micropython/extmod/modselect.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

629 lines
21 KiB
C
Raw Normal View History

/*
* This file is part of the MicroPython project, http://micropython.org/
*
* The MIT License (MIT)
*
* Copyright (c) 2014-2023 Damien P. George
* Copyright (c) 2015-2017 Paul Sokolovsky
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include "py/mpconfig.h"
#include "py/runtime.h"
#include "py/obj.h"
#include "py/objlist.h"
#include "py/stream.h"
#include "py/mperrno.h"
#include "py/mphal.h"
#if MICROPY_PY_SELECT
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_SELECT && MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
#error "select.select is not supported with MICROPY_PY_SELECT_POSIX_OPTIMISATIONS"
#endif
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
#include <poll.h>
#if !((MP_STREAM_POLL_RD) == (POLLIN) && \
(MP_STREAM_POLL_WR) == (POLLOUT) && \
(MP_STREAM_POLL_ERR) == (POLLERR) && \
(MP_STREAM_POLL_HUP) == (POLLHUP) && \
(MP_STREAM_POLL_NVAL) == (POLLNVAL))
#error "With MICROPY_PY_SELECT_POSIX_OPTIMISATIONS enabled, POLL constants must match"
#endif
// When non-file-descriptor objects are on the list to be polled (the polling of
// which involves repeatedly calling ioctl(MP_STREAM_POLL)), this variable sets
// the period between polling these objects.
#define MICROPY_PY_SELECT_IOCTL_CALL_PERIOD_MS (1)
#endif
// Flags for ipoll()
#define FLAG_ONESHOT (1)
// A single pollable object.
typedef struct _poll_obj_t {
mp_obj_t obj;
mp_uint_t (*ioctl)(mp_obj_t obj, mp_uint_t request, uintptr_t arg, int *errcode);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
// If the pollable object has an associated file descriptor, then pollfd points to an entry
// in poll_set_t::pollfds, and the events/revents fields for this object are stored in the
// pollfd entry (and the nonfd_* members are unused).
// Otherwise the object is a non-file-descriptor object and pollfd==NULL, and the events/
// revents fields are stored in the nonfd_* members (which are named as such so that code
// doesn't accidentally mix the use of these members when this optimisation is used).
struct pollfd *pollfd;
uint16_t nonfd_events;
uint16_t nonfd_revents;
#else
mp_uint_t events;
mp_uint_t revents;
#endif
} poll_obj_t;
// A set of pollable objects.
typedef struct _poll_set_t {
// Map containing a dict with key=object to poll, value=its corresponding poll_obj_t.
mp_map_t map;
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
// Array of pollfd entries for objects that have a file descriptor.
unsigned short alloc;
unsigned short len;
struct pollfd *pollfds;
#endif
} poll_set_t;
STATIC void poll_set_init(poll_set_t *poll_set, size_t n) {
mp_map_init(&poll_set->map, n);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
poll_set->alloc = 0;
poll_set->len = 0;
poll_set->pollfds = NULL;
#endif
}
#if MICROPY_PY_SELECT_SELECT
STATIC void poll_set_deinit(poll_set_t *poll_set) {
mp_map_deinit(&poll_set->map);
}
#endif
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
STATIC mp_uint_t poll_obj_get_events(poll_obj_t *poll_obj) {
assert(poll_obj->pollfd == NULL);
return poll_obj->nonfd_events;
}
STATIC void poll_obj_set_events(poll_obj_t *poll_obj, mp_uint_t events) {
if (poll_obj->pollfd != NULL) {
poll_obj->pollfd->events = events;
} else {
poll_obj->nonfd_events = events;
}
}
STATIC mp_uint_t poll_obj_get_revents(poll_obj_t *poll_obj) {
if (poll_obj->pollfd != NULL) {
return poll_obj->pollfd->revents;
} else {
return poll_obj->nonfd_revents;
}
}
STATIC void poll_obj_set_revents(poll_obj_t *poll_obj, mp_uint_t revents) {
if (poll_obj->pollfd != NULL) {
poll_obj->pollfd->revents = revents;
} else {
poll_obj->nonfd_revents = revents;
}
}
STATIC struct pollfd *poll_set_add_fd(poll_set_t *poll_set, int fd) {
struct pollfd *free_slot = NULL;
for (unsigned int i = 0; i < poll_set->len; ++i) {
struct pollfd *slot = &poll_set->pollfds[i];
if (slot->fd == -1) {
free_slot = slot;
break;
}
}
if (free_slot == NULL) {
if (poll_set->len >= poll_set->alloc) {
poll_set->pollfds = m_renew(struct pollfd, poll_set->pollfds, poll_set->alloc, poll_set->alloc + 4);
poll_set->alloc += 4;
}
free_slot = &poll_set->pollfds[poll_set->len++];
}
free_slot->fd = fd;
return free_slot;
}
static inline bool poll_set_all_are_fds(poll_set_t *poll_set) {
return poll_set->map.used == poll_set->len;
}
#else
static inline mp_uint_t poll_obj_get_events(poll_obj_t *poll_obj) {
return poll_obj->events;
}
static inline void poll_obj_set_events(poll_obj_t *poll_obj, mp_uint_t events) {
poll_obj->events = events;
}
static inline mp_uint_t poll_obj_get_revents(poll_obj_t *poll_obj) {
return poll_obj->revents;
}
static inline void poll_obj_set_revents(poll_obj_t *poll_obj, mp_uint_t revents) {
poll_obj->revents = revents;
}
#endif
STATIC void poll_set_add_obj(poll_set_t *poll_set, const mp_obj_t *obj, mp_uint_t obj_len, mp_uint_t events, bool or_events) {
for (mp_uint_t i = 0; i < obj_len; i++) {
mp_map_elem_t *elem = mp_map_lookup(&poll_set->map, mp_obj_id(obj[i]), MP_MAP_LOOKUP_ADD_IF_NOT_FOUND);
if (elem->value == MP_OBJ_NULL) {
// object not found; get its ioctl and add it to the poll list
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
// If an exception is raised below when adding the new object then the map entry for that
// object remains unpopulated, and methods like poll() may crash. This case is not handled.
poll_obj_t *poll_obj = m_new_obj(poll_obj_t);
poll_obj->obj = obj[i];
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
int fd = -1;
if (mp_obj_is_int(obj[i])) {
// A file descriptor integer passed in as the object, so use it directly.
fd = mp_obj_get_int(obj[i]);
if (fd < 0) {
mp_raise_ValueError(NULL);
}
poll_obj->ioctl = NULL;
} else {
// An object passed in. Check if it has a file descriptor.
const mp_stream_p_t *stream_p = mp_get_stream_raise(obj[i], MP_STREAM_OP_IOCTL);
poll_obj->ioctl = stream_p->ioctl;
int err;
mp_uint_t res = stream_p->ioctl(obj[i], MP_STREAM_GET_FILENO, 0, &err);
if (res != MP_STREAM_ERROR) {
fd = res;
}
}
if (fd >= 0) {
// Object has a file descriptor so add it to pollfds.
poll_obj->pollfd = poll_set_add_fd(poll_set, fd);
} else {
// Object doesn't have a file descriptor.
poll_obj->pollfd = NULL;
}
#else
const mp_stream_p_t *stream_p = mp_get_stream_raise(obj[i], MP_STREAM_OP_IOCTL);
poll_obj->ioctl = stream_p->ioctl;
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#endif
poll_obj_set_events(poll_obj, events);
poll_obj_set_revents(poll_obj, 0);
elem->value = MP_OBJ_FROM_PTR(poll_obj);
} else {
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
// object exists; update its events
poll_obj_t *poll_obj = (poll_obj_t *)MP_OBJ_TO_PTR(elem->value);
#if MICROPY_PY_SELECT_SELECT
if (or_events) {
events |= poll_obj_get_events(poll_obj);
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#else
(void)or_events;
#endif
poll_obj_set_events(poll_obj, events);
}
}
}
// For each object in the poll set, poll it once.
STATIC mp_uint_t poll_set_poll_once(poll_set_t *poll_set, size_t *rwx_num) {
mp_uint_t n_ready = 0;
for (mp_uint_t i = 0; i < poll_set->map.alloc; ++i) {
if (!mp_map_slot_is_filled(&poll_set->map, i)) {
continue;
}
poll_obj_t *poll_obj = MP_OBJ_TO_PTR(poll_set->map.table[i].value);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
if (poll_obj->pollfd != NULL) {
// Object has file descriptor so will be polled separately by poll().
continue;
}
#endif
int errcode;
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
mp_int_t ret = poll_obj->ioctl(poll_obj->obj, MP_STREAM_POLL, poll_obj_get_events(poll_obj), &errcode);
poll_obj_set_revents(poll_obj, ret);
if (ret == -1) {
// error doing ioctl
mp_raise_OSError(errcode);
}
if (ret != 0) {
// object is ready
n_ready += 1;
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#if MICROPY_PY_SELECT_SELECT
if (rwx_num != NULL) {
if (ret & MP_STREAM_POLL_RD) {
rwx_num[0] += 1;
}
if (ret & MP_STREAM_POLL_WR) {
rwx_num[1] += 1;
}
if ((ret & ~(MP_STREAM_POLL_RD | MP_STREAM_POLL_WR)) != 0) {
rwx_num[2] += 1;
}
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#else
(void)rwx_num;
#endif
}
}
return n_ready;
}
STATIC mp_uint_t poll_set_poll_until_ready_or_timeout(poll_set_t *poll_set, size_t *rwx_num, mp_uint_t timeout) {
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
mp_uint_t start_ticks = mp_hal_ticks_ms();
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
for (;;) {
MP_THREAD_GIL_EXIT();
// Compute the timeout.
int t = MICROPY_PY_SELECT_IOCTL_CALL_PERIOD_MS;
if (poll_set_all_are_fds(poll_set)) {
// All our pollables are file descriptors, so we can use a blocking
// poll and let it (the underlying system) handle the timeout.
if (timeout == (mp_uint_t)-1) {
t = -1;
} else {
mp_uint_t delta = mp_hal_ticks_ms() - start_ticks;
if (delta >= timeout) {
t = 0;
} else {
t = timeout - delta;
}
}
}
// Call system poll for those objects that have a file descriptor.
int n_ready = poll(poll_set->pollfds, poll_set->len, t);
MP_THREAD_GIL_ENTER();
// The call to poll() may have been interrupted, but per PEP 475 we must retry if the
// signal is EINTR (this implements a special case of calling MP_HAL_RETRY_SYSCALL()).
if (n_ready == -1) {
int err = errno;
if (err != EINTR) {
mp_raise_OSError(err);
}
n_ready = 0;
}
// Explicitly poll any objects that do not have a file descriptor.
if (!poll_set_all_are_fds(poll_set)) {
n_ready += poll_set_poll_once(poll_set, rwx_num);
}
// Return if an object is ready, or if the timeout expired.
if (n_ready > 0 || (timeout != (mp_uint_t)-1 && mp_hal_ticks_ms() - start_ticks >= timeout)) {
return n_ready;
}
// This would be MICROPY_EVENT_POLL_HOOK but the call to poll() above already includes a delay.
mp_handle_pending(true);
}
#else
for (;;) {
// poll the objects
mp_uint_t n_ready = poll_set_poll_once(poll_set, rwx_num);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if (n_ready > 0 || (timeout != (mp_uint_t)-1 && mp_hal_ticks_ms() - start_ticks >= timeout)) {
return n_ready;
}
MICROPY_EVENT_POLL_HOOK
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
#endif
}
#if MICROPY_PY_SELECT_SELECT
// select(rlist, wlist, xlist[, timeout])
STATIC mp_obj_t select_select(size_t n_args, const mp_obj_t *args) {
// get array data from tuple/list arguments
size_t rwx_len[3];
mp_obj_t *r_array, *w_array, *x_array;
mp_obj_get_array(args[0], &rwx_len[0], &r_array);
mp_obj_get_array(args[1], &rwx_len[1], &w_array);
mp_obj_get_array(args[2], &rwx_len[2], &x_array);
// get timeout
mp_uint_t timeout = -1;
if (n_args == 4) {
if (args[3] != mp_const_none) {
#if MICROPY_PY_BUILTINS_FLOAT
float timeout_f = mp_obj_get_float_to_f(args[3]);
if (timeout_f >= 0) {
timeout = (mp_uint_t)(timeout_f * 1000);
}
#else
timeout = mp_obj_get_int(args[3]) * 1000;
#endif
}
}
// merge separate lists and get the ioctl function for each object
poll_set_t poll_set;
poll_set_init(&poll_set, rwx_len[0] + rwx_len[1] + rwx_len[2]);
poll_set_add_obj(&poll_set, r_array, rwx_len[0], MP_STREAM_POLL_RD, true);
poll_set_add_obj(&poll_set, w_array, rwx_len[1], MP_STREAM_POLL_WR, true);
poll_set_add_obj(&poll_set, x_array, rwx_len[2], MP_STREAM_POLL_ERR | MP_STREAM_POLL_HUP, true);
// poll all objects
rwx_len[0] = rwx_len[1] = rwx_len[2] = 0;
poll_set_poll_until_ready_or_timeout(&poll_set, rwx_len, timeout);
// one or more objects are ready, or we had a timeout
mp_obj_t list_array[3];
list_array[0] = mp_obj_new_list(rwx_len[0], NULL);
list_array[1] = mp_obj_new_list(rwx_len[1], NULL);
list_array[2] = mp_obj_new_list(rwx_len[2], NULL);
rwx_len[0] = rwx_len[1] = rwx_len[2] = 0;
for (mp_uint_t i = 0; i < poll_set.map.alloc; ++i) {
if (!mp_map_slot_is_filled(&poll_set.map, i)) {
continue;
}
poll_obj_t *poll_obj = MP_OBJ_TO_PTR(poll_set.map.table[i].value);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if (poll_obj->revents & MP_STREAM_POLL_RD) {
((mp_obj_list_t *)MP_OBJ_TO_PTR(list_array[0]))->items[rwx_len[0]++] = poll_obj->obj;
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if (poll_obj->revents & MP_STREAM_POLL_WR) {
((mp_obj_list_t *)MP_OBJ_TO_PTR(list_array[1]))->items[rwx_len[1]++] = poll_obj->obj;
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if ((poll_obj->revents & ~(MP_STREAM_POLL_RD | MP_STREAM_POLL_WR)) != 0) {
((mp_obj_list_t *)MP_OBJ_TO_PTR(list_array[2]))->items[rwx_len[2]++] = poll_obj->obj;
}
}
poll_set_deinit(&poll_set);
return mp_obj_new_tuple(3, list_array);
}
MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(mp_select_select_obj, 3, 4, select_select);
#endif // MICROPY_PY_SELECT_SELECT
typedef struct _mp_obj_poll_t {
mp_obj_base_t base;
poll_set_t poll_set;
short iter_cnt;
short iter_idx;
int flags;
// callee-owned tuple
mp_obj_t ret_tuple;
} mp_obj_poll_t;
// register(obj[, eventmask])
STATIC mp_obj_t poll_register(size_t n_args, const mp_obj_t *args) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(args[0]);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
mp_uint_t events;
if (n_args == 3) {
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
events = mp_obj_get_int(args[2]);
} else {
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
events = MP_STREAM_POLL_RD | MP_STREAM_POLL_WR;
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
poll_set_add_obj(&self->poll_set, &args[1], 1, events, false);
return mp_const_none;
}
MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(poll_register_obj, 2, 3, poll_register);
// unregister(obj)
STATIC mp_obj_t poll_unregister(mp_obj_t self_in, mp_obj_t obj_in) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(self_in);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
mp_map_elem_t *elem = mp_map_lookup(&self->poll_set.map, mp_obj_id(obj_in), MP_MAP_LOOKUP_REMOVE_IF_FOUND);
#if MICROPY_PY_SELECT_POSIX_OPTIMISATIONS
if (elem != NULL) {
poll_obj_t *poll_obj = (poll_obj_t *)MP_OBJ_TO_PTR(elem->value);
if (poll_obj->pollfd != NULL) {
poll_obj->pollfd->fd = -1;
}
elem->value = MP_OBJ_NULL;
}
#else
(void)elem;
#endif
// TODO raise KeyError if obj didn't exist in map
return mp_const_none;
}
MP_DEFINE_CONST_FUN_OBJ_2(poll_unregister_obj, poll_unregister);
// modify(obj, eventmask)
STATIC mp_obj_t poll_modify(mp_obj_t self_in, mp_obj_t obj_in, mp_obj_t eventmask_in) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(self_in);
mp_map_elem_t *elem = mp_map_lookup(&self->poll_set.map, mp_obj_id(obj_in), MP_MAP_LOOKUP);
if (elem == NULL) {
mp_raise_OSError(MP_ENOENT);
}
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
poll_obj_set_events((poll_obj_t *)MP_OBJ_TO_PTR(elem->value), mp_obj_get_int(eventmask_in));
return mp_const_none;
}
MP_DEFINE_CONST_FUN_OBJ_3(poll_modify_obj, poll_modify);
STATIC mp_uint_t poll_poll_internal(uint n_args, const mp_obj_t *args) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(args[0]);
// work out timeout (its given already in ms)
mp_uint_t timeout = -1;
int flags = 0;
if (n_args >= 2) {
if (args[1] != mp_const_none) {
mp_int_t timeout_i = mp_obj_get_int(args[1]);
if (timeout_i >= 0) {
timeout = timeout_i;
}
}
if (n_args >= 3) {
flags = mp_obj_get_int(args[2]);
}
}
self->flags = flags;
return poll_set_poll_until_ready_or_timeout(&self->poll_set, NULL, timeout);
}
STATIC mp_obj_t poll_poll(size_t n_args, const mp_obj_t *args) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(args[0]);
mp_uint_t n_ready = poll_poll_internal(n_args, args);
// one or more objects are ready, or we had a timeout
mp_obj_list_t *ret_list = MP_OBJ_TO_PTR(mp_obj_new_list(n_ready, NULL));
n_ready = 0;
for (mp_uint_t i = 0; i < self->poll_set.map.alloc; ++i) {
if (!mp_map_slot_is_filled(&self->poll_set.map, i)) {
continue;
}
poll_obj_t *poll_obj = MP_OBJ_TO_PTR(self->poll_set.map.table[i].value);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if (poll_obj_get_revents(poll_obj) != 0) {
mp_obj_t tuple[2] = {poll_obj->obj, MP_OBJ_NEW_SMALL_INT(poll_obj_get_revents(poll_obj))};
ret_list->items[n_ready++] = mp_obj_new_tuple(2, tuple);
}
}
return MP_OBJ_FROM_PTR(ret_list);
}
MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(poll_poll_obj, 1, 2, poll_poll);
STATIC mp_obj_t poll_ipoll(size_t n_args, const mp_obj_t *args) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(args[0]);
if (self->ret_tuple == MP_OBJ_NULL) {
self->ret_tuple = mp_obj_new_tuple(2, NULL);
}
int n_ready = poll_poll_internal(n_args, args);
self->iter_cnt = n_ready;
self->iter_idx = 0;
return args[0];
}
MP_DEFINE_CONST_FUN_OBJ_VAR_BETWEEN(poll_ipoll_obj, 1, 3, poll_ipoll);
STATIC mp_obj_t poll_iternext(mp_obj_t self_in) {
mp_obj_poll_t *self = MP_OBJ_TO_PTR(self_in);
if (self->iter_cnt == 0) {
return MP_OBJ_STOP_ITERATION;
}
self->iter_cnt--;
for (mp_uint_t i = self->iter_idx; i < self->poll_set.map.alloc; ++i) {
self->iter_idx++;
if (!mp_map_slot_is_filled(&self->poll_set.map, i)) {
continue;
}
poll_obj_t *poll_obj = MP_OBJ_TO_PTR(self->poll_set.map.table[i].value);
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
if (poll_obj_get_revents(poll_obj) != 0) {
mp_obj_tuple_t *t = MP_OBJ_TO_PTR(self->ret_tuple);
t->items[0] = poll_obj->obj;
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
t->items[1] = MP_OBJ_NEW_SMALL_INT(poll_obj_get_revents(poll_obj));
if (self->flags & FLAG_ONESHOT) {
extmod/modselect: Add optimisation to use system poll when possible. A previous commit removed the unix-specific select module implementation and made unix use the common one. This commit adds an optimisation so that the system poll function is used when polling objects that have a file descriptor. With this optimisation enabled, if code registers both file-descriptor-based objects, and non- file-descriptor-based objects with select.poll() then the following occurs: - the system poll is called for all file-descriptor-based objects with a timeout of 1ms - then the bare-metal polling implementation is used for remaining objects, which calls into their ioctl method (which can be in C or Python) In the case where all objects have file descriptors, the system poll is called with the full timeout requested by the caller. That makes it as efficient as possible in the case everything has a file descriptor. Benefits of this approach: - all ports use the same select module implementation - the unix port now supports polling of all objects and matches bare metal implementations - it's still efficient for existing cases where only files and sockets are polled (on unix) - the bare metal implementation does not change - polling of SSL objects will now work on unix by calling in to the ioctl method on SSL objects (this is required for asyncio ssl support) Note that extmod/vfs_posix_file.c has poll disable when the optimisation is enabled, because the code is not reachable when the optimisation is used. Signed-off-by: Damien George <damien@micropython.org>
2023-08-01 03:39:39 +01:00
// Don't poll next time, until new event mask will be set explicitly
poll_obj_set_events(poll_obj, 0);
}
return MP_OBJ_FROM_PTR(t);
}
}
assert(!"inconsistent number of poll active entries");
self->iter_cnt = 0;
return MP_OBJ_STOP_ITERATION;
}
STATIC const mp_rom_map_elem_t poll_locals_dict_table[] = {
{ MP_ROM_QSTR(MP_QSTR_register), MP_ROM_PTR(&poll_register_obj) },
{ MP_ROM_QSTR(MP_QSTR_unregister), MP_ROM_PTR(&poll_unregister_obj) },
{ MP_ROM_QSTR(MP_QSTR_modify), MP_ROM_PTR(&poll_modify_obj) },
{ MP_ROM_QSTR(MP_QSTR_poll), MP_ROM_PTR(&poll_poll_obj) },
{ MP_ROM_QSTR(MP_QSTR_ipoll), MP_ROM_PTR(&poll_ipoll_obj) },
};
STATIC MP_DEFINE_CONST_DICT(poll_locals_dict, poll_locals_dict_table);
STATIC MP_DEFINE_CONST_OBJ_TYPE(
mp_type_poll,
MP_QSTR_poll,
MP_TYPE_FLAG_ITER_IS_ITERNEXT,
iter, poll_iternext,
locals_dict, &poll_locals_dict
);
// poll()
STATIC mp_obj_t select_poll(void) {
mp_obj_poll_t *poll = mp_obj_malloc(mp_obj_poll_t, &mp_type_poll);
poll_set_init(&poll->poll_set, 0);
poll->iter_cnt = 0;
poll->ret_tuple = MP_OBJ_NULL;
return MP_OBJ_FROM_PTR(poll);
}
MP_DEFINE_CONST_FUN_OBJ_0(mp_select_poll_obj, select_poll);
STATIC const mp_rom_map_elem_t mp_module_select_globals_table[] = {
{ MP_ROM_QSTR(MP_QSTR___name__), MP_ROM_QSTR(MP_QSTR_select) },
#if MICROPY_PY_SELECT_SELECT
{ MP_ROM_QSTR(MP_QSTR_select), MP_ROM_PTR(&mp_select_select_obj) },
#endif
{ MP_ROM_QSTR(MP_QSTR_poll), MP_ROM_PTR(&mp_select_poll_obj) },
{ MP_ROM_QSTR(MP_QSTR_POLLIN), MP_ROM_INT(MP_STREAM_POLL_RD) },
{ MP_ROM_QSTR(MP_QSTR_POLLOUT), MP_ROM_INT(MP_STREAM_POLL_WR) },
{ MP_ROM_QSTR(MP_QSTR_POLLERR), MP_ROM_INT(MP_STREAM_POLL_ERR) },
{ MP_ROM_QSTR(MP_QSTR_POLLHUP), MP_ROM_INT(MP_STREAM_POLL_HUP) },
};
STATIC MP_DEFINE_CONST_DICT(mp_module_select_globals, mp_module_select_globals_table);
const mp_obj_module_t mp_module_select = {
.base = { &mp_type_module },
.globals = (mp_obj_dict_t *)&mp_module_select_globals,
};
MP_REGISTER_EXTENSIBLE_MODULE(MP_QSTR_select, mp_module_select);
#endif // MICROPY_PY_SELECT