xref: /titanic_44/usr/src/uts/common/io/bnxe/577xx/drivers/common/lm/l4/lm_l4rx.c (revision d14abf155341d55053c76eeec58b787a456b753b)
1 
2 #include "lm5710.h"
3 #include "bd_chain.h"
4 #include "command.h"
5 #include "context.h"
6 #include "lm_l4fp.h"
7 #include "lm_l4sp.h"
8 #include "mm_l4if.h"
9 #include "mm.h"
10 
11 /* The maximum counter value for consumed count, if it exceeds this value we post it to firmware.
12  * FW holds 32bits for this counter. Therefore 100MB is OK (see L4 VBD spec) */
13 #define MAX_GRQ_COUNTER               0x6400000
14 #define IS_OOO_CQE(__cmd)   ((__cmd == CMP_OPCODE_TOE_GNI) \
15                           || (__cmd == CMP_OPCODE_TOE_GAIR) \
16                           || (__cmd == CMP_OPCODE_TOE_GAIL) \
17                           || (__cmd == CMP_OPCODE_TOE_GRI) \
18                           || (__cmd == CMP_OPCODE_TOE_GJ) \
19                           || (__cmd == CMP_OPCODE_TOE_DGI))
20 
21 typedef struct toe_rx_bd toe_rx_bd_t;
22 
23 static u16_t lm_squeeze_rx_buffer_list(
24     struct _lm_device_t * pdev,
25     lm_tcp_state_t      * tcp,
26     u16_t                 adjust_number,
27     lm_tcp_gen_buf_t   ** unwanted_gen_buf
28     );
29 
30 static lm_status_t _lm_tcp_rx_post_buf(
31     struct _lm_device_t *pdev,
32     lm_tcp_state_t      *tcp,
33     lm_tcp_buffer_t     *tcp_buf,
34     lm_frag_list_t      *frag_list
35     );
36 
37 static void lm_tcp_incr_consumed_gen(
38     struct _lm_device_t * pdev,
39     lm_tcp_state_t      * tcp,
40     u32_t                 nbytes
41     );
42 
43 static void lm_tcp_return_gen_bufs(
44     struct _lm_device_t * pdev,
45     lm_tcp_state_t      * tcp,
46     lm_tcp_gen_buf_t    * gen_buf,
47     u32_t                 flags,
48     u8_t                  grq_idx
49     );
50 
51 static void lm_tcp_return_list_of_gen_bufs(
52     struct _lm_device_t * pdev,
53     lm_tcp_state_t      * tcp,
54     d_list_t            * gen_buf_list,
55     u32_t                 flags,
56     u8_t                  grq_idx
57     );
58 
_lm_tcp_isle_get_free_list(struct _lm_device_t * pdev,u8_t grq_idx)59 static lm_isle_t * _lm_tcp_isle_get_free_list(
60     struct _lm_device_t * pdev,
61     u8_t                  grq_idx)
62 {
63     lm_isle_t * free_list = NULL;
64     lm_isle_t * isles_pool = pdev->toe_info.grqs[grq_idx].isles_pool;
65     u32_t   isle_pool_idx;
66     u32_t   isle_pool_size = pdev->params.l4_isles_pool_size;
67     DbgBreakIf(!isles_pool);
68     for (isle_pool_idx = 0; isle_pool_idx < isle_pool_size; isle_pool_idx++) {
69         if ((isles_pool[isle_pool_idx].isle_link.next == NULL) && (isles_pool[isle_pool_idx].isle_link.prev == NULL)) {
70             free_list = isles_pool + isle_pool_idx;
71             break;
72         }
73     }
74     DbgBreakIf(!free_list);
75     return free_list;
76 }
77 
_lm_tcp_isle_find(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u8_t num_isle)78 static lm_isle_t *  _lm_tcp_isle_find(
79     struct _lm_device_t * pdev,
80     lm_tcp_state_t      * tcp,
81     u8_t                  num_isle)
82 {
83     lm_isle_t * isle = NULL;
84     lm_tcp_con_rx_gen_info_t * gen_info;
85     u8_t        isle_cnt, isle_idx;
86 
87     DbgBreakIf(!(tcp && tcp->rx_con));
88     gen_info = &tcp->rx_con->u.rx.gen_info;
89     isle_cnt = (u8_t)d_list_entry_cnt(&gen_info->isles_list);
90     DbgBreakIf(!isle_cnt);
91     DbgBreakIf(num_isle > isle_cnt);
92     if (num_isle == gen_info->current_isle_number) {
93         isle = gen_info->current_isle;
94     } else {
95         isle = (lm_isle_t*)gen_info->isles_list.head;
96         for (isle_idx = 1; isle_idx < num_isle; isle_idx++) {
97             isle = (lm_isle_t*)d_list_next_entry(&isle->isle_link);
98         }
99         gen_info->current_isle_number = num_isle;
100         gen_info->current_isle = isle;
101     }
102     return isle;
103 }
104 
_lm_tcp_isle_remove(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u8_t grq_idx,u8_t num_isle,d_list_t * gen_buf_list)105 static u32_t _lm_tcp_isle_remove(
106     struct _lm_device_t * pdev,
107     lm_tcp_state_t      * tcp,
108     u8_t                  grq_idx,
109     u8_t                  num_isle,
110     d_list_t            * gen_buf_list)
111 {
112     u32_t       nbytes = 0;
113     lm_isle_t * new_current_isle = NULL;
114     lm_isle_t * isle = NULL;
115     lm_tcp_con_rx_gen_info_t * gen_info;
116     u8_t        isles_cnt;
117     u8_t        new_current_isle_num;
118 
119     DbgBreakIf(!(tcp && tcp->rx_con));
120     gen_info = &tcp->rx_con->u.rx.gen_info;
121     isles_cnt = (u8_t)d_list_entry_cnt(&gen_info->isles_list);
122     DbgBreakIf(!(num_isle && (num_isle <= isles_cnt)));
123     isle = _lm_tcp_isle_find(pdev,tcp,num_isle);
124 
125 //    DbgBreakIf((isles_cnt > 1) && (num_isle == 1));
126     if (isle->isle_link.next != NULL) {
127         new_current_isle = (lm_isle_t*)isle->isle_link.next;
128         new_current_isle_num = num_isle;
129     } else if (isle->isle_link.prev != NULL) {
130         new_current_isle = (lm_isle_t*)isle->isle_link.prev;
131         new_current_isle_num = num_isle - 1;
132     } else {
133         new_current_isle = NULL;
134         new_current_isle_num = 0;
135     }
136 
137 #if defined(_NTDDK_)
138 #pragma prefast (push)
139 #pragma prefast (disable:6011)
140 #endif //_NTDDK_
141     d_list_remove_entry(&gen_info->isles_list, &isle->isle_link);
142 #if defined(_NTDDK_)
143 #pragma prefast (pop)
144 #endif //_NTDDK_
145 
146     nbytes = isle->isle_nbytes;
147     d_list_add_tail(gen_buf_list, &isle->isle_gen_bufs_list_head);
148     d_list_init(&isle->isle_gen_bufs_list_head, NULL, NULL, 0);
149     if (new_current_isle_num) {
150         if (num_isle == 1) {
151 #if defined(_NTDDK_)
152 #pragma prefast (push)
153 #pragma prefast (disable:28182)
154 #endif //_NTDDK_
155             d_list_remove_entry(&gen_info->isles_list, &new_current_isle->isle_link);
156 #if defined(_NTDDK_)
157 #pragma prefast (pop)
158 #endif //_NTDDK_
159             d_list_add_tail(&isle->isle_gen_bufs_list_head, &new_current_isle->isle_gen_bufs_list_head);
160             d_list_push_head(&gen_info->isles_list, &isle->isle_link);
161             isle->isle_nbytes = new_current_isle->isle_nbytes;
162 #ifdef DEBUG_OOO_CQE
163             isle->dedicated_cid = new_current_isle->dedicated_cid;
164             isle->recent_ooo_combined_cqe  = new_current_isle->recent_ooo_combined_cqe;
165 #endif
166             isle = new_current_isle;
167             new_current_isle = &gen_info->first_isle;
168         }
169         mm_mem_zero(&isle->isle_gen_bufs_list_head, sizeof(lm_isle_t) - sizeof(d_list_entry_t));
170         isle->isle_link.next = isle->isle_link.prev = NULL;
171     }
172     gen_info->current_isle = new_current_isle;
173     gen_info->current_isle_number = new_current_isle_num;
174     return nbytes;
175 }
176 
177 u32_t lm_tcp_rx_peninsula_to_rq(lm_device_t * pdev, lm_tcp_state_t * tcp, u32_t max_num_bytes_to_copy, u8_t sb_idx);
178 
179 /* TODO: remove this temporary solution for solaris / linux compilation conflict, linux needs the
180  * first option, solaris the latter */
181 #if defined(__LINUX)
182 #define TOE_RX_INIT_ZERO {{0}}
183 #else
184 #define TOE_RX_INIT_ZERO {0}
185 #endif
186 
187 #define TOE_RX_DOORBELL(pdev,cid) do{\
188     struct doorbell db = TOE_RX_INIT_ZERO;\
189     db.header.data |= ((TOE_CONNECTION_TYPE << DOORBELL_HDR_T_CONN_TYPE_SHIFT) |\
190                     (DOORBELL_HDR_T_RX << DOORBELL_HDR_T_RX_SHIFT));\
191     DOORBELL((pdev), (cid), *((u32_t *)&db));\
192     } while(0)
193 
lm_tcp_rx_write_db(lm_device_t * pdev,lm_tcp_state_t * tcp)194 static __inline void lm_tcp_rx_write_db(
195     lm_device_t *pdev,
196     lm_tcp_state_t *tcp
197     )
198 {
199     lm_tcp_con_t *rx_con = tcp->rx_con;
200     volatile struct toe_rx_db_data *db_data = rx_con->db_data.rx;
201 
202     if (!(rx_con->flags & TCP_RX_DB_BLOCKED)) {
203         db_data->bds_prod += rx_con->db_more_bds;               /* nbds should be written before nbytes (FW assumption) */
204         db_data->bytes_prod += rx_con->db_more_bytes;
205 
206         DbgMessage(pdev, INFORMl4rx,
207                     "_lm_tcp_rx_write_db: cid=%d, (nbytes+=%d, nbds+=%d)\n",
208                     tcp->cid, rx_con->db_more_bytes, rx_con->db_more_bds);
209         TOE_RX_DOORBELL(pdev, tcp->cid);
210     }
211 
212     /* assert if the new addition will make the cyclic counter post_cnt smaller than comp_cnt */
213     DbgBreakIf(S64_SUB(rx_con->bytes_post_cnt + rx_con->db_more_bytes, rx_con->bytes_comp_cnt) < 0);
214     rx_con->bytes_post_cnt += rx_con->db_more_bytes;
215     rx_con->buffer_post_cnt += rx_con->db_more_bufs;
216     rx_con->db_more_bytes = rx_con->db_more_bds = rx_con->db_more_bufs = 0;
217     rx_con->fp_db_cnt++;
218 }
219 
220 /** Description
221  *  This function is used to increase the window-size. Window is increased in 3 cases:
222  *  1. RQ-placed bytes
223  *  2. GRQ-Indicated succesfully (short/long loop, doensn't matter)
224  *  3. Window-update from NDIS (initial rcv window increased)
225  *  4. This function also takes into account dwa: delayed window algorithm, and updates the
226  *     data structures accordingly, however, not all window-updates are part of the dwa algorithm,
227  *     specifically, (3) therefore, we need to know if the update is dwa-aware or not.
228  */
lm_tcp_rx_post_sws(lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_con_t * rx_con,u32_t nbytes,u8_t op)229 void lm_tcp_rx_post_sws (
230     lm_device_t    * pdev,
231     lm_tcp_state_t * tcp,
232     lm_tcp_con_t   * rx_con,
233     u32_t            nbytes,
234     u8_t             op
235     )
236 {
237     volatile struct toe_rx_db_data *db_data = rx_con->db_data.rx;
238     s32_t diff_to_fw;
239 
240     switch (op)
241     {
242     case TCP_RX_POST_SWS_INC:
243         /*DbgMessage(pdev, FATAL, "lm_tcp_rx_post_sws() INC: OLD drv_rcv_win_right_edge=%d, nbytes=%d, NEW drv_rcv_win_right_edge=%d FW right_edge=%d \n", rx_con->u.rx.sws_info.drv_rcv_win_right_edge, nbytes, rx_con->u.rx.sws_info.drv_rcv_win_right_edge + nbytes, db_data->rcv_win_right_edge);*/
244         if (rx_con->u.rx.sws_info.extra_bytes > nbytes) {
245             rx_con->u.rx.sws_info.extra_bytes -= nbytes;
246             nbytes = 0;
247         } else {
248             nbytes -= rx_con->u.rx.sws_info.extra_bytes;
249             rx_con->u.rx.sws_info.extra_bytes = 0;
250             rx_con->u.rx.sws_info.drv_rcv_win_right_edge += nbytes;
251             if (rx_con->u.rx.sws_info.drv_rcv_win_right_edge >= db_data->rcv_win_right_edge) {
252                   RESET_FLAGS(tcp->rx_con->db_data.rx->flags, TOE_RX_DB_DATA_IGNORE_WND_UPDATES);
253             }
254         }
255         break;
256     case TCP_RX_POST_SWS_DEC:
257         if (rx_con->u.rx.sws_info.extra_bytes) {
258             rx_con->u.rx.sws_info.extra_bytes += nbytes;
259             nbytes = 0;
260         }
261         /*DbgMessage(pdev, FATAL, "lm_tcp_rx_post_sws() DEC: OLD drv_rcv_win_right_edge=%d, nbytes=%d, NEW drv_rcv_win_right_edge=%d\n", rx_con->u.rx.sws_info.drv_rcv_win_right_edge, nbytes, rx_con->u.rx.sws_info.drv_rcv_win_right_edge - nbytes);*/
262         rx_con->u.rx.sws_info.drv_rcv_win_right_edge -= nbytes;
263         SET_FLAGS(db_data->flags, TOE_RX_DB_DATA_IGNORE_WND_UPDATES);
264         break;
265     case TCP_RX_POST_SWS_SET:
266         /*DbgMessage(pdev, FATAL, "lm_tcp_rx_post_sws() SET: nbytes=%d\n", nbytes);*/
267         db_data->rcv_win_right_edge = nbytes;
268         rx_con->u.rx.sws_info.extra_bytes = 0;;
269         break;
270     default:
271         DbgBreakMsg("lm_tcp_rx_post_sws: Invalid operation\n");
272         return;
273     }
274 
275     /* note that diff_to_fw could be negative due to possibility of window-decrease in LH */
276     diff_to_fw = S32_SUB(rx_con->u.rx.sws_info.drv_rcv_win_right_edge, db_data->rcv_win_right_edge);
277 
278     /* If this update isn't dwa_aware, it's good to go... */
279 
280     //DbgMessage(pdev, WARNl4, "###lm_tcp_rx_post_sws cid=%d num_bytes=%d diff_to_fw=%d \n", tcp->cid, nbytes, diff_to_fw );
281     /* we give the window only if diff_to_fw is larger than mss, which also means only in case it is negative... */
282     if ( ((diff_to_fw >= (s32_t)rx_con->u.rx.sws_info.mss) ||
283          (diff_to_fw >= (((s32_t)tcp->tcp_cached.initial_rcv_wnd) / 2)))) {
284         if (rx_con->u.rx.sws_info.timer_on) {
285             /* Vladz TBD: Cancel the timer */
286             rx_con->u.rx.sws_info.timer_on = 0;
287         }
288 
289         /* Ring the Advertise Window doorbell here */
290         if (!(tcp->rx_con->flags & TCP_RX_DB_BLOCKED) && !(tcp->rx_con->flags & TCP_RX_POST_BLOCKED)) {
291             db_data->rcv_win_right_edge = rx_con->u.rx.sws_info.drv_rcv_win_right_edge;
292             DbgMessage(pdev, INFORMl4rx,
293                 "_lm_tcp_adv_wnd_write_db: cid=%d, nbytes=%d\n",
294                 tcp->cid, diff_to_fw);
295             TOE_RX_DOORBELL(pdev, tcp->cid);
296         }
297     } else {
298         if ( ! rx_con->u.rx.sws_info.timer_on ) {
299             /* Vladz TBD: schedule the timer here */
300             rx_con->u.rx.sws_info.timer_on = 1;
301         }
302     }
303 }
304 
_lm_tcp_rx_set_bd(IN lm_frag_t * frag,IN u16_t flags,IN lm_bd_chain_t * rx_chain,IN u32_t dbg_bytes_prod)305 static __inline toe_rx_bd_t * _lm_tcp_rx_set_bd (
306     IN   lm_frag_t         * frag,
307     IN   u16_t               flags,
308     IN   lm_bd_chain_t     * rx_chain,
309     IN   u32_t               dbg_bytes_prod /* Used for synchronizing between fw and driver rq-available-bytes
310                                              * This is used only as a debug variable for asserting in the fw. */
311     )
312 {
313     struct toe_rx_bd * rx_bd;
314 
315     /* hw limit: each bd can point to a buffer with max size of 64KB */
316     DbgBreakIf(frag->size > TCP_MAX_SGE_SIZE || frag->size == 0);
317     rx_bd = (struct toe_rx_bd *)lm_toe_bd_chain_produce_bd(rx_chain);
318     rx_bd->addr_hi = frag->addr.as_u32.high;
319     rx_bd->addr_lo = frag->addr.as_u32.low;
320     rx_bd->flags = flags;
321     rx_bd->size = (u16_t)frag->size;
322     rx_bd->dbg_bytes_prod = dbg_bytes_prod;
323     DbgMessage(NULL, VERBOSEl4rx, "Setting Rx BD flags=0x%x, bd_addr=0x%p, size=%d\n", rx_bd->flags, rx_bd, frag->size);
324     return rx_bd;
325 }
326 
327 
328 /** Description
329  *  function completes nbytes on a single tcp buffer and completes the buffer if it is
330  *  completed.
331  * Assumptions:
332  *   fp-lock is taken.
333  *   It is only called from lm_tcp_rx_post_buf!!!
334  */
lm_tcp_complete_tcp_buf(lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_con_t * con,lm_tcp_buffer_t * tcp_buf,u32_t completed_bytes)335 static void lm_tcp_complete_tcp_buf(
336     lm_device_t * pdev, lm_tcp_state_t * tcp, lm_tcp_con_t * con, lm_tcp_buffer_t * tcp_buf, u32_t completed_bytes)
337 {
338     s_list_t completed_bufs;
339     s_list_entry_t * entry;
340 
341     DbgBreakIf(completed_bytes > tcp_buf->more_to_comp);
342     tcp_buf->more_to_comp -= completed_bytes;
343     con->app_buf_bytes_acc_comp += completed_bytes;
344 
345     if(tcp_buf->more_to_comp == 0 &&  GET_FLAGS(tcp_buf->flags, TCP_BUF_FLAG_L4_POST_END)) {
346         tcp_buf->app_buf_xferred = con->app_buf_bytes_acc_comp;
347         DbgBreakIf(tcp_buf->app_buf_xferred > tcp_buf->app_buf_size); /* this may be partial completion */
348         con->app_buf_bytes_acc_comp = 0;
349         if (GET_FLAGS(con->flags, TCP_POST_COMPLETE_SPLIT)) {
350             RESET_FLAGS(con->flags, TCP_POST_COMPLETE_SPLIT);
351         }
352     } else {
353         tcp_buf->app_buf_xferred = 0;
354     }
355 
356     if (tcp_buf->more_to_comp == 0) {
357         /* should have nothing in the active tb list except this buffer, if we're completing this buffer,
358          * it means that we had something in the peninsula, this means that at the end of the DPC there was
359          * nothing in the active-tb-list, and between DPCs all posted buffers 'occupied' bytes from the peninsula
360          * and were completed to the client. This means that there can be no RQ completions during the DPC that
361          * will try to access the active tb list w/o a lock
362          */
363         DbgBreakIf(s_list_entry_cnt(&con->active_tb_list) != 1);
364         lm_bd_chain_bds_consumed(&con->bd_chain, tcp_buf->bd_used);
365 
366         con->buffer_completed_cnt ++;
367         DbgMessage(pdev, VERBOSEl4fp,
368                     "cid=%d, completing tcp buf towards mm from post-flow, actual_completed_bytes=%d\n",
369                     tcp->cid, tcp_buf->size);
370         entry = s_list_pop_head(&con->active_tb_list);
371         DbgBreakIf(con->rq_nbytes < tcp_buf->size);
372         con->rq_nbytes -= tcp_buf->size;
373         s_list_init(&completed_bufs, entry, entry, 1);
374         con->rq_completion_calls++;
375         mm_tcp_complete_bufs(pdev, tcp, con, &completed_bufs, LM_STATUS_SUCCESS);
376     }
377 }
378 
379 
380 
lm_tcp_rx_cmp_process(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t completed_bytes,u8_t push)381 void lm_tcp_rx_cmp_process(
382     struct _lm_device_t * pdev,
383     lm_tcp_state_t      * tcp,
384     u32_t                 completed_bytes,
385     u8_t                  push
386     )
387 {
388     lm_tcp_con_t *rx_con;
389     u32_t actual_bytes_completed;
390     MM_INIT_TCP_LOCK_HANDLE();
391 
392     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_cmp_process, completed_bytes=%d, push=%d cid=%d\n", completed_bytes, push, tcp->cid);
393     DbgBreakIf(!(completed_bytes || push)); /* otherwise there is no point for this function to be called */
394 
395     rx_con = tcp->rx_con;
396     DbgBreakIf(! rx_con);
397 
398     if (!(rx_con->flags & TCP_DEFERRED_PROCESSING)) {
399         mm_acquire_tcp_lock(pdev, rx_con);
400     }
401     DbgBreakIf(rx_con->flags & TCP_RX_COMP_BLOCKED);
402 
403     /* RQ completions can't arrive while we have something in the peninsula (peninsula must either be completed or copied
404      * to the app-buffer before) An RQ_SKP within the dpc will always take care of previous RQs waiting to be copied to. */
405     DbgBreakIf(!d_list_is_empty(&rx_con->u.rx.gen_info.peninsula_list));
406     DbgBreakIf(!d_list_is_empty(&rx_con->u.rx.gen_info.dpc_peninsula_list));
407 
408     actual_bytes_completed = lm_tcp_complete_nbytes(pdev, tcp, rx_con, completed_bytes , push);
409 
410     rx_con->bytes_comp_cnt += actual_bytes_completed;
411     DbgBreakIf(S64_SUB(rx_con->bytes_post_cnt, rx_con->bytes_comp_cnt) < 0);
412     DbgMessage(pdev, VERBOSEl4rx, "lm_tcp_rx_comp, after comp: pending=%d, active_bufs=%d\n",
413                 S64_SUB(rx_con->bytes_post_cnt, rx_con->bytes_comp_cnt),
414                 s_list_entry_cnt(&rx_con->active_tb_list));
415 
416     if ( completed_bytes ) {
417         /* Vladz: TBD
418         lm_neigh_update_nic_reachability_time(tcp->path->neigh) */
419     }
420     if (!(rx_con->flags & TCP_DEFERRED_PROCESSING)) {
421         mm_release_tcp_lock(pdev, rx_con);
422     }
423 } /* lm_tcp_rx_comp */
424 
425 
lm_tcp_rx_skp_process(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t bytes_skipped,u8_t sb_idx)426 void lm_tcp_rx_skp_process(
427     struct _lm_device_t * pdev,
428     lm_tcp_state_t      * tcp,
429     u32_t                 bytes_skipped,
430     u8_t                  sb_idx
431     )
432 {
433     lm_tcp_con_t *rx_con;
434     u32_t comp_bytes;
435     MM_INIT_TCP_LOCK_HANDLE();
436 
437     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_skp_process, bytes_skipped=%d, cid=%d\n", bytes_skipped, tcp->cid);
438 
439     if (bytes_skipped == 0) {
440         /* nothing to do here - occurs on special fw case, where there is GRQ->RQ processing with no GRQ and no RQ,
441          * this will usually happen at the beginning or in special cases of the connection */
442         return;
443     }
444 
445     rx_con = tcp->rx_con;
446     DbgBreakIf(! rx_con);
447 
448     if (!GET_FLAGS(rx_con->flags, TCP_DEFERRED_PROCESSING)) {
449         mm_acquire_tcp_lock(pdev, rx_con);
450     }
451     DbgBreakIf(GET_FLAGS(rx_con->flags, TCP_RX_COMP_BLOCKED));
452 
453     comp_bytes = min(bytes_skipped, tcp->rx_con->u.rx.skp_bytes_copied);
454     if (comp_bytes) {
455         tcp->rx_con->bytes_comp_cnt += comp_bytes;
456         /* complete nbytes on buffers (dpc-flow ) */
457         lm_tcp_complete_nbytes(pdev, tcp, tcp->rx_con, comp_bytes, /* push=*/ 0);
458         bytes_skipped -= comp_bytes;
459         tcp->rx_con->u.rx.skp_bytes_copied -= comp_bytes;
460     }
461 
462     /* We know for sure, that all the application buffers we are about to access have already been posted
463      * before the dpc, and therefore are valid in the active_tb_list.
464      * TBA Michals: bypass FW
465      */
466     if (bytes_skipped) {
467         DbgBreakIf(!d_list_is_empty(&rx_con->u.rx.gen_info.peninsula_list));
468         DbgBreakIfAll(d_list_is_empty(&rx_con->u.rx.gen_info.dpc_peninsula_list));
469         DbgBreakIf(((lm_tcp_gen_buf_t *)d_list_peek_head(&rx_con->u.rx.gen_info.dpc_peninsula_list))->placed_bytes == 0);
470         rx_con->u.rx.gen_info.bytes_copied_cnt_in_process += lm_tcp_rx_peninsula_to_rq(pdev, tcp, bytes_skipped,sb_idx);
471     }
472 
473     if (!GET_FLAGS(rx_con->flags, TCP_DEFERRED_PROCESSING)) {
474         mm_release_tcp_lock(pdev, rx_con);
475     }
476 } /* lm_tcp_rx_skp */
477 
lm_tcp_rx_delete_isle(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u8_t sb_idx,u8_t num_isle,u32_t num_of_isles)478 void lm_tcp_rx_delete_isle(
479     struct _lm_device_t * pdev,
480     lm_tcp_state_t      * tcp,
481     u8_t                  sb_idx,
482     u8_t                  num_isle,
483     u32_t                 num_of_isles)
484 {
485     lm_tcp_con_t             * rx_con = tcp->rx_con;
486     lm_tcp_con_rx_gen_info_t * gen_info;
487     d_list_t                   removed_list;
488     u32_t                      isle_nbytes;
489 
490 
491 
492     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_delete_isle cid=%d isle=%d\n", tcp->cid, num_isle);
493     gen_info = &rx_con->u.rx.gen_info;
494     d_list_init(&removed_list, NULL, NULL, 0);
495 
496     while (num_of_isles) {
497         isle_nbytes = _lm_tcp_isle_remove(pdev, tcp, sb_idx, num_isle + (num_of_isles - 1), &removed_list);
498         pdev->toe_info.grqs[sb_idx].number_of_isles_delta--;
499         DbgBreakIf(isle_nbytes > gen_info->isle_nbytes);
500         gen_info->isle_nbytes -= isle_nbytes;
501         num_of_isles--;
502     }
503 
504     pdev->toe_info.grqs[sb_idx].gen_bufs_in_isles_delta -= (s32_t)d_list_entry_cnt(&removed_list);
505     if (!d_list_is_empty(&removed_list)) {
506         lm_tcp_return_list_of_gen_bufs(pdev,tcp ,&removed_list, MM_TCP_RGB_COLLECT_GEN_BUFS, sb_idx);
507         tcp->rx_con->droped_non_empty_isles++;
508     } else {
509         DbgBreak();
510         tcp->rx_con->droped_empty_isles++;
511     }
512     rx_con->dpc_info.dpc_flags &= ~(LM_TCP_DPC_TOO_BIG_ISLE | LM_TCP_DPC_TOO_MANY_ISLES);
513     return;
514 }
515 
lm_toe_is_rx_completion(lm_device_t * pdev,u8_t drv_toe_rss_id)516 u8_t lm_toe_is_rx_completion(lm_device_t *pdev, u8_t drv_toe_rss_id)
517 {
518     u8_t result = FALSE;
519     lm_tcp_rcq_t *rcq = NULL;
520 
521     DbgBreakIf(!(pdev && ARRSIZE(pdev->toe_info.rcqs) > drv_toe_rss_id));
522 
523     rcq = &pdev->toe_info.rcqs[drv_toe_rss_id];
524 
525     if ( rcq->hw_con_idx_ptr &&
526         *rcq->hw_con_idx_ptr != lm_bd_chain_cons_idx(&rcq->bd_chain) )
527     {
528         result = TRUE;
529     }
530     DbgMessage(pdev, INFORMl4int, "lm_toe_is_rx_completion(): result is:%s\n", result? "TRUE" : "FALSE");
531 
532     return result;
533 }
534 
535 /** Description
536  *  checks if the processing of a certain RCQ is suspended
537  */
lm_toe_is_rcq_suspended(lm_device_t * pdev,u8_t drv_toe_rss_id)538 u8_t lm_toe_is_rcq_suspended(lm_device_t *pdev, u8_t drv_toe_rss_id)
539 {
540     u8_t result = FALSE;
541     lm_tcp_rcq_t *rcq = NULL;
542 
543     if (drv_toe_rss_id < MAX_L4_RX_CHAIN)
544     {
545         rcq = &pdev->toe_info.rcqs[drv_toe_rss_id];
546         if (rcq->suspend_processing) {
547             result = TRUE;
548 	}
549     }
550     DbgMessage(pdev, INFORMl4int, "lm_toe_is_rcq_suspended(): sb_idx:%d, result is:%s\n", drv_toe_rss_id, result?"TRUE":"FALSE");
551     return result;
552 }
553 
554 
555 /** Description
556  *  Increment consumed generic counter for a connection.
557  *  To avoid rollover in the FW if the counter exceeds a maximum threshold, the driver should
558  *  not wait for application buffers and post 'receive window update' doorbell immediately.
559  *  The FW holds 32bits for this counter. Therefore a threshold of 100MB is OK.
560  */
lm_tcp_incr_consumed_gen(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t nbytes)561 static void lm_tcp_incr_consumed_gen(
562     struct _lm_device_t * pdev,
563     lm_tcp_state_t      * tcp,
564     u32_t                 nbytes
565     )
566 {
567     volatile struct toe_rx_db_data *db_data = tcp->rx_con->db_data.rx;
568 
569     db_data->consumed_grq_bytes += nbytes;
570 
571     /* theres no need to increased the consumed_cnt in two stages (one in the driver and one for FW db_data)
572      * we can always directly increase FW db_data, we need to decide whether we need to give a doorbell, basically
573      * we have two cases where doorbells are given: (1) buffer posted and bypasses fw (2) indication succeeded in which case
574      * window will also be increased, however, the window isn't always increased: if it's smaller than MSS, so, if we
575      * increase the consumed count by something smaller than mss - we'll give the doorbell here...   */
576 
577     if (nbytes < tcp->rx_con->u.rx.sws_info.mss) {
578         if (!(tcp->rx_con->flags & TCP_RX_DB_BLOCKED)) {
579             TOE_RX_DOORBELL(pdev, tcp->cid);
580         }
581     }
582 }
583 
584 /** Description
585  *  Copies as many bytes as possible from the peninsula to the single tcp buffer received
586  *  updates the peninsula.
587  *  This function can be called from two flows:
588  *  1. Post of a buffer
589  *  2. Completion of a dpc.
590  *  We need to know which flow it is called from to know which peninsula list to use:
591  *    dpc_peninsula_list / peninsula_list.
592  *  Post ALWAYS uses the peninsula_list, since it doesn't know about the dpc_peninsula
593  *  Completion ALWAYS uses the dpc_peninsula_list, and in this case peninsula_list MUST be empty
594  *  this is because there can be buffers in the active_tb_list ONLY if peninsula_list is empty.
595  *
596  *  first_buf_offset refers to the peninsula we're dealing with, at the end of the dpc the dpc_peninsula
597  *  is copied to the peninsula, therefore first_buf_offset will still be valid. copying from post means that
598  *  there is something in the peninsula which means theres nothing in the active_tb_list ==> won't be a copy from
599  *  dpc. Copying from dpc means theres something in the active-tb-list ==> nothing in the peninsula ==> won't be called
600  *  from post, mutual exclusion exists between the post/dpc of copying, therefore we can have only one first_buffer_offset
601  *  all other accesses (indication) are done under a lock.
602  * param: dpc - indicates if this is called from the dpc or not (post)
603  * Assumptions:
604  *  tcp_buf->more_to_comp is initialized
605  *  tcp_buf->size         is initialized
606  *  num_bufs_complete is initialized by caller (could differ from zero)
607  * Returns
608  *  the actual number of bytes copied
609  *  num_bufs_complete is the number of buffers that were completely copied to the pool and can be
610  *  returned to the pool.
611  */
lm_tcp_rx_peninsula_to_rq_copy(lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_buffer_t * tcp_buf,d_list_t * return_list,u32_t max_num_bytes_to_copy,u8_t dpc)612 static u32_t lm_tcp_rx_peninsula_to_rq_copy(
613     lm_device_t     * pdev,
614     lm_tcp_state_t  * tcp,
615     lm_tcp_buffer_t * tcp_buf,
616     d_list_t        * return_list,
617     u32_t             max_num_bytes_to_copy,
618     u8_t              dpc)
619 {
620     lm_tcp_gen_buf_t         * curr_gen_buf;
621     lm_tcp_con_rx_gen_info_t * gen_info;
622     d_list_t                 * peninsula;
623     u32_t                      tcp_offset;
624     u32_t                      ncopy;
625     u32_t                      bytes_left;
626     u32_t                      bytes_copied = 0;
627 
628     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_peninsula_to_rq_copy tcp_buf = 0x%x cid=%d\n", *((u32_t *)&tcp_buf), tcp->cid);
629 
630     gen_info = &tcp->rx_con->u.rx.gen_info;
631 
632     if (dpc) {
633         peninsula = &gen_info->dpc_peninsula_list;
634     } else {
635         peninsula = &gen_info->peninsula_list;
636     }
637 
638     curr_gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_head(peninsula);
639     tcp_offset = tcp_buf->size - tcp_buf->more_to_comp;
640     bytes_left = min(tcp_buf->more_to_comp, max_num_bytes_to_copy); /* copy to buffer only what's aloud...*/
641 
642     /* start copying as much as possible from peninsula to tcp buffer */
643     while (bytes_left && curr_gen_buf && curr_gen_buf->placed_bytes) {
644         ncopy = curr_gen_buf->placed_bytes - gen_info->first_buf_offset;
645         if (ncopy > bytes_left) {
646             ncopy = bytes_left;
647         }
648         if (mm_tcp_copy_to_tcp_buf(pdev, tcp, tcp_buf,
649                                    curr_gen_buf->buf_virt + gen_info->first_buf_offset, /* start of data in generic buffer */
650                                    tcp_offset, ncopy) != ncopy)
651         {
652             gen_info->copy_gen_buf_dmae_cnt++;
653 
654             /* If this is  generic buffer that has the free_when_done flag on it means it's non-cached memory and not physical
655              * memory -> so, we can't try and dmae to it... not likely to happen... */
656             if (!GET_FLAGS(curr_gen_buf->flags, GEN_FLAG_FREE_WHEN_DONE)) {
657                 if (mm_tcp_rx_peninsula_to_rq_copy_dmae(pdev,
658                                                         tcp,
659                                                         curr_gen_buf->buf_phys,
660                                                         gen_info->first_buf_offset, /* start of data in generic buffer */
661                                                         tcp_buf,
662                                                         tcp_offset,
663                                                         ncopy) != ncopy)
664                 {
665                     DbgBreakMsg("Unable To Copy");
666                     gen_info->copy_gen_buf_fail_cnt++;
667 
668                     break;
669                 }
670              } else {
671                     DbgBreakMsg("Unable To Copy");
672                     gen_info->copy_gen_buf_fail_cnt++;
673 
674                     break;
675              }
676         }
677 
678         /* update peninsula */
679         bytes_copied += ncopy;
680 
681         gen_info->first_buf_offset += (u16_t)ncopy;
682 
683         /* done with the generic buffer? - return it to the pool */
684         if (curr_gen_buf->placed_bytes == gen_info->first_buf_offset) {
685             curr_gen_buf = (lm_tcp_gen_buf_t *)d_list_pop_head(peninsula);
686             d_list_push_tail(return_list, &curr_gen_buf->link);
687             gen_info->first_buf_offset = 0;
688             gen_info->num_buffers_copied_grq++;
689             curr_gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_head(peninsula);
690         }
691 
692         /* update tcp buf stuff */
693         bytes_left -= ncopy;
694         tcp_offset += ncopy;
695     }
696 
697     if (dpc) {
698         gen_info->dpc_peninsula_nbytes -= bytes_copied;
699     } else {
700         gen_info->peninsula_nbytes -= bytes_copied;
701     }
702 
703     /* return the number of bytes actually copied */
704     return bytes_copied;
705 }
706 
707 /** Description
708  *  function copies data from the peninsula to tcp buffers already placed in the
709  *  active_tb_list. The function completes the buffers if a tcp buffer from active_tb_list
710  *  was partially/fully filled. This case simulates a call to lm_tcp_rx_comp
711  *    (i.e. a completion received from firmware)
712  * Assumptions:
713  */
lm_tcp_rx_peninsula_to_rq(lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t max_num_bytes_to_copy,u8_t sb_idx)714 u32_t lm_tcp_rx_peninsula_to_rq(lm_device_t * pdev, lm_tcp_state_t * tcp, u32_t max_num_bytes_to_copy, u8_t sb_idx)
715 {
716     lm_tcp_buffer_t          * curr_tcp_buf;
717     lm_tcp_con_rx_gen_info_t * gen_info;
718     d_list_t                   return_list;
719     u32_t                      copied_bytes = 0, currently_copied = 0;
720 
721     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_peninsula_to_rq cid=%d\n", tcp->cid);
722 
723     gen_info     = &tcp->rx_con->u.rx.gen_info;
724 
725     DbgBreakIf(gen_info->peninsula_blocked == TRUE); /* terminate was already called - no copying should be done */
726 
727     /* Copy data from dpc_peninsula to tcp buffer[s] */
728     d_list_init(&return_list, NULL, NULL, 0);
729 
730     curr_tcp_buf = lm_tcp_next_entry_dpc_active_list(tcp->rx_con);
731 
732     /* TBA Michals: FW Bypass First check if we can copy to bypass buffers */
733 
734     /* Copy the number of bytes received in SKP */
735     while (max_num_bytes_to_copy && gen_info->dpc_peninsula_nbytes  && curr_tcp_buf) {
736         currently_copied = lm_tcp_rx_peninsula_to_rq_copy(pdev, tcp, curr_tcp_buf, &return_list, max_num_bytes_to_copy, TRUE);
737         curr_tcp_buf = (lm_tcp_buffer_t *)s_list_next_entry(&curr_tcp_buf->link);
738         DbgBreakIf(max_num_bytes_to_copy < currently_copied);
739         max_num_bytes_to_copy -= currently_copied;
740         copied_bytes += currently_copied;
741     }
742 
743     if (!d_list_is_empty(&return_list)) {
744 
745         lm_tcp_return_list_of_gen_bufs(pdev,tcp , &return_list,
746                                (sb_idx != NON_EXISTENT_SB_IDX) ? MM_TCP_RGB_COLLECT_GEN_BUFS : 0, sb_idx);
747     }
748 
749     /* If we've copied to a buffer in the active_tb_list we need to complete it since fw knows
750      * the driver has the bytes and the driver will take care of copying them and completing them.
751      * this path simulates a call to lm_tcp_rx_comp (buffers taken from active_tb_list) */
752     /* Note that pending bytes here could reach a negative value if a partial
753      * application buffer was posted and the doorbell hasn't been given yet, however,
754      * once the doorbell is given for the application buffer the pending bytes will reach a non-negative
755      * value (>=0) */
756     tcp->rx_con->bytes_comp_cnt += copied_bytes;
757     /* complete nbytes on buffers (dpc-flow ) */
758     lm_tcp_complete_nbytes(pdev, tcp, tcp->rx_con, copied_bytes, /* push=*/ 0);
759 
760     DbgMessage(pdev, VERBOSEl4rx, "lm_tcp_rx_peninsula_to_rq copied %d bytes cid=%d\n", copied_bytes, tcp->cid);
761     return copied_bytes;
762 }
763 
764 /** Description
765  *  determines whether or not we can indicate.
766  *  Rules:
767  *  - Indication is not blocked
768  *  - we are not in the middle of completion a split-buffer
769  *     we can only indicate after an entire buffer has been completed/copied to.
770  *     we determine this by the app_buf_bytes_acc_comp. This is to avoid the
771  *     following data integrity race:
772  *     application buffer: app_start, app_end
773  *     app_start is posted, peninsula copied to app_start, app_start completed to
774  *     fw then the rest is indicated. fw receives app_end, fw thinks peninsula was
775  *     copied to buffer, application buffer misses data...
776  *   - our active_tb_list is empty... we HAVE to make sure to
777  *     always indicate after we've fully utilized our RQ
778  *     buffers...
779  */
_lm_tcp_ok_to_indicate(lm_tcp_con_t * rx_con)780 static __inline u8_t _lm_tcp_ok_to_indicate(lm_tcp_con_t * rx_con)
781 {
782     return (!(rx_con->flags & TCP_RX_IND_BLOCKED) && (rx_con->app_buf_bytes_acc_comp == 0) &&
783             (s_list_is_empty(&rx_con->active_tb_list)));
784 }
785 
786 /** Description
787  *  GA: add a buffer to the peninsula - nbytes represents the number of bytes in the previous buffer.
788  *  GR: release a buffer from the peninsula - nbytes represents the number of bytes in the current buffer.
789  * Assumption:
790  *  GR can only be called on a buffer that had been added using GA before
791  */
lm_tcp_rx_gen_peninsula_process(lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t nbytes,lm_tcp_gen_buf_t * gen_buf)792 void lm_tcp_rx_gen_peninsula_process(lm_device_t * pdev, lm_tcp_state_t * tcp, u32_t nbytes, lm_tcp_gen_buf_t * gen_buf)
793 {
794     lm_tcp_con_t             * rx_con = tcp->rx_con;
795     lm_tcp_con_rx_gen_info_t * gen_info;
796     lm_tcp_gen_buf_t         * last_gen_buf;
797 
798     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_gen_peninsula_process, nbytes=%d, cid=%d add=%s\n", nbytes, tcp->cid,
799                 (gen_buf)? "TRUE"  : "FALSE");
800 
801     DbgBreakIf(rx_con->flags & TCP_RX_COMP_BLOCKED);
802 
803     gen_info = &rx_con->u.rx.gen_info;
804 
805     /* update the previous buffer OR current buffer if this is a release operation. This function is always called
806      * from within a DPC and updates the dpc_peninsula */
807     if (nbytes) {
808         gen_info->dpc_peninsula_nbytes += nbytes;
809         last_gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_tail(&gen_info->dpc_peninsula_list);
810         DbgBreakIfAll(last_gen_buf == NULL);
811         DbgBreakIfAll(last_gen_buf->placed_bytes != 0);
812         DbgBreakIfAll(nbytes > LM_TCP_GEN_BUF_SIZE(pdev));
813         last_gen_buf->placed_bytes = (u16_t)nbytes;
814     }
815 
816     if (gen_buf /* add */) {
817         DbgBreakIf(SIG(gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
818         DbgBreakIf(END_SIG(gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
819 
820         d_list_push_tail(&gen_info->dpc_peninsula_list, &gen_buf->link);
821     }
822 
823 }
824 
lm_tcp_rx_gen_isle_create(lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_gen_buf_t * gen_buf,u8_t sb_idx,u8_t isle_num)825 void lm_tcp_rx_gen_isle_create(lm_device_t * pdev, lm_tcp_state_t * tcp, lm_tcp_gen_buf_t * gen_buf, u8_t sb_idx, u8_t isle_num)
826 {
827     lm_isle_t                * current_isle = NULL;
828     lm_isle_t                * next_isle = NULL;
829     lm_tcp_con_t             * rx_con = tcp->rx_con;
830     lm_tcp_con_rx_gen_info_t * gen_info;
831     u8_t                       isles_cnt;
832     d_list_entry_t           * isle_entry_prev = NULL;
833     d_list_entry_t           * isle_entry_next = NULL;
834 
835     gen_info = &rx_con->u.rx.gen_info;
836     isles_cnt = (u8_t)d_list_entry_cnt(&gen_info->isles_list);
837     if (isles_cnt) {
838         DbgBreakIf(isles_cnt == T_TCP_MAX_ISLES_PER_CONNECTION_TOE);
839         current_isle = _lm_tcp_isle_get_free_list(pdev, sb_idx);
840         DbgBreakIf(!current_isle);
841 #ifdef DEBUG_OOO_CQE
842         DbgBreakIf(current_isle->dedicated_cid != 0);
843         current_isle->dedicated_cid = tcp->cid;
844 #endif
845     } else {
846         current_isle = &gen_info->first_isle;
847     }
848 
849     d_list_push_head(&current_isle->isle_gen_bufs_list_head, &gen_buf->link);
850     current_isle->isle_nbytes = 0;
851     if (isle_num == 1) {
852         if (current_isle != &gen_info->first_isle) {
853             *current_isle = gen_info->first_isle;
854             d_list_init(&gen_info->first_isle.isle_gen_bufs_list_head, NULL, NULL, 0);
855             d_list_push_head(&gen_info->first_isle.isle_gen_bufs_list_head, &gen_buf->link);
856             gen_info->first_isle.isle_nbytes = 0;
857             isle_entry_prev = &gen_info->first_isle.isle_link;
858             isle_entry_next = gen_info->first_isle.isle_link.next;
859         }
860     } else if (isle_num <= isles_cnt) {
861         next_isle = _lm_tcp_isle_find(pdev,tcp,isle_num);
862         isle_entry_prev = next_isle->isle_link.prev;
863         isle_entry_next = &next_isle->isle_link;
864     } else if (isle_num == (isles_cnt + 1)) {
865         isle_entry_next = NULL;
866         isle_entry_prev = gen_info->isles_list.tail;
867     } else {
868         DbgBreak();
869     }
870 
871     d_list_insert_entry(&gen_info->isles_list, isle_entry_prev, isle_entry_next, &current_isle->isle_link);
872     if (isle_num == 1) {
873         current_isle = &gen_info->first_isle;
874     }
875 #ifdef DEBUG_OOO_CQE
876     SET_DEBUG_OOO_INFO(current_isle, CMP_OPCODE_TOE_GNI, 0);
877 #endif
878     gen_info->current_isle = current_isle;
879     gen_info->current_isle_number = isle_num;
880     pdev->toe_info.grqs[sb_idx].number_of_isles_delta++;
881     if (isles_cnt == gen_info->max_number_of_isles) {
882         gen_info->max_number_of_isles++;
883     }
884 }
885 
lm_tcp_rx_gen_isle_right_process(lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t nbytes,lm_tcp_gen_buf_t * gen_buf,u8_t sb_idx,u8_t isle_num)886 void lm_tcp_rx_gen_isle_right_process(lm_device_t * pdev, lm_tcp_state_t * tcp, u32_t nbytes, lm_tcp_gen_buf_t * gen_buf, u8_t sb_idx, u8_t isle_num)
887 {
888     lm_tcp_con_t             * rx_con = tcp->rx_con;
889     lm_tcp_con_rx_gen_info_t * gen_info;
890     lm_tcp_gen_buf_t         * last_gen_buf;
891     lm_isle_t                * requested_isle;
892 
893     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_gen_isle_process nbytes = %d cid=%d\n", nbytes, tcp->cid);
894 
895     gen_info = &rx_con->u.rx.gen_info;
896     requested_isle = _lm_tcp_isle_find(pdev,tcp,isle_num);
897     DbgBreakIf(!requested_isle);
898 
899     /* update the previous buffer */
900     last_gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_tail(&requested_isle->isle_gen_bufs_list_head);
901     DbgBreakIf(last_gen_buf == NULL);
902     if (nbytes) {
903         gen_info->isle_nbytes += nbytes;
904         requested_isle->isle_nbytes += nbytes;
905         DbgBreakIf(last_gen_buf->placed_bytes != 0);
906         DbgBreakIf(nbytes > 0xffff);
907         last_gen_buf->placed_bytes = (u16_t)nbytes;
908     } else {
909         DbgBreakIf(gen_buf == NULL);
910         DbgBreakIf(last_gen_buf->placed_bytes == 0);
911     }
912 
913     if (gen_buf) {
914         DbgBreakIf(SIG(gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
915         DbgBreakIf(END_SIG(gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
916 
917         d_list_push_tail(&requested_isle->isle_gen_bufs_list_head, &gen_buf->link);
918         pdev->toe_info.grqs[sb_idx].gen_bufs_in_isles_delta++;
919         if (pdev->params.l4_max_gen_bufs_in_isle
920                 && (d_list_entry_cnt(&requested_isle->isle_gen_bufs_list_head) > pdev->params.l4_max_gen_bufs_in_isle)) {
921             if (pdev->params.l4_limit_isles & L4_LI_NOTIFY) {
922                 DbgBreak();
923             }
924             if (pdev->params.l4_limit_isles & L4_LI_MAX_GEN_BUFS_IN_ISLE) {
925                 rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_TOO_BIG_ISLE;
926             }
927         }
928 #ifdef DEBUG_OOO_CQE
929         SET_DEBUG_OOO_INFO(requested_isle, CMP_OPCODE_TOE_GAIR, nbytes);
930     } else {
931         SET_DEBUG_OOO_INFO(requested_isle, CMP_OPCODE_TOE_GRI, nbytes);
932 #endif
933     }
934 }
935 
lm_tcp_rx_gen_isle_left_process(lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t nbytes,lm_tcp_gen_buf_t * gen_buf,u8_t sb_idx,u8_t isle_num)936 void lm_tcp_rx_gen_isle_left_process(lm_device_t * pdev, lm_tcp_state_t * tcp, u32_t nbytes, lm_tcp_gen_buf_t * gen_buf, u8_t sb_idx, u8_t isle_num)
937 {
938     lm_tcp_con_t             * rx_con = tcp->rx_con;
939     lm_tcp_con_rx_gen_info_t * gen_info;
940     lm_tcp_gen_buf_t         * last_gen_buf;
941     lm_isle_t                * requested_isle;
942 
943     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_gen_isle_process nbytes = %d cid=%d\n", nbytes, tcp->cid);
944 
945     gen_info = &rx_con->u.rx.gen_info;
946     requested_isle = _lm_tcp_isle_find(pdev,tcp,isle_num);
947     DbgBreakIf(!requested_isle);
948 
949     if (nbytes) {
950         DbgBreakIf(!gen_info->wait_for_isle_left);
951         DbgBreakIf(gen_buf != NULL);
952         gen_info->wait_for_isle_left = FALSE;
953         gen_info->isle_nbytes += nbytes;
954         requested_isle->isle_nbytes += nbytes;
955 #if defined(_NTDDK_)
956 #pragma prefast (push)
957 #pragma prefast (disable:28182) // If nbytes is larger that zero than ((returned_list_of_gen_bufs))->head is not NULL.
958 #endif //_NTDDK_
959         last_gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_head(&requested_isle->isle_gen_bufs_list_head);
960         DbgBreakIf(last_gen_buf->placed_bytes);
961         last_gen_buf->placed_bytes = (u16_t)nbytes;
962 #if defined(_NTDDK_)
963 #pragma prefast (pop)
964 #endif //_NTDDK_
965     } else {
966         DbgBreakIf(gen_info->wait_for_isle_left);
967         DbgBreakIf(gen_buf == NULL);
968         DbgBreakIf(SIG(gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
969         DbgBreakIf(END_SIG(gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
970         gen_info->wait_for_isle_left = TRUE;
971         d_list_push_head(&requested_isle->isle_gen_bufs_list_head, &gen_buf->link);
972         pdev->toe_info.grqs[sb_idx].gen_bufs_in_isles_delta++;
973     }
974 #ifdef DEBUG_OOO_CQE
975     SET_DEBUG_OOO_INFO(requested_isle, CMP_OPCODE_TOE_GAIL, nbytes);
976 #endif
977 }
978 
lm_tcp_rx_gen_join_process(lm_device_t * pdev,lm_tcp_state_t * tcp,u8_t sb_idx,u8_t isle_num)979 void lm_tcp_rx_gen_join_process(lm_device_t * pdev, lm_tcp_state_t * tcp, u8_t sb_idx, u8_t isle_num)
980 {
981     lm_tcp_con_t             * rx_con = tcp->rx_con;
982     lm_tcp_con_rx_gen_info_t * gen_info;
983     lm_isle_t                * start_isle;
984     d_list_t                   gen_buf_list;
985     u32_t                      isle_nbytes;
986     DbgMessage(pdev, VERBOSEl4rx, "##lm_tcp_rx_gen_join_process cid=%d\n", tcp->cid);
987 
988     gen_info = &rx_con->u.rx.gen_info;
989 
990 
991     if (!isle_num) {
992         /* break if peninsula list isn't empty and the last buffer in list isn't released yet */
993         DbgBreakIf(d_list_entry_cnt(&gen_info->dpc_peninsula_list) &&
994                    ((lm_tcp_gen_buf_t *)(d_list_peek_tail(&gen_info->dpc_peninsula_list)))->placed_bytes == 0);
995         d_list_init(&gen_buf_list, NULL, NULL, 0);
996         isle_nbytes = _lm_tcp_isle_remove(pdev, tcp, sb_idx, 1, &gen_buf_list);
997 //        DbgBreakIf(!(isle_nbytes && d_list_entry_cnt(&gen_buf_list)));
998         if (d_list_entry_cnt(&gen_buf_list) > 1) {
999             DbgBreakIf(((lm_tcp_gen_buf_t *)(d_list_peek_head(&gen_buf_list)))->placed_bytes == 0);
1000         }
1001         pdev->toe_info.grqs[sb_idx].gen_bufs_in_isles_delta -= (s32_t)d_list_entry_cnt(&gen_buf_list);
1002         pdev->toe_info.grqs[sb_idx].number_of_isles_delta--;
1003 
1004         if (!d_list_is_empty(&gen_buf_list)) {
1005         d_list_add_tail(&gen_info->dpc_peninsula_list, &gen_buf_list);
1006         }
1007         gen_info->dpc_peninsula_nbytes += isle_nbytes;
1008         gen_info->isle_nbytes -= isle_nbytes;
1009     } else {
1010         start_isle = _lm_tcp_isle_find(pdev,tcp,isle_num);
1011         d_list_init(&gen_buf_list, NULL, NULL, 0);
1012         isle_nbytes = _lm_tcp_isle_remove(pdev, tcp, sb_idx, isle_num + 1, &gen_buf_list);
1013 //        DbgBreakIf(!(isle_nbytes && d_list_entry_cnt(&gen_buf_list)));
1014         pdev->toe_info.grqs[sb_idx].number_of_isles_delta--;
1015         if (d_list_entry_cnt(&gen_buf_list) > 1) {
1016             DbgBreakIf(((lm_tcp_gen_buf_t *)(d_list_peek_head(&gen_buf_list)))->placed_bytes == 0);
1017         }
1018         DbgBreakIf(((lm_tcp_gen_buf_t *)(d_list_peek_tail(&start_isle->isle_gen_bufs_list_head)))->placed_bytes == 0);
1019         if (!d_list_is_empty(&gen_buf_list)) {
1020         d_list_add_tail(&start_isle->isle_gen_bufs_list_head, &gen_buf_list);
1021         }
1022         start_isle->isle_nbytes += isle_nbytes;
1023 #ifdef DEBUG_OOO_CQE
1024         SET_DEBUG_OOO_INFO(start_isle,CMP_OPCODE_TOE_GJ,0);
1025 #endif
1026     }
1027     rx_con->dpc_info.dpc_flags &= ~(LM_TCP_DPC_TOO_BIG_ISLE | LM_TCP_DPC_TOO_MANY_ISLES);
1028 
1029 }
1030 
lm_tcp_rx_next_grq_buf(lm_device_t * pdev,u8_t sb_idx)1031 static __inline lm_tcp_gen_buf_t * lm_tcp_rx_next_grq_buf(lm_device_t * pdev, u8_t sb_idx)
1032 {
1033     lm_tcp_gen_buf_t * gen_buf;
1034 
1035     /* 11/12/2008 - TODO: Enhance locking acquisition method,
1036      * TBD: aggragate cons, and active_gen_list updates */
1037     MM_ACQUIRE_TOE_GRQ_LOCK_DPC(pdev, sb_idx);
1038 
1039     /* Get the generic buffer for this completion */
1040     gen_buf = (lm_tcp_gen_buf_t *)d_list_pop_head(&pdev->toe_info.grqs[sb_idx].active_gen_list);
1041     if (ERR_IF(gen_buf == NULL)) {
1042         DbgBreakMsg("Received a fw GA/GAI without any generic buffers\n");
1043         return NULL;
1044     }
1045     DbgBreakIf(!gen_buf);
1046     DbgBreakIf(SIG(gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
1047     DbgBreakIf(END_SIG(gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
1048 
1049     /* each generic buffer is represented by ONE bd on the bd-chain */
1050     lm_bd_chain_bds_consumed(&pdev->toe_info.grqs[sb_idx].bd_chain, 1);
1051 
1052     MM_RELEASE_TOE_GRQ_LOCK_DPC(pdev, sb_idx);
1053 
1054     return gen_buf;
1055 }
1056 
1057 /** Description
1058  *   completes the fast-path operations for a certain connection
1059  *  Assumption:
1060  *   fp-rx lock is taken
1061  *   This function is mutual exclusive: there can only be one thread running it at a time.
1062  */
lm_tcp_rx_complete_tcp_fp(lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_con_t * con)1063 void lm_tcp_rx_complete_tcp_fp(lm_device_t * pdev, lm_tcp_state_t * tcp, lm_tcp_con_t * con)
1064 {
1065     lm_tcp_buffer_t * curr_tcp_buf;
1066     u32_t add_sws_bytes = 0;
1067 
1068     if (con->dpc_info.dpc_comp_blocked) {
1069         /* we will no longer receive a "skp" */
1070         SET_FLAGS(con->flags, TCP_POST_NO_SKP); /* so that new posts complete immediately... */
1071         /* complete any outstanding skp bytes... */
1072         if (tcp->rx_con->u.rx.skp_bytes_copied) {
1073             /* now we can complete these bytes that have already been copied... */
1074             tcp->rx_con->bytes_comp_cnt += tcp->rx_con->u.rx.skp_bytes_copied;
1075             /* complete nbytes on buffers (dpc-flow ) */
1076             lm_tcp_complete_nbytes(pdev, tcp, tcp->rx_con, tcp->rx_con->u.rx.skp_bytes_copied, /* push=*/ 0);
1077             tcp->rx_con->u.rx.skp_bytes_copied = 0;
1078         }
1079     }
1080 
1081     /* TBA Michals FW BYPASS...copy here */
1082     if (!d_list_is_empty(&con->u.rx.gen_info.dpc_peninsula_list)) {
1083         /* only copy if this is the end... otherwise, we will wait for that SKP... */
1084         if (lm_tcp_next_entry_dpc_active_list(con) && con->u.rx.gen_info.dpc_peninsula_nbytes && con->dpc_info.dpc_comp_blocked) {
1085             /* couldn't have been posted buffers if peninsula exists... */
1086             DbgBreakIf(!d_list_is_empty(&con->u.rx.gen_info.peninsula_list));
1087             con->u.rx.gen_info.bytes_copied_cnt_in_comp += lm_tcp_rx_peninsula_to_rq(pdev, tcp, 0xffffffff,NON_EXISTENT_SB_IDX);
1088         }
1089 
1090         /* check if we still have something in the peninsula after the copying AND our active tb list is empty... otherwise, it's intended
1091          * for that and we'll wait for the next RQ_SKP in the next DPC. UNLESS, we've got completion block, in which case RQ_SKP won't make it
1092          * way ever... */
1093         curr_tcp_buf = lm_tcp_next_entry_dpc_active_list(con);
1094         DbgBreakIf(!d_list_is_empty(&con->u.rx.gen_info.dpc_peninsula_list) && curr_tcp_buf && con->dpc_info.dpc_comp_blocked);
1095         if (!d_list_is_empty(&con->u.rx.gen_info.dpc_peninsula_list) && !curr_tcp_buf) {
1096             d_list_add_tail(&con->u.rx.gen_info.peninsula_list, &con->u.rx.gen_info.dpc_peninsula_list);
1097             con->u.rx.gen_info.peninsula_nbytes += con->u.rx.gen_info.dpc_peninsula_nbytes;
1098             con->u.rx.gen_info.dpc_peninsula_nbytes = 0;
1099 
1100             /* we want to leave any non-released buffer in the dpc_peninsula (so that we don't access the list w/o a lock) */
1101             if (((lm_tcp_gen_buf_t *)d_list_peek_tail(&con->u.rx.gen_info.peninsula_list))->placed_bytes == 0) {
1102                 lm_tcp_gen_buf_t * gen_buf;
1103                 gen_buf = (lm_tcp_gen_buf_t *)d_list_pop_tail(&con->u.rx.gen_info.peninsula_list);
1104                 if CHK_NULL(gen_buf)
1105                 {
1106                     DbgBreakIfAll( !gen_buf ) ;
1107                     return;
1108                 }
1109                 d_list_init(&con->u.rx.gen_info.dpc_peninsula_list, &gen_buf->link, &gen_buf->link, 1);
1110             } else {
1111                 d_list_clear(&con->u.rx.gen_info.dpc_peninsula_list);
1112             }
1113 
1114         }
1115     }
1116 
1117     /**** Client completing :  may result in lock-release *****/
1118     /* during lock-release, due to this function being called from service_deferred, more
1119      * cqes can be processed. We don't want to mix. This function is mutually exclusive, so
1120      * any processing makes it's way to being completed by calling this function.
1121      * the following define a "fast-path completion"
1122      * (i)   RQ buffers to be completed
1123      *       defined by dpc_completed_tail and are collected during lm_tcp_complete_bufs BEFORE lock
1124      *       is released, so no more buffer processing can make it's way into this buffer completion.
1125      * (ii)  GRQ buffers to be indicated
1126      *       Are taken from peninsula, and not dpc_peninsula, so no NEW generic buffers can make their
1127      *       way to this indication
1128      * (iii) Fin to be indicated
1129      *       determined by the flags, since dpc_flags CAN be modified during processing we copy
1130      *       them to a snapshot_flags parameter, which is initialized in this function only, so no fin
1131      *       can can make its way in while we release the lock.
1132      * (iv)  Remainders for sp
1133      *       all sp operations are logged in dpc_flags. for the same reason as (iii) no sp commands can
1134      *       make their way in during this fp-completion, all sp-processing after will relate to this point in time.
1135      */
1136     /* NDC is the only fp flag: determining that we should complete all the processed cqes. Therefore, we can
1137      * turn it off here. We should turn it off, since if no sp flags are on, the sp-complete function shouldn't be called
1138      */
1139 //    RESET_FLAGS(con->dpc_info.dpc_flags, LM_TCP_DPC_NDC);
1140     con->dpc_info.snapshot_flags = con->dpc_info.dpc_flags;
1141     con->dpc_info.dpc_flags = 0;
1142 
1143     /* compensate fw-window with the rq-placed bytes */
1144     if (con->dpc_info.dpc_rq_placed_bytes) {
1145         add_sws_bytes += con->dpc_info.dpc_rq_placed_bytes;
1146         con->dpc_info.dpc_rq_placed_bytes = 0;
1147     }
1148 
1149 
1150     /* check if we completed a buffer that as a result unblocks the um from posting more (a split buffer that
1151      * was placed on the last bd). If this occured - we should not have any other RQs!!! */
1152     if (con->dpc_info.dpc_unblock_post) {
1153         RESET_FLAGS(con->flags, TCP_POST_DELAYED);
1154         con->dpc_info.dpc_unblock_post = 0;
1155     }
1156 
1157     /* NOTE: AFTER THIS STAGE DO NOT ACCESS DPC-INFO ANYMORE - for deferred cqes issue */
1158 
1159     /* complete buffers to client */
1160     if (con->dpc_info.dpc_completed_tail != NULL) {
1161         lm_tcp_complete_bufs(pdev,tcp,con);
1162     }
1163 
1164     /* Is there something left to indicate? */
1165     if (!d_list_is_empty(&con->u.rx.gen_info.peninsula_list) && _lm_tcp_ok_to_indicate(con)) {
1166         mm_tcp_rx_indicate_gen(pdev,tcp);
1167         add_sws_bytes += tcp->rx_con->u.rx.gen_info.add_sws_bytes; /* any bytes we need to update will be aggregated here during indicate */
1168         tcp->rx_con->u.rx.gen_info.add_sws_bytes = 0;
1169     }
1170 
1171     if (add_sws_bytes) {
1172         lm_tcp_rx_post_sws(pdev, tcp, con, add_sws_bytes, TCP_RX_POST_SWS_INC);
1173     }
1174 
1175 }
1176 
1177 
1178 /** Description
1179  *  processes a single cqe.
1180  */
lm_tcp_rx_process_cqe(lm_device_t * pdev,struct toe_rx_cqe * cqe,lm_tcp_state_t * tcp,u8_t sb_idx)1181 void lm_tcp_rx_process_cqe(
1182     lm_device_t       * pdev,
1183     struct toe_rx_cqe * cqe,
1184     lm_tcp_state_t    * tcp,
1185     u8_t                sb_idx)
1186 {
1187     u32_t   nbytes;
1188     u8_t    cmd;
1189     u8_t    isle_num = 0;
1190 
1191     cmd = ((cqe->params1 & TOE_RX_CQE_COMPLETION_OPCODE) >> TOE_RX_CQE_COMPLETION_OPCODE_SHIFT);
1192 
1193 
1194     /* Check that the cqe nbytes make sense, we could have got here by chance... */
1195     /* update completion has a different usage for nbyts which is a sequence -so any number is valid*/
1196     if(IS_OOO_CQE(cmd)) {
1197         nbytes = (cqe->data.ooo_params.ooo_params & TOE_RX_CQE_OOO_PARAMS_NBYTES) >> TOE_RX_CQE_OOO_PARAMS_NBYTES_SHIFT;
1198         isle_num = (cqe->data.ooo_params.ooo_params & TOE_RX_CQE_OOO_PARAMS_ISLE_NUM) >> TOE_RX_CQE_OOO_PARAMS_ISLE_NUM_SHIFT;
1199         if (((isle_num == 0) && (cmd != CMP_OPCODE_TOE_GJ)) || (isle_num > T_TCP_MAX_ISLES_PER_CONNECTION_TOE)) {
1200             DbgMessage(pdev, FATAL, "Isle number %d is not valid for OOO CQE %d\n", isle_num, cmd);
1201             DbgBreak();
1202         }
1203     } else if (cmd == RAMROD_OPCODE_TOE_UPDATE) {
1204         nbytes = cqe->data.raw_data;
1205     } else {
1206         nbytes = (cqe->data.in_order_params.in_order_params & TOE_RX_CQE_IN_ORDER_PARAMS_NBYTES) >> TOE_RX_CQE_IN_ORDER_PARAMS_NBYTES_SHIFT;
1207         DbgBreakIfAll(nbytes & 0xc0000000); /* two upper bits on show a completion larger than 1GB - a bit odd...*/
1208         DbgBreakIf(nbytes && tcp->rx_con->dpc_info.dpc_comp_blocked);
1209     }
1210     if (pdev->toe_info.archipelago.l4_decrease_archipelago
1211             && d_list_entry_cnt(&tcp->rx_con->u.rx.gen_info.first_isle.isle_gen_bufs_list_head)) {
1212         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_TOO_MANY_ISLES;
1213     }
1214     switch(cmd)
1215     {
1216     case CMP_OPCODE_TOE_SRC_ERR:
1217         DbgMessage(pdev, FATAL, "ERROR: NO SEARCHER ENTRY!\n");
1218         DbgBreakIfAll(TRUE);
1219         return;
1220     case CMP_OPCODE_TOE_GA:
1221         //DbgMessage(pdev, WARN, "GenericAdd cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1222         lm_tcp_rx_gen_peninsula_process(pdev, tcp, nbytes,
1223                                         lm_tcp_rx_next_grq_buf(pdev, sb_idx));
1224         return;
1225     case CMP_OPCODE_TOE_GNI:
1226         //DbgMessage(pdev, WARN, "GenericCreateIsle cid=%d isle_num=%d!\n", tcp->cid, isle_num);
1227         DbgBreakIf(nbytes);
1228         lm_tcp_rx_gen_isle_create(pdev, tcp,
1229                                   lm_tcp_rx_next_grq_buf(pdev, sb_idx), sb_idx, isle_num);
1230         return;
1231     case CMP_OPCODE_TOE_GAIR:
1232         //DbgMessage(pdev, WARN, "GenericAddIsleR cid=%d isle_num=%d nbytes=%d!\n", tcp->cid, isle_num, nbytes);
1233         lm_tcp_rx_gen_isle_right_process(pdev, tcp, nbytes,
1234                                    lm_tcp_rx_next_grq_buf(pdev, sb_idx), sb_idx, isle_num);
1235         return;
1236     case CMP_OPCODE_TOE_GAIL:
1237         DbgMessage(pdev, WARN, "GenericAddIsleL cid=%d isle_num=%d nbytes=%d!\n", tcp->cid, isle_num, nbytes);
1238         if (nbytes)
1239         {
1240             lm_tcp_rx_gen_isle_left_process(pdev, tcp, nbytes,
1241                                        NULL, sb_idx, isle_num);
1242         }
1243         else
1244         {
1245             lm_tcp_rx_gen_isle_left_process(pdev, tcp, 0,
1246                                        lm_tcp_rx_next_grq_buf(pdev, sb_idx), sb_idx, isle_num);
1247         }
1248         return;
1249     case CMP_OPCODE_TOE_GRI:
1250 //        DbgMessage(pdev, WARN, "GenericReleaseIsle cid=%d isle_num=%d nbytes=%d!\n", tcp->cid, isle_num, nbytes);
1251         lm_tcp_rx_gen_isle_right_process(pdev, tcp, nbytes, NULL, sb_idx, isle_num);
1252         return;
1253     case CMP_OPCODE_TOE_GR:
1254         //DbgMessage(pdev, WARN, "GenericRelease cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1255         lm_tcp_rx_gen_peninsula_process(pdev, tcp, nbytes, NULL);
1256         return;
1257     case CMP_OPCODE_TOE_GJ:
1258         //DbgMessage(pdev, WARN, "GenericJoin cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1259         lm_tcp_rx_gen_join_process(pdev, tcp, sb_idx, isle_num);
1260         return;
1261     case CMP_OPCODE_TOE_CMP:
1262         //DbgMessage(pdev, WARN, "Cmp(push) cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1263         /* Add fast path handler here */
1264         lm_tcp_rx_cmp_process(pdev, tcp, nbytes, 1);
1265         return;
1266     case CMP_OPCODE_TOE_REL:
1267         //DbgMessage(pdev, WARN, "Rel(nopush) cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1268         lm_tcp_rx_cmp_process(pdev, tcp, nbytes, 0);
1269         return;
1270 	case CMP_OPCODE_TOE_SKP:
1271         //DbgMessage(pdev, WARN, "Skp cid=%d nbytes=%d!\n", tcp->cid, cqe->nbytes);
1272         lm_tcp_rx_skp_process(pdev, tcp, nbytes, sb_idx);
1273         return;
1274     case CMP_OPCODE_TOE_DGI:
1275         DbgMessage(pdev, WARN, "Delete Isle cid=%d!\n", tcp->cid);
1276         lm_tcp_rx_delete_isle(pdev, tcp, sb_idx, isle_num, nbytes);
1277         return;
1278     }
1279 
1280     /* for the rest of the commands, if we have nbytes, we need to complete them (generic/app) */
1281     /* unless it's an update completion, in which case the nbytes has a different meaning. */
1282     if ((cmd != RAMROD_OPCODE_TOE_UPDATE) && nbytes) {
1283         lm_tcp_gen_buf_t * gen_buf;
1284         gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_tail(&tcp->rx_con->u.rx.gen_info.dpc_peninsula_list);
1285         if(gen_buf && (gen_buf->placed_bytes == 0)) {
1286             lm_tcp_rx_gen_peninsula_process(pdev, tcp, nbytes, NULL);
1287         } else {
1288             /* if we're here - we will no longer see a RQ_SKP so, let's simulate one...note if we didn't get nbytes here.. we still need
1289              * to take care of this later if it's a blocking completion the skip will have to be everything in the peninsula
1290             * we can access skp_bytes here lockless, because the only time it will be accessed in post is if there is something in the peninsula, if we got a RQ_SKP here, there can't be...*/
1291             DbgBreakIf(!d_list_is_empty(&tcp->rx_con->u.rx.gen_info.peninsula_list));
1292             DbgBreakIf(tcp->rx_con->rq_nbytes <= tcp->rx_con->u.rx.gen_info.dpc_peninsula_nbytes+tcp->rx_con->u.rx.skp_bytes_copied); // we got a  RQ completion here... so peninsula CAN;T cover RQ!!!
1293             lm_tcp_rx_skp_process(pdev, tcp, tcp->rx_con->u.rx.gen_info.dpc_peninsula_nbytes+tcp->rx_con->u.rx.skp_bytes_copied, sb_idx);
1294 
1295             /* We give push=1 here, this will seperate between 'received' data and 'aborted' bufs. we won't
1296              * have any buffers left that need to be aborted that have partial completed data on them  */
1297             lm_tcp_rx_cmp_process(pdev, tcp, nbytes, 2 /* push as result of sp-completion*/);
1298         }
1299     }
1300 
1301     switch (cmd) {
1302     case CMP_OPCODE_TOE_FIN_RCV:
1303         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_FIN_RECV;
1304         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_REMOTE_FIN_RECEIVED */
1305         return;
1306     case CMP_OPCODE_TOE_FIN_UPL:
1307         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_FIN_RECV_UPL;
1308         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_REMOTE_FIN_RECEIVED + Request to upload the connection */
1309         return;
1310     case CMP_OPCODE_TOE_RST_RCV:
1311         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RESET_RECV;
1312         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_REMOTE_RST_RECEIVED */
1313         return;
1314     case RAMROD_OPCODE_TOE_UPDATE:
1315         DbgBreakIf( (tcp->hdr.status != STATE_STATUS_NORMAL) && (tcp->hdr.status != STATE_STATUS_ABORTED));
1316         DbgBreakIf(tcp->sp_request == NULL);
1317         DbgBreakIf((tcp->sp_request->type != SP_REQUEST_UPDATE_NEIGH) &&
1318                    (tcp->sp_request->type != SP_REQUEST_UPDATE_PATH) &&
1319                    (tcp->sp_request->type != SP_REQUEST_UPDATE_TCP) &&
1320                    (tcp->sp_request->type != SP_REQUEST_UPDATE_PATH_RELINK));
1321         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1322 
1323         /*DbgMessage(pdev, FATAL, "lm_tcp_rx_process_cqe() RAMROD_OPCODE_TOE_UPDATE: IGNORE_WND_UPDATES=%d, cqe->nbytes=%d\n", GET_FLAGS(tcp->rx_con->db_data.rx->flags, TOE_RX_DB_DATA_IGNORE_WND_UPDATES), cqe->nbytes);*/
1324 
1325         if ((tcp->sp_request->type == SP_REQUEST_UPDATE_TCP) && (GET_FLAGS(tcp->rx_con->db_data.rx->flags, TOE_RX_DB_DATA_IGNORE_WND_UPDATES)))
1326         {
1327             tcp->rx_con->dpc_info.dpc_fw_wnd_after_dec = nbytes;
1328         }
1329         return;
1330     case CMP_OPCODE_TOE_URG:
1331         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_URG;
1332         return;
1333     case CMP_OPCODE_TOE_MAX_RT:
1334         DbgMessage(pdev, WARNl4, "lm_tcp_rx_process_cqe: CMP_OPCODE_TOE_MAX_RT cid=%d\n", tcp->cid);
1335         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RT_TO;
1336         return;
1337     case CMP_OPCODE_TOE_RT_TO:
1338         DbgMessage(pdev, WARNl4, "lm_tcp_rx_process_cqe: CMP_OPCODE_TOE_RT_TO cid=%d\n", tcp->cid);
1339         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RT_TO;
1340         return;
1341     case CMP_OPCODE_TOE_KA_TO:
1342         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_KA_TO;
1343         return;
1344     case CMP_OPCODE_TOE_DBT_RE:
1345         /* LH Inbox specification:  Black Hole detection (RFC 2923)
1346          * TCP Chimney target MUST upload the connection if the TCPDoubtReachabilityRetransmissions threshold is hit.
1347          * SPARTA test scripts and tests that will fail if not implemented: All tests in Tcp_BlackholeDetection.wsf, we cause
1348          * the upload by giving L4_UPLOAD_REASON_UPLOAD_REQUEST (same as Teton) */
1349         DbgMessage(pdev, INFORMl4, "lm_tcp_rx_process_cqe: RCQE CMP_OPCODE_TOE_DBT_RE, cid=%d\n", tcp->cid);
1350         DbgMessage(pdev, WARNl4, "lm_tcp_rx_process_cqe: RCQE CMP_OPCODE_TOE_DBT_RE, cid=%d IGNORING!!!\n", tcp->cid);
1351         /* We add this here only for windows and not ediag */
1352         #if (!defined(DOS)) && (!defined(__LINUX))
1353         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_DBT_RE;
1354         #endif
1355         return;
1356     case CMP_OPCODE_TOE_SYN:
1357     case CMP_OPCODE_TOE_FW2_TO:
1358         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_UPLD_CLOSE;
1359         return;
1360     case CMP_OPCODE_TOE_2WY_CLS:
1361         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_UPLD_CLOSE;
1362         return;
1363     case CMP_OPCODE_TOE_OPT_ERR:
1364         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_OPT_ERR;
1365         return;
1366     case RAMROD_OPCODE_TOE_QUERY:
1367         DbgBreakIf(! tcp->sp_request );
1368         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_QUERY);
1369         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1370         return;
1371     case RAMROD_OPCODE_TOE_SEARCHER_DELETE:
1372         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_TERMINATE_OFFLOAD);
1373         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1374         return;
1375     case RAMROD_OPCODE_TOE_RESET_SEND:
1376         DbgBreakIf(! tcp->sp_request);
1377         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_ABORTIVE_DISCONNECT);
1378         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1379         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_RST_REQ_COMPLETED */
1380         return;
1381     case RAMROD_OPCODE_TOE_INVALIDATE:
1382         DbgBreakIf(! tcp->sp_request);
1383         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_INVALIDATE);
1384         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1385         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_INV_REQ_COMPLETED */
1386         return;
1387     case RAMROD_OPCODE_TOE_TERMINATE:
1388         DbgBreakIf(! tcp->sp_request);
1389         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_TERMINATE1_OFFLOAD);
1390         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1391         tcp->rx_con->dpc_info.dpc_comp_blocked = TRUE; /* TCP_TRM_REQ_COMPLETED */
1392         return;
1393     case RAMROD_OPCODE_TOE_EMPTY_RAMROD:
1394         DbgBreakIf(nbytes);
1395         DbgBreakIf(! tcp->sp_request );
1396         DbgBreakIf((tcp->sp_request->type != SP_REQUEST_PENDING_LOCAL_FIN_DISCONNECT) &&
1397                    (tcp->sp_request->type != SP_REQUEST_PENDING_REMOTE_DISCONNECT) &&
1398                    (tcp->sp_request->type != SP_REQUEST_PENDING_TX_RST));
1399         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1400         return;
1401     case RAMROD_OPCODE_TOE_INITIATE_OFFLOAD:
1402         DbgBreakIf(nbytes);
1403         DbgBreakIf(! tcp->sp_request );
1404         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_INITIATE_OFFLOAD);
1405 
1406         /* 13/08/08 NirV: bugbug, temp workaround for dpc watch dog bug,
1407          * complete ofld request here - assumption: tcp lock is NOT taken by caller */
1408         lm_tcp_comp_initiate_offload_request(pdev, tcp, LM_STATUS_SUCCESS);
1409         lm_sq_complete(pdev, CMD_PRIORITY_NORMAL, RAMROD_OPCODE_TOE_INITIATE_OFFLOAD, tcp->ulp_type, tcp->cid);
1410 
1411         return;
1412     case CMP_OPCODE_TOE_LCN_ERR:
1413         DbgBreakIf(! tcp->sp_request );
1414         DbgBreakIf(tcp->sp_request->type != SP_REQUEST_INITIATE_OFFLOAD);
1415         tcp->rx_con->dpc_info.dpc_flags |= LM_TCP_DPC_RAMROD_CMP;
1416         return;
1417     default:
1418         DbgMessage(pdev, FATAL, "unexpected rx cqe opcode=%d\n", cmd);
1419         DbgBreakIfAll(TRUE);
1420     }
1421 }
1422 
lm_tcp_rx_process_cqes(lm_device_t * pdev,u8_t drv_toe_rss_id,s_list_t * connections)1423 u8_t lm_tcp_rx_process_cqes(lm_device_t *pdev, u8_t drv_toe_rss_id, s_list_t * connections)
1424 {
1425     lm_tcp_rcq_t *rcq;
1426     lm_tcp_grq_t *grq;
1427     struct toe_rx_cqe *cqe, *hist_cqe;
1428     lm_tcp_state_t *tcp = NULL;
1429     u32_t cid;
1430     u32_t avg_dpc_cnt;
1431     u16_t cq_new_idx;
1432     u16_t cq_old_idx;
1433     u16_t num_to_reproduce = 0;
1434     u8_t  defer_cqe;
1435     u8_t  process_rss_upd_later = FALSE;
1436     MM_INIT_TCP_LOCK_HANDLE();
1437 
1438     DbgMessage(pdev, VERBOSEl4int , "###lm_tcp_rx_process_cqes START\n");
1439 
1440     rcq = &pdev->toe_info.rcqs[drv_toe_rss_id];
1441     grq = &pdev->toe_info.grqs[drv_toe_rss_id];
1442     cq_new_idx = *(rcq->hw_con_idx_ptr);
1443     cq_old_idx = lm_bd_chain_cons_idx(&rcq->bd_chain);
1444     DbgBreakIf(S16_SUB(cq_new_idx, cq_old_idx) < 0);
1445 
1446     /* save statistics */
1447     rcq->num_cqes_last_dpc = S16_SUB(cq_new_idx, cq_old_idx);
1448     DbgMessage(pdev, VERBOSEl4int, "###lm_tcp_rx_process_cqes num_cqes=%d\n", rcq->num_cqes_last_dpc);
1449 
1450     if (rcq->num_cqes_last_dpc) {  /* Exclude zeroed value from statistics*/
1451         if(rcq->max_cqes_per_dpc < rcq->num_cqes_last_dpc) {
1452             rcq->max_cqes_per_dpc = rcq->num_cqes_last_dpc;
1453         }
1454         /* we don't want to wrap around...*/
1455         if ((rcq->sum_cqes_last_x_dpcs + rcq->num_cqes_last_dpc) < rcq->sum_cqes_last_x_dpcs) {
1456             rcq->avg_dpc_cnt = 0;
1457             rcq->sum_cqes_last_x_dpcs = 0;
1458         }
1459         rcq->sum_cqes_last_x_dpcs += rcq->num_cqes_last_dpc;
1460         rcq->avg_dpc_cnt++;
1461         avg_dpc_cnt = rcq->avg_dpc_cnt;
1462         if (avg_dpc_cnt) { /*Prevent division by 0*/
1463             rcq->avg_cqes_per_dpc = rcq->sum_cqes_last_x_dpcs / avg_dpc_cnt;
1464         } else {
1465             rcq->sum_cqes_last_x_dpcs = 0;
1466         }
1467     }
1468 
1469 
1470     /* if we are suspended, we need to check if we can resume processing */
1471     if (rcq->suspend_processing == TRUE) {
1472         lm_tcp_rss_update_suspend_rcq(pdev, rcq);
1473         if (rcq->suspend_processing == TRUE) {
1474             /* skip the consumption loop */
1475             cq_new_idx = cq_old_idx;
1476             DbgMessage(pdev, VERBOSEl4int, "lm_tcp_rx_process_cqes(): rcq suspended - idx:%d\n", drv_toe_rss_id);
1477         }
1478     }
1479 
1480     while(cq_old_idx != cq_new_idx) {
1481         u32_t update_stats_type;
1482         u8_t opcode;
1483 
1484         DbgBreakIf(S16_SUB(cq_new_idx, cq_old_idx) <= 0);
1485 
1486         /* get next consumed cqe */
1487         cqe = lm_toe_bd_chain_consume_bd(&rcq->bd_chain);
1488         update_stats_type = cqe->data.raw_data;
1489         DbgBreakIf(!cqe);
1490         num_to_reproduce++;
1491 
1492         /* get cid and opcode from cqe */
1493         cid = SW_CID(((cqe->params1 & TOE_RX_CQE_CID) >> TOE_RX_CQE_CID_SHIFT));
1494         opcode = (cqe->params1 & TOE_RX_CQE_COMPLETION_OPCODE) >> TOE_RX_CQE_COMPLETION_OPCODE_SHIFT;
1495 
1496         if (opcode == RAMROD_OPCODE_TOE_RSS_UPDATE) {
1497 
1498             /* update the saved consumer */
1499             cq_old_idx = lm_bd_chain_cons_idx(&rcq->bd_chain);
1500 
1501             /* rss update ramrod */
1502             DbgMessage(pdev, INFORMl4int, "lm_tcp_rx_process_cqes(): calling lm_tcp_rss_update_ramrod_comp - drv_toe_rss_id:%d\n", drv_toe_rss_id);
1503             if (num_to_reproduce > 1) {
1504                 process_rss_upd_later = TRUE;
1505                 lm_tcp_rss_update_ramrod_comp(pdev, rcq, cid, update_stats_type, FALSE);
1506                 break;
1507             }
1508             lm_tcp_rss_update_ramrod_comp(pdev, rcq, cid, update_stats_type, TRUE);
1509 
1510             /* suspend further RCQ processing (if needed) */
1511             if (rcq->suspend_processing == TRUE)
1512                 break;
1513             else
1514                 continue;
1515 
1516         }
1517 
1518         if (cid < MAX_ETH_REG_CONS) {
1519             /* toe init ramrod */
1520             DbgBreakIf(((cqe->params1 & TOE_RX_CQE_COMPLETION_OPCODE) >> TOE_RX_CQE_COMPLETION_OPCODE_SHIFT)
1521                        != RAMROD_OPCODE_TOE_INIT);
1522             lm_tcp_init_ramrod_comp(pdev);
1523             cq_old_idx = lm_bd_chain_cons_idx(&rcq->bd_chain);
1524             DbgBreakIf(cq_old_idx != cq_new_idx);
1525             /* We need to update the slow-path ring. This is usually done in the lm_tcp_rx_complete_sp_cqes,
1526              * but we won't get there since this completion is not associated with a connection. USUALLY we
1527              * have to update the sp-ring only AFTER we've written the CQ producer, this is to promise that there
1528              * will always be an empty entry for another ramrod completion, but in this case we're safe, since only
1529              * one CQE is occupied anyway */
1530             lm_sq_complete(pdev, CMD_PRIORITY_NORMAL, RAMROD_OPCODE_TOE_INIT, TOE_CONNECTION_TYPE, LM_SW_LEADING_RSS_CID(pdev));
1531             break;
1532         }
1533 
1534         tcp = lm_cid_cookie(pdev, TOE_CONNECTION_TYPE, cid);
1535         DbgBreakIf(!tcp);
1536         /* save cqe in history_cqes */
1537         hist_cqe = (struct toe_rx_cqe *)lm_tcp_qe_buffer_next_cqe_override(&tcp->rx_con->history_cqes);
1538         *hist_cqe = *cqe;
1539 
1540         /* ASSUMPTION: if COMP_DEFERRED changes from FALSE to TRUE, the change occurs only in DPC
1541          * o/w it can only change from TRUE to FALSE.
1542          *
1543          * Read flag w/o lock. Flag may change by the time we call rx_defer_cqe
1544          * Need to check again under lock. We want to avoid acquiring the lock every DPC */
1545         defer_cqe = ((tcp->rx_con->flags & TCP_RX_COMP_DEFERRED) == TCP_RX_COMP_DEFERRED);
1546         if (defer_cqe) {
1547             /* if we're deferring completions - just store the cqe and continue to the next one
1548              * Assumptions: ALL commands need to be deferred, we aren't expecting any command on
1549              * L4 that we should pay attention to for this connection ( only one outstanding sp at a time ) */
1550             /* Return if we are still deferred (may have changed since initial check was w/o a lock */
1551             mm_acquire_tcp_lock(pdev, tcp->rx_con);
1552             /* check again under lock if we're deferred */
1553             defer_cqe = ((tcp->rx_con->flags & TCP_RX_COMP_DEFERRED) == TCP_RX_COMP_DEFERRED);
1554             if (defer_cqe) {
1555                 tcp->rx_con->flags |= TCP_DEFERRED_PROCESSING;
1556 
1557                 /* 13/08/08 NirV: bugbug, temp workaround for dpc watch dog bug,
1558                  * release the tcp lock if cqe is offload complete */
1559                 if (((cqe->params1 & TOE_RX_CQE_COMPLETION_OPCODE) >> TOE_RX_CQE_COMPLETION_OPCODE_SHIFT) == RAMROD_OPCODE_TOE_INITIATE_OFFLOAD)
1560                 {
1561                     mm_release_tcp_lock(pdev, tcp->rx_con);
1562                 }
1563 
1564                 lm_tcp_rx_process_cqe(pdev,cqe,tcp,drv_toe_rss_id);
1565             }
1566 
1567             /* 13/08/08 NirV: bugbug, temp workaround for dpc watch dog bug,
1568              * release the tcp lock if cqe is not offload complete (was released earlier) */
1569             if (((cqe->params1 & TOE_RX_CQE_COMPLETION_OPCODE) >> TOE_RX_CQE_COMPLETION_OPCODE_SHIFT) != RAMROD_OPCODE_TOE_INITIATE_OFFLOAD)
1570             {
1571                 mm_release_tcp_lock(pdev, tcp->rx_con);
1572             }
1573         }
1574 
1575         if (!defer_cqe) {
1576             /* connections will always be initialized to a dummy, so once a tcp connection is added to the
1577              * list, it's link will be initialized to point to another link other than NULL */
1578             if (s_list_next_entry(&tcp->rx_con->dpc_info.link) == NULL) {
1579                 s_list_push_head(connections, &tcp->rx_con->dpc_info.link);
1580             }
1581             lm_tcp_rx_process_cqe(pdev, cqe, tcp, drv_toe_rss_id);
1582         }
1583 
1584         cq_old_idx = lm_bd_chain_cons_idx(&rcq->bd_chain);
1585     }
1586 
1587     /* We may have nothing to reproduce if we were called from a sw_dpc */
1588     if (num_to_reproduce) {
1589         lm_toe_bd_chain_bds_produced(&rcq->bd_chain, num_to_reproduce);
1590 
1591         /* GilR 5/13/2006 - TBA - save some stats? */
1592 
1593         /* notify the fw of the prod of the RCQ */
1594         LM_INTMEM_WRITE16(pdev, USTORM_TOE_CQ_PROD_OFFSET(LM_TOE_FW_RSS_ID(pdev,drv_toe_rss_id) , PORT_ID(pdev)),
1595                           lm_bd_chain_prod_idx(&rcq->bd_chain),  BAR_USTRORM_INTMEM);
1596 
1597         if (pdev->params.enable_dynamic_hc[HC_INDEX_TOE_RX_CQ_CONS]) {
1598             u32_t l4_quasi_byte_counter;
1599             u16_t prod_idx_diff = lm_bd_chain_prod_idx(&rcq->bd_chain) - rcq->bd_chain.bds_per_page * rcq->bd_chain.page_cnt;
1600             l4_quasi_byte_counter = prod_idx_diff;
1601             l4_quasi_byte_counter <<= 16;
1602             //fIXME
1603             LM_INTMEM_WRITE32(pdev, rcq->hc_sb_info.iro_dhc_offset, l4_quasi_byte_counter, BAR_CSTRORM_INTMEM);
1604         }
1605     }
1606     DbgMessage(pdev, VERBOSEl4int , "###lm_tcp_rx_process_cqes END\n");
1607     return process_rss_upd_later;
1608 }
1609 
1610 /** Description
1611  *  compensate the grq
1612  *  Assumption:
1613  *    called under the GRQ LOCK
1614  */
lm_tcp_rx_compensate_grq(lm_device_t * pdev,u8_t drv_toe_rss_id)1615 void lm_tcp_rx_compensate_grq(lm_device_t * pdev, u8_t drv_toe_rss_id)
1616 {
1617     d_list_t * collected_gen_bufs_list = &pdev->toe_info.grqs[drv_toe_rss_id].aux_gen_list;
1618 
1619     MM_ACQUIRE_TOE_GRQ_LOCK_DPC(pdev, drv_toe_rss_id);
1620     if (lm_tcp_rx_fill_grq(pdev, drv_toe_rss_id, collected_gen_bufs_list,FILL_GRQ_FULL)) {
1621         DbgMessage(pdev, INFORMl4rx, "lm_toe_service_rx_intr: Updating GRQ producer\n");
1622         /* notify the fw of the prod of the GRQ */
1623         LM_INTMEM_WRITE16(pdev, USTORM_TOE_GRQ_PROD_OFFSET(LM_TOE_FW_RSS_ID(pdev,drv_toe_rss_id) , PORT_ID(pdev)),
1624                           lm_bd_chain_prod_idx(&pdev->toe_info.grqs[drv_toe_rss_id].bd_chain), BAR_USTRORM_INTMEM);
1625     }
1626     /* check if occupancy is above threshold */
1627     if (pdev->toe_info.grqs[drv_toe_rss_id].bd_chain.capacity - pdev->toe_info.grqs[drv_toe_rss_id].bd_chain.bd_left < GRQ_XON_TH) {
1628         pdev->toe_info.grqs[drv_toe_rss_id].grq_compensate_on_alloc = TRUE;
1629     } else {
1630         pdev->toe_info.grqs[drv_toe_rss_id].grq_compensate_on_alloc = FALSE;
1631     }
1632 
1633     MM_RELEASE_TOE_GRQ_LOCK_DPC(pdev, drv_toe_rss_id);
1634 
1635     if (!d_list_is_empty(collected_gen_bufs_list)) {
1636         mm_tcp_return_list_of_gen_bufs(pdev,collected_gen_bufs_list,0, NON_EXISTENT_SB_IDX);
1637         d_list_clear(collected_gen_bufs_list);
1638     }
1639 }
1640 
lm_tcp_rx_lock_grq(lm_device_t * pdev,u8_t drv_toe_rss_id)1641 static __inline void lm_tcp_rx_lock_grq(lm_device_t *pdev, u8_t drv_toe_rss_id)
1642 {
1643     /* If we've asked for compensation on allocation (which is only set from within a dpc)
1644      * there is a risk of the grq being accessed from a different context (the alloc context)
1645      * therefore, we cancel this option. Needs to be under lock in case alloc context is already
1646      * compensating */
1647     if (pdev->toe_info.grqs[drv_toe_rss_id].grq_compensate_on_alloc) {
1648         MM_ACQUIRE_TOE_GRQ_LOCK_DPC(pdev, drv_toe_rss_id);
1649         pdev->toe_info.grqs[drv_toe_rss_id].grq_compensate_on_alloc = FALSE;
1650         MM_RELEASE_TOE_GRQ_LOCK_DPC(pdev, drv_toe_rss_id);
1651     }
1652 }
1653 
lm_toe_service_rx_intr(lm_device_t * pdev,u8_t drv_toe_rss_id)1654 void lm_toe_service_rx_intr(lm_device_t *pdev, u8_t drv_toe_rss_id)
1655 {
1656     s_list_t         connections;
1657     s_list_entry_t   dummy;
1658     lm_tcp_con_t   * con;
1659     lm_tcp_state_t * tcp;
1660     u32_t            dbg_loop_cnt = 0;
1661     u8_t             process_rss_upd;
1662 
1663     MM_INIT_TCP_LOCK_HANDLE();
1664 
1665     DbgMessage(pdev, VERBOSEl4int , "###lm_toe_service_rx_intr START\n");
1666     DbgBreakIf(!(pdev && ARRSIZE(pdev->toe_info.rcqs) > drv_toe_rss_id));
1667 
1668     /* lock the grq from external access: i.e.. allocation compensation */
1669     lm_tcp_rx_lock_grq(pdev, drv_toe_rss_id);
1670 
1671     while (TRUE) {
1672         dbg_loop_cnt++;
1673         s_list_clear(&connections);
1674         s_list_push_head(&connections, &dummy);
1675         /* process the cqes and initialize connections with all the connections that appeared
1676          * in the DPC */
1677         process_rss_upd = lm_tcp_rx_process_cqes(pdev,drv_toe_rss_id,&connections);
1678 
1679         /* Compensate the GRQ with generic buffers from the pool : process_cqes takes buffers from the grq */
1680         lm_tcp_rx_compensate_grq(pdev,drv_toe_rss_id);
1681 
1682         /* FP: traverse the connections. remember to ignore the last one */
1683         con = (lm_tcp_con_t *)s_list_peek_head(&connections);
1684         tcp = con->tcp_state;
1685         while (s_list_next_entry(&con->dpc_info.link) != NULL) {
1686             mm_acquire_tcp_lock(pdev, con);
1687             lm_tcp_rx_complete_tcp_fp(pdev, con->tcp_state, con);
1688             mm_release_tcp_lock(pdev, con);
1689             con = (lm_tcp_con_t *)s_list_next_entry(&con->dpc_info.link);
1690             tcp = con->tcp_state;
1691         }
1692 
1693         /* SP : traverse the connections. remember to ignore the last one */
1694         con = (lm_tcp_con_t *)s_list_pop_head(&connections);
1695         s_list_next_entry(&con->dpc_info.link) = NULL;
1696         tcp = con->tcp_state;
1697         while (s_list_entry_cnt(&connections) > 0) {
1698             /* we access snapshot and not dpc, since once the dpc_flags were copied
1699              * to snapshot they were zeroized */
1700             if (con->dpc_info.snapshot_flags) {
1701                 lm_tcp_rx_complete_tcp_sp(pdev, tcp, con);
1702             }
1703             con = (lm_tcp_con_t *)s_list_pop_head(&connections);
1704             s_list_next_entry(&con->dpc_info.link) = NULL;
1705             tcp = con->tcp_state;
1706         }
1707 
1708         if (process_rss_upd) {
1709             lm_tcp_rss_update_suspend_rcq(pdev,&pdev->toe_info.rcqs[drv_toe_rss_id]);
1710             if (!pdev->toe_info.rcqs[drv_toe_rss_id].suspend_processing) {
1711                 pdev->toe_info.rcqs[drv_toe_rss_id].rss_update_processing_continued++;
1712                 continue;
1713             }
1714         }
1715         break;
1716     }
1717     if (pdev->toe_info.rcqs[drv_toe_rss_id].rss_update_processing_max_continued < dbg_loop_cnt) {
1718         pdev->toe_info.rcqs[drv_toe_rss_id].rss_update_processing_max_continued = dbg_loop_cnt;
1719     }
1720 
1721     if (pdev->toe_info.grqs[drv_toe_rss_id].number_of_isles_delta || pdev->toe_info.grqs[drv_toe_rss_id].gen_bufs_in_isles_delta) {
1722         MM_ACQUIRE_ISLES_CONTROL_LOCK_DPC(pdev);
1723         lm_tcp_update_isles_cnts(pdev, pdev->toe_info.grqs[drv_toe_rss_id].number_of_isles_delta,
1724                                  pdev->toe_info.grqs[drv_toe_rss_id].gen_bufs_in_isles_delta);
1725         MM_RELEASE_ISLES_CONTROL_LOCK_DPC(pdev);
1726         pdev->toe_info.grqs[drv_toe_rss_id].number_of_isles_delta = pdev->toe_info.grqs[drv_toe_rss_id].gen_bufs_in_isles_delta = 0;
1727     }
1728 
1729     DbgMessage(pdev, VERBOSEl4int , "###lm_toe_service_rx_intr END\n");
1730 }
1731 
1732 /** Description:
1733  *  Post a single tcp buffer to the Rx bd chain
1734  * Assumptions:
1735  *  - caller initiated tcp_buf->flags field with BUFFER_START/BUFFER_END/PUSH appropriately
1736  * Returns:
1737  *  - SUCCESS - tcp buf was successfully attached to the bd chain
1738  *  - RESOURCE - not enough available BDs on bd chain for given tcp buf
1739  *  - CONNECTION_CLOSED - whenever connection's flag are marked as 'POST BLOCKED' */
lm_tcp_rx_post_buf(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_buffer_t * tcp_buf,lm_frag_list_t * frag_list)1740 lm_status_t lm_tcp_rx_post_buf(
1741     struct _lm_device_t *pdev,
1742     lm_tcp_state_t      *tcp,
1743     lm_tcp_buffer_t     *tcp_buf,
1744     lm_frag_list_t      *frag_list
1745     )
1746 {
1747     lm_tcp_con_t             * rx_con;
1748     lm_tcp_con_rx_gen_info_t * gen_info;
1749     lm_status_t                lm_stat      = LM_STATUS_SUCCESS;
1750     d_list_t                   return_list; /* buffers to return to pool in case of copying to buffer */
1751     u32_t                      copied_bytes = 0;
1752     u32_t                      add_sws_bytes = 0;
1753     u8_t                       split_buffer = FALSE;
1754 
1755     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_post_buf cid=%d\n", tcp->cid);
1756     DbgBreakIf(!(pdev && tcp));
1757     DbgBreakIf(tcp->cid && (tcp != lm_cid_cookie(pdev, TOE_CONNECTION_TYPE, tcp->cid)));
1758     /* (tcp_buf==NULL <=> frag_list==NULL) && (frag_list!= NULL => frag_list->cnt != 0) */
1759     DbgBreakIf( ( ! ( ( (!tcp_buf) && (!frag_list) ) || (tcp_buf && frag_list) ) ) ||
1760                 ( frag_list && (frag_list->cnt == 0) ) );
1761 
1762     rx_con = tcp->rx_con;
1763     if ( GET_FLAGS(rx_con->flags, TCP_RX_POST_BLOCKED) ) {
1764 //        DbgBreakIf(!tcp_buf); /* (lm_tcp_rx_post_buf design guides VBD doc) */
1765         if (!tcp_buf) {
1766             tcp->rx_con->zb_rx_post_blocked++;
1767             return LM_STATUS_SUCCESS;
1768         } else {
1769             tcp->rx_con->rx_post_blocked++;
1770             return LM_STATUS_CONNECTION_CLOSED;
1771         }
1772     }
1773 
1774     /* TCP_POST_DELAYED is turned on when the lm can not process new buffers for some reason, but not permanently
1775      * Assumption: UM will eventually try to repost this buffer... */
1776     if ( GET_FLAGS(rx_con->flags, TCP_POST_DELAYED)) {
1777         return LM_STATUS_FAILURE;
1778     }
1779 
1780     RESET_FLAGS(rx_con->flags, TCP_INDICATE_REJECTED);
1781 
1782     /* set tcp_buf fields */
1783     if (tcp_buf) {
1784         /* check bd chain availability  */
1785         if(lm_bd_chain_avail_bds(&rx_con->bd_chain) < frag_list->cnt) {
1786             DbgBreakIf(s_list_is_empty(&rx_con->active_tb_list));
1787             /* Check if the last placed BD was part of a split buffer (no end flag) if so, mark is at special split-end
1788              * and give a doorbell as if it was with END. Also, block UM from giving us more buffers until we've completed
1789              * this one (See L4 VBD Spec for more details on "Large Application Buffers" */
1790             if (!(GET_FLAGS(rx_con->u.rx.last_rx_bd->flags , TOE_RX_BD_END))) {
1791                 SET_FLAGS(rx_con->u.rx.last_rx_bd->flags, (TOE_RX_BD_END | TOE_RX_BD_SPLIT));
1792                 /* Mark the last buffer in active-tb-list as 'special' so that we know when we complete it that we can
1793                  * unblock UM... */
1794                 tcp_buf = (lm_tcp_buffer_t *)s_list_peek_tail(&rx_con->active_tb_list);
1795                 SET_FLAGS(tcp_buf->flags, TCP_BUF_FLAG_L4_SPLIT);
1796                 SET_FLAGS(rx_con->flags, TCP_POST_DELAYED);
1797                 lm_tcp_rx_write_db(pdev, tcp);
1798             }
1799             DbgMessage(pdev, INFORMl4rx, "post rx buf failed, rx chain is full (cid=%d, avail bds=%d, buf nfrags=%d)\n",
1800                         tcp->cid, lm_bd_chain_avail_bds(&rx_con->bd_chain), frag_list->cnt);
1801             return LM_STATUS_RESOURCE;
1802         }
1803 
1804         tcp_buf->size = tcp_buf->more_to_comp = (u32_t)frag_list->size;
1805         tcp_buf->bd_used = 0; /* will be modified if buffer will be posted */
1806         DbgBreakIf(!(tcp_buf->flags & TCP_BUF_FLAG_L4_POST_START ?
1807                  rx_con->app_buf_bytes_acc_post == 0 :
1808                  rx_con->app_buf_bytes_acc_post > 0));
1809         rx_con->app_buf_bytes_acc_post += tcp_buf->size;
1810 
1811         /* special care in case of last tcp buffer of an application buffer */
1812         if(tcp_buf->flags & TCP_BUF_FLAG_L4_POST_END) {
1813             tcp_buf->app_buf_xferred = 0; /* just for safety */
1814             tcp_buf->app_buf_size = rx_con->app_buf_bytes_acc_post;
1815             rx_con->app_buf_bytes_acc_post = 0;
1816         }
1817         split_buffer = !(GET_FLAGS(tcp_buf->flags, TCP_BUF_FLAG_L4_POST_START) && GET_FLAGS(tcp_buf->flags, TCP_BUF_FLAG_L4_POST_END));
1818     } else {
1819         /* zero-byte request */
1820         rx_con->u.rx.rx_zero_byte_recv_reqs++;
1821     }
1822 
1823     /* we could be in the middle of completing a split-buffer... this is in case the previous split buffer was posted partially and we got a
1824      * cmp with push... need to complete it here. */
1825     if (GET_FLAGS(rx_con->flags, TCP_POST_COMPLETE_SPLIT)) {
1826         DbgBreakIf(!split_buffer); /* we can only be in this state if we're completing split buffers... */
1827         rx_con->bytes_push_skip_cnt += tcp_buf->more_to_comp; /* how many bytes did we skip? */
1828         tcp_buf->more_to_comp = 0;
1829         rx_con->partially_completed_buf_cnt++;
1830         /* complete buffer */
1831         s_list_push_tail(&(tcp->rx_con->active_tb_list), &(tcp_buf->link));
1832         rx_con->rq_nbytes += tcp_buf->size;
1833         rx_con->buffer_skip_post_cnt++;
1834         lm_tcp_complete_tcp_buf(pdev, tcp, rx_con,tcp_buf,0);
1835         return LM_STATUS_SUCCESS;
1836     }
1837 
1838     gen_info = &rx_con->u.rx.gen_info;
1839 
1840     if ( gen_info->peninsula_nbytes ) {
1841         DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_post_buf WITH GENERIC, cid=%d, tcp_buf=%p, buf_size=%d, buf_flags=%d, peninsula_nbytes=%d\n",
1842                 tcp->cid, tcp_buf, frag_list ? frag_list->size : 0, tcp_buf ? tcp_buf->flags : 0, rx_con->u.rx.gen_info.peninsula_nbytes);
1843         if (tcp_buf) {
1844             d_list_init(&return_list, NULL, NULL, 0);
1845             copied_bytes = lm_tcp_rx_peninsula_to_rq_copy(pdev,tcp,tcp_buf,&return_list, 0xffffffff, FALSE);
1846             gen_info->bytes_copied_cnt_in_post += copied_bytes;
1847             if (!d_list_is_empty(&return_list)) {
1848                 lm_tcp_return_list_of_gen_bufs(pdev,tcp,&return_list, MM_TCP_RGB_COMPENSATE_GRQS, NON_EXISTENT_SB_IDX);
1849             }
1850             if ((copied_bytes == tcp_buf->size) && !split_buffer && s_list_is_empty(&rx_con->active_tb_list)) {
1851                 /* consumed_cnt: our way of telling fw we bypassed it */
1852                 lm_tcp_incr_consumed_gen(pdev, tcp, tcp_buf->size);
1853                 /* simulate a _lm_tcp_rx_post_buf for lm_tcp_complete_bufs */
1854                 s_list_push_tail(&(tcp->rx_con->active_tb_list), &(tcp_buf->link));
1855                 rx_con->rq_nbytes += tcp_buf->size;
1856                 rx_con->buffer_skip_post_cnt++;
1857                 rx_con->bytes_skip_post_cnt += copied_bytes;
1858                 /* If we copied some bytes to the RQ, we can now compensate FW-Window with these copied bytes. */
1859                 add_sws_bytes += copied_bytes;
1860                 /* this function completes nbytes on the tcp buf and may complete the buffer if more_to_comp = 0*/
1861                 lm_tcp_complete_tcp_buf(pdev, tcp, rx_con,tcp_buf,copied_bytes);
1862             } else {
1863                 /* will be posted and therefore get a SKP at some stage. */
1864                 if (!GET_FLAGS(rx_con->flags, TCP_POST_NO_SKP)) {
1865                     rx_con->u.rx.skp_bytes_copied += copied_bytes;
1866                 }
1867                 lm_stat = _lm_tcp_rx_post_buf(pdev, tcp, tcp_buf, frag_list);
1868                 DbgBreakIf(lm_stat != LM_STATUS_SUCCESS);
1869                 if (copied_bytes && GET_FLAGS(rx_con->flags, TCP_POST_NO_SKP)) {
1870                     lm_tcp_rx_write_db(pdev, tcp); /* for the case of split buffer in which bytes/bds are accumulated in bd_more* fields. bd_more* fields must be cleaned at this phase  */
1871                     rx_con->bytes_comp_cnt += copied_bytes;
1872                     /* If we copied some bytes to the RQ, we can now compensate FW-Window with these copied bytes. */
1873                     add_sws_bytes += copied_bytes;
1874                     /* this function completes nbytes on the tcp buf and may complete the buffer if more_to_comp = 0*/
1875                     lm_tcp_complete_tcp_buf(pdev, tcp, rx_con,tcp_buf,copied_bytes);
1876                 }
1877             }
1878         }
1879         /* if we have something to indicate after copying and it's ok to indicate... - indicate it */
1880         if (gen_info->peninsula_nbytes && _lm_tcp_ok_to_indicate(rx_con)) {
1881             DbgBreakIf(frag_list && (frag_list->size != copied_bytes)); /* can't have bytes left with free space in tcp buf */
1882             mm_tcp_rx_indicate_gen(pdev, tcp);
1883             add_sws_bytes += gen_info->add_sws_bytes; /* any bytes we need to update will be aggregated here during indicate */
1884             gen_info->add_sws_bytes = 0;
1885 
1886         }
1887     } else if (tcp_buf) {
1888         DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_post_buf NO COPY, cid=%d, tcp_buf=%p, buf_size=%d, buf_flags=%d, peninsula_nbytes=%d\n",
1889             tcp->cid, tcp_buf, frag_list->size, tcp_buf->flags, rx_con->u.rx.gen_info.peninsula_nbytes);
1890         lm_stat = _lm_tcp_rx_post_buf(pdev, tcp, tcp_buf, frag_list);
1891         DbgBreakIf(lm_stat != LM_STATUS_SUCCESS);
1892     }
1893 
1894     if (add_sws_bytes) {
1895         lm_tcp_rx_post_sws(pdev, tcp, rx_con, add_sws_bytes, TCP_RX_POST_SWS_INC);
1896     }
1897 
1898 
1899     return lm_stat;
1900 }
1901 
1902 
1903 /* Assumptions:
1904  *  - caller initiated appropriately the following fields:
1905  *      - tcp_buf->flags
1906  *      - tcp_buf->size, tcp_buf->more_to_comp
1907  *      - tcp_buf->app_buf_size, tcp_buf->app_buf_xferred
1908  *  - caller verified that there is enough availabe BDs in the BD chain for the given buffer */
_lm_tcp_rx_post_buf(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_buffer_t * tcp_buf,lm_frag_list_t * frag_list)1909 static lm_status_t _lm_tcp_rx_post_buf(
1910     struct _lm_device_t *pdev,
1911     lm_tcp_state_t      *tcp,
1912     lm_tcp_buffer_t     *tcp_buf,
1913     lm_frag_list_t      *frag_list
1914     )
1915 {
1916     lm_tcp_con_t *rx_con = tcp->rx_con;
1917     lm_bd_chain_t * rx_chain;
1918     u16_t old_prod, new_prod;
1919     struct toe_rx_bd * rx_bd;
1920     lm_frag_t * frag = frag_list->frag_arr;
1921     u32_t  dbg_buf_size = 0;
1922     u32_t  bd_bytes_prod; /* Each bd is initialized with a cyclic counter of bytes prod until that bd. */
1923     u16_t  flags = 0;
1924     u32_t i;
1925 
1926     /* Number of fragments of entire application buffer can't be bigger
1927      * than size of the BD chain (entire application buffer since we can't
1928      * post partial application buffer to the FW , db_more_bds however includes the "next" bd, so we need
1929      * to take that into consideration as well */
1930     DbgBreakIfAll( (rx_con->db_more_bds + frag_list->cnt) > (u32_t)(rx_con->bd_chain.capacity + rx_con->bd_chain.page_cnt));
1931 
1932     rx_chain = &rx_con->bd_chain;
1933     DbgBreakIf(lm_bd_chain_avail_bds(rx_chain) < frag_list->cnt);
1934 
1935     old_prod = lm_bd_chain_prod_idx(rx_chain);
1936 
1937     /* First BD should have the START flag */
1938     if(tcp_buf->flags & TCP_BUF_FLAG_L4_POST_START) {
1939         flags = TOE_RX_BD_START;
1940     }
1941 
1942     /* Set NO_PUSH flag if needed */
1943     if ( tcp_buf->flags & TCP_BUF_FLAG_L4_RX_NO_PUSH ) {
1944         flags |= TOE_RX_BD_NO_PUSH;
1945     }
1946     if (tcp_buf->flags & TCP_BUF_FLAG_L4_PARTIAL_FILLED) {
1947         if (!rx_con->partially_filled_buf_sent && !rx_con->rq_completion_calls) {
1948             SET_FLAGS(rx_con->db_data.rx->flags, TOE_RX_DB_DATA_PARTIAL_FILLED_BUF);
1949         } else {
1950             RESET_FLAGS(tcp_buf->flags, TCP_BUF_FLAG_L4_PARTIAL_FILLED);
1951         }
1952         rx_con->partially_filled_buf_sent++;
1953     }
1954     /* Attach the first frag to the BD-chain */
1955     bd_bytes_prod = rx_con->db_data.rx->bytes_prod + rx_con->db_more_bytes;
1956     rx_bd = _lm_tcp_rx_set_bd(frag, flags, rx_chain, bd_bytes_prod);
1957     bd_bytes_prod += frag->size;
1958     dbg_buf_size += frag->size;
1959     flags &= ~TOE_RX_BD_START;
1960     frag++;
1961 
1962      /* "attach" the frags to the bd chain */
1963     for(i = 1; i < frag_list->cnt; i++, frag++) {
1964         rx_bd = _lm_tcp_rx_set_bd(frag, flags, rx_chain, bd_bytes_prod);
1965         dbg_buf_size += frag->size;
1966         bd_bytes_prod += frag->size;
1967     }
1968     tcp->rx_con->u.rx.last_rx_bd = rx_bd;
1969 
1970     /* The last BD must have the END flag */
1971     if(tcp_buf->flags & TCP_BUF_FLAG_L4_POST_END) {
1972         rx_bd->flags |= TOE_RX_BD_END;
1973         DbgMessage(NULL, VERBOSEl4rx, "Setting Rx last BD flags=0x%x\n", rx_bd->flags);
1974     }
1975 
1976     DbgBreakIf(frag_list->cnt > TCP_MAX_SGL_SIZE);
1977     tcp_buf->bd_used = frag_list->cnt & TCP_MAX_SGL_SIZE;
1978     DbgBreakIf(tcp_buf->size != dbg_buf_size);
1979 
1980     /* Perpare data for a DoorBell */
1981     rx_con->db_more_bytes += tcp_buf->size;
1982     new_prod = lm_bd_chain_prod_idx(rx_chain);
1983     DbgBreakIf(S16_SUB(new_prod, old_prod) < tcp_buf->bd_used);
1984     rx_con->db_more_bds += S16_SUB(new_prod, old_prod);
1985     rx_con->db_more_bufs++;
1986 
1987     /* Enqueue the buffer to the active_tb_list */
1988     s_list_push_tail(&(rx_con->active_tb_list), &(tcp_buf->link));
1989     rx_con->rq_nbytes += tcp_buf->size;
1990 
1991     if(tcp_buf->flags & TCP_BUF_FLAG_L4_POST_END) {
1992         lm_tcp_rx_write_db(pdev, tcp);
1993     }
1994 
1995 
1996     return LM_STATUS_SUCCESS;
1997 }
1998 
_lm_tcp_rx_get_buffered_data(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_frag_list_t ** frag_list,lm_tcp_gen_buf_t ** gen_buf)1999 static lm_status_t _lm_tcp_rx_get_buffered_data(
2000     struct _lm_device_t * pdev,
2001     lm_tcp_state_t      * tcp,
2002     lm_frag_list_t     ** frag_list,  /* if *frag_list is NULL, the rx con pre-allocaed will be used */
2003     lm_tcp_gen_buf_t   ** gen_buf
2004     )
2005 {
2006     lm_tcp_con_t             * rx_con = tcp->rx_con;
2007     lm_tcp_con_rx_gen_info_t * gen_info;
2008     lm_tcp_gen_buf_t         * head_of_indication;
2009     d_list_t                   indicate_list;
2010     d_list_entry_t           * entry;
2011     lm_tcp_gen_buf_t         * curr_gen_buf;
2012     u32_t                      gen_offset, i;
2013     u32_t                      num_bufs_to_indicate;
2014     u32_t                      ind_nbufs=0, ind_nbytes=0;
2015     u8_t                       dont_send_to_system_more_then_rwin;
2016     DbgMessage(pdev, VERBOSEl4rx, "###_lm_tcp_rx_get_buffered_data cid=%d\n", tcp->cid);
2017 
2018     gen_info = &rx_con->u.rx.gen_info;
2019 
2020 
2021     if ((u16_t)tcp->tcp_cached.rcv_indication_size != 0) {
2022         DbgBreakMsg("MichalS rcv_indication_size != 0 not implemented\n");
2023        /* MichalS TBA: RcvIndicationSize > 0 will change following block quite a lot */
2024     }
2025 
2026     num_bufs_to_indicate = d_list_entry_cnt(&gen_info->peninsula_list);
2027 
2028     /* The buffers in peninsula_list are ALWAYS released, unreleased buffers are in the dpc_peninsula_list. */
2029     DbgBreakIf(((lm_tcp_gen_buf_t *)d_list_peek_tail(&gen_info->peninsula_list))->placed_bytes == 0);
2030 
2031     if (*frag_list == NULL) {
2032         *frag_list = gen_info->frag_list;
2033         (*frag_list)->cnt = gen_info->max_frag_count;
2034     }
2035 
2036     if (num_bufs_to_indicate > (*frag_list)->cnt) {
2037         DbgMessage(pdev, WARNl4rx, "_lm_tcp_rx_get_buffered_data: number of buffers to indicate[%d] is larger than frag_cnt[%d] cid=%d\n",
2038                 num_bufs_to_indicate, (*frag_list)->cnt, tcp->cid);
2039         num_bufs_to_indicate = gen_info->max_frag_count;
2040         gen_info->num_non_full_indications++;
2041     }
2042     d_list_init(&indicate_list, NULL, NULL, 0);
2043     dont_send_to_system_more_then_rwin = (u8_t)gen_info->dont_send_to_system_more_then_rwin;
2044     while (num_bufs_to_indicate--) {
2045         entry = d_list_pop_head(&gen_info->peninsula_list);
2046         DbgBreakIf(entry == NULL);
2047         if (dont_send_to_system_more_then_rwin) {
2048             if ((ind_nbytes + ((lm_tcp_gen_buf_t *)entry)->placed_bytes)
2049                     > tcp->tcp_cached.initial_rcv_wnd) {
2050                 if (ind_nbytes) {
2051                     d_list_push_head(&gen_info->peninsula_list, entry);
2052                     break;
2053                 } else {
2054                     dont_send_to_system_more_then_rwin = FALSE;
2055                 }
2056             }
2057         }
2058         d_list_push_tail(&indicate_list, entry);
2059         ind_nbufs ++;
2060         ind_nbytes += ((lm_tcp_gen_buf_t *)entry)->placed_bytes;
2061     }
2062 
2063     ind_nbytes -= gen_info->first_buf_offset;
2064 
2065     head_of_indication = (lm_tcp_gen_buf_t *)d_list_peek_head(&indicate_list);
2066 
2067     if CHK_NULL(head_of_indication)
2068     {
2069         DbgBreakIfAll( !head_of_indication ) ;
2070         return LM_STATUS_FAILURE ;
2071     }
2072 
2073     head_of_indication->tcp = tcp;
2074     head_of_indication->ind_nbufs = ind_nbufs;
2075     head_of_indication->ind_bytes = ind_nbytes;
2076     DbgBreakIf(gen_info->peninsula_nbytes < ind_nbytes);
2077     gen_info->peninsula_nbytes -= ind_nbytes;
2078 
2079     /* initialize frag list */
2080     (*frag_list)->cnt  = ind_nbufs;
2081     (*frag_list)->size = ind_nbytes;
2082     curr_gen_buf = head_of_indication;
2083 
2084     gen_offset = gen_info->first_buf_offset;
2085     for (i = 0; i < (*frag_list)->cnt; i++ ) {
2086         (*frag_list)->frag_arr[i].addr.as_ptr = curr_gen_buf->buf_virt + gen_offset;
2087         (*frag_list)->frag_arr[i].size = curr_gen_buf->placed_bytes - gen_offset;
2088         curr_gen_buf = NEXT_GEN_BUF(curr_gen_buf);
2089         gen_offset = 0;  /* only first buffer can have an offset */
2090         /* we don't touch gen_info->first_buf_offset - this is handled in lm_tcp_rx_buffered_data_indicated */
2091     }
2092     *gen_buf = head_of_indication;
2093     DbgMessage(pdev, VERBOSEl4rx, "###_lm_tcp_rx_get_buffered_data ind_bytes = %d\n", (*frag_list)->size);
2094 
2095     mm_atomic_inc(&pdev->toe_info.stats.total_indicated);
2096     return LM_STATUS_SUCCESS;
2097 }
2098 
lm_tcp_rx_get_buffered_data_from_terminate(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_frag_list_t ** frag_list,lm_tcp_gen_buf_t ** gen_buf)2099 lm_status_t lm_tcp_rx_get_buffered_data_from_terminate (
2100     struct _lm_device_t * pdev,
2101     lm_tcp_state_t      * tcp,
2102     lm_frag_list_t     ** frag_list,
2103     lm_tcp_gen_buf_t   ** gen_buf
2104     )
2105 {
2106     lm_tcp_con_t             * rx_con           = tcp->rx_con;
2107     lm_tcp_con_rx_gen_info_t * gen_info;
2108     u16_t                      buff_cnt;
2109     lm_tcp_gen_buf_t         * unwanted_gen_buf = NULL;
2110     lm_tcp_gen_buf_t         * temp_gen_buf     = NULL;
2111     lm_status_t                lm_status        = LM_STATUS_SUCCESS;
2112 
2113     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_get_buffered_data_from_terminate cid=%d\n", tcp->cid);
2114 
2115     gen_info = &rx_con->u.rx.gen_info;
2116 
2117     /* make sure ALL the peninsula is released */
2118     DbgBreakIf(!d_list_is_empty(&gen_info->peninsula_list) &&
2119                (((lm_tcp_gen_buf_t *)d_list_peek_tail(&gen_info->peninsula_list))->placed_bytes == 0));
2120 
2121     *frag_list = NULL;
2122     if (gen_info->peninsula_nbytes == 0) {
2123         return LM_STATUS_SUCCESS;
2124     }
2125 
2126 /*  DbgBreakIf(gen_info->peninsula_nbytes > tcp->tcp_cached.initial_rcv_wnd);*/
2127     gen_info->dont_send_to_system_more_then_rwin = FALSE;
2128     if ((buff_cnt = (u16_t)d_list_entry_cnt(&gen_info->peninsula_list)) > gen_info->max_frag_count) {
2129         lm_bd_chain_t *bd_chain = &tcp->rx_con->bd_chain;
2130         u16_t possible_frag_count, decreased_count;
2131         possible_frag_count = (/*bd_chain->page_cnt**/
2132                                LM_PAGE_SIZE - sizeof(lm_frag_list_t)) / sizeof(lm_frag_t);
2133         DbgMessage(pdev, WARNl4rx | WARNl4sp,
2134                     "###lm_tcp_rx_get_buffered_data_from_terminate cid=%d: peninsula_list cnt (%d) > max frag_count (%d)\n",
2135                     tcp->cid, buff_cnt, gen_info->max_frag_count);
2136 
2137         if (possible_frag_count > gen_info->max_frag_count) {
2138             /* This solution is ugly:
2139                 since there will not be any further buffered data indications to the client, we must be able to
2140                 indicate all the buffered data now. But the preallocated frag list in the rx con is too short!
2141                 So instead of the pre-allocated frag list we need to use a larger memory. Our options:
2142                 1. allocate memory here and release it later.
2143                 2. use other pre-allocated memory that is not in use anymore (e.g. the bd chain) [chosen solution]
2144                 In any case both solutions may fail: memory allocation can fail and the other pre-allocated memory
2145                 might also be too short. the fallback from this is:
2146                 - don't indicate anything and release the peninsula (NOT IMPLEMENTED)
2147             DbgBreakIfAll((u16_t)(sizeof(lm_frag_list_t) + sizeof(lm_frag_t)*buff_cnt) > bd_chain->page_cnt*LM_PAGE_SIZE); */
2148             if (possible_frag_count < buff_cnt) {
2149                 decreased_count = possible_frag_count;
2150                 DbgMessage(pdev, WARNl4rx | WARNl4sp,
2151                             "###lm_tcp_rx_get_buffered_data_from_terminate cid=%d: peninsula_list cnt (%d) > aux.frag_cnt (%d)\n",
2152                             tcp->cid, buff_cnt, possible_frag_count);
2153             } else {
2154                 decreased_count = 0;
2155                 DbgMessage(pdev, WARNl4rx | WARNl4sp,
2156                             "###lm_tcp_rx_get_buffered_data_from_terminate cid=%d: aux.frag_cnt (%d) is enough for %d buffs\n",
2157                             tcp->cid, possible_frag_count, buff_cnt);
2158             }
2159             *frag_list = (lm_frag_list_t*)bd_chain->bd_chain_virt;
2160             (*frag_list)->cnt = possible_frag_count;
2161             (*frag_list)->size = 0;
2162         } else {
2163             decreased_count = (u16_t)gen_info->max_frag_count;
2164         }
2165         if (decreased_count) {
2166             u16_t returned_buff_cnt = lm_squeeze_rx_buffer_list(pdev, tcp, decreased_count, &unwanted_gen_buf);
2167             if (decreased_count < returned_buff_cnt) {
2168                 lm_frag_list_t* new_frag_list;
2169                 u32_t mem_size_for_new_frag_list = returned_buff_cnt * sizeof(lm_frag_t) + sizeof(lm_frag_list_t);
2170 //                new_frag_list = (lm_frag_list_t*)mm_alloc_mem(pdev, mem_size_for_new_frag_list, LM_RESOURCE_NDIS);
2171                 new_frag_list = (lm_frag_list_t*)mm_rt_alloc_mem(pdev, mem_size_for_new_frag_list, LM_RESOURCE_NDIS);
2172 
2173                 if (new_frag_list != NULL) {
2174                     tcp->type_of_aux_memory = TCP_CON_AUX_RT_MEM;
2175                     tcp->aux_memory = new_frag_list;
2176                     tcp->aux_mem_size = mem_size_for_new_frag_list;
2177                     *frag_list = new_frag_list;
2178                     (*frag_list)->cnt = returned_buff_cnt;
2179                     (*frag_list)->size = 0;
2180                     tcp->aux_mem_flag = TCP_CON_AUX_RT_MEM_SUCCSESS_ALLOCATION;
2181                 } else {
2182                     /* No way. Let's send up only part of data. Data distortion is unavoidable.
2183                        TODO: prevent data distortion by termination the connection itself at least */
2184                     lm_status = LM_STATUS_RESOURCE;
2185                     tcp->aux_mem_flag = TCP_CON_AUX_RT_MEM_FAILED_ALLOCATION;
2186                     /* Get rid of whatever remains in the peninsula...add it to unwanted... */
2187                     if (unwanted_gen_buf)
2188                     {
2189                         temp_gen_buf = (lm_tcp_gen_buf_t*)d_list_peek_tail(&gen_info->peninsula_list);
2190                         if (temp_gen_buf)
2191                         {
2192                             temp_gen_buf->link.next = &(unwanted_gen_buf->link);
2193                             unwanted_gen_buf->link.prev = &(temp_gen_buf->link);
2194                             unwanted_gen_buf = (lm_tcp_gen_buf_t*)d_list_peek_head(&gen_info->peninsula_list);
2195                         }
2196                     }
2197                     else
2198                     {
2199                         unwanted_gen_buf = (lm_tcp_gen_buf_t*)d_list_peek_head(&gen_info->peninsula_list);
2200                     }
2201                     d_list_clear(&gen_info->peninsula_list);
2202 
2203                 }
2204             }
2205         }
2206     }
2207     if (lm_status == LM_STATUS_SUCCESS)
2208     {
2209         _lm_tcp_rx_get_buffered_data(pdev, tcp, frag_list, gen_buf);
2210 
2211         /* for cleaness: lm_tcp_rx_buffered_data_indicated will not be called
2212          * indication is 'succesfull' */
2213         gen_info->num_bytes_indicated += (u32_t)(*frag_list)->size;
2214         gen_info->first_buf_offset = 0;
2215         gen_info->num_buffers_indicated += (*gen_buf)->ind_nbufs;
2216     }
2217 
2218     gen_info->peninsula_blocked = TRUE;
2219 
2220     if (unwanted_gen_buf) {
2221          lm_tcp_return_gen_bufs(pdev, tcp, unwanted_gen_buf,MM_TCP_RGB_COMPENSATE_GRQS, NON_EXISTENT_SB_IDX);
2222     }
2223 
2224     if (*gen_buf) {
2225         /* with data taken from terminate, we can always act as in 'short-loop' since the bytes for
2226          * this connection won't increase the window anyway... */
2227         (*gen_buf)->flags &= ~GEN_FLAG_SWS_UPDATE;
2228     }
2229 
2230     return lm_status;
2231 }
2232 
lm_tcp_rx_get_buffered_data(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_frag_list_t ** frag_list,lm_tcp_gen_buf_t ** gen_buf)2233 lm_status_t lm_tcp_rx_get_buffered_data(
2234     struct _lm_device_t * pdev,
2235     lm_tcp_state_t      * tcp,
2236     lm_frag_list_t     ** frag_list,
2237     lm_tcp_gen_buf_t   ** gen_buf
2238     )
2239 {
2240     lm_tcp_con_t             * rx_con = tcp->rx_con;
2241     lm_tcp_con_rx_gen_info_t * gen_info;
2242     lm_status_t                lm_status;
2243 
2244     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_get_buffered_data cid=%d\n", tcp->cid);
2245     gen_info = &rx_con->u.rx.gen_info;
2246 
2247     DbgBreakIf(gen_info->peninsula_blocked == TRUE); /* terminate was already called */
2248 
2249     if (gen_info->peninsula_nbytes == 0 || (rx_con->flags & TCP_RX_IND_BLOCKED)) {
2250         return LM_STATUS_FAILURE;
2251     }
2252 
2253     *frag_list = NULL;
2254     lm_status = _lm_tcp_rx_get_buffered_data(pdev, tcp, frag_list, gen_buf);
2255     if (*gen_buf) {
2256         if (gen_info->update_window_mode == LM_TOE_UPDATE_MODE_LONG_LOOP) {
2257             gen_info->pending_indicated_bytes += (*gen_buf)->ind_bytes;
2258             /* We need to increase the number of pending return indications here, since once we return
2259              * we are basically pending for the return of this specific indication. There are two cases
2260              * that require decreasing the pending return indications. The first is if the indication failed
2261              * the second is if it succeeded AND the buffers returned... */
2262              gen_info->pending_return_indications++;
2263             (*gen_buf)->flags |= GEN_FLAG_SWS_UPDATE;
2264         } else {
2265             (*gen_buf)->flags &= ~GEN_FLAG_SWS_UPDATE;
2266         }
2267     }
2268 
2269     return LM_STATUS_SUCCESS;
2270 }
2271 
lm_tcp_rx_buffered_data_indicated(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u32_t accepted_bytes,lm_tcp_gen_buf_t * gen_buf)2272 void lm_tcp_rx_buffered_data_indicated(
2273     struct _lm_device_t * pdev,
2274     lm_tcp_state_t      * tcp,
2275     u32_t                 accepted_bytes,
2276     lm_tcp_gen_buf_t    * gen_buf /* head of indications generic buffer NULL if indication succeeded */
2277     )
2278 {
2279     lm_tcp_con_rx_gen_info_t * gen_info = &tcp->rx_con->u.rx.gen_info;
2280 
2281     DbgMessage(pdev, VERBOSEl4rx , "###lm_tcp_rx_buffered_data_indicated accepted_bytes = %d cid=%d\n", accepted_bytes, tcp->cid);
2282 
2283     DbgBreakIf(gen_info->peninsula_blocked == TRUE); /* terminate was already called */
2284 
2285     lm_tcp_incr_consumed_gen(pdev, tcp, accepted_bytes);
2286     gen_info->num_bytes_indicated += accepted_bytes;
2287 
2288     if (gen_buf == NULL) { /* succesfull indication */
2289         gen_info->first_buf_offset = 0;
2290         if (gen_info->update_window_mode == LM_TOE_UPDATE_MODE_SHORT_LOOP) {
2291             gen_info->add_sws_bytes += accepted_bytes;
2292         }
2293         gen_info->num_success_indicates++;
2294         gen_info->bytes_indicated_accepted += accepted_bytes;
2295         tcp->rx_con->u.rx.zero_byte_posted_during_ind = FALSE;
2296     } else {  /* complete rejection / partial success, gen_buf remains in our control  */
2297         /* indication failed */
2298         lm_tcp_gen_buf_t * curr_gen_buf, * ret_buf;
2299         d_list_t          return_to_pool_list;
2300         d_list_t          return_to_peninsula_list;
2301         u32_t             nbytes;
2302         DbgBreakIf(accepted_bytes > gen_buf->ind_bytes);
2303         gen_info->peninsula_nbytes += gen_buf->ind_bytes - accepted_bytes;
2304 
2305         gen_info->num_failed_indicates++;
2306         gen_info->bytes_indicated_accepted+= accepted_bytes;
2307         gen_info->bytes_indicated_rejected+= gen_buf->ind_bytes - accepted_bytes;
2308 
2309         DbgMessage(pdev, INFORMl4rx, "GENERIC: %s Indication for cid=%d accepted_bytes=%d\n",
2310                     (accepted_bytes == 0)? "Rejected" : "Partial", tcp->cid, accepted_bytes);
2311 
2312         d_list_init(&return_to_pool_list, NULL, NULL, 0);
2313         d_list_init(&return_to_peninsula_list, NULL, NULL, 0);
2314 
2315         DbgBreakIf(gen_buf->tcp->rx_con->flags & TCP_INDICATE_REJECTED);
2316         if (tcp->rx_con->u.rx.zero_byte_posted_during_ind) {
2317             tcp->rx_con->u.rx.zero_byte_posted_during_ind = FALSE;
2318         } else {
2319             gen_buf->tcp->rx_con->flags |= TCP_INDICATE_REJECTED;
2320         }
2321 
2322         curr_gen_buf = gen_buf;
2323 
2324         /* indicated bytes are in fact 'freed up' space: so we can make the sws_bytes larger,
2325          * this is always true here luxury-mode or not */
2326         gen_info->add_sws_bytes += accepted_bytes;
2327 
2328         /* buffer was returned to us so it is no longer pending return...if we increased the 'pending' we have
2329          * to decrease */
2330         if (gen_buf->flags & GEN_FLAG_SWS_UPDATE) {
2331             gen_info->pending_return_indications--;
2332             gen_info->pending_indicated_bytes-=gen_buf->ind_bytes;
2333         }
2334         mm_atomic_inc(&pdev->toe_info.stats.total_indicated_returned); /* stats */
2335 
2336         /* return buffers that were fully indicated to the generic pool, ones that we're not, to the peninsula */
2337         while (accepted_bytes) {
2338             nbytes = ((lm_tcp_gen_buf_t *)curr_gen_buf)->placed_bytes - gen_info->first_buf_offset;
2339             if (accepted_bytes >= nbytes) {
2340                 /* the buffer was completely accepted */
2341                 accepted_bytes -= nbytes;
2342                 ret_buf = curr_gen_buf;
2343                 curr_gen_buf = NEXT_GEN_BUF(curr_gen_buf);
2344                 d_list_push_tail(&return_to_pool_list, &ret_buf->link);
2345                 gen_info->num_buffers_indicated++;
2346                 gen_info->first_buf_offset = 0;
2347             } else {
2348                 gen_info->first_buf_offset += (u16_t)accepted_bytes;
2349                 accepted_bytes = 0;
2350             }
2351         }
2352 
2353         /* is there anything to return to the peninsula ? (i.e. return_head moved) */
2354         while (curr_gen_buf) {
2355             curr_gen_buf->ind_bytes = 0;
2356             curr_gen_buf->ind_nbufs = 0;
2357             ret_buf = curr_gen_buf;
2358             curr_gen_buf = NEXT_GEN_BUF(curr_gen_buf);
2359             gen_info->bufs_indicated_rejected++;
2360             d_list_push_tail(&return_to_peninsula_list, &ret_buf->link);
2361         }
2362 
2363         if (!d_list_is_empty(&return_to_pool_list)) {
2364             lm_tcp_return_list_of_gen_bufs(pdev, tcp, &return_to_pool_list, MM_TCP_RGB_COMPENSATE_GRQS, NON_EXISTENT_SB_IDX);
2365         }
2366 
2367         /* There must be at least something to return to the peninsula since this was partial indication */
2368         DbgBreakIf(d_list_is_empty(&return_to_peninsula_list));
2369         /* re-insert generic buffers to the peninsula.
2370          * we need to re-insert the buffers to the head of the peninsula */
2371         d_list_add_head(&gen_info->peninsula_list, &return_to_peninsula_list);
2372 
2373     }
2374 
2375 }
2376 
2377 /** Description
2378  * returns the buffers to the generic pool
2379  */
lm_tcp_return_gen_bufs(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_gen_buf_t * gen_buf,u32_t flags,u8_t grq_idx)2380 void lm_tcp_return_gen_bufs(struct _lm_device_t * pdev, lm_tcp_state_t * tcp, lm_tcp_gen_buf_t * gen_buf,u32_t flags, u8_t grq_idx)
2381 {
2382     lm_tcp_gen_buf_t * curr_gen_buf = gen_buf;
2383 
2384     #if DBG
2385     gen_buf->ind_nbufs = 0; /* for debugging purposes will count how many buffers are in our list */
2386     while (curr_gen_buf) {
2387         DbgBreakIf(SIG(curr_gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
2388         DbgBreakIf(END_SIG(curr_gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
2389         /* We increase the bytes for both pool-buffers, and buffered-data buffers because when the OS
2390          * gives posted buffers the window is smaller */
2391         curr_gen_buf = NEXT_GEN_BUF(curr_gen_buf);
2392         gen_buf->ind_nbufs++;
2393     }
2394     #endif
2395 
2396     mm_tcp_return_gen_bufs(pdev, gen_buf,flags,grq_idx);
2397 }
2398 
2399 /** Description
2400  * returns the buffers to the generic pool
2401  */
lm_tcp_return_list_of_gen_bufs(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,d_list_t * gen_buf_list,u32_t flags,u8_t grq_idx)2402 void lm_tcp_return_list_of_gen_bufs(struct _lm_device_t * pdev, lm_tcp_state_t * tcp, d_list_t * gen_buf_list,u32_t flags, u8_t grq_idx)
2403 {
2404     lm_tcp_gen_buf_t * gen_buf = (lm_tcp_gen_buf_t *)d_list_peek_head(gen_buf_list);
2405     lm_tcp_gen_buf_t * curr_gen_buf = gen_buf;
2406 
2407     #if DBG
2408     gen_buf->ind_nbufs = 0; /* for debugging purposes will count how many buffers are in our list */
2409     while (curr_gen_buf) {
2410         DbgBreakIf(SIG(curr_gen_buf->buf_virt) != L4GEN_BUFFER_SIG);
2411         DbgBreakIf(END_SIG(curr_gen_buf->buf_virt, LM_TCP_GEN_BUF_SIZE(pdev)) != L4GEN_BUFFER_SIG_END);
2412         /* We increase the bytes for both pool-buffers, and buffered-data buffers because when the OS
2413          * gives posted buffers the window is smaller */
2414         curr_gen_buf = NEXT_GEN_BUF(curr_gen_buf);
2415         gen_buf->ind_nbufs++;
2416     }
2417     #endif
2418 
2419     mm_tcp_return_list_of_gen_bufs(pdev, gen_buf_list,flags,grq_idx);
2420 }
2421 
lm_tcp_rx_indication_returned(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,lm_tcp_gen_buf_t * gen_buf)2422 void lm_tcp_rx_indication_returned(struct _lm_device_t *pdev, lm_tcp_state_t * tcp, lm_tcp_gen_buf_t * gen_buf)
2423 {
2424     DbgMessage(pdev, VERBOSEl4rx, "###lm_tcp_rx_con_indication_returned cid=%d\n", tcp->cid);
2425 
2426     DbgBreakIf(tcp != gen_buf->tcp);
2427     DbgBreakIf(tcp->cid && (tcp != lm_cid_cookie(pdev, TOE_CONNECTION_TYPE, tcp->cid)));
2428 
2429     /* TBA fix in case of RcvIndicateSize > 0 */
2430     DbgBreakIf(gen_buf->refcnt != 0);
2431 
2432     tcp->rx_con->u.rx.gen_info.pending_return_indications--;
2433     tcp->rx_con->u.rx.gen_info.pending_indicated_bytes -= gen_buf->ind_bytes;
2434 
2435     /* Update the sws bytes according to the ind number of bytes this function is only called if in fact
2436      * this is a buffer that is marked as an 'update buffer' otherwise this function isn't called.  */
2437     DbgBreakIfAll(!(gen_buf->flags & GEN_FLAG_SWS_UPDATE));
2438     lm_tcp_rx_post_sws(pdev, tcp, tcp->rx_con, gen_buf->ind_bytes, TCP_RX_POST_SWS_INC);
2439     lm_tcp_return_gen_bufs(pdev, tcp, gen_buf, MM_TCP_RGB_COMPENSATE_GRQS, NON_EXISTENT_SB_IDX);
2440 }
2441 
lm_tcp_is_tcp_dead(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u8_t op)2442 u8_t lm_tcp_is_tcp_dead(struct _lm_device_t * pdev, lm_tcp_state_t * tcp, u8_t op)
2443 {
2444     UNREFERENCED_PARAMETER_(pdev);
2445 
2446     if(op == TCP_IS_DEAD_OP_UPLD_COMP) {
2447         DbgBreakIf(tcp->hdr.status != STATE_STATUS_UPLOAD_PENDING);
2448         tcp->hdr.status = STATE_STATUS_UPLOAD_DONE;
2449     }
2450     if (GET_FLAGS(tcp->rx_con->flags, TCP_COMP_DEFERRED)) {
2451         /* we can't kill the connection here! it's still being handled by deferred function which will
2452          * access it... killing will be done from that context... */
2453         return FALSE;
2454     }
2455     if (tcp->rx_con->u.rx.gen_info.pending_return_indications == 0) {
2456         /* If the function is called from offload completion flow, we might have completions on the RCQ
2457            that we haven't processed yet so haven't completed / indicated bufs,
2458            so there are bytes in the peninsula and this state is legal */
2459         DbgBreakIf(!(tcp->rx_con->flags & TCP_RX_IND_BLOCKED) &&
2460                     (tcp->rx_con->u.rx.gen_info.peninsula_nbytes != 0) &&
2461                     (op != TCP_IS_DEAD_OP_OFLD_COMP_DFRD));
2462         if (tcp->hdr.status == STATE_STATUS_UPLOAD_DONE) {
2463             return TRUE;
2464         }
2465     }
2466     return FALSE;
2467 }
2468 
lm_tcp_con_status(struct _lm_device_t * pdev,lm_tcp_con_t * rx_con)2469 lm_status_t lm_tcp_con_status(struct _lm_device_t * pdev, lm_tcp_con_t * rx_con)
2470 {
2471     UNREFERENCED_PARAMETER_(pdev);
2472 
2473     if (rx_con->flags & TCP_RX_POST_BLOCKED) {
2474         return LM_STATUS_CONNECTION_CLOSED;
2475     }
2476     return LM_STATUS_SUCCESS;
2477 }
2478 
lm_tcp_calc_gen_buf_size(struct _lm_device_t * pdev)2479 u32_t lm_tcp_calc_gen_buf_size(struct _lm_device_t * pdev)
2480 {
2481     u32_t       gen_buf_size = 0;
2482     u32_t const chain_idx    = LM_SW_LEADING_RSS_CID(pdev);
2483 
2484     /* determine size of buffer: in steps of pages, larger than the minimum and
2485      * the mtu */
2486     if(CHK_NULL(pdev) ||
2487        ERR_IF((ARRSIZE(pdev->params.l2_cli_con_params) <= chain_idx) ||
2488               (CHIP_IS_E1H(pdev) && (chain_idx >= ETH_MAX_RX_CLIENTS_E1H)) || /* TODO E2 add IS_E2*/
2489               (CHIP_IS_E1(pdev) && (chain_idx >= ETH_MAX_RX_CLIENTS_E1)) ))
2490     {
2491         DbgBreakIf(1);
2492         return 0;
2493     }
2494 
2495     if (pdev->params.l4_gen_buf_size < pdev->params.l2_cli_con_params[chain_idx].mtu)
2496     {
2497         gen_buf_size = pdev->params.l2_cli_con_params[chain_idx].mtu;
2498 
2499     }
2500     else
2501     {
2502         gen_buf_size = pdev->params.l4_gen_buf_size;
2503     }
2504     /* bring to page-size boundary */
2505     gen_buf_size = (gen_buf_size + (LM_PAGE_SIZE-1)) & ~(LM_PAGE_SIZE-1);
2506 
2507     return gen_buf_size;
2508 }
2509 
lm_squeeze_rx_buffer_list(struct _lm_device_t * pdev,lm_tcp_state_t * tcp,u16_t adjust_number,lm_tcp_gen_buf_t ** unwanted_gen_buf)2510 u16_t lm_squeeze_rx_buffer_list(
2511     struct _lm_device_t * pdev,
2512     lm_tcp_state_t      * tcp,
2513     u16_t                 adjust_number,
2514     lm_tcp_gen_buf_t   ** unwanted_gen_buf
2515     )
2516 {
2517     u32_t                      gen_buff_size         = lm_tcp_calc_gen_buf_size(pdev);
2518     lm_tcp_con_t             * rx_con                =  tcp->rx_con;
2519     lm_tcp_con_rx_gen_info_t * gen_info              = &rx_con->u.rx.gen_info;
2520     d_list_t                   unwanted_list         = {0};
2521     lm_tcp_gen_buf_t         * gen_buf_copy_to       = NULL;
2522     lm_tcp_gen_buf_t         * gen_buf_copy_from     = NULL, *next_buffer   = NULL;
2523     u16_t                      free_bytes_to_copy    = 0,     bytes_to_copy = 0, gen_buf_offset = 0;
2524     u8_t                       force_buffer_division = FALSE;
2525     u16_t                      buffers_number        = (u16_t)d_list_entry_cnt(&gen_info->peninsula_list);
2526 
2527     *unwanted_gen_buf = NULL;
2528 
2529     if ((adjust_number * gen_buff_size) >= gen_info->peninsula_nbytes) {
2530         d_list_init(&unwanted_list, NULL, NULL, 0);
2531         gen_buf_copy_to = (lm_tcp_gen_buf_t*)d_list_peek_head(&gen_info->peninsula_list);
2532         next_buffer = NEXT_GEN_BUF(gen_buf_copy_to);
2533         free_bytes_to_copy = gen_buff_size - gen_buf_copy_to->placed_bytes;
2534         while (buffers_number > adjust_number) {
2535             gen_buf_copy_from = next_buffer;
2536             if (gen_buf_copy_from != NULL) {
2537                 next_buffer = NEXT_GEN_BUF(gen_buf_copy_from);
2538                 bytes_to_copy = gen_buf_copy_from->placed_bytes;
2539                 if (bytes_to_copy <= free_bytes_to_copy) {
2540                     mm_memcpy(gen_buf_copy_to->buf_virt + gen_buf_copy_to->placed_bytes,
2541                               gen_buf_copy_from->buf_virt, bytes_to_copy);
2542                     free_bytes_to_copy -= bytes_to_copy;
2543                     gen_buf_copy_to->placed_bytes += bytes_to_copy;
2544                     d_list_remove_entry(&gen_info->peninsula_list, &gen_buf_copy_from->link);
2545                     d_list_push_tail(&unwanted_list, &gen_buf_copy_from->link);
2546                     buffers_number--;
2547                     continue;
2548                 } else {
2549                     if (force_buffer_division) {
2550                         if (free_bytes_to_copy) {
2551                             mm_memcpy(gen_buf_copy_to->buf_virt + gen_buf_copy_to->placed_bytes,
2552                                       gen_buf_copy_from->buf_virt, free_bytes_to_copy);
2553                             gen_buf_copy_to->placed_bytes += free_bytes_to_copy;
2554                             mm_memcpy(gen_buf_copy_from->buf_virt,
2555                                       gen_buf_copy_from->buf_virt + free_bytes_to_copy, bytes_to_copy - free_bytes_to_copy);
2556                             gen_buf_copy_from->placed_bytes -= free_bytes_to_copy;
2557                         }
2558                     }
2559                     gen_buf_copy_to = gen_buf_copy_from;
2560                     next_buffer = NEXT_GEN_BUF(gen_buf_copy_from);
2561                     free_bytes_to_copy = gen_buff_size - gen_buf_copy_to->placed_bytes;
2562                     continue;
2563                 }
2564             } else {
2565                 if (!force_buffer_division) {
2566                     force_buffer_division = TRUE;
2567                     gen_buf_copy_to = (lm_tcp_gen_buf_t*)d_list_peek_head(&gen_info->peninsula_list);
2568                     next_buffer = NEXT_GEN_BUF(gen_buf_copy_to);
2569                     gen_buf_offset = gen_info->first_buf_offset;
2570                     if (gen_buf_offset) {
2571                         /* move to start of buffer*/
2572                         mm_memcpy(gen_buf_copy_to->buf_virt,
2573                                   gen_buf_copy_to->buf_virt + gen_buf_offset, gen_buf_copy_to->placed_bytes - gen_buf_offset);
2574                         gen_buf_copy_to->placed_bytes -= gen_buf_offset;
2575                         gen_buf_offset = gen_info->first_buf_offset = 0;
2576                     }
2577                     free_bytes_to_copy = gen_buff_size - gen_buf_copy_to->placed_bytes;
2578                     continue;
2579                 } else {
2580                     DbgMessage(pdev, WARNl4rx | WARNl4sp,
2581                                 "###lm_squeeze_rx_buffer_list cid=%d: peninsula_list cnt (%d) is still more frag_count (%d)\n",
2582                                 tcp->cid, buffers_number, adjust_number);
2583                     break;
2584                 }
2585             }
2586         }
2587         *unwanted_gen_buf = (lm_tcp_gen_buf_t*)d_list_peek_head(&unwanted_list);
2588         DbgMessage(pdev, WARNl4rx | WARNl4sp,
2589                     "###lm_squeeze_rx_buffer_list cid=%d(%d,%d,%d): peninsula_list cnt is decreased till %d\n",
2590                     tcp->cid, tcp->tcp_cached.initial_rcv_wnd, tcp->tcp_cached.rcv_indication_size, gen_buff_size, buffers_number);
2591     } else {
2592         DbgMessage(pdev, WARNl4rx | WARNl4sp,
2593                     "###lm_squeeze_rx_buffer_list cid=%d(%d,%d): could not replace %dB (%d bufs) into %d frags of %dB each\n",
2594                     tcp->cid, tcp->tcp_cached.initial_rcv_wnd, tcp->tcp_cached.rcv_indication_size,
2595                     gen_info->peninsula_nbytes, buffers_number, adjust_number, gen_buff_size);
2596     }
2597     return  buffers_number;
2598 }
2599 
lm_tcp_rx_clear_isles(struct _lm_device_t * pdev,lm_tcp_state_t * tcp_state,d_list_t * isles_list)2600 void lm_tcp_rx_clear_isles(struct _lm_device_t * pdev, lm_tcp_state_t * tcp_state, d_list_t * isles_list)
2601 {
2602     lm_tcp_con_rx_gen_info_t * gen_info;
2603     u8_t        isle_cnt;
2604 
2605     DbgBreakIf(!(tcp_state && tcp_state->rx_con));
2606     gen_info = &tcp_state->rx_con->u.rx.gen_info;
2607     while ((isle_cnt = (u8_t)d_list_entry_cnt(&gen_info->isles_list))) {
2608         d_list_t aux_isles_list;
2609         d_list_init(&aux_isles_list, NULL, NULL, 0);
2610         _lm_tcp_isle_remove(pdev, tcp_state, NON_EXISTENT_SB_IDX, isle_cnt, &aux_isles_list);
2611         if (!d_list_is_empty(&aux_isles_list)) {
2612         d_list_add_head(isles_list, &aux_isles_list);
2613     }
2614     }
2615     return;
2616 }
2617 
2618