diff --git a/components/esp_hw_support/lowpower/cpu_retention/port/esp32p4/sleep_cpu.c b/components/esp_hw_support/lowpower/cpu_retention/port/esp32p4/sleep_cpu.c
index f27af1d5dc..f907f9ad32 100644
--- a/components/esp_hw_support/lowpower/cpu_retention/port/esp32p4/sleep_cpu.c
+++ b/components/esp_hw_support/lowpower/cpu_retention/port/esp32p4/sleep_cpu.c
@@ -586,6 +586,7 @@ void sleep_smp_cpu_sleep_prepare(void)
     esp_ipc_isr_call((esp_ipc_isr_func_t)smp_core_do_retention, NULL);
 #else
     esp_ipc_isr_stall_other_cpu();
+    esp_ipc_isr_stall_pause();
 #endif
 }
 
@@ -600,6 +601,7 @@ void sleep_smp_cpu_wakeup_prepare(void)
     }
     atomic_store(&s_smp_retention_state[core_id], SMP_IDLE);
 #else
+    esp_ipc_isr_stall_resume();
     esp_ipc_isr_release_other_cpu();
 #endif
 }
diff --git a/components/esp_hw_support/mspi_timing_tuning.c b/components/esp_hw_support/mspi_timing_tuning.c
index 239d3b30de..bfab81120a 100644
--- a/components/esp_hw_support/mspi_timing_tuning.c
+++ b/components/esp_hw_support/mspi_timing_tuning.c
@@ -28,6 +28,10 @@
 #include "hal/spimem_flash_ll.h"
 #endif
 
+#if SOC_CACHE_INTERNAL_MEM_VIA_L1CACHE
+#include "esp_ipc_isr.h"
+#endif
+
 #if CONFIG_ESPTOOLPY_FLASHFREQ_120M
 #define FLASH_FREQUENCY_MHZ 120
 #elif CONFIG_ESPTOOLPY_FLASHFREQ_80M
@@ -517,6 +521,13 @@ void mspi_timing_enter_high_speed_mode(bool control_spi1)
 
 void mspi_timing_change_speed_mode_cache_safe(bool switch_down)
 {
+#if SOC_CACHE_INTERNAL_MEM_VIA_L1CACHE && !CONFIG_FREERTOS_UNICORE
+    // For esp chips with two levels of Cache, if another core attempts to access SPI Flash or PSRAM after the
+    // cache is freeze, the access will fail and will keep retrying. This will completely block the L1 Cache,
+    // causing the current core to be unable to access the stack and data in the L2 RAM, which will causes a
+    // deadlock, so we need to stall another core at first.
+    esp_ipc_isr_stall_other_cpu();
+#endif
     /**
      * If a no-cache-freeze-supported chip needs timing tuning, add a protection way:
      * - spinlock
@@ -539,6 +550,10 @@ void mspi_timing_change_speed_mode_cache_safe(bool switch_down)
 #if SOC_CACHE_FREEZE_SUPPORTED
     cache_hal_unfreeze(CACHE_LL_LEVEL_EXT_MEM, CACHE_TYPE_ALL);
 #endif  //#if SOC_CACHE_FREEZE_SUPPORTED
+
+#if SOC_CACHE_INTERNAL_MEM_VIA_L1CACHE && !CONFIG_FREERTOS_UNICORE
+    esp_ipc_isr_release_other_cpu();
+#endif
 }
 
 /*------------------------------------------------------------------------------
diff --git a/components/esp_hw_support/sleep_modes.c b/components/esp_hw_support/sleep_modes.c
index 312d7b004b..027e7a675a 100644
--- a/components/esp_hw_support/sleep_modes.c
+++ b/components/esp_hw_support/sleep_modes.c
@@ -1030,6 +1030,7 @@ static esp_err_t IRAM_ATTR deep_sleep_start(bool allow_sleep_rejection)
      */
     portENTER_CRITICAL(&spinlock_rtc_deep_sleep);
     esp_ipc_isr_stall_other_cpu();
+    esp_ipc_isr_stall_pause();
 
     // record current RTC time
     s_config.rtc_ticks_at_sleep_start = rtc_time_get();
@@ -1096,6 +1097,7 @@ static esp_err_t IRAM_ATTR deep_sleep_start(bool allow_sleep_rejection)
         }
     }
     // Never returns here, except that the sleep is rejected.
+    esp_ipc_isr_stall_resume();
     esp_ipc_isr_release_other_cpu();
     portEXIT_CRITICAL(&spinlock_rtc_deep_sleep);
     return err;
@@ -1236,6 +1238,7 @@ esp_err_t esp_light_sleep_start(void)
     sleep_smp_cpu_sleep_prepare();
 #else
     esp_ipc_isr_stall_other_cpu();
+    esp_ipc_isr_stall_pause();
 #endif
 #endif
 
@@ -1389,6 +1392,7 @@ esp_err_t esp_light_sleep_start(void)
 #if CONFIG_PM_POWER_DOWN_CPU_IN_LIGHT_SLEEP && SOC_PM_CPU_RETENTION_BY_SW
     sleep_smp_cpu_wakeup_prepare();
 #else
+    esp_ipc_isr_stall_resume();
     esp_ipc_isr_release_other_cpu();
 #endif
 #endif
diff --git a/components/esp_pm/test_apps/esp_pm/main/test_pm.c b/components/esp_pm/test_apps/esp_pm/main/test_pm.c
index d65bf44c0f..e52c786c4e 100644
--- a/components/esp_pm/test_apps/esp_pm/main/test_pm.c
+++ b/components/esp_pm/test_apps/esp_pm/main/test_pm.c
@@ -241,7 +241,6 @@ TEST_CASE("Can wake up from automatic light sleep by GPIO", "[pm][ignore]")
 #endif //!TEMPORARY_DISABLED_FOR_TARGETS(ESP32S2, ESP32S3)
 #endif //CONFIG_ULP_COPROC_TYPE_FSM
 
-#if !TEMPORARY_DISABLED_FOR_TARGETS(ESP32P4) //TODO: IDF-9628
 typedef struct {
     int delay_us;
     int result;
@@ -295,7 +294,6 @@ TEST_CASE("vTaskDelay duration is correct with light sleep enabled", "[pm]")
 
     light_sleep_disable();
 }
-#endif
 
 /* This test is similar to the one in test_esp_timer.c, but since we can't use
  * ref_clock, this test uses RTC clock for timing. Also enables automatic