diff options
Diffstat (limited to 'arch')
59 files changed, 2374 insertions, 365 deletions
diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig index 32b9da4622ce..0c7fb52727ba 100644 --- a/arch/loongarch/Kconfig +++ b/arch/loongarch/Kconfig @@ -5,6 +5,7 @@ config LOONGARCH select ACPI select ACPI_GENERIC_GSI if ACPI select ACPI_MCFG if ACPI + select ACPI_PPTT if ACPI select ACPI_SYSTEM_POWER_STATES_SUPPORT if ACPI select ARCH_BINFMT_ELF_STATE select ARCH_ENABLE_MEMORY_HOTPLUG @@ -49,6 +50,8 @@ config LOONGARCH select ARCH_SUPPORTS_ACPI select ARCH_SUPPORTS_ATOMIC_RMW select ARCH_SUPPORTS_HUGETLBFS + select ARCH_SUPPORTS_LTO_CLANG + select ARCH_SUPPORTS_LTO_CLANG_THIN select ARCH_SUPPORTS_NUMA_BALANCING select ARCH_USE_BUILTIN_BSWAP select ARCH_USE_CMPXCHG_LOCKREF @@ -81,9 +84,12 @@ config LOONGARCH select GENERIC_SCHED_CLOCK select GENERIC_SMP_IDLE_THREAD select GENERIC_TIME_VSYSCALL + select GENERIC_VDSO_TIME_NS select GPIOLIB select HAS_IOPORT select HAVE_ARCH_AUDITSYSCALL + select HAVE_ARCH_JUMP_LABEL + select HAVE_ARCH_JUMP_LABEL_RELATIVE select HAVE_ARCH_MMAP_RND_BITS if MMU select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_TRACEHOOK @@ -91,6 +97,7 @@ config LOONGARCH select HAVE_ASM_MODVERSIONS select HAVE_CONTEXT_TRACKING_USER select HAVE_C_RECORDMCOUNT + select HAVE_DEBUG_KMEMLEAK select HAVE_DEBUG_STACKOVERFLOW select HAVE_DMA_CONTIGUOUS select HAVE_DYNAMIC_FTRACE @@ -121,6 +128,7 @@ config LOONGARCH select HAVE_PERF_REGS select HAVE_PERF_USER_STACK_DUMP select HAVE_REGS_AND_STACK_ACCESS_API + select HAVE_RETHOOK select HAVE_RSEQ select HAVE_SAMPLE_FTRACE_DIRECT select HAVE_SAMPLE_FTRACE_DIRECT_MULTI @@ -163,14 +171,6 @@ config 32BIT config 64BIT def_bool y -config CPU_HAS_FPU - bool - default y - -config CPU_HAS_PREFETCH - bool - default y - config GENERIC_BUG def_bool y depends on BUG @@ -243,6 +243,15 @@ config SCHED_OMIT_FRAME_POINTER config AS_HAS_EXPLICIT_RELOCS def_bool $(as-instr,x:pcalau12i \$t0$(comma)%pc_hi20(x)) +config AS_HAS_FCSR_CLASS + def_bool $(as-instr,movfcsr2gr \$t0$(comma)\$fcsr0) + +config AS_HAS_LSX_EXTENSION + def_bool $(as-instr,vld \$vr0$(comma)\$a0$(comma)0) + +config AS_HAS_LASX_EXTENSION + def_bool $(as-instr,xvld \$xr0$(comma)\$a0$(comma)0) + menu "Kernel type and options" source "kernel/Kconfig.hz" @@ -374,6 +383,13 @@ config EFI_STUB This kernel feature allows the kernel to be loaded directly by EFI firmware without the use of a bootloader. +config SCHED_SMT + bool "SMT scheduler support" + default y + help + Improves scheduler's performance when there are multiple + threads in one physical core. + config SMP bool "Multi-Processing support" help @@ -483,6 +499,43 @@ config ARCH_STRICT_ALIGN to run kernel only on systems with h/w unaligned access support in order to optimise for performance. +config CPU_HAS_FPU + bool + default y + +config CPU_HAS_LSX + bool "Support for the Loongson SIMD Extension" + depends on AS_HAS_LSX_EXTENSION + help + Loongson SIMD Extension (LSX) introduces 128 bit wide vector registers + and a set of SIMD instructions to operate on them. When this option + is enabled the kernel will support allocating & switching LSX + vector register contexts. If you know that your kernel will only be + running on CPUs which do not support LSX or that your userland will + not be making use of it then you may wish to say N here to reduce + the size & complexity of your kernel. + + If unsure, say Y. + +config CPU_HAS_LASX + bool "Support for the Loongson Advanced SIMD Extension" + depends on CPU_HAS_LSX + depends on AS_HAS_LASX_EXTENSION + help + Loongson Advanced SIMD Extension (LASX) introduces 256 bit wide vector + registers and a set of SIMD instructions to operate on them. When this + option is enabled the kernel will support allocating & switching LASX + vector register contexts. If you know that your kernel will only be + running on CPUs which do not support LASX or that your userland will + not be making use of it then you may wish to say N here to reduce + the size & complexity of your kernel. + + If unsure, say Y. + +config CPU_HAS_PREFETCH + bool + default y + config KEXEC bool "Kexec system call" select KEXEC_CORE @@ -592,6 +645,9 @@ config ARCH_MMAP_RND_BITS_MIN config ARCH_MMAP_RND_BITS_MAX default 18 +config ARCH_SUPPORTS_UPROBES + def_bool y + menu "Power management options" config ARCH_SUSPEND_POSSIBLE diff --git a/arch/loongarch/Makefile b/arch/loongarch/Makefile index a27e264bdaa5..09ba338a64de 100644 --- a/arch/loongarch/Makefile +++ b/arch/loongarch/Makefile @@ -46,8 +46,8 @@ ld-emul = $(64bit-emul) cflags-y += -mabi=lp64s endif -cflags-y += -G0 -pipe -msoft-float -LDFLAGS_vmlinux += -G0 -static -n -nostdlib +cflags-y += -pipe -msoft-float +LDFLAGS_vmlinux += -static -n -nostdlib # When the assembler supports explicit relocation hint, we must use it. # GCC may have -mexplicit-relocs off by default if it was built with an old @@ -56,13 +56,18 @@ LDFLAGS_vmlinux += -G0 -static -n -nostdlib # When the assembler does not supports explicit relocation hint, we can't use # it. Disable it if the compiler supports it. # -# If you've seen "unknown reloc hint" message building the kernel and you are -# now wondering why "-mexplicit-relocs" is not wrapped with cc-option: the -# combination of a "new" assembler and "old" compiler is not supported. Either -# upgrade the compiler or downgrade the assembler. +# The combination of a "new" assembler and "old" GCC is not supported, given +# the rarity of this combo and the extra complexity needed to make it work. +# Either upgrade the compiler or downgrade the assembler; the build will error +# out if it is the case (by probing for the model attribute; all supported +# compilers in this case would have support). +# +# Also, -mdirect-extern-access is useful in case of building with explicit +# relocs, for avoiding unnecessary GOT accesses. It is harmless to not have +# support though. ifdef CONFIG_AS_HAS_EXPLICIT_RELOCS -cflags-y += -mexplicit-relocs -KBUILD_CFLAGS_KERNEL += -mdirect-extern-access +cflags-y += $(call cc-option,-mexplicit-relocs) +KBUILD_CFLAGS_KERNEL += $(call cc-option,-mdirect-extern-access) else cflags-y += $(call cc-option,-mno-explicit-relocs) KBUILD_AFLAGS_KERNEL += -Wa,-mla-global-with-pcrel @@ -107,7 +112,7 @@ KBUILD_CFLAGS += -isystem $(shell $(CC) -print-file-name=include) KBUILD_LDFLAGS += -m $(ld-emul) ifdef CONFIG_LOONGARCH -CHECKFLAGS += $(shell $(CC) $(KBUILD_CFLAGS) -dM -E -x c /dev/null | \ +CHECKFLAGS += $(shell $(CC) $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS) -dM -E -x c /dev/null | \ grep -E -vw '__GNUC_(MINOR_|PATCHLEVEL_)?_' | \ sed -e "s/^\#define /-D'/" -e "s/ /'='/" -e "s/$$/'/" -e 's/\$$/&&/g') endif diff --git a/arch/loongarch/include/asm/Kbuild b/arch/loongarch/include/asm/Kbuild index 77ad8e6f0906..6b222f227342 100644 --- a/arch/loongarch/include/asm/Kbuild +++ b/arch/loongarch/include/asm/Kbuild @@ -5,7 +5,6 @@ generic-y += mcs_spinlock.h generic-y += parport.h generic-y += early_ioremap.h generic-y += qrwlock.h -generic-y += qspinlock.h generic-y += rwsem.h generic-y += segment.h generic-y += user.h diff --git a/arch/loongarch/include/asm/acpi.h b/arch/loongarch/include/asm/acpi.h index 976a810352c6..8de6c4b83a61 100644 --- a/arch/loongarch/include/asm/acpi.h +++ b/arch/loongarch/include/asm/acpi.h @@ -8,11 +8,14 @@ #ifndef _ASM_LOONGARCH_ACPI_H #define _ASM_LOONGARCH_ACPI_H +#include <asm/suspend.h> + #ifdef CONFIG_ACPI extern int acpi_strict; extern int acpi_disabled; extern int acpi_pci_disabled; extern int acpi_noirq; +extern int pptt_enabled; #define acpi_os_ioremap acpi_os_ioremap void __iomem *acpi_os_ioremap(acpi_physical_address phys, acpi_size size); @@ -30,6 +33,14 @@ static inline bool acpi_has_cpu_in_madt(void) } extern struct list_head acpi_wakeup_device_list; +extern struct acpi_madt_core_pic acpi_core_pic[NR_CPUS]; + +extern int __init parse_acpi_topology(void); + +static inline u32 get_acpi_id_for_cpu(unsigned int cpu) +{ + return acpi_core_pic[cpu_logical_map(cpu)].processor_id; +} #endif /* !CONFIG_ACPI */ @@ -37,12 +48,10 @@ extern struct list_head acpi_wakeup_device_list; extern int loongarch_acpi_suspend(void); extern int (*acpi_suspend_lowlevel)(void); -extern void loongarch_suspend_enter(void); static inline unsigned long acpi_get_wakeup_address(void) { #ifdef CONFIG_SUSPEND - extern void loongarch_wakeup_start(void); return (unsigned long)loongarch_wakeup_start; #endif return 0UL; diff --git a/arch/loongarch/include/asm/asmmacro.h b/arch/loongarch/include/asm/asmmacro.h index c51a1b43acb4..79e1d53fea89 100644 --- a/arch/loongarch/include/asm/asmmacro.h +++ b/arch/loongarch/include/asm/asmmacro.h @@ -270,6 +270,399 @@ fld.d $f31, \tmp, THREAD_FPR31 - THREAD_FPR0 .endm + .macro lsx_save_data thread tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \tmp, \thread, \tmp + vst $vr0, \tmp, THREAD_FPR0 - THREAD_FPR0 + vst $vr1, \tmp, THREAD_FPR1 - THREAD_FPR0 + vst $vr2, \tmp, THREAD_FPR2 - THREAD_FPR0 + vst $vr3, \tmp, THREAD_FPR3 - THREAD_FPR0 + vst $vr4, \tmp, THREAD_FPR4 - THREAD_FPR0 + vst $vr5, \tmp, THREAD_FPR5 - THREAD_FPR0 + vst $vr6, \tmp, THREAD_FPR6 - THREAD_FPR0 + vst $vr7, \tmp, THREAD_FPR7 - THREAD_FPR0 + vst $vr8, \tmp, THREAD_FPR8 - THREAD_FPR0 + vst $vr9, \tmp, THREAD_FPR9 - THREAD_FPR0 + vst $vr10, \tmp, THREAD_FPR10 - THREAD_FPR0 + vst $vr11, \tmp, THREAD_FPR11 - THREAD_FPR0 + vst $vr12, \tmp, THREAD_FPR12 - THREAD_FPR0 + vst $vr13, \tmp, THREAD_FPR13 - THREAD_FPR0 + vst $vr14, \tmp, THREAD_FPR14 - THREAD_FPR0 + vst $vr15, \tmp, THREAD_FPR15 - THREAD_FPR0 + vst $vr16, \tmp, THREAD_FPR16 - THREAD_FPR0 + vst $vr17, \tmp, THREAD_FPR17 - THREAD_FPR0 + vst $vr18, \tmp, THREAD_FPR18 - THREAD_FPR0 + vst $vr19, \tmp, THREAD_FPR19 - THREAD_FPR0 + vst $vr20, \tmp, THREAD_FPR20 - THREAD_FPR0 + vst $vr21, \tmp, THREAD_FPR21 - THREAD_FPR0 + vst $vr22, \tmp, THREAD_FPR22 - THREAD_FPR0 + vst $vr23, \tmp, THREAD_FPR23 - THREAD_FPR0 + vst $vr24, \tmp, THREAD_FPR24 - THREAD_FPR0 + vst $vr25, \tmp, THREAD_FPR25 - THREAD_FPR0 + vst $vr26, \tmp, THREAD_FPR26 - THREAD_FPR0 + vst $vr27, \tmp, THREAD_FPR27 - THREAD_FPR0 + vst $vr28, \tmp, THREAD_FPR28 - THREAD_FPR0 + vst $vr29, \tmp, THREAD_FPR29 - THREAD_FPR0 + vst $vr30, \tmp, THREAD_FPR30 - THREAD_FPR0 + vst $vr31, \tmp, THREAD_FPR31 - THREAD_FPR0 + .endm + + .macro lsx_restore_data thread tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \tmp, \thread, \tmp + vld $vr0, \tmp, THREAD_FPR0 - THREAD_FPR0 + vld $vr1, \tmp, THREAD_FPR1 - THREAD_FPR0 + vld $vr2, \tmp, THREAD_FPR2 - THREAD_FPR0 + vld $vr3, \tmp, THREAD_FPR3 - THREAD_FPR0 + vld $vr4, \tmp, THREAD_FPR4 - THREAD_FPR0 + vld $vr5, \tmp, THREAD_FPR5 - THREAD_FPR0 + vld $vr6, \tmp, THREAD_FPR6 - THREAD_FPR0 + vld $vr7, \tmp, THREAD_FPR7 - THREAD_FPR0 + vld $vr8, \tmp, THREAD_FPR8 - THREAD_FPR0 + vld $vr9, \tmp, THREAD_FPR9 - THREAD_FPR0 + vld $vr10, \tmp, THREAD_FPR10 - THREAD_FPR0 + vld $vr11, \tmp, THREAD_FPR11 - THREAD_FPR0 + vld $vr12, \tmp, THREAD_FPR12 - THREAD_FPR0 + vld $vr13, \tmp, THREAD_FPR13 - THREAD_FPR0 + vld $vr14, \tmp, THREAD_FPR14 - THREAD_FPR0 + vld $vr15, \tmp, THREAD_FPR15 - THREAD_FPR0 + vld $vr16, \tmp, THREAD_FPR16 - THREAD_FPR0 + vld $vr17, \tmp, THREAD_FPR17 - THREAD_FPR0 + vld $vr18, \tmp, THREAD_FPR18 - THREAD_FPR0 + vld $vr19, \tmp, THREAD_FPR19 - THREAD_FPR0 + vld $vr20, \tmp, THREAD_FPR20 - THREAD_FPR0 + vld $vr21, \tmp, THREAD_FPR21 - THREAD_FPR0 + vld $vr22, \tmp, THREAD_FPR22 - THREAD_FPR0 + vld $vr23, \tmp, THREAD_FPR23 - THREAD_FPR0 + vld $vr24, \tmp, THREAD_FPR24 - THREAD_FPR0 + vld $vr25, \tmp, THREAD_FPR25 - THREAD_FPR0 + vld $vr26, \tmp, THREAD_FPR26 - THREAD_FPR0 + vld $vr27, \tmp, THREAD_FPR27 - THREAD_FPR0 + vld $vr28, \tmp, THREAD_FPR28 - THREAD_FPR0 + vld $vr29, \tmp, THREAD_FPR29 - THREAD_FPR0 + vld $vr30, \tmp, THREAD_FPR30 - THREAD_FPR0 + vld $vr31, \tmp, THREAD_FPR31 - THREAD_FPR0 + .endm + + .macro lsx_save_all thread tmp0 tmp1 + fpu_save_cc \thread, \tmp0, \tmp1 + fpu_save_csr \thread, \tmp0 + lsx_save_data \thread, \tmp0 + .endm + + .macro lsx_restore_all thread tmp0 tmp1 + lsx_restore_data \thread, \tmp0 + fpu_restore_cc \thread, \tmp0, \tmp1 + fpu_restore_csr \thread, \tmp0 + .endm + + .macro lsx_save_upper vd base tmp off + vpickve2gr.d \tmp, \vd, 1 + st.d \tmp, \base, (\off+8) + .endm + + .macro lsx_save_all_upper thread base tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \base, \thread, \tmp + lsx_save_upper $vr0, \base, \tmp, (THREAD_FPR0-THREAD_FPR0) + lsx_save_upper $vr1, \base, \tmp, (THREAD_FPR1-THREAD_FPR0) + lsx_save_upper $vr2, \base, \tmp, (THREAD_FPR2-THREAD_FPR0) + lsx_save_upper $vr3, \base, \tmp, (THREAD_FPR3-THREAD_FPR0) + lsx_save_upper $vr4, \base, \tmp, (THREAD_FPR4-THREAD_FPR0) + lsx_save_upper $vr5, \base, \tmp, (THREAD_FPR5-THREAD_FPR0) + lsx_save_upper $vr6, \base, \tmp, (THREAD_FPR6-THREAD_FPR0) + lsx_save_upper $vr7, \base, \tmp, (THREAD_FPR7-THREAD_FPR0) + lsx_save_upper $vr8, \base, \tmp, (THREAD_FPR8-THREAD_FPR0) + lsx_save_upper $vr9, \base, \tmp, (THREAD_FPR9-THREAD_FPR0) + lsx_save_upper $vr10, \base, \tmp, (THREAD_FPR10-THREAD_FPR0) + lsx_save_upper $vr11, \base, \tmp, (THREAD_FPR11-THREAD_FPR0) + lsx_save_upper $vr12, \base, \tmp, (THREAD_FPR12-THREAD_FPR0) + lsx_save_upper $vr13, \base, \tmp, (THREAD_FPR13-THREAD_FPR0) + lsx_save_upper $vr14, \base, \tmp, (THREAD_FPR14-THREAD_FPR0) + lsx_save_upper $vr15, \base, \tmp, (THREAD_FPR15-THREAD_FPR0) + lsx_save_upper $vr16, \base, \tmp, (THREAD_FPR16-THREAD_FPR0) + lsx_save_upper $vr17, \base, \tmp, (THREAD_FPR17-THREAD_FPR0) + lsx_save_upper $vr18, \base, \tmp, (THREAD_FPR18-THREAD_FPR0) + lsx_save_upper $vr19, \base, \tmp, (THREAD_FPR19-THREAD_FPR0) + lsx_save_upper $vr20, \base, \tmp, (THREAD_FPR20-THREAD_FPR0) + lsx_save_upper $vr21, \base, \tmp, (THREAD_FPR21-THREAD_FPR0) + lsx_save_upper $vr22, \base, \tmp, (THREAD_FPR22-THREAD_FPR0) + lsx_save_upper $vr23, \base, \tmp, (THREAD_FPR23-THREAD_FPR0) + lsx_save_upper $vr24, \base, \tmp, (THREAD_FPR24-THREAD_FPR0) + lsx_save_upper $vr25, \base, \tmp, (THREAD_FPR25-THREAD_FPR0) + lsx_save_upper $vr26, \base, \tmp, (THREAD_FPR26-THREAD_FPR0) + lsx_save_upper $vr27, \base, \tmp, (THREAD_FPR27-THREAD_FPR0) + lsx_save_upper $vr28, \base, \tmp, (THREAD_FPR28-THREAD_FPR0) + lsx_save_upper $vr29, \base, \tmp, (THREAD_FPR29-THREAD_FPR0) + lsx_save_upper $vr30, \base, \tmp, (THREAD_FPR30-THREAD_FPR0) + lsx_save_upper $vr31, \base, \tmp, (THREAD_FPR31-THREAD_FPR0) + .endm + + .macro lsx_restore_upper vd base tmp off + ld.d \tmp, \base, (\off+8) + vinsgr2vr.d \vd, \tmp, 1 + .endm + + .macro lsx_restore_all_upper thread base tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \base, \thread, \tmp + lsx_restore_upper $vr0, \base, \tmp, (THREAD_FPR0-THREAD_FPR0) + lsx_restore_upper $vr1, \base, \tmp, (THREAD_FPR1-THREAD_FPR0) + lsx_restore_upper $vr2, \base, \tmp, (THREAD_FPR2-THREAD_FPR0) + lsx_restore_upper $vr3, \base, \tmp, (THREAD_FPR3-THREAD_FPR0) + lsx_restore_upper $vr4, \base, \tmp, (THREAD_FPR4-THREAD_FPR0) + lsx_restore_upper $vr5, \base, \tmp, (THREAD_FPR5-THREAD_FPR0) + lsx_restore_upper $vr6, \base, \tmp, (THREAD_FPR6-THREAD_FPR0) + lsx_restore_upper $vr7, \base, \tmp, (THREAD_FPR7-THREAD_FPR0) + lsx_restore_upper $vr8, \base, \tmp, (THREAD_FPR8-THREAD_FPR0) + lsx_restore_upper $vr9, \base, \tmp, (THREAD_FPR9-THREAD_FPR0) + lsx_restore_upper $vr10, \base, \tmp, (THREAD_FPR10-THREAD_FPR0) + lsx_restore_upper $vr11, \base, \tmp, (THREAD_FPR11-THREAD_FPR0) + lsx_restore_upper $vr12, \base, \tmp, (THREAD_FPR12-THREAD_FPR0) + lsx_restore_upper $vr13, \base, \tmp, (THREAD_FPR13-THREAD_FPR0) + lsx_restore_upper $vr14, \base, \tmp, (THREAD_FPR14-THREAD_FPR0) + lsx_restore_upper $vr15, \base, \tmp, (THREAD_FPR15-THREAD_FPR0) + lsx_restore_upper $vr16, \base, \tmp, (THREAD_FPR16-THREAD_FPR0) + lsx_restore_upper $vr17, \base, \tmp, (THREAD_FPR17-THREAD_FPR0) + lsx_restore_upper $vr18, \base, \tmp, (THREAD_FPR18-THREAD_FPR0) + lsx_restore_upper $vr19, \base, \tmp, (THREAD_FPR19-THREAD_FPR0) + lsx_restore_upper $vr20, \base, \tmp, (THREAD_FPR20-THREAD_FPR0) + lsx_restore_upper $vr21, \base, \tmp, (THREAD_FPR21-THREAD_FPR0) + lsx_restore_upper $vr22, \base, \tmp, (THREAD_FPR22-THREAD_FPR0) + lsx_restore_upper $vr23, \base, \tmp, (THREAD_FPR23-THREAD_FPR0) + lsx_restore_upper $vr24, \base, \tmp, (THREAD_FPR24-THREAD_FPR0) + lsx_restore_upper $vr25, \base, \tmp, (THREAD_FPR25-THREAD_FPR0) + lsx_restore_upper $vr26, \base, \tmp, (THREAD_FPR26-THREAD_FPR0) + lsx_restore_upper $vr27, \base, \tmp, (THREAD_FPR27-THREAD_FPR0) + lsx_restore_upper $vr28, \base, \tmp, (THREAD_FPR28-THREAD_FPR0) + lsx_restore_upper $vr29, \base, \tmp, (THREAD_FPR29-THREAD_FPR0) + lsx_restore_upper $vr30, \base, \tmp, (THREAD_FPR30-THREAD_FPR0) + lsx_restore_upper $vr31, \base, \tmp, (THREAD_FPR31-THREAD_FPR0) + .endm + + .macro lsx_init_upper vd tmp + vinsgr2vr.d \vd, \tmp, 1 + .endm + + .macro lsx_init_all_upper tmp + not \tmp, zero + lsx_init_upper $vr0 \tmp + lsx_init_upper $vr1 \tmp + lsx_init_upper $vr2 \tmp + lsx_init_upper $vr3 \tmp + lsx_init_upper $vr4 \tmp + lsx_init_upper $vr5 \tmp + lsx_init_upper $vr6 \tmp + lsx_init_upper $vr7 \tmp + lsx_init_upper $vr8 \tmp + lsx_init_upper $vr9 \tmp + lsx_init_upper $vr10 \tmp + lsx_init_upper $vr11 \tmp + lsx_init_upper $vr12 \tmp + lsx_init_upper $vr13 \tmp + lsx_init_upper $vr14 \tmp + lsx_init_upper $vr15 \tmp + lsx_init_upper $vr16 \tmp + lsx_init_upper $vr17 \tmp + lsx_init_upper $vr18 \tmp + lsx_init_upper $vr19 \tmp + lsx_init_upper $vr20 \tmp + lsx_init_upper $vr21 \tmp + lsx_init_upper $vr22 \tmp + lsx_init_upper $vr23 \tmp + lsx_init_upper $vr24 \tmp + lsx_init_upper $vr25 \tmp + lsx_init_upper $vr26 \tmp + lsx_init_upper $vr27 \tmp + lsx_init_upper $vr28 \tmp + lsx_init_upper $vr29 \tmp + lsx_init_upper $vr30 \tmp + lsx_init_upper $vr31 \tmp + .endm + + .macro lasx_save_data thread tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \tmp, \thread, \tmp + xvst $xr0, \tmp, THREAD_FPR0 - THREAD_FPR0 + xvst $xr1, \tmp, THREAD_FPR1 - THREAD_FPR0 + xvst $xr2, \tmp, THREAD_FPR2 - THREAD_FPR0 + xvst $xr3, \tmp, THREAD_FPR3 - THREAD_FPR0 + xvst $xr4, \tmp, THREAD_FPR4 - THREAD_FPR0 + xvst $xr5, \tmp, THREAD_FPR5 - THREAD_FPR0 + xvst $xr6, \tmp, THREAD_FPR6 - THREAD_FPR0 + xvst $xr7, \tmp, THREAD_FPR7 - THREAD_FPR0 + xvst $xr8, \tmp, THREAD_FPR8 - THREAD_FPR0 + xvst $xr9, \tmp, THREAD_FPR9 - THREAD_FPR0 + xvst $xr10, \tmp, THREAD_FPR10 - THREAD_FPR0 + xvst $xr11, \tmp, THREAD_FPR11 - THREAD_FPR0 + xvst $xr12, \tmp, THREAD_FPR12 - THREAD_FPR0 + xvst $xr13, \tmp, THREAD_FPR13 - THREAD_FPR0 + xvst $xr14, \tmp, THREAD_FPR14 - THREAD_FPR0 + xvst $xr15, \tmp, THREAD_FPR15 - THREAD_FPR0 + xvst $xr16, \tmp, THREAD_FPR16 - THREAD_FPR0 + xvst $xr17, \tmp, THREAD_FPR17 - THREAD_FPR0 + xvst $xr18, \tmp, THREAD_FPR18 - THREAD_FPR0 + xvst $xr19, \tmp, THREAD_FPR19 - THREAD_FPR0 + xvst $xr20, \tmp, THREAD_FPR20 - THREAD_FPR0 + xvst $xr21, \tmp, THREAD_FPR21 - THREAD_FPR0 + xvst $xr22, \tmp, THREAD_FPR22 - THREAD_FPR0 + xvst $xr23, \tmp, THREAD_FPR23 - THREAD_FPR0 + xvst $xr24, \tmp, THREAD_FPR24 - THREAD_FPR0 + xvst $xr25, \tmp, THREAD_FPR25 - THREAD_FPR0 + xvst $xr26, \tmp, THREAD_FPR26 - THREAD_FPR0 + xvst $xr27, \tmp, THREAD_FPR27 - THREAD_FPR0 + xvst $xr28, \tmp, THREAD_FPR28 - THREAD_FPR0 + xvst $xr29, \tmp, THREAD_FPR29 - THREAD_FPR0 + xvst $xr30, \tmp, THREAD_FPR30 - THREAD_FPR0 + xvst $xr31, \tmp, THREAD_FPR31 - THREAD_FPR0 + .endm + + .macro lasx_restore_data thread tmp + li.w \tmp, THREAD_FPR0 + PTR_ADD \tmp, \thread, \tmp + xvld $xr0, \tmp, THREAD_FPR0 - THREAD_FPR0 + xvld $xr1, \tmp, THREAD_FPR1 - THREAD_FPR0 + xvld $xr2, \tmp, THREAD_FPR2 - THREAD_FPR0 + xvld $xr3, \tmp, THREAD_FPR3 - THREAD_FPR0 + xvld $xr4, \tmp, THREAD_FPR4 - THREAD_FPR0 + xvld $xr5, \tmp, THREAD_FPR5 - THREAD_FPR0 + xvld $xr6, \tmp, THREAD_FPR6 - THREAD_FPR0 + xvld $xr7, \tmp, THREAD_FPR7 - THREAD_FPR0 + xvld $xr8, \tmp, THREAD_FPR8 - THREAD_FPR0 + xvld $xr9, \tmp, THREAD_FPR9 - THREAD_FPR0 + xvld $xr10, \tmp, THREAD_FPR10 - THREAD_FPR0 + xvld $xr11, \tmp, THREAD_FPR11 - THREAD_FPR0 + xvld $xr12, \tmp, THREAD_FPR12 - THREAD_FPR0 + xvld $xr13, \tmp, THREAD_FPR13 - THREAD_FPR0 + xvld $xr14, \tmp, THREAD_FPR14 - THREAD_FPR0 + xvld $xr15, \tmp, THREAD_FPR15 - THREAD_FPR0 + xvld $xr16, \tmp, THREAD_FPR16 - THREAD_FPR0 + xvld $xr17, \tmp, THREAD_FPR17 - THREAD_FPR0 + xvld $xr18, \tmp, THREAD_FPR18 - THREAD_FPR0 + xvld $xr19, \tmp, THREAD_FPR19 - THREAD_FPR0 + xvld $xr20, \tmp, THREAD_FPR20 - THREAD_FPR0 + xvld $xr21, \tmp, THREAD_FPR21 - THREAD_FPR0 + xvld $xr22, \tmp, THREAD_FPR22 - THREAD_FPR0 + xvld $xr23, \tmp, THREAD_FPR23 - THREAD_FPR0 + xvld $xr24, \tmp, THREAD_FPR24 - THREAD_FPR0 + xvld $xr25, \tmp, THREAD_FPR25 - THREAD_FPR0 + xvld $xr26, \tmp, THREAD_FPR26 - THREAD_FPR0 + xvld $xr27, \tmp, THREAD_FPR27 - THREAD_FPR0 + xvld $xr28, \tmp, THREAD_FPR28 - THREAD_FPR0 + xvld $xr29, \tmp, THREAD_FPR29 - THREAD_FPR0 + xvld $xr30, \tmp, THREAD_FPR30 - THREAD_FPR0 + xvld $xr31, \tmp, THREAD_FPR31 - THREAD_FPR0 + .endm + + .macro lasx_save_all thread tmp0 tmp1 + fpu_save_cc \thread, \tmp0, \tmp1 + fpu_save_csr \thread, \tmp0 + lasx_save_data \thread, \tmp0 + .endm + + .macro lasx_restore_all thread tmp0 tmp1 + lasx_restore_data \thread, \tmp0 + fpu_restore_cc \thread, \tmp0, \tmp1 + fpu_restore_csr |