code,
1668                                   EFX_QWORD_VAL(event));
1669                 }
1670         }
1671
1672 out:
1673         channel->eventq_read_ptr = read_ptr;
1674         return spent;
1675 }
1676
1677 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
1678 {
1679         struct efx_nic *efx = channel->efx;
1680         efx_dword_t rptr;
1681
1682         if (EFX_EF10_WORKAROUND_35388(efx)) {
1683                 BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
1684                              (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
1685                 BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
1686                              (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
1687
1688                 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
1689                                      EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
1690                                      ERF_DD_EVQ_IND_RPTR,
1691                                      (channel->eventq_read_ptr &
1692                                       channel->eventq_mask) >>
1693                                      ERF_DD_EVQ_IND_RPTR_WIDTH);
1694                 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
1695                                 channel->channel);
1696                 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
1697                                      EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
1698                                      ERF_DD_EVQ_IND_RPTR,
1699                                      channel->eventq_read_ptr &
1700                                      ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
1701                 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
1702                                 channel->channel);
1703         } else {
1704                 EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
1705                                      channel->eventq_read_ptr &
1706                                      channel->eventq_mask);
1707                 efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
1708         }
1709 }
1710
1711 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
1712 {
1713         MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1714         struct efx_nic *efx = channel->efx;
1715         efx_qword_t event;
1716         int rc;
1717
1718         EFX_POPULATE_QWORD_2(event,
1719                              ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1720                              ESF_DZ_EV_DATA, EFX_EF10_TEST);
1721
1722         MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1723
1724         /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1725          * already swapped the data to little-endian order.
1726          */
1727         memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1728                sizeof(efx_qword_t));
1729
1730         rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
1731                           NULL, 0, NULL);
1732         if (rc != 0)
1733                 goto fail;
1734
1735         return;
1736
1737 fail:
1738         WARN_ON(true);
1739         netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1740 }
1741
1742 void efx_ef10_handle_drain_event(struct efx_nic *efx)
1743 {
1744         if (atomic_dec_and_test(&efx->active_queues))
1745                 wake_up(&efx->flush_wq);
1746
1747         WARN_ON(atomic_read(&efx->active_queues) < 0);
1748 }
1749
1750 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
1751 {
1752         struct efx_ef10_nic_data *nic_data = efx->nic_data;
1753         struct efx_channel *channel;
1754         struct efx_tx_queue *tx_queue;
1755         struct efx_rx_queue *rx_queue;
1756         int pending;
1757
1758         /* If the MC has just rebooted, the TX/RX queues will have already been
1759          * torn down, but efx->active_queues needs to be set to zero.
1760          */
1761         if (nic_data->must_realloc_vis) {
1762                 atomic_set(&efx->active_queues, 0);
1763                 return 0;
1764         }
1765
1766         /* Do not attempt to write to the NIC during EEH recovery */
1767         if (efx->state != STATE_RECOVERY) {
1768                 efx_for_each_channel(channel, efx) {
1769                         efx_for_each_channel_rx_queue(rx_queue, channel)
1770                                 efx_ef10_rx_fini(rx_queue);
1771                         efx_for_each_channel_tx_queue(tx_queue, channel)
1772                                 efx_ef10_tx_fini(tx_queue);
1773                 }
1774
1775                 wait_event_timeout(efx->flush_wq,
1776                                    atomic_read(&efx->active_queues) == 0,
1777                                    msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
1778                 pending = atomic_read(&efx->active_queues);
1779                 if (pending) {
1780                         netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
1781                                   pending);
1782                         return -ETIMEDOUT;
1783                 }
1784         }
1785
1786         return 0;
1787 }
1788
1789 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
1790                                   const struct efx_filter_spec *right)
1791 {
1792         if ((left->match_flags ^ right->match_flags) |
1793             ((left->flags ^ right->flags) &
1794              (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
1795                 return false;
1796
1797         return memcmp(&left->outer_vid, &right->outer_vid,
1798                       sizeof(struct efx_filter_spec) -
1799                       offsetof(struct efx_filter_spec, outer_vid)) == 0;
1800 }
1801
1802 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
1803 {
1804         BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
1805         return jhash2((const u32 *)&spec->outer_vid,
1806                       (sizeof(struct efx_filter_spec) -
1807                        offsetof(struct efx_filter_spec, outer_vid)) / 4,
1808                       0);
1809         /* XXX should we randomise the initval? */
1810 }
1811
1812 /* Decide whether a filter should be exclusive or else should allow
1813  * delivery to additional recipients.  Currently we decide that
1814  * filters for specific local unicast MAC and IP addresses are
1815  * exclusive.
1816  */
1817 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
1818 {
1819         if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
1820             !is_multicast_ether_addr(spec->loc_mac))
1821                 return true;
1822
1823         if ((spec->match_flags &
1824              (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
1825             (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
1826                 if (spec->ether_type == htons(ETH_P_IP) &&
1827                     !ipv4_is_multicast(spec->loc_host[0]))
1828                         return true;
1829                 if (spec->ether_type == htons(ETH_P_IPV6) &&
1830                     ((const u8 *)spec->loc_host)[0] != 0xff)
1831                         return true;
1832         }
1833
1834         return false;
1835 }
1836
1837 static struct efx_filter_spec *
1838 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
1839                            unsigned int filter_idx)
1840 {
1841         return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
1842                                           ~EFX_EF10_FILTER_FLAGS);
1843 }
1844
1845 static unsigned int
1846 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
1847                            unsigned int filter_idx)
1848 {
1849         return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
1850 }
1851
1852 static void
1853 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
1854                           unsigned int filter_idx,
1855                           const struct efx_filter_spec *spec,
1856                           unsigned int flags)
1857 {
1858         table->entry[filter_idx].spec = (unsigned long)spec | flags;
1859 }
1860
1861 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
1862                                       const struct efx_filter_spec *spec,
1863                                       efx_dword_t *inbuf, u64 handle,
1864                                       bool replacing)
1865 {
1866         struct efx_ef10_nic_data *nic_data = efx->nic_data;
1867
1868         memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
1869
1870         if (replacing) {
1871                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
1872                                MC_CMD_FILTER_OP_IN_OP_REPLACE);
1873                 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
1874         } else {
1875                 u32 match_fields = 0;
1876
1877                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
1878                                efx_ef10_filter_is_exclusive(spec) ?
1879                                MC_CMD_FILTER_OP_IN_OP_INSERT :
1880                                MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
1881
1882                 /* Convert match flags and values.  Unlike almost
1883                  * everything else in MCDI, these fields are in
1884                  * network byte order.
1885                  */
1886                 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
1887                         match_fields |=
1888                                 is_multicast_ether_addr(spec->loc_mac) ?
1889                                 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
1890                                 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
1891 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)                          \
1892                 if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
1893                         match_fields |=                                      \
1894                                 1 << MC_CMD_FILTER_OP_IN_MATCH_ ##           \
1895                                 mcdi_field ## _LBN;                          \
1896                         BUILD_BUG_ON(                                        \
1897                                 MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
1898                                 sizeof(spec->gen_field));                    \
1899                         memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ## mcdi_field), \
1900                                &spec->gen_field, sizeof(spec->gen_field));   \
1901                 }
1902                 COPY_FIELD(REM_HOST, rem_host, SRC_IP);
1903                 COPY_FIELD(LOC_HOST, loc_host, DST_IP);
1904                 COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
1905                 COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
1906                 COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
1907                 COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
1908                 COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
1909                 COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
1910                 COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
1911                 COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
1912 #undef COPY_FIELD
1913                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
1914                                match_fields);
1915         }
1916
1917         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1918         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
1919                        spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
1920                        MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
1921                        MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
1922         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
1923                        MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
1924         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE, spec->dmaq_id);
1925         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
1926                        (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
1927                        MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
1928                        MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
1929         if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
1930                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
1931                                spec->rss_context !=
1932                                EFX_FILTER_RSS_CONTEXT_DEFAULT ?
1933                                spec->rss_context : nic_data->rx_rss_context);
1934 }
1935
1936 static int efx_ef10_filter_push(struct efx_nic *efx,
1937                                 const struct efx_filter_spec *spec,
1938                                 u64 *handle, bool replacing)
1939 {
1940         MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
1941         MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
1942         int rc;
1943
1944         efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
1945         rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
1946                           outbuf, sizeof(outbuf), NULL);
1947         if (rc == 0)
1948                 *handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
1949         return rc;
1950 }
1951
1952 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
1953                                         enum efx_filter_match_flags match_flags)
1954 {
1955         unsigned int match_pri;
1956
1957         for (match_pri = 0;
1958              match_pri < table->rx_match_count;
1959              match_pri++)
1960                 if (table->rx_match_flags[match_pri] == match_flags)
1961                         return match_pri;
1962
1963         return -EPROTONOSUPPORT;
1964 }
1965
1966 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
1967                                   struct efx_filter_spec *spec,
1968                                   bool replace_equal)
1969 {
1970         struct efx_ef10_filter_table *table = efx->filter_state;
1971         DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
1972         struct efx_filter_spec *saved_spec;
1973         unsigned int match_pri, hash;
1974         unsigned int priv_flags;
1975         bool replacing = false;
1976         int ins_index = -1;
1977         DEFINE_WAIT(wait);
1978         bool is_mc_recip;
1979         s32 rc;
1980
1981         /* For now, only support RX filters */
1982         if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
1983             EFX_FILTER_FLAG_RX)
1984                 return -EINVAL;
1985
1986         rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
1987         if (rc < 0)
1988                 return rc;
1989         match_pri = rc;
1990
1991         hash = efx_ef10_filter_hash(spec);
1992         is_mc_recip = efx_filter_is_mc_recipient(spec);
1993         if (is_mc_recip)
1994                 bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
1995
1996         /* Find any existing filters with the same match tuple or
1997          * else a free slot to insert at.  If any of them are busy,
1998          * we have to wait and retry.
1999          */
2000         for (;;) {
2001                 unsigned int depth = 1;
2002                 unsigned int i;
2003
2004                 spin_lock_bh(&efx->filter_lock);
2005
2006                 for (;;) {
2007                         i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2008                         saved_spec = efx_ef10_filter_entry_spec(table, i);
2009
2010                         if (!saved_spec) {
2011                                 if (ins_index < 0)
2012                                         ins_index = i;
2013                         } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2014                                 if (table->entry[i].spec &
2015                                     EFX_EF10_FILTER_FLAG_BUSY)
2016                                         break;
2017                                 if (spec->priority < saved_spec->priority &&
2018                                     !(saved_spec->priority ==
2019                                       EFX_FILTER_PRI_REQUIRED &&
2020                                       saved_spec->flags &
2021                                       EFX_FILTER_FLAG_RX_STACK)) {
2022                                         rc = -EPERM;
2023                                         goto out_unlock;
2024                                 }
2025                                 if (!is_mc_recip) {
2026                                         /* This is the only one */
2027                                         if (spec->priority ==
2028                                             saved_spec->priority &&
2029                                             !replace_equal) {
2030                                                 rc = -EEXIST;
2031                                                 goto out_unlock;
2032                                         }
2033                                         ins_index = i;
2034                                         goto found;
2035                                 } else if (spec->priority >
2036                                            saved_spec->priority ||
2037                                            (spec->priority ==
2038                                             saved_spec->priority &&
2039                                             replace_equal)) {
2040                                         if (ins_index < 0)
2041                                                 ins_index = i;
2042                                         else
2043                                                 __set_bit(depth, mc_rem_map);
2044                                 }
2045                         }
2046
2047                         /* Once we reach the maximum search depth, use
2048                          * the first suitable slot or return -EBUSY if
2049                          * there was none
2050                          */
2051                         if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2052                                 if (ins_index < 0) {
2053                                         rc = -EBUSY;
2054                                         goto out_unlock;
2055                                 }
2056                                 goto found;
2057                         }
2058
2059                         ++depth;
2060                 }
2061
2062                 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2063                 spin_unlock_bh(&efx->filter_lock);
2064                 schedule();
2065         }
2066
2067 found:
2068         /* Create a software table entry if necessary, and mark it
2069          * busy.  We might yet fail to insert, but any attempt to
2070          * insert a conflicting filter while we're waiting for the
2071          * firmware must find the busy entry.
2072          */
2073         saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2074         if (saved_spec) {
2075                 if (spec->flags & EFX_FILTER_FLAG_RX_STACK) {
2076                         /* Just make sure it won't be removed */
2077                         saved_spec->flags |= EFX_FILTER_FLAG_RX_STACK;
2078                         table->entry[ins_index].spec &=
2079                                 ~EFX_EF10_FILTER_FLAG_STACK_OLD;
2080                         rc = ins_index;
2081                         goto out_unlock;
2082                 }
2083                 replacing = true;
2084                 priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
2085         } else {
2086                 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2087                 if (!saved_spec) {
2088                         rc = -ENOMEM;
2089                         goto out_unlock;
2090                 }
2091                 *saved_spec = *spec;
2092                 priv_flags = 0;
2093         }
2094         efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2095                                   priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
2096
2097         /* Mark lower-priority multicast recipients busy prior to removal */
2098         if (is_mc_recip) {
2099                 unsigned int depth, i;
2100
2101                 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2102                         i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2103                         if (test_bit(depth, mc_rem_map))
2104                                 table->entry[i].spec |=
2105                                         EFX_EF10_FILTER_FLAG_BUSY;
2106                 }
2107         }
2108
2109         spin_unlock_bh(&efx->filter_lock);
2110
2111         rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
2112                                   replacing);
2113
2114         /* Finalise the software table entry */
2115         spin_lock_bh(&efx->filter_lock);
2116         if (rc == 0) {
2117                 if (replacing) {
2118                         /* Update the fields that may differ */
2119                         saved_spec->priority = spec->priority;
2120                         saved_spec->flags &= EFX_FILTER_FLAG_RX_STACK;
2121                         saved_spec->flags |= spec->flags;
2122                         saved_spec->rss_context = spec->rss_context;
2123                         saved_spec->dmaq_id = spec->dmaq_id;
2124                 }
2125         } else if (!replacing) {
2126                 kfree(saved_spec);
2127                 saved_spec = NULL;
2128         }
2129         efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
2130
2131         /* Remove and finalise entries for lower-priority multicast
2132          * recipients
2133          */
2134         if (is_mc_recip) {
2135                 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2136                 unsigned int depth, i;
2137
2138                 memset(inbuf, 0, sizeof(inbuf));
2139
2140                 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2141                         if (!test_bit(depth, mc_rem_map))
2142                                 continue;
2143
2144                         i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2145                         saved_spec = efx_ef10_filter_entry_spec(table, i);
2146                         priv_flags = efx_ef10_filter_entry_flags(table, i);
2147
2148                         if (rc == 0) {
2149                                 spin_unlock_bh(&efx->filter_lock);
2150                                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2151                                                MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2152                                 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2153                                                table->entry[i].handle);
2154                                 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2155                                                   inbuf, sizeof(inbuf),
2156                                                   NULL, 0, NULL);
2157                                 spin_lock_bh(&efx->filter_lock);
2158                         }
2159
2160                         if (rc == 0) {
2161                                 kfree(saved_spec);
2162                                 saved_spec = NULL;
2163                                 priv_flags = 0;
2164                         } else {
2165                                 priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
2166                         }
2167                         efx_ef10_filter_set_entry(table, i, saved_spec,
2168                                                   priv_flags);
2169                 }
2170         }
2171
2172         /* If successful, return the inserted filter ID */
2173         if (rc == 0)
2174                 rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
2175
2176         wake_up_all(&table->waitq);
2177 out_unlock:
2178         spin_unlock_bh(&efx->filter_lock);
2179         finish_wait(&table->waitq, &wait);
2180         return rc;
2181 }
2182
2183 void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
2184 {
2185         /* no need to do anything here on EF10 */
2186 }
2187
2188 /* Remove a filter.
2189  * If !stack_requested, remove by ID
2190  * If stack_requested, remove by index
2191  * Filter ID may come from userland and must be range-checked.
2192  */
2193 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
2194                                            enum efx_filter_priority priority,
2195                                            u32 filter_id, bool stack_requested)
2196 {
2197         unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2198         struct efx_ef10_filter_table *table = efx->filter_state;
2199         MCDI_DECLARE_BUF(inbuf,
2200                          MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2201                          MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2202         struct efx_filter_spec *spec;
2203         DEFINE_WAIT(wait);
2204         int rc;
2205
2206         /* Find the software table entry and mark it busy.  Don't
2207          * remove it yet; any attempt to update while we're waiting
2208          * for the firmware must find the busy entry.
2209          */
2210         for (;;) {
2211                 spin_lock_bh(&efx->filter_lock);
2212                 if (!(table->entry[filter_idx].spec &
2213                       EFX_EF10_FILTER_FLAG_BUSY))
2214                         break;
2215                 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2216                 spin_unlock_bh(&efx->filter_lock);
2217                 schedule();
2218         }
2219         spec = efx_ef10_filter_entry_spec(table, filter_idx);
2220         if (!spec || spec->priority > priority ||
2221             (!stack_requested &&
2222              efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
2223              filter_id / HUNT_FILTER_TBL_ROWS)) {
2224                 rc = -ENOENT;
2225                 goto out_unlock;
2226         }
2227         table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2228         spin_unlock_bh(&efx->filter_lock);
2229
2230         if (spec->flags & EFX_FILTER_FLAG_RX_STACK && !stack_requested) {
2231                 /* Reset steering of a stack-owned filter */
2232
2233                 struct efx_filter_spec new_spec = *spec;
2234
2235                 new_spec.priority = EFX_FILTER_PRI_REQUIRED;
2236                 new_spec.flags = (EFX_FILTER_FLAG_RX |
2237                                   EFX_FILTER_FLAG_RX_RSS |
2238                                   EFX_FILTER_FLAG_RX_STACK);
2239                 new_spec.dmaq_id = 0;
2240                 new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
2241                 rc = efx_ef10_filter_push(efx, &new_spec,
2242                                           &table->entry[filter_idx].handle,
2243                                           true);
2244
2245                 spin_lock_bh(&efx->filter_lock);
2246                 if (rc == 0)
2247                         *spec = new_spec;
2248         } else {
2249                 /* Really remove the filter */
2250
2251                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2252                                efx_ef10_filter_is_exclusive(spec) ?
2253                                MC_CMD_FILTER_OP_IN_OP_REMOVE :
2254                                MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2255                 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2256                                table->entry[filter_idx].handle);
2257                 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2258                                   inbuf, sizeof(inbuf), NULL, 0, NULL);
2259
2260                 spin_lock_bh(&efx->filter_lock);
2261                 if (rc == 0) {
2262                         kfree(spec);
2263                         efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2264                 }
2265         }
2266         table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2267         wake_up_all(&table->waitq);
2268 out_unlock:
2269         spin_unlock_bh(&efx->filter_lock);
2270         finish_wait(&table->waitq, &wait);
2271         return rc;
2272 }
2273
2274 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
2275                                        enum efx_filter_priority priority,
2276                                        u32 filter_id)
2277 {
2278         return efx_ef10_filter_remove_internal(efx, priority, filter_id, false);
2279 }
2280
2281 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
2282                                     enum efx_filter_priority priority,
2283                                     u32 filter_id, struct efx_filter_spec *spec)
2284 {
2285         unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2286         struct efx_ef10_filter_table *table = efx->filter_state;
2287         const struct efx_filter_spec *saved_spec;
2288         int rc;
2289
2290         spin_lock_bh(&efx->filter_lock);
2291         saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
2292         if (saved_spec && saved_spec->priority == priority &&
2293             efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
2294             filter_id / HUNT_FILTER_TBL_ROWS) {
2295                 *spec = *saved_spec;
2296                 rc = 0;
2297         } else {
2298                 rc = -ENOENT;
2299         }
2300         spin_unlock_bh(&efx->filter_lock);
2301         return rc;
2302 }
2303
2304 static void efx_ef10_filter_clear_rx(struct efx_nic *efx,
2305                                      enum efx_filter_priority priority)
2306 {
2307         /* TODO */
2308 }
2309
2310 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
2311                                          enum efx_filter_priority priority)
2312 {
2313         struct efx_ef10_filter_table *table = efx->filter_state;
2314         unsigned int filter_idx;
2315         s32 count = 0;
2316
2317         spin_lock_bh(&efx->filter_lock);
2318         for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2319                 if (table->entry[filter_idx].spec &&
2320                     efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
2321                     priority)
2322                         ++count;
2323         }
2324         spin_unlock_bh(&efx->filter_lock);
2325         return count;
2326 }
2327
2328 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
2329 {
2330         struct efx_ef10_filter_table *table = efx->filter_state;
2331
2332         return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
2333 }
2334
2335 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
2336                                       enum efx_filter_priority priority,
2337                                       u32 *buf, u32 size)
2338 {
2339         struct efx_ef10_filter_table *table = efx->filter_state;
2340         struct efx_filter_spec *spec;
2341         unsigned int filter_idx;
2342         s32 count = 0;
2343
2344         spin_lock_bh(&efx->filter_lock);
2345         for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2346                 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2347                 if (spec && spec->priority == priority) {
2348                         if (count == size) {
2349                                 count = -EMSGSIZE;
2350                                 break;
2351                         }
2352                         buf[count++] = (efx_ef10_filter_rx_match_pri(
2353                                                 table, spec->match_flags) *
2354                                         HUNT_FILTER_TBL_ROWS +
2355                                         filter_idx);
2356                 }
2357         }
2358         spin_unlock_bh(&efx->filter_lock);
2359         return count;
2360 }
2361
2362 #ifdef CONFIG_RFS_ACCEL
2363
2364 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
2365
2366 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
2367                                       struct efx_filter_spec *spec)
2368 {
2369         struct efx_ef10_filter_table *table = efx->filter_state;
2370         MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2371         struct efx_filter_spec *saved_spec;
2372         unsigned int hash, i, depth = 1;
2373         bool replacing = false;
2374         int ins_index = -1;
2375         u64 cookie;
2376         s32 rc;
2377
2378         /* Must be an RX filter without RSS and not for a multicast
2379          * destination address (RFS only works for connected sockets).
2380          * These restrictions allow us to pass only a tiny amount of
2381          * data through to the completion function.
2382          */
2383         EFX_WARN_ON_PARANOID(spec->flags !=
2384                              (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
2385         EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
2386         EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
2387
2388         hash = efx_ef10_filter_hash(spec);
2389
2390         spin_lock_bh(&efx->filter_lock);
2391
2392         /* Find any existing filter with the same match tuple or else
2393          * a free slot to insert at.  If an existing filter is busy,
2394          * we have to give up.
2395          */
2396         for (;;) {
2397                 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2398                 saved_spec = efx_ef10_filter_entry_spec(table, i);
2399
2400                 if (!saved_spec) {
2401                         if (ins_index < 0)
2402                                 ins_index = i;
2403                 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2404                         if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
2405                                 rc = -EBUSY;
2406                                 goto fail_unlock;
2407                         }
2408                         EFX_WARN_ON_PARANOID(saved_spec->flags &
2409                                              EFX_FILTER_FLAG_RX_STACK);
2410                         if (spec->priority < saved_spec->priority) {
2411                                 rc = -EPERM;
2412                                 goto fail_unlock;
2413                         }
2414                         ins_index = i;
2415                         break;
2416                 }
2417
2418                 /* Once we reach the maximum search depth, use the
2419                  * first suitable slot or return -EBUSY if there was
2420                  * none
2421                  */
2422                 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2423                         if (ins_index < 0) {
2424                                 rc = -EBUSY;
2425                                 goto fail_unlock;
2426                         }
2427                         break;
2428                 }
2429
2430                 ++depth;
2431         }
2432
2433         /* Create a software table entry if necessary, and mark it
2434          * busy.  We might yet fail to insert, but any attempt to
2435          * insert a conflicting filter while we're waiting for the
2436          * firmware must find the busy entry.
2437          */
2438         saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2439         if (saved_spec) {
2440                 replacing = true;
2441         } else {
2442                 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2443                 if (!saved_spec) {
2444                         rc = -ENOMEM;
2445                         goto fail_unlock;
2446                 }
2447                 *saved_spec = *spec;
2448         }
2449         efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2450                                   EFX_EF10_FILTER_FLAG_BUSY);
2451
2452         spin_unlock_bh(&efx->filter_lock);
2453
2454         /* Pack up the variables needed on completion */
2455         cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
2456
2457         efx_ef10_filter_push_prep(efx, spec, inbuf,
2458                                   table->entry[ins_index].handle, replacing);
2459         efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2460                            MC_CMD_FILTER_OP_OUT_LEN,
2461                            efx_ef10_filter_rfs_insert_complete, cookie);
2462
2463         return ins_index;
2464
2465 fail_unlock:
2466         spin_unlock_bh(&efx->filter_lock);
2467         return rc;
2468 }
2469
2470 static void
2471 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
2472                                     int rc, efx_dword_t *outbuf,
2473                                     size_t outlen_actual)
2474 {
2475         struct efx_ef10_filter_table *table = efx->filter_state;
2476         unsigned int ins_index, dmaq_id;
2477         struct efx_filter_spec *spec;
2478         bool replacing;
2479
2480         /* Unpack the cookie */
2481         replacing = cookie >> 31;
2482         ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
2483         dmaq_id = cookie & 0xffff;
2484
2485         spin_lock_bh(&efx->filter_lock);
2486         spec = efx_ef10_filter_entry_spec(table, ins_index);
2487         if (rc == 0) {
2488                 table->entry[ins_index].handle =
2489                         MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2490                 if (replacing)
2491                         spec->dmaq_id = dmaq_id;
2492         } else if (!replacing) {
2493                 kfree(spec);
2494                 spec = NULL;
2495         }
2496         efx_ef10_filter_set_entry(table, ins_index, spec, 0);
2497         spin_unlock_bh(&efx->filter_lock);
2498
2499         wake_up_all(&table->waitq);
2500 }
2501
2502 static void
2503 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2504                                     unsigned long filter_idx,
2505                                     int rc, efx_dword_t *outbuf,
2506                                     size_t outlen_actual);
2507
2508 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2509                                            unsigned int filter_idx)
2510 {
2511         struct efx_ef10_filter_table *table = efx->filter_state;
2512         struct efx_filter_spec *spec =
2513                 efx_ef10_filter_entry_spec(table, filter_idx);
2514         MCDI_DECLARE_BUF(inbuf,
2515                          MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2516                          MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2517
2518         if (!spec ||
2519             (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
2520             spec->priority != EFX_FILTER_PRI_HINT ||
2521             !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
2522                                  flow_id, filter_idx))
2523                 return false;
2524
2525         MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2526                        MC_CMD_FILTER_OP_IN_OP_REMOVE);
2527         MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2528                        table->entry[filter_idx].handle);
2529         if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
2530                                efx_ef10_filter_rfs_expire_complete, filter_idx))
2531                 return false;
2532
2533         table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2534         return true;
2535 }
2536
2537 static void
2538 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2539                                     unsigned long filter_idx,
2540                                     int rc, efx_dword_t *outbuf,
2541                                     size_t outlen_actual)
2542 {
2543         struct efx_ef10_filter_table *table = efx->filter_state;
2544         struct efx_filter_spec *spec =
2545                 efx_ef10_filter_entry_spec(table, filter_idx);
2546
2547         spin_lock_bh(&efx->filter_lock);
2548         if (rc == 0) {
2549                 kfree(spec);
2550                 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2551         }
2552         table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2553         wake_up_all(&table->waitq);
2554         spin_unlock_bh(&efx->filter_lock);
2555 }
2556
2557 #endif /* CONFIG_RFS_ACCEL */
2558
2559 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
2560 {
2561         int match_flags = 0;
2562
2563 #define MAP_FLAG(gen_flag, mcdi_field) {                                \
2564                 u32 old_mcdi_flags = mcdi_flags;                        \
2565                 mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##      \
2566                                 mcdi_field ## _LBN);                    \
2567                 if (mcdi_flags != old_mcdi_flags)                       \
2568                         match_flags |= EFX_FILTER_MATCH_ ## gen_flag;   \
2569         }
2570         MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
2571         MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
2572         MAP_FLAG(REM_HOST, SRC_IP);
2573         MAP_FLAG(LOC_HOST, DST_IP);
2574         MAP_FLAG(REM_MAC, SRC_MAC);
2575         MAP_FLAG(REM_PORT, SRC_PORT);
2576         MAP_FLAG(LOC_MAC, DST_MAC);
2577         MAP_FLAG(LOC_PORT, DST_PORT);
2578         MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
2579         MAP_FLAG(INNER_VID, INNER_VLAN);
2580         MAP_FLAG(OUTER_VID, OUTER_VLAN);
2581         MAP_FLAG(IP_PROTO, IP_PROTO);
2582 #undef MAP_FLAG
2583
2584         /* Did we map them all? */
2585         if (mcdi_flags)
2586                 return -EINVAL;
2587
2588         return match_flags;
2589 }
2590
2591 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2592 {
2593         MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
2594         MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
2595         unsigned int pd_match_pri, pd_match_count;
2596         struct efx_ef10_filter_table *table;
2597         size_t outlen;
2598         int rc;
2599
2600         table = kzalloc(sizeof(*table), GFP_KERNEL);
2601         if (!table)
2602                 return -ENOMEM;
2603
2604         /* Find out which RX filter types are supported, and their priorities */
2605         MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
2606                        MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
2607         rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
2608                           inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
2609                           &outlen);
2610         if (rc)
2611                 goto fail;
2612         pd_match_count = MCDI_VAR_ARRAY_LEN(
2613                 outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
2614         table->rx_match_count = 0;
2615
2616         for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
2617                 u32 mcdi_flags =
2618                         MCDI_ARRAY_DWORD(
2619                                 outbuf,
2620                                 GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
2621                                 pd_match_pri);
2622                 rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
2623                 if (rc < 0) {
2624                         netif_dbg(efx, probe, efx->net_dev,
2625                                   "%s: fw flags %#x pri %u not supported in driver\n",
2626                                   __func__, mcdi_flags, pd_match_pri);
2627                 } else {
2628                         netif_dbg(efx, probe, efx->net_dev,
2629                                   "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
2630                                   __func__, mcdi_flags, pd_match_pri,
2631                                   rc, table->rx_match_count);
2632                         table->rx_match_flags[table->rx_match_count++] = rc;
2633                 }
2634         }
2635
2636         table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
2637         if (!table->entry) {
2638                 rc = -ENOMEM;
2639                 goto fail;
2640         }
2641
2642         efx->filter_state = table;
2643         init_waitqueue_head(&table->waitq);
2644         return 0;
2645
2646 fail:
2647         kfree(table);
2648         return rc;
2649 }
2650
2651 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
2652 {
2653         struct efx_ef10_filter_table *table = efx->filter_state;
2654         struct efx_ef10_nic_data *nic_data = efx->nic_data;
2655         struct efx_filter_spec *spec;
2656         unsigned int filter_idx;
2657         bool failed = false;
2658         int rc;
2659
2660         if (!nic_data->must_restore_filters)
2661                 return;
2662
2663         spin_lock_bh(&efx->filter_lock);
2664
2665         for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2666                 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2667                 if (!spec)
2668                         continue;
2669
2670                 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2671                 spin_unlock_bh(&efx->filter_lock);
2672
2673                 rc = efx_ef10_filter_push(efx, spec,
2674                                           &table->entry[filter_idx].handle,
2675                                           false);
2676                 if (rc)
2677                         failed = true;
2678
2679                 spin_lock_bh(&efx->filter_lock);
2680                 if (rc) {
2681                         kfree(spec);
2682                         efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2683                 } else {
2684                         table->entry[filter_idx].spec &=
2685                                 ~EFX_EF10_FILTER_FLAG_BUSY;
2686                 }
2687         }
2688
2689         spin_unlock_bh(&efx->filter_lock);
2690
2691         if (failed)
2692                 netif_err(efx, hw, efx->net_dev,
2693                           "unable to restore all filters\n");
2694         else
2695                 nic_data->must_restore_filters = false;
2696 }
2697
2698 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
2699 {
2700         struct efx_ef10_filter_table *table = efx->filter_state;
2701         MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2702         struct efx_filter_spec *spec;
2703         unsigned int filter_idx;
2704         int rc;
2705
2706         for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2707                 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2708                 if (!spec)
2709                         continue;
2710
2711                 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2712                                efx_ef10_filter_is_exclusive(spec) ?
2713                                MC_CMD_FILTER_OP_IN_OP_REMOVE :
2714                                MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2715                 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2716                                table->entry[filter_idx].handle);
2717                 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2718                                   NULL, 0, NULL);
2719
2720                 WARN_ON(rc != 0);
2721                 kfree(spec);
2722         }
2723
2724         vfree(table->entry);
2725         kfree(table);
2726 }
2727
2728 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
2729 {
2730         struct efx_ef10_filter_table *table = efx->filter_state;
2731         struct net_device *net_dev = efx->net_dev;
2732         struct efx_filter_spec spec;
2733         bool remove_failed = false;
2734         struct netdev_hw_addr *uc;
2735         struct netdev_hw_addr *mc;
2736         unsigned int filter_idx;
2737         int i, n, rc;
2738
2739         if (!efx_dev_registered(efx))
2740                 return;
2741
2742         /* Mark old filters that may need to be removed */
2743         spin_lock_bh(&efx->filter_lock);
2744         n = table->stack_uc_count < 0 ? 1 : table->stack_uc_count;
2745         for (i = 0; i < n; i++) {
2746                 filter_idx = table->stack_uc_list[i].id % HUNT_FILTER_TBL_ROWS;
2747                 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
2748         }
2749         n = table->stack_mc_count < 0 ? 1 : table->stack_mc_count;
2750         for (i = 0; i < n; i++) {
2751                 filter_idx = table->stack_mc_list[i].id % HUNT_FILTER_TBL_ROWS;
2752                 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
2753         }
2754         spin_unlock_bh(&efx->filter_lock);
2755
2756         /* Copy/convert the address lists; add the primary station
2757          * address and broadcast address
2758          */
2759         netif_addr_lock_bh(net_dev);
2760         if (net_dev->flags & IFF_PROMISC ||
2761             netdev_uc_count(net_dev) >= EFX_EF10_FILTER_STACK_UC_MAX) {
2762                 table->stack_uc_count = -1;
2763         } else {
2764                 table->stack_uc_count = 1 + netdev_uc_count(net_dev);
2765                 memcpy(table->stack_uc_list[0].addr, net_dev->dev_addr,
2766                        ETH_ALEN);
2767                 i = 1;
2768                 netdev_for_each_uc_addr(uc, net_dev) {
2769                         memcpy(table->stack_uc_list[i].addr,
2770                                uc->addr, ETH_ALEN);
2771                         i++;
2772                 }
2773         }
2774         if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI) ||
2775             netdev_mc_count(net_dev) >= EFX_EF10_FILTER_STACK_MC_MAX) {
2776                 table->stack_mc_count = -1;
2777         } else {
2778                 table->stack_mc_count = 1 + netdev_mc_count(net_dev);
2779                 eth_broadcast_addr(table->stack_mc_list[0].addr);
2780                 i = 1;
2781                 netdev_for_each_mc_addr(mc, net_dev) {
2782                         memcpy(table->stack_mc_list[i].addr,
2783                                mc->addr, ETH_ALEN);
2784                         i++;
2785                 }
2786         }
2787         netif_addr_unlock_bh(net_dev);
2788
2789         /* Insert/renew unicast filters */
2790         if (table->stack_uc_count >= 0) {
2791                 for (i = 0; i < table->stack_uc_count; i++) {
2792                         efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
2793                                            EFX_FILTER_FLAG_RX_RSS |
2794                                            EFX_FILTER_FLAG_RX_STACK,
2795                                            0);
2796                         efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
2797                                                  table->stack_uc_list[i].addr);
2798                         rc = efx_ef10_filter_insert(efx, &spec, true);
2799                         if (rc < 0) {
2800                                 /* Fall back to unicast-promisc */
2801                                 while (i--)
2802                                         efx_ef10_filter_remove_safe(
2803                                                 efx, EFX_FILTER_PRI_REQUIRED,
2804                                                 table->stack_uc_list[i].id);
2805                                 table->stack_uc_count = -1;
2806                                 break;
2807                         }
2808                         table->stack_uc_list[i].id = rc;
2809                 }
2810         }
2811         if (table->stack_uc_count < 0) {
2812                 efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
2813                                    EFX_FILTER_FLAG_RX_RSS |
2814                                    EFX_FILTER_FLAG_RX_STACK,
2815                                    0);
2816                 efx_filter_set_uc_def(&spec);
2817                 rc = efx_ef10_filter_insert(efx, &spec, true);
2818                 if (rc < 0) {
2819                         WARN_ON(1);
2820                         table->stack_uc_count = 0;
2821                 } else {
2822                         table->stack_uc_list[0].id = rc;
2823                 }
2824         }
2825
2826         /* Insert/renew multicast filters */
2827         if (table->stack_mc_count >= 0) {
2828                 for (i = 0; i < table->stack_mc_count; i++) {
2829                         efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
2830                                            EFX_FILTER_FLAG_RX_RSS |
2831                                            EFX_FILTER_FLAG_RX_STACK,
2832                                            0);
2833                         efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
2834                                                  table->stack_mc_list[i].addr);
2835                         rc = efx_ef10_filter_insert(efx, &spec, true);
2836                         if (rc < 0) {
2837                                 /* Fall back to multicast-promisc */
2838                                 while (i--)
2839                                         efx_ef10_filter_remove_safe(
2840                                                 efx, EFX_FILTER_PRI_REQUIRED,
2841                                                 table->stack_mc_list[i].id);
2842                                 table->stack_mc_count = -1;
2843                                 break;
2844                         }
2845                         table->stack_mc_list[i].id = rc;
2846                 }
2847         }
2848         if (table->stack_mc_count < 0) {
2849                 efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
2850                                    EFX_FILTER_FLAG_RX_RSS |
2851                                    EFX_FILTER_FLAG_RX_STACK,
2852                                    0);
2853                 efx_filter_set_mc_def(&spec);
2854                 rc = efx_ef10_filter_insert(efx, &spec, true);
2855                 if (rc < 0) {
2856                         WARN_ON(1);
2857                         table->stack_mc_count = 0;
2858                 } else {
2859                         table->stack_mc_list[0].id = rc;
2860                 }
2861         }
2862
2863         /* Remove filters that weren't renewed.  Since nothing else
2864          * changes the STACK_OLD flag or removes these filters, we
2865          * don't need to hold the filter_lock while scanning for
2866          * these filters.
2867          */
2868         for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
2869                 if (ACCESS_ONCE(table->entry[i].spec) &
2870                     EFX_EF10_FILTER_FLAG_STACK_OLD) {
2871                         if (efx_ef10_filter_remove_internal(efx,
2872                                         EFX_FILTER_PRI_REQUIRED,
2873                                         i, true) < 0)
2874                                 remove_failed = true;
2875                 }
2876         }
2877         WARN_ON(remove_failed);
2878 }
2879
2880 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
2881 {
2882         efx_ef10_filter_sync_rx_mode(efx);
2883
2884         return efx_mcdi_set_mac(efx);
2885 }
2886
2887 #ifdef CONFIG_SFC_MTD
2888
2889 struct efx_ef10_nvram_type_info {
2890         u16 type, type_mask;
2891         u8 port;
2892         const char *name;
2893 };
2894
2895 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
2896         { NVRAM_PARTITION_TYPE_MC_FIRMWARE,        0,    0, "sfc_mcfw" },
2897         { NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
2898         { NVRAM_PARTITION_TYPE_EXPANSION_ROM,      0,    0, "sfc_exp_rom" },
2899         { NVRAM_PARTITION_TYPE_STATIC_CONFIG,      0,    0, "sfc_static_cfg" },
2900         { NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,     0,    0, "sfc_dynamic_cfg" },
2901         { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
2902         { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
2903         { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
2904         { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
2905         { NVRAM_PARTITION_TYPE_PHY_MIN,            0xff, 0, "sfc_phy_fw" },
2906 };
2907
2908 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
2909                                         struct efx_mcdi_mtd_partition *part,
2910                                         unsigned int type)
2911 {
2912         MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
2913         MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
2914         const struct efx_ef10_nvram_type_info *info;
2915         size_t size, erase_size, outlen;
2916         bool protected;
2917         int rc;
2918
2919         for (info = efx_ef10_nvram_types; ; info++) {
2920                 if (info ==
2921                     efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
2922                         return -ENODEV;
2923                 if ((type & ~info->type_mask) == info->type)
2924                         break;
2925         }
2926         if (info->port != efx_port_num(efx))
2927                 return -ENODEV;
2928
2929         rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
2930         if (rc)
2931                 return rc;
2932         if (protected)
2933                 return -ENODEV; /* hide it */
2934
2935         part->nvram_type = type;
2936
2937         MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
2938         rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
2939                           outbuf, sizeof(outbuf), &outlen);
2940         if (rc)
2941                 return rc;
2942         if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
2943                 return -EIO;
2944         if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
2945             (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
2946                 part->fw_subtype = MCDI_DWORD(outbuf,
2947                                               NVRAM_METADATA_OUT_SUBTYPE);
2948
2949         part->common.dev_type_name = "EF10 NVRAM manager";
2950         part->common.type_name = info->name;
2951
2952         part->common.mtd.type = MTD_NORFLASH;
2953         part->common.mtd.flags = MTD_CAP_NORFLASH;
2954         part->common.mtd.size = size;
2955         part->common.mtd.erasesize = erase_size;
2956
2957         return 0;
2958 }
2959
2960 static int efx_ef10_mtd_probe(struct efx_nic *efx)
2961 {
2962         MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
2963         struct efx_mcdi_mtd_partition *parts;
2964         size_t outlen, n_parts_total, i, n_parts;
2965         unsigned int type;
2966         int rc;
2967
2968         ASSERT_RTNL();
2969
2970         BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
2971         rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
2972                           outbuf, sizeof(outbuf), &outlen);
2973         if (rc)
2974                 return rc;
2975         if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
2976                 return -EIO;
2977
2978         n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
2979         if (n_parts_total >
2980             MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
2981                 return -EIO;
2982
2983         parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
2984         if (!parts)
2985                 return -ENOMEM;
2986
2987         n_parts = 0;
2988         for (i = 0; i < n_parts_total; i++) {
2989                 type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
2990                                         i);
2991                 rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
2992                 if (rc == 0)
2993                         n_parts++;
2994                 else if (rc != -ENODEV)
2995                         goto fail;
2996         }
2997
2998         rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
2999 fail:
3000         if (rc)
3001                 kfree(parts);
3002         return rc;
3003 }
3004
3005 #endif /* CONFIG_SFC_MTD */
3006
3007 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3008 {
3009         _efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3010 }
3011
3012 const struct efx_nic_type efx_hunt_a0_nic_type = {
3013         .mem_map_size = efx_ef10_mem_map_size,
3014         .probe = efx_ef10_probe,
3015         .remove = efx_ef10_remove,
3016         .dimension_resources = efx_ef10_dimension_resources,
3017         .init = efx_ef10_init_nic,
3018         .fini = efx_port_dummy_op_void,
3019         .map_reset_reason = efx_mcdi_map_reset_reason,
3020         .map_reset_flags = efx_ef10_map_reset_flags,
3021         .reset = efx_mcdi_reset,
3022         .probe_port = efx_mcdi_port_probe,
3023         .remove_port = efx_mcdi_port_remove,
3024         .fini_dmaq = efx_ef10_fini_dmaq,
3025         .describe_stats = efx_ef10_describe_stats,
3026         .update_stats = efx_ef10_update_stats,
3027         .start_stats = efx_mcdi_mac_start_stats,
3028         .stop_stats = efx_mcdi_mac_stop_stats,
3029         .set_id_led = efx_mcdi_set_id_led,
3030         .push_irq_moderation = efx_ef10_push_irq_moderation,
3031         .reconfigure_mac = efx_ef10_mac_reconfigure,
3032         .check_mac_fault = efx_mcdi_mac_check_fault,
3033         .reconfigure_port = efx_mcdi_port_reconfigure,
3034         .get_wol = efx_ef10_get_wol,
3035         .set_wol = efx_ef10_set_wol,
3036         .resume_wol = efx_port_dummy_op_void,
3037         /* TODO: test_chip */
3038         .test_nvram = efx_mcdi_nvram_test_all,
3039         .mcdi_request = efx_ef10_mcdi_request,
3040         .mcdi_poll_response = efx_ef10_mcdi_poll_response,
3041         .mcdi_read_response = efx_ef10_mcdi_read_response,
3042         .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
3043         .irq_enable_master = efx_port_dummy_op_void,
3044         .irq_test_generate = efx_ef10_irq_test_generate,
3045         .irq_disable_non_ev = efx_port_dummy_op_void,
3046         .irq_handle_msi = efx_ef10_msi_interrupt,
3047         .irq_handle_legacy = efx_ef10_legacy_interrupt,
3048         .tx_probe = efx_ef10_tx_probe,
3049         .tx_init = efx_ef10_tx_init,
3050         .tx_remove = efx_ef10_tx_remove,
3051         .tx_write = efx_ef10_tx_write,
3052         .rx_push_indir_table = efx_ef10_rx_push_indir_table,
3053         .rx_probe = efx_ef10_rx_probe,
3054         .rx_init = efx_ef10_rx_init,
3055         .rx_remove = efx_ef10_rx_remove,
3056         .rx_write = efx_ef10_rx_write,
3057         .rx_defer_refill = efx_ef10_rx_defer_refill,
3058         .ev_probe = efx_ef10_ev_probe,
3059         .ev_init = efx_ef10_ev_init,
3060         .ev_fini = efx_ef10_ev_fini,
3061         .ev_remove = efx_ef10_ev_remove,
3062         .ev_process = efx_ef10_ev_process,
3063         .ev_read_ack = efx_ef10_ev_read_ack,
3064         .ev_test_generate = efx_ef10_ev_test_generate,
3065         .filter_table_probe = efx_ef10_filter_table_probe,
3066         .filter_table_restore = efx_ef10_filter_table_restore,
3067         .filter_table_remove = efx_ef10_filter_table_remove,
3068         .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
3069         .filter_insert = efx_ef10_filter_insert,
3070         .filter_remove_safe = efx_ef10_filter_remove_safe,
3071         .filter_get_safe = efx_ef10_filter_get_safe,
3072         .filter_clear_rx = efx_ef10_filter_clear_rx,
3073         .filter_count_rx_used = efx_ef10_filter_count_rx_used,
3074         .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
3075         .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
3076 #ifdef CONFIG_RFS_ACCEL
3077         .filter_rfs_insert = efx_ef10_filter_rfs_insert,
3078         .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
3079 #endif
3080 #ifdef CONFIG_SFC_MTD
3081         .mtd_probe = efx_ef10_mtd_probe,
3082         .mtd_rename = efx_mcdi_mtd_rename,
3083         .mtd_read = efx_mcdi_mtd_read,
3084         .mtd_erase = efx_mcdi_mtd_erase,
3085         .mtd_write = efx_mcdi_mtd_write,
3086         .mtd_sync = efx_mcdi_mtd_sync,
3087 #endif
3088         .ptp_write_host_time = efx_ef10_ptp_write_host_time,
3089
3090         .revision = EFX_REV_HUNT_A0,
3091         .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
3092         .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
3093         .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
3094         .can_rx_scatter = true,
3095         .always_rx_scatter = true,
3096         .max_interrupt_mode = EFX_INT_MODE_MSIX,
3097         .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
3098         .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3099                              NETIF_F_RXHASH | NETIF_F_NTUPLE),
3100         .mcdi_max_ver = 2,
3101         .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
3102 };