From 2c3c1048746a4622d8c89a29670120dc8fab93c4 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 7 Apr 2024 20:49:45 +0200 Subject: Adding upstream version 6.1.76. Signed-off-by: Daniel Baumann --- arch/arc/kernel/fpu.c | 82 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 82 insertions(+) create mode 100644 arch/arc/kernel/fpu.c (limited to 'arch/arc/kernel/fpu.c') diff --git a/arch/arc/kernel/fpu.c b/arch/arc/kernel/fpu.c new file mode 100644 index 000000000..ec640219d --- /dev/null +++ b/arch/arc/kernel/fpu.c @@ -0,0 +1,82 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * fpu.c - save/restore of Floating Point Unit Registers on task switch + * + * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com) + */ + +#include +#include + +#ifdef CONFIG_ISA_ARCOMPACT + +/* + * To save/restore FPU regs, simplest scheme would use LR/SR insns. + * However since SR serializes the pipeline, an alternate "hack" can be used + * which uses the FPU Exchange insn (DEXCL) to r/w FPU regs. + * + * Store to 64bit dpfp1 reg from a pair of core regs: + * dexcl1 0, r1, r0 ; where r1:r0 is the 64 bit val + * + * Read from dpfp1 into pair of core regs (w/o clobbering dpfp1) + * mov_s r3, 0 + * daddh11 r1, r3, r3 ; get "hi" into r1 (dpfp1 unchanged) + * dexcl1 r0, r1, r3 ; get "low" into r0 (dpfp1 low clobbered) + * dexcl1 0, r1, r0 ; restore dpfp1 to orig value + * + * However we can tweak the read, so that read-out of outgoing task's FPU regs + * and write of incoming task's regs happen in one shot. So all the work is + * done before context switch + */ + +void fpu_save_restore(struct task_struct *prev, struct task_struct *next) +{ + unsigned int *saveto = &prev->thread.fpu.aux_dpfp[0].l; + unsigned int *readfrom = &next->thread.fpu.aux_dpfp[0].l; + + const unsigned int zero = 0; + + __asm__ __volatile__( + "daddh11 %0, %2, %2\n" + "dexcl1 %1, %3, %4\n" + : "=&r" (*(saveto + 1)), /* early clobber must here */ + "=&r" (*(saveto)) + : "r" (zero), "r" (*(readfrom + 1)), "r" (*(readfrom)) + ); + + __asm__ __volatile__( + "daddh22 %0, %2, %2\n" + "dexcl2 %1, %3, %4\n" + : "=&r"(*(saveto + 3)), /* early clobber must here */ + "=&r"(*(saveto + 2)) + : "r" (zero), "r" (*(readfrom + 3)), "r" (*(readfrom + 2)) + ); +} + +#else + +void fpu_init_task(struct pt_regs *regs) +{ + const unsigned int fwe = 0x80000000; + + /* default rounding mode */ + write_aux_reg(ARC_REG_FPU_CTRL, 0x100); + + /* Initialize to zero: setting requires FWE be set */ + write_aux_reg(ARC_REG_FPU_STATUS, fwe); +} + +void fpu_save_restore(struct task_struct *prev, struct task_struct *next) +{ + struct arc_fpu *save = &prev->thread.fpu; + struct arc_fpu *restore = &next->thread.fpu; + const unsigned int fwe = 0x80000000; + + save->ctrl = read_aux_reg(ARC_REG_FPU_CTRL); + save->status = read_aux_reg(ARC_REG_FPU_STATUS); + + write_aux_reg(ARC_REG_FPU_CTRL, restore->ctrl); + write_aux_reg(ARC_REG_FPU_STATUS, (fwe | restore->status)); +} + +#endif -- cgit v1.2.3