1 /*
2  * Copyright (C) 2016-2018 Netronome Systems, Inc.
3  *
4  * This software is dual licensed under the GNU General License Version 2,
5  * June 1991 as shown in the file COPYING in the top-level directory of this
6  * source tree or the BSD 2-Clause License provided below.  You have the
7  * option to license this software under the complete terms of either license.
8  *
9  * The BSD 2-Clause License:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      1. Redistributions of source code must retain the above
16  *         copyright notice, this list of conditions and the following
17  *         disclaimer.
18  *
19  *      2. Redistributions in binary form must reproduce the above
20  *         copyright notice, this list of conditions and the following
21  *         disclaimer in the documentation and/or other materials
22  *         provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #ifndef __NFP_BPF_H__
35 #define __NFP_BPF_H__ 1
36 
37 #include <linux/bitfield.h>
38 #include <linux/bpf.h>
39 #include <linux/bpf_verifier.h>
40 #include <linux/kernel.h>
41 #include <linux/list.h>
42 #include <linux/rhashtable.h>
43 #include <linux/skbuff.h>
44 #include <linux/types.h>
45 #include <linux/wait.h>
46 
47 #include "../nfp_asm.h"
48 #include "fw.h"
49 
50 #define cmsg_warn(bpf, msg...)	nn_dp_warn(&(bpf)->app->ctrl->dp, msg)
51 
52 /* For relocation logic use up-most byte of branch instruction as scratch
53  * area.  Remember to clear this before sending instructions to HW!
54  */
55 #define OP_RELO_TYPE	0xff00000000000000ULL
56 
57 enum nfp_relo_type {
58 	RELO_NONE = 0,
59 	/* standard internal jumps */
60 	RELO_BR_REL,
61 	/* internal jumps to parts of the outro */
62 	RELO_BR_GO_OUT,
63 	RELO_BR_GO_ABORT,
64 	/* external jumps to fixed addresses */
65 	RELO_BR_NEXT_PKT,
66 	RELO_BR_HELPER,
67 	/* immediate relocation against load address */
68 	RELO_IMMED_REL,
69 };
70 
71 /* To make absolute relocated branches (branches other than RELO_BR_REL)
72  * distinguishable in user space dumps from normal jumps, add a large offset
73  * to them.
74  */
75 #define BR_OFF_RELO		15000
76 
77 enum static_regs {
78 	STATIC_REG_IMMA		= 20, /* Bank AB */
79 	STATIC_REG_IMM		= 21, /* Bank AB */
80 	STATIC_REG_STACK	= 22, /* Bank A */
81 	STATIC_REG_PKT_LEN	= 22, /* Bank B */
82 };
83 
84 enum pkt_vec {
85 	PKT_VEC_PKT_LEN		= 0,
86 	PKT_VEC_PKT_PTR		= 2,
87 	PKT_VEC_QSEL_SET	= 4,
88 	PKT_VEC_QSEL_VAL	= 6,
89 };
90 
91 #define PKT_VEL_QSEL_SET_BIT	4
92 
93 #define pv_len(np)	reg_lm(1, PKT_VEC_PKT_LEN)
94 #define pv_ctm_ptr(np)	reg_lm(1, PKT_VEC_PKT_PTR)
95 #define pv_qsel_set(np)	reg_lm(1, PKT_VEC_QSEL_SET)
96 #define pv_qsel_val(np)	reg_lm(1, PKT_VEC_QSEL_VAL)
97 
98 #define stack_reg(np)	reg_a(STATIC_REG_STACK)
99 #define stack_imm(np)	imm_b(np)
100 #define plen_reg(np)	reg_b(STATIC_REG_PKT_LEN)
101 #define pptr_reg(np)	pv_ctm_ptr(np)
102 #define imm_a(np)	reg_a(STATIC_REG_IMM)
103 #define imm_b(np)	reg_b(STATIC_REG_IMM)
104 #define imma_a(np)	reg_a(STATIC_REG_IMMA)
105 #define imma_b(np)	reg_b(STATIC_REG_IMMA)
106 #define imm_both(np)	reg_both(STATIC_REG_IMM)
107 
108 #define NFP_BPF_ABI_FLAGS	reg_imm(0)
109 #define   NFP_BPF_ABI_FLAG_MARK	1
110 
111 /**
112  * struct nfp_app_bpf - bpf app priv structure
113  * @app:		backpointer to the app
114  *
115  * @bpf_dev:		BPF offload device handle
116  *
117  * @tag_allocator:	bitmap of control message tags in use
118  * @tag_alloc_next:	next tag bit to allocate
119  * @tag_alloc_last:	next tag bit to be freed
120  *
121  * @cmsg_replies:	received cmsg replies waiting to be consumed
122  * @cmsg_wq:		work queue for waiting for cmsg replies
123  *
124  * @cmsg_key_sz:	size of key in cmsg element array
125  * @cmsg_val_sz:	size of value in cmsg element array
126  *
127  * @map_list:		list of offloaded maps
128  * @maps_in_use:	number of currently offloaded maps
129  * @map_elems_in_use:	number of elements allocated to offloaded maps
130  *
131  * @maps_neutral:	hash table of offload-neutral maps (on pointer)
132  *
133  * @abi_version:	global BPF ABI version
134  *
135  * @adjust_head:	adjust head capability
136  * @adjust_head.flags:		extra flags for adjust head
137  * @adjust_head.off_min:	minimal packet offset within buffer required
138  * @adjust_head.off_max:	maximum packet offset within buffer required
139  * @adjust_head.guaranteed_sub:	negative adjustment guaranteed possible
140  * @adjust_head.guaranteed_add:	positive adjustment guaranteed possible
141  *
142  * @maps:		map capability
143  * @maps.types:			supported map types
144  * @maps.max_maps:		max number of maps supported
145  * @maps.max_elems:		max number of entries in each map
146  * @maps.max_key_sz:		max size of map key
147  * @maps.max_val_sz:		max size of map value
148  * @maps.max_elem_sz:		max size of map entry (key + value)
149  *
150  * @helpers:		helper addressess for various calls
151  * @helpers.map_lookup:		map lookup helper address
152  * @helpers.map_update:		map update helper address
153  * @helpers.map_delete:		map delete helper address
154  * @helpers.perf_event_output:	output perf event to a ring buffer
155  *
156  * @pseudo_random:	FW initialized the pseudo-random machinery (CSRs)
157  * @queue_select:	BPF can set the RX queue ID in packet vector
158  * @adjust_tail:	BPF can simply trunc packet size for adjust tail
159  */
160 struct nfp_app_bpf {
161 	struct nfp_app *app;
162 
163 	struct bpf_offload_dev *bpf_dev;
164 
165 	DECLARE_BITMAP(tag_allocator, U16_MAX + 1);
166 	u16 tag_alloc_next;
167 	u16 tag_alloc_last;
168 
169 	struct sk_buff_head cmsg_replies;
170 	struct wait_queue_head cmsg_wq;
171 
172 	unsigned int cmsg_key_sz;
173 	unsigned int cmsg_val_sz;
174 
175 	struct list_head map_list;
176 	unsigned int maps_in_use;
177 	unsigned int map_elems_in_use;
178 
179 	struct rhashtable maps_neutral;
180 
181 	u32 abi_version;
182 
183 	struct nfp_bpf_cap_adjust_head {
184 		u32 flags;
185 		int off_min;
186 		int off_max;
187 		int guaranteed_sub;
188 		int guaranteed_add;
189 	} adjust_head;
190 
191 	struct {
192 		u32 types;
193 		u32 max_maps;
194 		u32 max_elems;
195 		u32 max_key_sz;
196 		u32 max_val_sz;
197 		u32 max_elem_sz;
198 	} maps;
199 
200 	struct {
201 		u32 map_lookup;
202 		u32 map_update;
203 		u32 map_delete;
204 		u32 perf_event_output;
205 	} helpers;
206 
207 	bool pseudo_random;
208 	bool queue_select;
209 	bool adjust_tail;
210 };
211 
212 enum nfp_bpf_map_use {
213 	NFP_MAP_UNUSED = 0,
214 	NFP_MAP_USE_READ,
215 	NFP_MAP_USE_WRITE,
216 	NFP_MAP_USE_ATOMIC_CNT,
217 };
218 
219 /**
220  * struct nfp_bpf_map - private per-map data attached to BPF maps for offload
221  * @offmap:	pointer to the offloaded BPF map
222  * @bpf:	back pointer to bpf app private structure
223  * @tid:	table id identifying map on datapath
224  * @l:		link on the nfp_app_bpf->map_list list
225  * @use_map:	map of how the value is used (in 4B chunks)
226  */
227 struct nfp_bpf_map {
228 	struct bpf_offloaded_map *offmap;
229 	struct nfp_app_bpf *bpf;
230 	u32 tid;
231 	struct list_head l;
232 	enum nfp_bpf_map_use use_map[];
233 };
234 
235 struct nfp_bpf_neutral_map {
236 	struct rhash_head l;
237 	struct bpf_map *ptr;
238 	u32 map_id;
239 	u32 count;
240 };
241 
242 extern const struct rhashtable_params nfp_bpf_maps_neutral_params;
243 
244 struct nfp_prog;
245 struct nfp_insn_meta;
246 typedef int (*instr_cb_t)(struct nfp_prog *, struct nfp_insn_meta *);
247 
248 #define nfp_prog_first_meta(nfp_prog)					\
249 	list_first_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
250 #define nfp_prog_last_meta(nfp_prog)					\
251 	list_last_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
252 #define nfp_meta_next(meta)	list_next_entry(meta, l)
253 #define nfp_meta_prev(meta)	list_prev_entry(meta, l)
254 
255 /**
256  * struct nfp_bpf_reg_state - register state for calls
257  * @reg: BPF register state from latest path
258  * @var_off: for stack arg - changes stack offset on different paths
259  */
260 struct nfp_bpf_reg_state {
261 	struct bpf_reg_state reg;
262 	bool var_off;
263 };
264 
265 #define FLAG_INSN_IS_JUMP_DST			BIT(0)
266 #define FLAG_INSN_IS_SUBPROG_START		BIT(1)
267 
268 /**
269  * struct nfp_insn_meta - BPF instruction wrapper
270  * @insn: BPF instruction
271  * @ptr: pointer type for memory operations
272  * @ldst_gather_len: memcpy length gathered from load/store sequence
273  * @paired_st: the paired store insn at the head of the sequence
274  * @ptr_not_const: pointer is not always constant
275  * @pkt_cache: packet data cache information
276  * @pkt_cache.range_start: start offset for associated packet data cache
277  * @pkt_cache.range_end: end offset for associated packet data cache
278  * @pkt_cache.do_init: this read needs to initialize packet data cache
279  * @xadd_over_16bit: 16bit immediate is not guaranteed
280  * @xadd_maybe_16bit: 16bit immediate is possible
281  * @jmp_dst: destination info for jump instructions
282  * @jump_neg_op: jump instruction has inverted immediate, use ADD instead of SUB
283  * @func_id: function id for call instructions
284  * @arg1: arg1 for call instructions
285  * @arg2: arg2 for call instructions
286  * @umin_src: copy of core verifier umin_value for src opearnd.
287  * @umax_src: copy of core verifier umax_value for src operand.
288  * @umin_dst: copy of core verifier umin_value for dst opearnd.
289  * @umax_dst: copy of core verifier umax_value for dst operand.
290  * @off: index of first generated machine instruction (in nfp_prog.prog)
291  * @n: eBPF instruction number
292  * @flags: eBPF instruction extra optimization flags
293  * @skip: skip this instruction (optimized out)
294  * @double_cb: callback for second part of the instruction
295  * @l: link on nfp_prog->insns list
296  */
297 struct nfp_insn_meta {
298 	struct bpf_insn insn;
299 	union {
300 		/* pointer ops (ld/st/xadd) */
301 		struct {
302 			struct bpf_reg_state ptr;
303 			struct bpf_insn *paired_st;
304 			s16 ldst_gather_len;
305 			bool ptr_not_const;
306 			struct {
307 				s16 range_start;
308 				s16 range_end;
309 				bool do_init;
310 			} pkt_cache;
311 			bool xadd_over_16bit;
312 			bool xadd_maybe_16bit;
313 		};
314 		/* jump */
315 		struct {
316 			struct nfp_insn_meta *jmp_dst;
317 			bool jump_neg_op;
318 		};
319 		/* function calls */
320 		struct {
321 			u32 func_id;
322 			struct bpf_reg_state arg1;
323 			struct nfp_bpf_reg_state arg2;
324 		};
325 		/* We are interested in range info for operands of ALU
326 		 * operations. For example, shift amount, multiplicand and
327 		 * multiplier etc.
328 		 */
329 		struct {
330 			u64 umin_src;
331 			u64 umax_src;
332 			u64 umin_dst;
333 			u64 umax_dst;
334 		};
335 	};
336 	unsigned int off;
337 	unsigned short n;
338 	unsigned short flags;
339 	bool skip;
340 	instr_cb_t double_cb;
341 
342 	struct list_head l;
343 };
344 
345 #define BPF_SIZE_MASK	0x18
346 
347 static inline u8 mbpf_class(const struct nfp_insn_meta *meta)
348 {
349 	return BPF_CLASS(meta->insn.code);
350 }
351 
352 static inline u8 mbpf_src(const struct nfp_insn_meta *meta)
353 {
354 	return BPF_SRC(meta->insn.code);
355 }
356 
357 static inline u8 mbpf_op(const struct nfp_insn_meta *meta)
358 {
359 	return BPF_OP(meta->insn.code);
360 }
361 
362 static inline u8 mbpf_mode(const struct nfp_insn_meta *meta)
363 {
364 	return BPF_MODE(meta->insn.code);
365 }
366 
367 static inline bool is_mbpf_alu(const struct nfp_insn_meta *meta)
368 {
369 	return mbpf_class(meta) == BPF_ALU64 || mbpf_class(meta) == BPF_ALU;
370 }
371 
372 static inline bool is_mbpf_load(const struct nfp_insn_meta *meta)
373 {
374 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_LDX | BPF_MEM);
375 }
376 
377 static inline bool is_mbpf_store(const struct nfp_insn_meta *meta)
378 {
379 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_STX | BPF_MEM);
380 }
381 
382 static inline bool is_mbpf_load_pkt(const struct nfp_insn_meta *meta)
383 {
384 	return is_mbpf_load(meta) && meta->ptr.type == PTR_TO_PACKET;
385 }
386 
387 static inline bool is_mbpf_store_pkt(const struct nfp_insn_meta *meta)
388 {
389 	return is_mbpf_store(meta) && meta->ptr.type == PTR_TO_PACKET;
390 }
391 
392 static inline bool is_mbpf_classic_load(const struct nfp_insn_meta *meta)
393 {
394 	u8 code = meta->insn.code;
395 
396 	return BPF_CLASS(code) == BPF_LD &&
397 	       (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND);
398 }
399 
400 static inline bool is_mbpf_classic_store(const struct nfp_insn_meta *meta)
401 {
402 	u8 code = meta->insn.code;
403 
404 	return BPF_CLASS(code) == BPF_ST && BPF_MODE(code) == BPF_MEM;
405 }
406 
407 static inline bool is_mbpf_classic_store_pkt(const struct nfp_insn_meta *meta)
408 {
409 	return is_mbpf_classic_store(meta) && meta->ptr.type == PTR_TO_PACKET;
410 }
411 
412 static inline bool is_mbpf_xadd(const struct nfp_insn_meta *meta)
413 {
414 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_STX | BPF_XADD);
415 }
416 
417 static inline bool is_mbpf_mul(const struct nfp_insn_meta *meta)
418 {
419 	return is_mbpf_alu(meta) && mbpf_op(meta) == BPF_MUL;
420 }
421 
422 static inline bool is_mbpf_div(const struct nfp_insn_meta *meta)
423 {
424 	return is_mbpf_alu(meta) && mbpf_op(meta) == BPF_DIV;
425 }
426 
427 static inline bool is_mbpf_helper_call(const struct nfp_insn_meta *meta)
428 {
429 	struct bpf_insn insn = meta->insn;
430 
431 	return insn.code == (BPF_JMP | BPF_CALL) &&
432 		insn.src_reg != BPF_PSEUDO_CALL;
433 }
434 
435 /**
436  * struct nfp_bpf_subprog_info - nfp BPF sub-program (a.k.a. function) info
437  * @stack_depth:	maximum stack depth used by this sub-program
438  */
439 struct nfp_bpf_subprog_info {
440 	u16 stack_depth;
441 };
442 
443 /**
444  * struct nfp_prog - nfp BPF program
445  * @bpf: backpointer to the bpf app priv structure
446  * @prog: machine code
447  * @prog_len: number of valid instructions in @prog array
448  * @__prog_alloc_len: alloc size of @prog array
449  * @verifier_meta: temporary storage for verifier's insn meta
450  * @type: BPF program type
451  * @last_bpf_off: address of the last instruction translated from BPF
452  * @tgt_out: jump target for normal exit
453  * @tgt_abort: jump target for abort (e.g. access outside of packet buffer)
454  * @n_translated: number of successfully translated instructions (for errors)
455  * @error: error code if something went wrong
456  * @stack_frame_depth: max stack depth for current frame
457  * @adjust_head_location: if program has single adjust head call - the insn no.
458  * @map_records_cnt: the number of map pointers recorded for this prog
459  * @subprog_cnt: number of sub-programs, including main function
460  * @map_records: the map record pointers from bpf->maps_neutral
461  * @subprog: pointer to an array of objects holding info about sub-programs
462  * @insns: list of BPF instruction wrappers (struct nfp_insn_meta)
463  */
464 struct nfp_prog {
465 	struct nfp_app_bpf *bpf;
466 
467 	u64 *prog;
468 	unsigned int prog_len;
469 	unsigned int __prog_alloc_len;
470 
471 	struct nfp_insn_meta *verifier_meta;
472 
473 	enum bpf_prog_type type;
474 
475 	unsigned int last_bpf_off;
476 	unsigned int tgt_out;
477 	unsigned int tgt_abort;
478 
479 	unsigned int n_translated;
480 	int error;
481 
482 	unsigned int stack_frame_depth;
483 	unsigned int adjust_head_location;
484 
485 	unsigned int map_records_cnt;
486 	unsigned int subprog_cnt;
487 	struct nfp_bpf_neutral_map **map_records;
488 	struct nfp_bpf_subprog_info *subprog;
489 
490 	struct list_head insns;
491 };
492 
493 /**
494  * struct nfp_bpf_vnic - per-vNIC BPF priv structure
495  * @tc_prog:	currently loaded cls_bpf program
496  * @start_off:	address of the first instruction in the memory
497  * @tgt_done:	jump target to get the next packet
498  */
499 struct nfp_bpf_vnic {
500 	struct bpf_prog *tc_prog;
501 	unsigned int start_off;
502 	unsigned int tgt_done;
503 };
504 
505 void nfp_bpf_jit_prepare(struct nfp_prog *nfp_prog, unsigned int cnt);
506 int nfp_bpf_jit(struct nfp_prog *prog);
507 bool nfp_bpf_supported_opcode(u8 code);
508 
509 extern const struct bpf_prog_offload_ops nfp_bpf_analyzer_ops;
510 
511 struct netdev_bpf;
512 struct nfp_app;
513 struct nfp_net;
514 
515 int nfp_ndo_bpf(struct nfp_app *app, struct nfp_net *nn,
516 		struct netdev_bpf *bpf);
517 int nfp_net_bpf_offload(struct nfp_net *nn, struct bpf_prog *prog,
518 			bool old_prog, struct netlink_ext_ack *extack);
519 
520 struct nfp_insn_meta *
521 nfp_bpf_goto_meta(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
522 		  unsigned int insn_idx, unsigned int n_insns);
523 
524 void *nfp_bpf_relo_for_vnic(struct nfp_prog *nfp_prog, struct nfp_bpf_vnic *bv);
525 
526 unsigned int nfp_bpf_ctrl_cmsg_mtu(struct nfp_app_bpf *bpf);
527 long long int
528 nfp_bpf_ctrl_alloc_map(struct nfp_app_bpf *bpf, struct bpf_map *map);
529 void
530 nfp_bpf_ctrl_free_map(struct nfp_app_bpf *bpf, struct nfp_bpf_map *nfp_map);
531 int nfp_bpf_ctrl_getfirst_entry(struct bpf_offloaded_map *offmap,
532 				void *next_key);
533 int nfp_bpf_ctrl_update_entry(struct bpf_offloaded_map *offmap,
534 			      void *key, void *value, u64 flags);
535 int nfp_bpf_ctrl_del_entry(struct bpf_offloaded_map *offmap, void *key);
536 int nfp_bpf_ctrl_lookup_entry(struct bpf_offloaded_map *offmap,
537 			      void *key, void *value);
538 int nfp_bpf_ctrl_getnext_entry(struct bpf_offloaded_map *offmap,
539 			       void *key, void *next_key);
540 
541 int nfp_bpf_event_output(struct nfp_app_bpf *bpf, const void *data,
542 			 unsigned int len);
543 
544 void nfp_bpf_ctrl_msg_rx(struct nfp_app *app, struct sk_buff *skb);
545 void
546 nfp_bpf_ctrl_msg_rx_raw(struct nfp_app *app, const void *data,
547 			unsigned int len);
548 #endif
549