[socket] Cache lwip_sock pointers and inline ready() chain (#14408)

esphome/components/socket/bsd_sockets_impl.cpp

@@ -11,11 +11,15 @@ namespace esphome::socket {
 
 BSDSocketImpl::BSDSocketImpl(int fd, bool monitor_loop) {
   this->fd_ = fd;
-  // Register new socket with the application for select() if monitoring requested
-  if (monitor_loop && this->fd_ >= 0) {
-    // Only set loop_monitored_ to true if registration succeeds
-    this->loop_monitored_ = App.register_socket_fd(this->fd_);
-  }
+  if (!monitor_loop || this->fd_ < 0)
+    return;
+#ifdef USE_LWIP_FAST_SELECT
+  // Cache lwip_sock pointer and register for monitoring (hooks callback internally)
+  this->cached_sock_ = esphome_lwip_get_sock(this->fd_);
+  this->loop_monitored_ = App.register_socket(this->cached_sock_);
+#else
+  this->loop_monitored_ = App.register_socket_fd(this->fd_);
+#endif
 }
 
 BSDSocketImpl::~BSDSocketImpl() {
@@ -26,10 +30,17 @@ BSDSocketImpl::~BSDSocketImpl() {
 
 int BSDSocketImpl::close() {
   if (!this->closed_) {
-    // Unregister from select() before closing if monitored
+    // Unregister before closing to avoid dangling pointer in monitored set
+#ifdef USE_LWIP_FAST_SELECT
+    if (this->loop_monitored_) {
+      App.unregister_socket(this->cached_sock_);
+      this->cached_sock_ = nullptr;
+    }
+#else
     if (this->loop_monitored_) {
       App.unregister_socket_fd(this->fd_);
     }
+#endif
     int ret = ::close(this->fd_);
     this->closed_ = true;
     return ret;
@@ -48,8 +59,6 @@ int BSDSocketImpl::setblocking(bool blocking) {
   return 0;
 }
 
-bool BSDSocketImpl::ready() const { return socket_ready_fd(this->fd_, this->loop_monitored_); }
-
 size_t BSDSocketImpl::getpeername_to(std::span<char, SOCKADDR_STR_LEN> buf) {
   struct sockaddr_storage storage;
   socklen_t len = sizeof(storage);
@@ -86,14 +95,6 @@ std::unique_ptr<Socket> socket_loop_monitored(int domain, int type, int protocol
   return create_socket(domain, type, protocol, true);
 }
 
-std::unique_ptr<ListenSocket> socket_listen(int domain, int type, int protocol) {
-  return create_socket(domain, type, protocol, false);
-}
-
-std::unique_ptr<ListenSocket> socket_listen_loop_monitored(int domain, int type, int protocol) {
-  return create_socket(domain, type, protocol, true);
-}
-
 }  // namespace esphome::socket
 
 #endif  // USE_SOCKET_IMPL_BSD_SOCKETS

esphome/components/socket/bsd_sockets_impl.h

@@ -13,6 +13,10 @@
 #include <lwip/sockets.h>
 #endif
 
+#ifdef USE_LWIP_FAST_SELECT
+struct lwip_sock;
+#endif
+
 namespace esphome::socket {
 
 class BSDSocketImpl {
@@ -105,6 +109,9 @@ class BSDSocketImpl {
 
  protected:
   int fd_{-1};
+#ifdef USE_LWIP_FAST_SELECT
+  struct lwip_sock *cached_sock_{nullptr};  // Cached for direct rcvevent read in ready()
+#endif
   bool closed_{false};
   bool loop_monitored_{false};
 };

esphome/components/socket/lwip_sockets_impl.cpp

@@ -11,11 +11,15 @@ namespace esphome::socket {
 
 LwIPSocketImpl::LwIPSocketImpl(int fd, bool monitor_loop) {
   this->fd_ = fd;
-  // Register new socket with the application for select() if monitoring requested
-  if (monitor_loop && this->fd_ >= 0) {
-    // Only set loop_monitored_ to true if registration succeeds
-    this->loop_monitored_ = App.register_socket_fd(this->fd_);
-  }
+  if (!monitor_loop || this->fd_ < 0)
+    return;
+#ifdef USE_LWIP_FAST_SELECT
+  // Cache lwip_sock pointer and register for monitoring (hooks callback internally)
+  this->cached_sock_ = esphome_lwip_get_sock(this->fd_);
+  this->loop_monitored_ = App.register_socket(this->cached_sock_);
+#else
+  this->loop_monitored_ = App.register_socket_fd(this->fd_);
+#endif
 }
 
 LwIPSocketImpl::~LwIPSocketImpl() {
@@ -26,10 +30,17 @@ LwIPSocketImpl::~LwIPSocketImpl() {
 
 int LwIPSocketImpl::close() {
   if (!this->closed_) {
-    // Unregister from select() before closing if monitored
+    // Unregister before closing to avoid dangling pointer in monitored set
+#ifdef USE_LWIP_FAST_SELECT
+    if (this->loop_monitored_) {
+      App.unregister_socket(this->cached_sock_);
+      this->cached_sock_ = nullptr;
+    }
+#else
     if (this->loop_monitored_) {
       App.unregister_socket_fd(this->fd_);
     }
+#endif
     int ret = lwip_close(this->fd_);
     this->closed_ = true;
     return ret;
@@ -48,8 +59,6 @@ int LwIPSocketImpl::setblocking(bool blocking) {
   return 0;
 }
 
-bool LwIPSocketImpl::ready() const { return socket_ready_fd(this->fd_, this->loop_monitored_); }
-
 size_t LwIPSocketImpl::getpeername_to(std::span<char, SOCKADDR_STR_LEN> buf) {
   struct sockaddr_storage storage;
   socklen_t len = sizeof(storage);
@@ -86,14 +95,6 @@ std::unique_ptr<Socket> socket_loop_monitored(int domain, int type, int protocol
   return create_socket(domain, type, protocol, true);
 }
 
-std::unique_ptr<ListenSocket> socket_listen(int domain, int type, int protocol) {
-  return create_socket(domain, type, protocol, false);
-}
-
-std::unique_ptr<ListenSocket> socket_listen_loop_monitored(int domain, int type, int protocol) {
-  return create_socket(domain, type, protocol, true);
-}
-
 }  // namespace esphome::socket
 
 #endif  // USE_SOCKET_IMPL_LWIP_SOCKETS

esphome/components/socket/lwip_sockets_impl.h

@@ -9,6 +9,10 @@
 #include "esphome/core/helpers.h"
 #include "headers.h"
 
+#ifdef USE_LWIP_FAST_SELECT
+struct lwip_sock;
+#endif
+
 namespace esphome::socket {
 
 class LwIPSocketImpl {
@@ -71,6 +75,9 @@ class LwIPSocketImpl {
 
  protected:
   int fd_{-1};
+#ifdef USE_LWIP_FAST_SELECT
+  struct lwip_sock *cached_sock_{nullptr};  // Cached for direct rcvevent read in ready()
+#endif
   bool closed_{false};
   bool loop_monitored_{false};
 };

esphome/components/socket/socket.cpp

@@ -8,7 +8,7 @@
 
 namespace esphome::socket {
 
-#ifdef USE_SOCKET_SELECT_SUPPORT
+#if defined(USE_SOCKET_SELECT_SUPPORT) && !defined(USE_LWIP_FAST_SELECT)
 // Shared ready() implementation for fd-based socket implementations (BSD and LWIP sockets).
 // Checks if the Application's select() loop has marked this fd as ready.
 bool socket_ready_fd(int fd, bool loop_monitored) { return !loop_monitored || App.is_socket_ready_(fd); }
@@ -89,6 +89,9 @@ std::unique_ptr<Socket> socket_ip(int type, int protocol) {
 #endif /* USE_NETWORK_IPV6 */
 }
 
+#ifdef USE_SOCKET_IMPL_LWIP_TCP
+// LWIP_TCP has separate Socket/ListenSocket types — needs out-of-line factory.
+// BSD and LWIP_SOCKETS define this inline in socket.h.
 std::unique_ptr<ListenSocket> socket_ip_loop_monitored(int type, int protocol) {
 #if USE_NETWORK_IPV6
   return socket_listen_loop_monitored(AF_INET6, type, protocol);
@@ -96,6 +99,7 @@ std::unique_ptr<ListenSocket> socket_ip_loop_monitored(int type, int protocol) {
   return socket_listen_loop_monitored(AF_INET, type, protocol);
 #endif /* USE_NETWORK_IPV6 */
 }
+#endif
 
 socklen_t set_sockaddr(struct sockaddr *addr, socklen_t addrlen, const char *ip_address, uint16_t port) {
 #if USE_NETWORK_IPV6

esphome/components/socket/socket.h

@@ -6,6 +6,10 @@
 #include "esphome/core/optional.h"
 #include "headers.h"
 
+#ifdef USE_LWIP_FAST_SELECT
+#include "esphome/core/lwip_fast_select.h"
+#endif
+
 #if defined(USE_SOCKET_IMPL_LWIP_TCP) || defined(USE_SOCKET_IMPL_LWIP_SOCKETS) || defined(USE_SOCKET_IMPL_BSD_SOCKETS)
 
 // Include only the active implementation's header.
@@ -36,12 +40,29 @@ using Socket = LWIPRawImpl;
 using ListenSocket = LWIPRawListenImpl;
 #endif
 
-#ifdef USE_SOCKET_SELECT_SUPPORT
+#ifdef USE_LWIP_FAST_SELECT
+/// Shared ready() helper using cached lwip_sock pointer for direct rcvevent read.
+inline bool socket_ready(struct lwip_sock *cached_sock, bool loop_monitored) {
+  return !loop_monitored || (cached_sock != nullptr && esphome_lwip_socket_has_data(cached_sock));
+}
+#elif defined(USE_SOCKET_SELECT_SUPPORT)
 /// Shared ready() helper for fd-based socket implementations.
 /// Checks if the Application's select() loop has marked this fd as ready.
 bool socket_ready_fd(int fd, bool loop_monitored);
 #endif
 
+// Inline ready() — defined here because it depends on socket_ready/socket_ready_fd
+// declared above, while the impl headers are included before those declarations.
+#if defined(USE_SOCKET_IMPL_BSD_SOCKETS) || defined(USE_SOCKET_IMPL_LWIP_SOCKETS)
+inline bool Socket::ready() const {
+#ifdef USE_LWIP_FAST_SELECT
+  return socket_ready(this->cached_sock_, this->loop_monitored_);
+#else
+  return socket_ready_fd(this->fd_, this->loop_monitored_);
+#endif
+}
+#endif
+
 /// Create a socket of the given domain, type and protocol.
 std::unique_ptr<Socket> socket(int domain, int type, int protocol);
 /// Create a socket in the newest available IP domain (IPv6 or IPv4) of the given type and protocol.
@@ -56,11 +77,29 @@ std::unique_ptr<Socket> socket_ip(int type, int protocol);
 std::unique_ptr<Socket> socket_loop_monitored(int domain, int type, int protocol);
 
 /// Create a listening socket of the given domain, type and protocol.
-std::unique_ptr<ListenSocket> socket_listen(int domain, int type, int protocol);
 /// Create a listening socket and monitor it for data in the main loop.
-std::unique_ptr<ListenSocket> socket_listen_loop_monitored(int domain, int type, int protocol);
 /// Create a listening socket in the newest available IP domain and monitor it.
+#ifdef USE_SOCKET_IMPL_LWIP_TCP
+// LWIP_TCP has separate Socket/ListenSocket types — needs distinct factory functions.
+std::unique_ptr<ListenSocket> socket_listen(int domain, int type, int protocol);
+std::unique_ptr<ListenSocket> socket_listen_loop_monitored(int domain, int type, int protocol);
 std::unique_ptr<ListenSocket> socket_ip_loop_monitored(int type, int protocol);
+#else
+// BSD and LWIP_SOCKETS: Socket == ListenSocket, so listen variants just delegate.
+inline std::unique_ptr<ListenSocket> socket_listen(int domain, int type, int protocol) {
+  return socket(domain, type, protocol);
+}
+inline std::unique_ptr<ListenSocket> socket_listen_loop_monitored(int domain, int type, int protocol) {
+  return socket_loop_monitored(domain, type, protocol);
+}
+inline std::unique_ptr<ListenSocket> socket_ip_loop_monitored(int type, int protocol) {
+#if USE_NETWORK_IPV6
+  return socket_loop_monitored(AF_INET6, type, protocol);
+#else
+  return socket_loop_monitored(AF_INET, type, protocol);
+#endif
+}
+#endif
 
 /// Set a sockaddr to the specified address and port for the IP version used by socket_ip().
 /// @param addr Destination sockaddr structure

esphome/core/application.cpp

@@ -552,7 +552,32 @@ void Application::after_loop_tasks_() {
   this->in_loop_ = false;
 }
 
-#ifdef USE_SOCKET_SELECT_SUPPORT
+#ifdef USE_LWIP_FAST_SELECT
+bool Application::register_socket(struct lwip_sock *sock) {
+  // It modifies monitored_sockets_ without locking — must only be called from the main loop.
+  if (sock == nullptr)
+    return false;
+  esphome_lwip_hook_socket(sock);
+  this->monitored_sockets_.push_back(sock);
+  return true;
+}
+
+void Application::unregister_socket(struct lwip_sock *sock) {
+  // It modifies monitored_sockets_ without locking — must only be called from the main loop.
+  for (size_t i = 0; i < this->monitored_sockets_.size(); i++) {
+    if (this->monitored_sockets_[i] != sock)
+      continue;
+
+    // Swap with last element and pop - O(1) removal since order doesn't matter.
+    // No need to unhook the netconn callback — all LwIP sockets share the same
+    // static event_callback, and the socket will be closed by the caller.
+    if (i < this->monitored_sockets_.size() - 1)
+      this->monitored_sockets_[i] = this->monitored_sockets_.back();
+    this->monitored_sockets_.pop_back();
+    return;
+  }
+}
+#elif defined(USE_SOCKET_SELECT_SUPPORT)
 bool Application::register_socket_fd(int fd) {
   // WARNING: This function is NOT thread-safe and must only be called from the main loop
   // It modifies socket_fds_ and related variables without locking
@@ -571,15 +596,10 @@ bool Application::register_socket_fd(int fd) {
 #endif
 
   this->socket_fds_.push_back(fd);
-#ifdef USE_LWIP_FAST_SELECT
-  // Hook the socket's netconn callback for instant wake on receive events
-  esphome_lwip_hook_socket(fd);
-#else
   this->socket_fds_changed_ = true;
   if (fd > this->max_fd_) {
     this->max_fd_ = fd;
   }
-#endif
 
   return true;
 }
@@ -595,13 +615,9 @@ void Application::unregister_socket_fd(int fd) {
       continue;
 
     // Swap with last element and pop - O(1) removal since order doesn't matter.
-    // No need to unhook the netconn callback on fast select platforms — all LwIP
-    // sockets share the same static event_callback, and the socket will be closed
-    // by the caller.
     if (i < this->socket_fds_.size() - 1)
       this->socket_fds_[i] = this->socket_fds_.back();
     this->socket_fds_.pop_back();
-#ifndef USE_LWIP_FAST_SELECT
     this->socket_fds_changed_ = true;
     // Only recalculate max_fd if we removed the current max
     if (fd == this->max_fd_) {
@@ -611,7 +627,6 @@ void Application::unregister_socket_fd(int fd) {
           this->max_fd_ = sock_fd;
       }
     }
-#endif
     return;
   }
 }
@@ -621,7 +636,7 @@ void Application::unregister_socket_fd(int fd) {
 void Application::yield_with_select_(uint32_t delay_ms) {
   // Delay while monitoring sockets. When delay_ms is 0, always yield() to ensure other tasks run.
 #if defined(USE_SOCKET_SELECT_SUPPORT) && defined(USE_LWIP_FAST_SELECT)
-  // Fast path (ESP32/LibreTiny): reads rcvevent directly via lwip_socket_dbg_get_socket().
+  // Fast path (ESP32/LibreTiny): reads rcvevent directly from cached lwip_sock pointers.
   // Safe because this runs on the main loop which owns socket lifetime (create, read, close).
   if (delay_ms == 0) [[unlikely]] {
     yield();
@@ -632,8 +647,8 @@ void Application::yield_with_select_(uint32_t delay_ms) {
   // If a socket still has unread data (rcvevent > 0) but the task notification was already
   // consumed, ulTaskNotifyTake would block until timeout — adding up to delay_ms latency.
   // This scan preserves select() semantics: return immediately when any fd is ready.
-  for (int fd : this->socket_fds_) {
-    if (esphome_lwip_socket_has_data(fd)) {
+  for (struct lwip_sock *sock : this->monitored_sockets_) {
+    if (esphome_lwip_socket_has_data(sock)) {
       yield();
       return;
     }

esphome/core/application.h

@@ -500,14 +500,20 @@ class Application {
 
   Scheduler scheduler;
 
-  /// Register/unregister a socket file descriptor to be monitored for read events.
-#ifdef USE_SOCKET_SELECT_SUPPORT
-  /// These functions update the fd_set used by select() in the main loop.
+  /// Register/unregister a socket to be monitored for read events.
   /// WARNING: These functions are NOT thread-safe. They must only be called from the main loop.
+#ifdef USE_LWIP_FAST_SELECT
+  /// Fast select path: hooks netconn callback and registers for monitoring.
+  /// @return true if registration was successful, false if sock is null
+  bool register_socket(struct lwip_sock *sock);
+  void unregister_socket(struct lwip_sock *sock);
+#elif defined(USE_SOCKET_SELECT_SUPPORT)
+  /// Fallback select() path: monitors file descriptors.
   /// NOTE: File descriptors >= FD_SETSIZE (typically 10 on ESP) will be rejected with an error.
   /// @return true if registration was successful, false if fd exceeds limits
   bool register_socket_fd(int fd);
   void unregister_socket_fd(int fd);
+#endif
 
 #ifdef USE_WAKE_LOOP_THREADSAFE
   /// Wake the main event loop from another FreeRTOS task.
@@ -532,7 +538,6 @@ class Application {
   static void IRAM_ATTR wake_loop_any_context() { esphome_lwip_wake_main_loop_any_context(); }
 #endif
 #endif
-#endif
 
 #if defined(USE_ESP8266) && defined(USE_SOCKET_IMPL_LWIP_TCP)
   /// Wake the main event loop from any context (ISR, thread, or main loop).
@@ -542,23 +547,14 @@ class Application {
 
  protected:
   friend Component;
-#ifdef USE_SOCKET_SELECT_SUPPORT
+#if defined(USE_SOCKET_SELECT_SUPPORT) && !defined(USE_LWIP_FAST_SELECT)
   friend bool socket::socket_ready_fd(int fd, bool loop_monitored);
 #endif
   friend void ::setup();
   friend void ::original_setup();
 
-#ifdef USE_SOCKET_SELECT_SUPPORT
-  /// Fast path for Socket::ready() via friendship - skips negative fd check.
-  /// Main loop only — with USE_LWIP_FAST_SELECT, reads rcvevent via
-  /// lwip_socket_dbg_get_socket(), which has no refcount; safe only because
-  /// the main loop owns socket lifetime (creates, reads, and closes sockets
-  /// on the same thread).
-#ifdef USE_LWIP_FAST_SELECT
-  bool is_socket_ready_(int fd) const { return esphome_lwip_socket_has_data(fd); }
-#else
+#if defined(USE_SOCKET_SELECT_SUPPORT) && !defined(USE_LWIP_FAST_SELECT)
   bool is_socket_ready_(int fd) const { return FD_ISSET(fd, &this->read_fds_); }
-#endif
 #endif
 
   /// Register a component, detecting loop() override at compile time.
@@ -620,8 +616,12 @@ class Application {
   //   and active_end_ is incremented
   // - This eliminates branch mispredictions from flag checking in the hot loop
   FixedVector<Component *> looping_components_{};
-#ifdef USE_SOCKET_SELECT_SUPPORT
+#ifdef USE_LWIP_FAST_SELECT
+  std::vector<struct lwip_sock *> monitored_sockets_;  // Cached lwip_sock pointers for direct rcvevent read
+#elif defined(USE_SOCKET_SELECT_SUPPORT)
   std::vector<int> socket_fds_;  // Vector of all monitored socket file descriptors
+#endif
+#ifdef USE_SOCKET_SELECT_SUPPORT
 #if defined(USE_WAKE_LOOP_THREADSAFE) && !defined(USE_LWIP_FAST_SELECT)
   int wake_socket_fd_{-1};  // Shared wake notification socket for waking main loop from tasks
 #endif

esphome/core/lwip_fast_select.c

@@ -140,8 +140,10 @@
 _Static_assert(sizeof(TaskHandle_t) <= 4, "TaskHandle_t must be <= 4 bytes for atomic access");
 _Static_assert(sizeof(netconn_callback) <= 4, "netconn_callback must be <= 4 bytes for atomic access");
 
-// rcvevent must fit in a single atomic read
-_Static_assert(sizeof(((struct lwip_sock *) 0)->rcvevent) <= 4, "rcvevent must be <= 4 bytes for atomic access");
+// rcvevent must be exactly 2 bytes (s16_t) — the inline in lwip_fast_select.h reads it as int16_t.
+// If lwIP changes this to int or similar, the offset assert would still pass but the load width would be wrong.
+_Static_assert(sizeof(((struct lwip_sock *) 0)->rcvevent) == 2,
+               "rcvevent size changed — update int16_t cast in esphome_lwip_socket_has_data() in lwip_fast_select.h");
 
 // Struct member alignment — natural alignment guarantees atomicity on Xtensa/RISC-V/ARM.
 // Misaligned access would not be atomic even if the size is <= 4 bytes.
@@ -150,6 +152,10 @@ _Static_assert(offsetof(struct netconn, callback) % sizeof(netconn_callback) ==
 _Static_assert(offsetof(struct lwip_sock, rcvevent) % sizeof(((struct lwip_sock *) 0)->rcvevent) == 0,
                "lwip_sock.rcvevent must be naturally aligned for atomic access");
 
+// Verify the hardcoded offset used in the header's inline esphome_lwip_socket_has_data().
+_Static_assert(offsetof(struct lwip_sock, rcvevent) == ESPHOME_LWIP_SOCK_RCVEVENT_OFFSET,
+               "lwip_sock.rcvevent offset changed — update ESPHOME_LWIP_SOCK_RCVEVENT_OFFSET in lwip_fast_select.h");
+
 // Task handle for the main loop — written once in init(), read from TCP/IP and background tasks.
 static TaskHandle_t s_main_loop_task = NULL;
 
@@ -194,23 +200,11 @@ static inline struct lwip_sock *get_sock(int fd) {
   return sock;
 }
 
-bool esphome_lwip_socket_has_data(int fd) {
-  struct lwip_sock *sock = get_sock(fd);
-  if (sock == NULL)
-    return false;
-  // volatile prevents the compiler from caching/reordering this cross-thread read.
-  // The write side (TCP/IP thread) commits via SYS_ARCH_UNPROTECT which releases a
-  // FreeRTOS mutex (ESP32) or resumes the scheduler (LibreTiny), ensuring the value
-  // is visible. Aligned 16-bit reads are single-instruction loads (L16SI/LH/LDRH) on
-  // Xtensa/RISC-V/ARM and cannot produce torn values.
-  return *(volatile s16_t *) &sock->rcvevent > 0;
+struct lwip_sock *esphome_lwip_get_sock(int fd) {
+  return get_sock(fd);
 }
 
-void esphome_lwip_hook_socket(int fd) {
-  struct lwip_sock *sock = get_sock(fd);
-  if (sock == NULL)
-    return;
-
+void esphome_lwip_hook_socket(struct lwip_sock *sock) {
   // Save original callback once — all LwIP sockets share the same static event_callback
   // (DEFAULT_SOCKET_EVENTCB in sockets.c, used for SOCK_RAW, SOCK_DGRAM, and SOCK_STREAM).
   if (s_original_callback == NULL) {

esphome/core/lwip_fast_select.h

@@ -4,6 +4,17 @@
 // Replaces lwip_select() with direct rcvevent reads and FreeRTOS task notifications.
 
 #include <stdbool.h>
+#include <stdint.h>
+
+// Forward declare lwip_sock for C++ callers that store cached pointers.
+// The full definition is only available in the .c file (lwip/priv/sockets_priv.h
+// conflicts with C++ compilation units).
+struct lwip_sock;
+
+// Byte offset of rcvevent (s16_t) within struct lwip_sock.
+// Verified at compile time in lwip_fast_select.c via _Static_assert.
+// Anonymous enum for a compile-time constant that works in both C and C++.
+enum { ESPHOME_LWIP_SOCK_RCVEVENT_OFFSET = 8 };
 
 #ifdef __cplusplus
 extern "C" {
@@ -13,16 +24,38 @@ extern "C" {
 /// Saves the current task handle for xTaskNotifyGive() wake notifications.
 void esphome_lwip_fast_select_init(void);
 
-/// Check if a LwIP socket has data ready via direct rcvevent read (~215 ns per socket).
-/// Uses lwip_socket_dbg_get_socket() — a direct array lookup without the refcount that
-/// get_socket()/done_socket() uses. Safe because the caller owns the socket lifetime:
-/// both has_data reads and socket close/unregister happen on the main loop thread.
-bool esphome_lwip_socket_has_data(int fd);
+/// Look up a LwIP socket struct from a file descriptor.
+/// Returns NULL if fd is invalid or the socket/netconn is not initialized.
+/// Use this at registration time to cache the pointer for esphome_lwip_socket_has_data().
+struct lwip_sock *esphome_lwip_get_sock(int fd);
+
+/// Check if a cached LwIP socket has data ready via unlocked hint read of rcvevent.
+/// This avoids lwIP core lock contention between the main loop (CPU0) and
+/// streaming/networking work (CPU1). Correctness is preserved because callers
+/// already handle EWOULDBLOCK on nonblocking sockets — a stale hint simply causes
+/// a harmless retry on the next loop iteration. In practice, stale reads have not
+/// been observed across multi-day testing, but the design does not depend on that.
+///
+/// The sock pointer must have been obtained from esphome_lwip_get_sock() and must
+/// remain valid (caller owns socket lifetime — no concurrent close).
+/// Hot path: inlined volatile 16-bit load — no function call overhead.
+/// Uses offset-based access because lwip/priv/sockets_priv.h conflicts with C++.
+/// The offset and size are verified at compile time in lwip_fast_select.c.
+static inline bool esphome_lwip_socket_has_data(struct lwip_sock *sock) {
+  // Unlocked hint read — no lwIP core lock needed.
+  // volatile prevents the compiler from caching/reordering this cross-thread read.
+  // The write side (TCP/IP thread) commits via SYS_ARCH_UNPROTECT which releases a
+  // FreeRTOS mutex (ESP32) or resumes the scheduler (LibreTiny), ensuring the value
+  // is visible. Aligned 16-bit reads are single-instruction loads (L16SI/LH/LDRH) on
+  // Xtensa/RISC-V/ARM and cannot produce torn values.
+  return *(volatile int16_t *) ((char *) sock + (int) ESPHOME_LWIP_SOCK_RCVEVENT_OFFSET) > 0;
+}
 
 /// Hook a socket's netconn callback to notify the main loop task on receive events.
 /// Wraps the original event_callback with one that also calls xTaskNotifyGive().
 /// Must be called from the main loop after socket creation.
-void esphome_lwip_hook_socket(int fd);
+/// The sock pointer must have been obtained from esphome_lwip_get_sock().
+void esphome_lwip_hook_socket(struct lwip_sock *sock);
 
 /// Wake the main loop task from another FreeRTOS task — costs <1 us.
 /// NOT ISR-safe — must only be called from task context.