1 /* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
2 /* Copyright (C) 2016-2018 Netronome Systems, Inc. */
3 
4 #ifndef __NFP_BPF_H__
5 #define __NFP_BPF_H__ 1
6 
7 #include <linux/bitfield.h>
8 #include <linux/bpf.h>
9 #include <linux/bpf_verifier.h>
10 #include <linux/kernel.h>
11 #include <linux/list.h>
12 #include <linux/rhashtable.h>
13 #include <linux/skbuff.h>
14 #include <linux/types.h>
15 #include <linux/wait.h>
16 
17 #include "../nfp_asm.h"
18 #include "fw.h"
19 
20 #define cmsg_warn(bpf, msg...)	nn_dp_warn(&(bpf)->app->ctrl->dp, msg)
21 
22 /* For relocation logic use up-most byte of branch instruction as scratch
23  * area.  Remember to clear this before sending instructions to HW!
24  */
25 #define OP_RELO_TYPE	0xff00000000000000ULL
26 
27 enum nfp_relo_type {
28 	RELO_NONE = 0,
29 	/* standard internal jumps */
30 	RELO_BR_REL,
31 	/* internal jumps to parts of the outro */
32 	RELO_BR_GO_OUT,
33 	RELO_BR_GO_ABORT,
34 	RELO_BR_GO_CALL_PUSH_REGS,
35 	RELO_BR_GO_CALL_POP_REGS,
36 	/* external jumps to fixed addresses */
37 	RELO_BR_NEXT_PKT,
38 	RELO_BR_HELPER,
39 	/* immediate relocation against load address */
40 	RELO_IMMED_REL,
41 };
42 
43 /* To make absolute relocated branches (branches other than RELO_BR_REL)
44  * distinguishable in user space dumps from normal jumps, add a large offset
45  * to them.
46  */
47 #define BR_OFF_RELO		15000
48 
49 enum static_regs {
50 	STATIC_REG_IMMA		= 20, /* Bank AB */
51 	STATIC_REG_IMM		= 21, /* Bank AB */
52 	STATIC_REG_STACK	= 22, /* Bank A */
53 	STATIC_REG_PKT_LEN	= 22, /* Bank B */
54 };
55 
56 enum pkt_vec {
57 	PKT_VEC_PKT_LEN		= 0,
58 	PKT_VEC_PKT_PTR		= 2,
59 	PKT_VEC_QSEL_SET	= 4,
60 	PKT_VEC_QSEL_VAL	= 6,
61 };
62 
63 #define PKT_VEL_QSEL_SET_BIT	4
64 
65 #define pv_len(np)	reg_lm(1, PKT_VEC_PKT_LEN)
66 #define pv_ctm_ptr(np)	reg_lm(1, PKT_VEC_PKT_PTR)
67 #define pv_qsel_set(np)	reg_lm(1, PKT_VEC_QSEL_SET)
68 #define pv_qsel_val(np)	reg_lm(1, PKT_VEC_QSEL_VAL)
69 
70 #define stack_reg(np)	reg_a(STATIC_REG_STACK)
71 #define stack_imm(np)	imm_b(np)
72 #define plen_reg(np)	reg_b(STATIC_REG_PKT_LEN)
73 #define pptr_reg(np)	pv_ctm_ptr(np)
74 #define imm_a(np)	reg_a(STATIC_REG_IMM)
75 #define imm_b(np)	reg_b(STATIC_REG_IMM)
76 #define imma_a(np)	reg_a(STATIC_REG_IMMA)
77 #define imma_b(np)	reg_b(STATIC_REG_IMMA)
78 #define imm_both(np)	reg_both(STATIC_REG_IMM)
79 #define ret_reg(np)	imm_a(np)
80 
81 #define NFP_BPF_ABI_FLAGS	reg_imm(0)
82 #define   NFP_BPF_ABI_FLAG_MARK	1
83 
84 /**
85  * struct nfp_app_bpf - bpf app priv structure
86  * @app:		backpointer to the app
87  *
88  * @bpf_dev:		BPF offload device handle
89  *
90  * @tag_allocator:	bitmap of control message tags in use
91  * @tag_alloc_next:	next tag bit to allocate
92  * @tag_alloc_last:	next tag bit to be freed
93  *
94  * @cmsg_replies:	received cmsg replies waiting to be consumed
95  * @cmsg_wq:		work queue for waiting for cmsg replies
96  *
97  * @cmsg_key_sz:	size of key in cmsg element array
98  * @cmsg_val_sz:	size of value in cmsg element array
99  *
100  * @map_list:		list of offloaded maps
101  * @maps_in_use:	number of currently offloaded maps
102  * @map_elems_in_use:	number of elements allocated to offloaded maps
103  *
104  * @maps_neutral:	hash table of offload-neutral maps (on pointer)
105  *
106  * @abi_version:	global BPF ABI version
107  *
108  * @adjust_head:	adjust head capability
109  * @adjust_head.flags:		extra flags for adjust head
110  * @adjust_head.off_min:	minimal packet offset within buffer required
111  * @adjust_head.off_max:	maximum packet offset within buffer required
112  * @adjust_head.guaranteed_sub:	negative adjustment guaranteed possible
113  * @adjust_head.guaranteed_add:	positive adjustment guaranteed possible
114  *
115  * @maps:		map capability
116  * @maps.types:			supported map types
117  * @maps.max_maps:		max number of maps supported
118  * @maps.max_elems:		max number of entries in each map
119  * @maps.max_key_sz:		max size of map key
120  * @maps.max_val_sz:		max size of map value
121  * @maps.max_elem_sz:		max size of map entry (key + value)
122  *
123  * @helpers:		helper addressess for various calls
124  * @helpers.map_lookup:		map lookup helper address
125  * @helpers.map_update:		map update helper address
126  * @helpers.map_delete:		map delete helper address
127  * @helpers.perf_event_output:	output perf event to a ring buffer
128  *
129  * @pseudo_random:	FW initialized the pseudo-random machinery (CSRs)
130  * @queue_select:	BPF can set the RX queue ID in packet vector
131  * @adjust_tail:	BPF can simply trunc packet size for adjust tail
132  */
133 struct nfp_app_bpf {
134 	struct nfp_app *app;
135 
136 	struct bpf_offload_dev *bpf_dev;
137 
138 	DECLARE_BITMAP(tag_allocator, U16_MAX + 1);
139 	u16 tag_alloc_next;
140 	u16 tag_alloc_last;
141 
142 	struct sk_buff_head cmsg_replies;
143 	struct wait_queue_head cmsg_wq;
144 
145 	unsigned int cmsg_key_sz;
146 	unsigned int cmsg_val_sz;
147 
148 	struct list_head map_list;
149 	unsigned int maps_in_use;
150 	unsigned int map_elems_in_use;
151 
152 	struct rhashtable maps_neutral;
153 
154 	u32 abi_version;
155 
156 	struct nfp_bpf_cap_adjust_head {
157 		u32 flags;
158 		int off_min;
159 		int off_max;
160 		int guaranteed_sub;
161 		int guaranteed_add;
162 	} adjust_head;
163 
164 	struct {
165 		u32 types;
166 		u32 max_maps;
167 		u32 max_elems;
168 		u32 max_key_sz;
169 		u32 max_val_sz;
170 		u32 max_elem_sz;
171 	} maps;
172 
173 	struct {
174 		u32 map_lookup;
175 		u32 map_update;
176 		u32 map_delete;
177 		u32 perf_event_output;
178 	} helpers;
179 
180 	bool pseudo_random;
181 	bool queue_select;
182 	bool adjust_tail;
183 };
184 
185 enum nfp_bpf_map_use {
186 	NFP_MAP_UNUSED = 0,
187 	NFP_MAP_USE_READ,
188 	NFP_MAP_USE_WRITE,
189 	NFP_MAP_USE_ATOMIC_CNT,
190 };
191 
192 struct nfp_bpf_map_word {
193 	unsigned char type		:4;
194 	unsigned char non_zero_update	:1;
195 };
196 
197 /**
198  * struct nfp_bpf_map - private per-map data attached to BPF maps for offload
199  * @offmap:	pointer to the offloaded BPF map
200  * @bpf:	back pointer to bpf app private structure
201  * @tid:	table id identifying map on datapath
202  * @l:		link on the nfp_app_bpf->map_list list
203  * @use_map:	map of how the value is used (in 4B chunks)
204  */
205 struct nfp_bpf_map {
206 	struct bpf_offloaded_map *offmap;
207 	struct nfp_app_bpf *bpf;
208 	u32 tid;
209 	struct list_head l;
210 	struct nfp_bpf_map_word use_map[];
211 };
212 
213 struct nfp_bpf_neutral_map {
214 	struct rhash_head l;
215 	struct bpf_map *ptr;
216 	u32 map_id;
217 	u32 count;
218 };
219 
220 extern const struct rhashtable_params nfp_bpf_maps_neutral_params;
221 
222 struct nfp_prog;
223 struct nfp_insn_meta;
224 typedef int (*instr_cb_t)(struct nfp_prog *, struct nfp_insn_meta *);
225 
226 #define nfp_prog_first_meta(nfp_prog)					\
227 	list_first_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
228 #define nfp_prog_last_meta(nfp_prog)					\
229 	list_last_entry(&(nfp_prog)->insns, struct nfp_insn_meta, l)
230 #define nfp_meta_next(meta)	list_next_entry(meta, l)
231 #define nfp_meta_prev(meta)	list_prev_entry(meta, l)
232 
233 /**
234  * struct nfp_bpf_reg_state - register state for calls
235  * @reg: BPF register state from latest path
236  * @var_off: for stack arg - changes stack offset on different paths
237  */
238 struct nfp_bpf_reg_state {
239 	struct bpf_reg_state reg;
240 	bool var_off;
241 };
242 
243 #define FLAG_INSN_IS_JUMP_DST			BIT(0)
244 #define FLAG_INSN_IS_SUBPROG_START		BIT(1)
245 #define FLAG_INSN_PTR_CALLER_STACK_FRAME	BIT(2)
246 /* Instruction is pointless, noop even on its own */
247 #define FLAG_INSN_SKIP_NOOP			BIT(3)
248 /* Instruction is optimized out based on preceding instructions */
249 #define FLAG_INSN_SKIP_PREC_DEPENDENT		BIT(4)
250 /* Instruction is optimized by the verifier */
251 #define FLAG_INSN_SKIP_VERIFIER_OPT		BIT(5)
252 
253 #define FLAG_INSN_SKIP_MASK		(FLAG_INSN_SKIP_NOOP | \
254 					 FLAG_INSN_SKIP_PREC_DEPENDENT | \
255 					 FLAG_INSN_SKIP_VERIFIER_OPT)
256 
257 /**
258  * struct nfp_insn_meta - BPF instruction wrapper
259  * @insn: BPF instruction
260  * @ptr: pointer type for memory operations
261  * @ldst_gather_len: memcpy length gathered from load/store sequence
262  * @paired_st: the paired store insn at the head of the sequence
263  * @ptr_not_const: pointer is not always constant
264  * @pkt_cache: packet data cache information
265  * @pkt_cache.range_start: start offset for associated packet data cache
266  * @pkt_cache.range_end: end offset for associated packet data cache
267  * @pkt_cache.do_init: this read needs to initialize packet data cache
268  * @xadd_over_16bit: 16bit immediate is not guaranteed
269  * @xadd_maybe_16bit: 16bit immediate is possible
270  * @jmp_dst: destination info for jump instructions
271  * @jump_neg_op: jump instruction has inverted immediate, use ADD instead of SUB
272  * @num_insns_after_br: number of insns following a branch jump, used for fixup
273  * @func_id: function id for call instructions
274  * @arg1: arg1 for call instructions
275  * @arg2: arg2 for call instructions
276  * @umin_src: copy of core verifier umin_value for src opearnd.
277  * @umax_src: copy of core verifier umax_value for src operand.
278  * @umin_dst: copy of core verifier umin_value for dst opearnd.
279  * @umax_dst: copy of core verifier umax_value for dst operand.
280  * @off: index of first generated machine instruction (in nfp_prog.prog)
281  * @n: eBPF instruction number
282  * @flags: eBPF instruction extra optimization flags
283  * @subprog_idx: index of subprogram to which the instruction belongs
284  * @double_cb: callback for second part of the instruction
285  * @l: link on nfp_prog->insns list
286  */
287 struct nfp_insn_meta {
288 	struct bpf_insn insn;
289 	union {
290 		/* pointer ops (ld/st/xadd) */
291 		struct {
292 			struct bpf_reg_state ptr;
293 			struct bpf_insn *paired_st;
294 			s16 ldst_gather_len;
295 			bool ptr_not_const;
296 			struct {
297 				s16 range_start;
298 				s16 range_end;
299 				bool do_init;
300 			} pkt_cache;
301 			bool xadd_over_16bit;
302 			bool xadd_maybe_16bit;
303 		};
304 		/* jump */
305 		struct {
306 			struct nfp_insn_meta *jmp_dst;
307 			bool jump_neg_op;
308 			u32 num_insns_after_br; /* only for BPF-to-BPF calls */
309 		};
310 		/* function calls */
311 		struct {
312 			u32 func_id;
313 			struct bpf_reg_state arg1;
314 			struct nfp_bpf_reg_state arg2;
315 		};
316 		/* We are interested in range info for operands of ALU
317 		 * operations. For example, shift amount, multiplicand and
318 		 * multiplier etc.
319 		 */
320 		struct {
321 			u64 umin_src;
322 			u64 umax_src;
323 			u64 umin_dst;
324 			u64 umax_dst;
325 		};
326 	};
327 	unsigned int off;
328 	unsigned short n;
329 	unsigned short flags;
330 	unsigned short subprog_idx;
331 	instr_cb_t double_cb;
332 
333 	struct list_head l;
334 };
335 
336 #define BPF_SIZE_MASK	0x18
337 
338 static inline u8 mbpf_class(const struct nfp_insn_meta *meta)
339 {
340 	return BPF_CLASS(meta->insn.code);
341 }
342 
343 static inline u8 mbpf_src(const struct nfp_insn_meta *meta)
344 {
345 	return BPF_SRC(meta->insn.code);
346 }
347 
348 static inline u8 mbpf_op(const struct nfp_insn_meta *meta)
349 {
350 	return BPF_OP(meta->insn.code);
351 }
352 
353 static inline u8 mbpf_mode(const struct nfp_insn_meta *meta)
354 {
355 	return BPF_MODE(meta->insn.code);
356 }
357 
358 static inline bool is_mbpf_alu(const struct nfp_insn_meta *meta)
359 {
360 	return mbpf_class(meta) == BPF_ALU64 || mbpf_class(meta) == BPF_ALU;
361 }
362 
363 static inline bool is_mbpf_load(const struct nfp_insn_meta *meta)
364 {
365 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_LDX | BPF_MEM);
366 }
367 
368 static inline bool is_mbpf_jmp32(const struct nfp_insn_meta *meta)
369 {
370 	return mbpf_class(meta) == BPF_JMP32;
371 }
372 
373 static inline bool is_mbpf_jmp64(const struct nfp_insn_meta *meta)
374 {
375 	return mbpf_class(meta) == BPF_JMP;
376 }
377 
378 static inline bool is_mbpf_jmp(const struct nfp_insn_meta *meta)
379 {
380 	return is_mbpf_jmp32(meta) || is_mbpf_jmp64(meta);
381 }
382 
383 static inline bool is_mbpf_store(const struct nfp_insn_meta *meta)
384 {
385 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_STX | BPF_MEM);
386 }
387 
388 static inline bool is_mbpf_load_pkt(const struct nfp_insn_meta *meta)
389 {
390 	return is_mbpf_load(meta) && meta->ptr.type == PTR_TO_PACKET;
391 }
392 
393 static inline bool is_mbpf_store_pkt(const struct nfp_insn_meta *meta)
394 {
395 	return is_mbpf_store(meta) && meta->ptr.type == PTR_TO_PACKET;
396 }
397 
398 static inline bool is_mbpf_classic_load(const struct nfp_insn_meta *meta)
399 {
400 	u8 code = meta->insn.code;
401 
402 	return BPF_CLASS(code) == BPF_LD &&
403 	       (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND);
404 }
405 
406 static inline bool is_mbpf_classic_store(const struct nfp_insn_meta *meta)
407 {
408 	u8 code = meta->insn.code;
409 
410 	return BPF_CLASS(code) == BPF_ST && BPF_MODE(code) == BPF_MEM;
411 }
412 
413 static inline bool is_mbpf_classic_store_pkt(const struct nfp_insn_meta *meta)
414 {
415 	return is_mbpf_classic_store(meta) && meta->ptr.type == PTR_TO_PACKET;
416 }
417 
418 static inline bool is_mbpf_xadd(const struct nfp_insn_meta *meta)
419 {
420 	return (meta->insn.code & ~BPF_SIZE_MASK) == (BPF_STX | BPF_XADD);
421 }
422 
423 static inline bool is_mbpf_mul(const struct nfp_insn_meta *meta)
424 {
425 	return is_mbpf_alu(meta) && mbpf_op(meta) == BPF_MUL;
426 }
427 
428 static inline bool is_mbpf_div(const struct nfp_insn_meta *meta)
429 {
430 	return is_mbpf_alu(meta) && mbpf_op(meta) == BPF_DIV;
431 }
432 
433 static inline bool is_mbpf_cond_jump(const struct nfp_insn_meta *meta)
434 {
435 	u8 op;
436 
437 	if (is_mbpf_jmp32(meta))
438 		return true;
439 
440 	if (!is_mbpf_jmp64(meta))
441 		return false;
442 
443 	op = mbpf_op(meta);
444 	return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL;
445 }
446 
447 static inline bool is_mbpf_helper_call(const struct nfp_insn_meta *meta)
448 {
449 	struct bpf_insn insn = meta->insn;
450 
451 	return insn.code == (BPF_JMP | BPF_CALL) &&
452 		insn.src_reg != BPF_PSEUDO_CALL;
453 }
454 
455 static inline bool is_mbpf_pseudo_call(const struct nfp_insn_meta *meta)
456 {
457 	struct bpf_insn insn = meta->insn;
458 
459 	return insn.code == (BPF_JMP | BPF_CALL) &&
460 		insn.src_reg == BPF_PSEUDO_CALL;
461 }
462 
463 #define STACK_FRAME_ALIGN 64
464 
465 /**
466  * struct nfp_bpf_subprog_info - nfp BPF sub-program (a.k.a. function) info
467  * @stack_depth:	maximum stack depth used by this sub-program
468  * @needs_reg_push:	whether sub-program uses callee-saved registers
469  */
470 struct nfp_bpf_subprog_info {
471 	u16 stack_depth;
472 	u8 needs_reg_push : 1;
473 };
474 
475 /**
476  * struct nfp_prog - nfp BPF program
477  * @bpf: backpointer to the bpf app priv structure
478  * @prog: machine code
479  * @prog_len: number of valid instructions in @prog array
480  * @__prog_alloc_len: alloc size of @prog array
481  * @stack_size: total amount of stack used
482  * @verifier_meta: temporary storage for verifier's insn meta
483  * @type: BPF program type
484  * @last_bpf_off: address of the last instruction translated from BPF
485  * @tgt_out: jump target for normal exit
486  * @tgt_abort: jump target for abort (e.g. access outside of packet buffer)
487  * @tgt_call_push_regs: jump target for subroutine for saving R6~R9 to stack
488  * @tgt_call_pop_regs: jump target for subroutine used for restoring R6~R9
489  * @n_translated: number of successfully translated instructions (for errors)
490  * @error: error code if something went wrong
491  * @stack_frame_depth: max stack depth for current frame
492  * @adjust_head_location: if program has single adjust head call - the insn no.
493  * @map_records_cnt: the number of map pointers recorded for this prog
494  * @subprog_cnt: number of sub-programs, including main function
495  * @map_records: the map record pointers from bpf->maps_neutral
496  * @subprog: pointer to an array of objects holding info about sub-programs
497  * @n_insns: number of instructions on @insns list
498  * @insns: list of BPF instruction wrappers (struct nfp_insn_meta)
499  */
500 struct nfp_prog {
501 	struct nfp_app_bpf *bpf;
502 
503 	u64 *prog;
504 	unsigned int prog_len;
505 	unsigned int __prog_alloc_len;
506 
507 	unsigned int stack_size;
508 
509 	struct nfp_insn_meta *verifier_meta;
510 
511 	enum bpf_prog_type type;
512 
513 	unsigned int last_bpf_off;
514 	unsigned int tgt_out;
515 	unsigned int tgt_abort;
516 	unsigned int tgt_call_push_regs;
517 	unsigned int tgt_call_pop_regs;
518 
519 	unsigned int n_translated;
520 	int error;
521 
522 	unsigned int stack_frame_depth;
523 	unsigned int adjust_head_location;
524 
525 	unsigned int map_records_cnt;
526 	unsigned int subprog_cnt;
527 	struct nfp_bpf_neutral_map **map_records;
528 	struct nfp_bpf_subprog_info *subprog;
529 
530 	unsigned int n_insns;
531 	struct list_head insns;
532 };
533 
534 /**
535  * struct nfp_bpf_vnic - per-vNIC BPF priv structure
536  * @tc_prog:	currently loaded cls_bpf program
537  * @start_off:	address of the first instruction in the memory
538  * @tgt_done:	jump target to get the next packet
539  */
540 struct nfp_bpf_vnic {
541 	struct bpf_prog *tc_prog;
542 	unsigned int start_off;
543 	unsigned int tgt_done;
544 };
545 
546 bool nfp_is_subprog_start(struct nfp_insn_meta *meta);
547 void nfp_bpf_jit_prepare(struct nfp_prog *nfp_prog);
548 int nfp_bpf_jit(struct nfp_prog *prog);
549 bool nfp_bpf_supported_opcode(u8 code);
550 
551 int nfp_verify_insn(struct bpf_verifier_env *env, int insn_idx,
552 		    int prev_insn_idx);
553 int nfp_bpf_finalize(struct bpf_verifier_env *env);
554 
555 int nfp_bpf_opt_replace_insn(struct bpf_verifier_env *env, u32 off,
556 			     struct bpf_insn *insn);
557 int nfp_bpf_opt_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
558 
559 extern const struct bpf_prog_offload_ops nfp_bpf_dev_ops;
560 
561 struct netdev_bpf;
562 struct nfp_app;
563 struct nfp_net;
564 
565 int nfp_ndo_bpf(struct nfp_app *app, struct nfp_net *nn,
566 		struct netdev_bpf *bpf);
567 int nfp_net_bpf_offload(struct nfp_net *nn, struct bpf_prog *prog,
568 			bool old_prog, struct netlink_ext_ack *extack);
569 
570 struct nfp_insn_meta *
571 nfp_bpf_goto_meta(struct nfp_prog *nfp_prog, struct nfp_insn_meta *meta,
572 		  unsigned int insn_idx);
573 
574 void *nfp_bpf_relo_for_vnic(struct nfp_prog *nfp_prog, struct nfp_bpf_vnic *bv);
575 
576 unsigned int nfp_bpf_ctrl_cmsg_mtu(struct nfp_app_bpf *bpf);
577 long long int
578 nfp_bpf_ctrl_alloc_map(struct nfp_app_bpf *bpf, struct bpf_map *map);
579 void
580 nfp_bpf_ctrl_free_map(struct nfp_app_bpf *bpf, struct nfp_bpf_map *nfp_map);
581 int nfp_bpf_ctrl_getfirst_entry(struct bpf_offloaded_map *offmap,
582 				void *next_key);
583 int nfp_bpf_ctrl_update_entry(struct bpf_offloaded_map *offmap,
584 			      void *key, void *value, u64 flags);
585 int nfp_bpf_ctrl_del_entry(struct bpf_offloaded_map *offmap, void *key);
586 int nfp_bpf_ctrl_lookup_entry(struct bpf_offloaded_map *offmap,
587 			      void *key, void *value);
588 int nfp_bpf_ctrl_getnext_entry(struct bpf_offloaded_map *offmap,
589 			       void *key, void *next_key);
590 
591 int nfp_bpf_event_output(struct nfp_app_bpf *bpf, const void *data,
592 			 unsigned int len);
593 
594 void nfp_bpf_ctrl_msg_rx(struct nfp_app *app, struct sk_buff *skb);
595 void
596 nfp_bpf_ctrl_msg_rx_raw(struct nfp_app *app, const void *data,
597 			unsigned int len);
598 #endif
599