This adds an api to register and unregister the napi for io-uring. If the arg value is specified when unregistering, the current napi setting for the busy poll timeout is copied into the user structure. If this is not required, NULL can be passed as the arg value.
Unified · Split
| @@ -571,6 +571,10 @@ enum { · 570 unmodified lines | ||
| 571 | 571 | /* return status information for a buffer group */ |
| 572 | 572 | IORING_REGISTER_PBUF_STATUS = 26, |
| 573 | 573 | |
| 574 | + /* set/clear busy poll settings */ | |
| 575 | + IORING_REGISTER_NAPI = 27, | |
| 576 | + IORING_UNREGISTER_NAPI = 28, | |
| 577 | + | |
| 574 | 578 | /* this goes last */ |
| 575 | 579 | IORING_REGISTER_LAST, |
| 576 | 580 | |
| @@ -704,6 +708,14 @@ struct io_uring_buf_status { · 127 unmodified lines | ||
| 704 | 708 | __u32 resv[8]; |
| 705 | 709 | }; |
| 706 | 710 | |
| 711 | +/* argument for IORING_(UN)REGISTER_NAPI */ | |
| 712 | +struct io_uring_napi { | |
| 713 | + __u32 busy_poll_to; | |
| 714 | + __u8 prefer_busy_poll; | |
| 715 | + __u8 pad[3]; | |
| 716 | + __u64 resv; | |
| 717 | +}; | |
| 718 | + | |
| 707 | 719 | /* |
| 708 | 720 | * io_uring_restriction->opcode values |
| 709 | 721 | */ |
| @@ -207,6 +207,58 @@ void io_napi_free(struct io_ring_ctx *ctx) · 206 unmodified lines | ||
| 207 | 207 | spin_unlock(&ctx->napi_lock); |
| 208 | 208 | } |
| 209 | 209 | |
| 210 | +/* | |
| 211 | + * io_napi_register() - Register napi with io-uring | |
| 212 | + * @ctx: pointer to io-uring context structure | |
| 213 | + * @arg: pointer to io_uring_napi structure | |
| 214 | + * | |
| 215 | + * Register napi in the io-uring context. | |
| 216 | + */ | |
| 217 | +int io_register_napi(struct io_ring_ctx *ctx, void __user *arg) | |
| 218 | +{ | |
| 219 | + const struct io_uring_napi curr = { | |
| 220 | + .busy_poll_to = ctx->napi_busy_poll_to, | |
| 221 | + .prefer_busy_poll = ctx->napi_prefer_busy_poll | |
| 222 | + }; | |
| 223 | + struct io_uring_napi napi; | |
| 224 | + | |
| 225 | + if (copy_from_user(&napi, arg, sizeof(napi))) | |
| 226 | + return -EFAULT; | |
| 227 | + if (napi.pad[0] || napi.pad[1] || napi.pad[2] || napi.resv) | |
| 228 | + return -EINVAL; | |
| 229 | + | |
| 230 | + WRITE_ONCE(ctx->napi_busy_poll_to, napi.busy_poll_to); | |
| 231 | + WRITE_ONCE(ctx->napi_prefer_busy_poll, !!napi.prefer_busy_poll); | |
| 232 | + | |
| 233 | + if (copy_to_user(arg, &curr, sizeof(curr))) | |
| 234 | + return -EFAULT; | |
| 235 | + | |
| 236 | + return 0; | |
| 237 | +} | |
| 238 | + | |
| 239 | +/* | |
| 240 | + * io_napi_unregister() - Unregister napi with io-uring | |
| 241 | + * @ctx: pointer to io-uring context structure | |
| 242 | + * @arg: pointer to io_uring_napi structure | |
| 243 | + * | |
| 244 | + * Unregister napi. If arg has been specified copy the busy poll timeout and | |
| 245 | + * prefer busy poll setting to the passed in structure. | |
| 246 | + */ | |
| 247 | +int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg) | |
| 248 | +{ | |
| 249 | + const struct io_uring_napi curr = { | |
| 250 | + .busy_poll_to = ctx->napi_busy_poll_to, | |
| 251 | + .prefer_busy_poll = ctx->napi_prefer_busy_poll | |
| 252 | + }; | |
| 253 | + | |
| 254 | + if (arg && copy_to_user(arg, &curr, sizeof(curr))) | |
| 255 | + return -EFAULT; | |
| 256 | + | |
| 257 | + WRITE_ONCE(ctx->napi_busy_poll_to, 0); | |
| 258 | + WRITE_ONCE(ctx->napi_prefer_busy_poll, false); | |
| 259 | + return 0; | |
| 260 | +} | |
| 261 | + | |
| 210 | 262 | /* |
| 211 | 263 | * __io_napi_adjust_timeout() - Add napi id to the busy poll list |
| 212 | 264 | * @ctx: pointer to io-uring context structure |
| @@ -12,6 +12,9 @@ · 11 unmodified lines | ||
| 12 | 12 | void io_napi_init(struct io_ring_ctx *ctx); |
| 13 | 13 | void io_napi_free(struct io_ring_ctx *ctx); |
| 14 | 14 | |
| 15 | +int io_register_napi(struct io_ring_ctx *ctx, void __user *arg); | |
| 16 | +int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg); | |
| 17 | + | |
| 15 | 18 | void __io_napi_add(struct io_ring_ctx *ctx, struct socket *sock); |
| 16 | 19 | |
| 17 | 20 | void __io_napi_adjust_timeout(struct io_ring_ctx *ctx, |
| @@ -68,6 +71,14 @@ static inline void io_napi_init(struct io_ring_ctx *ctx) · 50 unmodified lines | ||
| 68 | 71 | static inline void io_napi_free(struct io_ring_ctx *ctx) |
| 69 | 72 | { |
| 70 | 73 | } |
| 74 | +static inline int io_register_napi(struct io_ring_ctx *ctx, void __user *arg) | |
| 75 | +{ | |
| 76 | + return -EOPNOTSUPP; | |
| 77 | +} | |
| 78 | +static inline int io_unregister_napi(struct io_ring_ctx *ctx, void __user *arg) | |
| 79 | +{ | |
| 80 | + return -EOPNOTSUPP; | |
| 81 | +} | |
| 71 | 82 | static inline bool io_napi(struct io_ring_ctx *ctx) |
| 72 | 83 | { |
| 73 | 84 | return false; |
| @@ -26,6 +26,7 @@ · 25 unmodified lines | ||
| 26 | 26 | #include "register.h" |
| 27 | 27 | #include "cancel.h" |
| 28 | 28 | #include "kbuf.h" |
| 29 | +#include "napi.h" | |
| 29 | 30 | |
| 30 | 31 | #define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \ |
| 31 | 32 | IORING_REGISTER_LAST + IORING_OP_LAST) |
| @@ -550,6 +551,18 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, · 518 unmodified lines | ||
| 550 | 551 | break; |
| 551 | 552 | ret = io_register_pbuf_status(ctx, arg); |
| 552 | 553 | break; |
| 554 | + case IORING_REGISTER_NAPI: | |
| 555 | + ret = -EINVAL; | |
| 556 | + if (!arg || nr_args != 1) | |
| 557 | + break; | |
| 558 | + ret = io_register_napi(ctx, arg); | |
| 559 | + break; | |
| 560 | + case IORING_UNREGISTER_NAPI: | |
| 561 | + ret = -EINVAL; | |
| 562 | + if (nr_args != 1) | |
| 563 | + break; | |
| 564 | + ret = io_unregister_napi(ctx, arg); | |
| 565 | + break; | |
| 553 | 566 | default: |
| 554 | 567 | ret = -EINVAL; |
| 555 | 568 | break; |