soc/intel_adsp/ace: custom arch_spin_relax and some syscalls related changes.

soc/intel/intel_adsp/ace/CMakeLists.txt

@@ -18,6 +18,10 @@ zephyr_include_directories(include)
 zephyr_include_directories(include/${CONFIG_SOC})
 zephyr_library_sources_ifdef(CONFIG_SOC_INTEL_COMM_WIDGET comm_widget.c)
 zephyr_library_sources_ifdef(CONFIG_SOC_INTEL_COMM_WIDGET comm_widget_messages.c)
+zephyr_library_sources_ifdef(
+  CONFIG_SOC_SERIES_INTEL_ADSP_ACE_CUSTOM_MORE_SPIN_RELAX_NOPS
+  spin_relax.c
+)
 
 if (CONFIG_XTENSA_MMU)
   zephyr_library_sources_ifdef(CONFIG_SOC_INTEL_ACE30 mmu_ace30.c)

soc/intel/intel_adsp/ace/Kconfig

@@ -31,3 +31,24 @@ config SRAM_RETENTION_MODE
 	  When this option is enabled, the SRAM retention mode will be
 	  activated during the firmware boot-up process. If disabled,
 	  the retention mode will not be activated.
+
+config SOC_SERIES_INTEL_ADSP_ACE_CUSTOM_MORE_SPIN_RELAX_NOPS
+	bool "Use Intel Audio DSP specific arch_spin_relax() with more NOPs"
+	depends on !XTENSA_MORE_SPIN_RELAX_NOPS
+	default y if SMP && MP_MAX_NUM_CPUS > 1
+	help
+	  Add some NOPs after failure to lock a spinlock. This gives
+	  the bus extra time to synchronize the RCW transaction
+	  among CPUs.
+
+config SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS
+	int "Number of NOPs to be used in Intel Audio DSP specific arch_spin_relax()"
+	depends on SOC_SERIES_INTEL_ADSP_ACE_CUSTOM_MORE_SPIN_RELAX_NOPS
+	default 32 if MP_MAX_NUM_CPUS = 1
+	default 64 if MP_MAX_NUM_CPUS = 2
+	default 96 if MP_MAX_NUM_CPUS = 3
+	default 128 if MP_MAX_NUM_CPUS = 4
+	default 160 if MP_MAX_NUM_CPUS = 5
+	help
+	  Specify the number of NOPs in Intel Audio DSP specific
+	  arch_spin_relax().

soc/intel/intel_adsp/ace/Kconfig.defconfig.series

@@ -57,20 +57,6 @@ config LOG_BACKEND_ADSP
 
 endif # LOG
 
-config XTENSA_MORE_SPIN_RELAX_NOPS
-	default y if SMP && MP_MAX_NUM_CPUS > 1
-
-if XTENSA_MORE_SPIN_RELAX_NOPS
-
-config XTENSA_NUM_SPIN_RELAX_NOPS
-	default 32 if MP_MAX_NUM_CPUS = 1
-	default 64 if MP_MAX_NUM_CPUS = 2
-	default 96 if MP_MAX_NUM_CPUS = 3
-	default 128 if MP_MAX_NUM_CPUS = 4
-	default 160 if MP_MAX_NUM_CPUS = 5
-
-endif # XTENSA_MORE_SPIN_RELAX_NOPS
-
 if KERNEL_VM_SUPPORT
 
 config KERNEL_VM_SIZE

soc/intel/intel_adsp/ace/ace-link.ld

@@ -297,6 +297,8 @@ SECTIONS {
 
     *libarch__xtensa__core.a:userspace.S.obj(.literal.xtensa_do_syscall)
     *libarch__xtensa__core.a:userspace.S.obj(.text.xtensa_do_syscall)
+    *libarch__xtensa__core.a:ptables.c.obj(.literal.xtensa_swap_update_page_tables)
+    *libarch__xtensa__core.a:ptables.c.obj(.text.xtensa_swap_update_page_tables)
 
 #ifdef CONFIG_XTENSA_SYSCALL_USE_HELPER
     *libarch__xtensa__core.a:syscall_helper.c.obj(.text.xtensa_syscall_helper*)
@@ -503,6 +505,24 @@ SECTIONS {
   .bss SEGSTART_UNCACHED (NOLOAD) :
   {
     _bss_start = .;
+
+#if defined(CONFIG_ZTEST) && defined(CONFIG_SIMULATOR_XTENSA)
+    /* For some weird unknown reasons, the simulator really do not
+     * like these cpuhold_* variables to be tightly packed together.
+     * This results in cpuhold_spawned not being updated, and we
+     * will be stuck in the while loop for it to be set.
+     * Workaround by explicitly aligning these variables.
+     */
+    . = ALIGN(16);
+    *:ztest.c.obj(.bss.cpuhold_sem)
+    . = ALIGN(16);
+    *:ztest.c.obj(.bss.cpuhold_active)
+    . = ALIGN(16);
+    *:ztest.c.obj(.bss.cpuhold_spawned)
+    . = ALIGN(16);
+    *:ztest.c.obj(.bss.cpuhold_pool_items)
+#endif /* CONFIG_ZTEST && CONFIG_SIMULATOR_XTENSA */
+
     *(.dynsbss)
     *(.sbss)
     *(.sbss.*)

soc/intel/intel_adsp/ace/spin_relax.c

@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2025 Intel Corporation
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include <stdint.h>
+
+#include <zephyr/toolchain.h>
+#include <zephyr/sys/util_macro.h>
+
+#ifdef CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS
+void arch_spin_relax(void)
+{
+	register uint32_t remaining = CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS;
+
+	while (remaining > 0) {
+#if (CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS % 4) == 0
+		remaining -= 4;
+
+		/*
+		 * Note the xcc/xt-clang likes to "truncate"
+		 * continuous NOPs to max 4 NOPs. So this is
+		 * the most we can do in one loop.
+		 */
+		__asm__("nop.n; nop.n; nop.n; nop.n;");
+#else
+		remaining--;
+		__asm__("nop.n");
+#endif
+	}
+}
+#endif /* CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS */