Index: head/sys/arm/arm/vfp.c =================================================================== --- head/sys/arm/arm/vfp.c (revision 288982) +++ head/sys/arm/arm/vfp.c (revision 288983) @@ -1,302 +1,312 @@ /*- * Copyright (c) 2014 Ian Lepore * Copyright (c) 2012 Mark Tinguely * * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #ifdef VFP #include #include #include #include #include #include #include +#include #include #include #include /* function prototypes */ static int vfp_bounce(u_int, u_int, struct trapframe *, int); static void vfp_restore(struct vfp_state *); extern int vfp_exists; static struct undefined_handler vfp10_uh, vfp11_uh; /* If true the VFP unit has 32 double registers, otherwise it has 16 */ static int is_d32; /* * About .fpu directives in this file... * * We should need simply .fpu vfpv3, but clang 3.5 has a quirk where setting * vfpv3 doesn't imply that vfp2 features are also available -- both have to be * explicitly set to get all the features of both. This is probably a bug in * clang, so it may get fixed and require changes here some day. Other changes * are probably coming in clang too, because there is email and open PRs * indicating they want to completely disable the ability to use .fpu and * similar directives in inline asm. That would be catastrophic for us, * hopefully they come to their senses. There was also some discusion of a new * syntax such as .push fpu=vfpv3; ...; .pop fpu; and that would be ideal for * us, better than what we have now really. * * For gcc, each .fpu directive completely overrides the prior directive, unlike * with clang, but luckily on gcc saying v3 implies all the v2 features as well. */ #define fmxr(reg, val) \ __asm __volatile(" .fpu vfpv2\n .fpu vfpv3\n" \ " vmsr " __STRING(reg) ", %0" :: "r"(val)); #define fmrx(reg) \ ({ u_int val = 0;\ __asm __volatile(" .fpu vfpv2\n .fpu vfpv3\n" \ " vmrs %0, " __STRING(reg) : "=r"(val)); \ val; \ }) static u_int get_coprocessorACR(void) { u_int val; __asm __volatile("mrc p15, 0, %0, c1, c0, 2" : "=r" (val) : : "cc"); return val; } static void set_coprocessorACR(u_int val) { __asm __volatile("mcr p15, 0, %0, c1, c0, 2\n\t" : : "r" (val) : "cc"); isb(); } /* called for each cpu */ void vfp_init(void) { u_int fpsid, fpexc, tmp; u_int coproc, vfp_arch; coproc = get_coprocessorACR(); coproc |= COPROC10 | COPROC11; set_coprocessorACR(coproc); fpsid = fmrx(fpsid); /* read the vfp system id */ fpexc = fmrx(fpexc); /* read the vfp exception reg */ if (!(fpsid & VFPSID_HARDSOFT_IMP)) { vfp_exists = 1; is_d32 = 0; PCPU_SET(vfpsid, fpsid); /* save the fpsid */ vfp_arch = (fpsid & VFPSID_SUBVERSION2_MASK) >> VFPSID_SUBVERSION_OFF; if (vfp_arch >= VFP_ARCH3) { tmp = fmrx(mvfr0); PCPU_SET(vfpmvfr0, tmp); if ((tmp & VMVFR0_RB_MASK) == 2) is_d32 = 1; tmp = fmrx(mvfr1); PCPU_SET(vfpmvfr1, tmp); + + if (PCPU_GET(cpuid) == 0) { + if ((tmp & VMVFR1_FZ_MASK) == 0x1) { + /* Denormals arithmetic support */ + initial_fpscr &= ~VFPSCR_FZ; + thread0.td_pcb->pcb_vfpstate.fpscr = + initial_fpscr; + } + } } /* initialize the coprocess 10 and 11 calls * These are called to restore the registers and enable * the VFP hardware. */ if (vfp10_uh.uh_handler == NULL) { vfp10_uh.uh_handler = vfp_bounce; vfp11_uh.uh_handler = vfp_bounce; install_coproc_handler_static(10, &vfp10_uh); install_coproc_handler_static(11, &vfp11_uh); } } } SYSINIT(vfp, SI_SUB_CPU, SI_ORDER_ANY, vfp_init, NULL); /* start VFP unit, restore the vfp registers from the PCB and retry * the instruction */ static int vfp_bounce(u_int addr, u_int insn, struct trapframe *frame, int code) { u_int cpu, fpexc; struct pcb *curpcb; ksiginfo_t ksi; if ((code & FAULT_USER) == 0) panic("undefined floating point instruction in supervisor mode"); critical_enter(); /* * If the VFP is already on and we got an undefined instruction, then * something tried to executate a truly invalid instruction that maps to * the VFP. */ fpexc = fmrx(fpexc); if (fpexc & VFPEXC_EN) { /* Clear any exceptions */ fmxr(fpexc, fpexc & ~(VFPEXC_EX | VFPEXC_FP2V)); /* kill the process - we do not handle emulation */ critical_exit(); if (fpexc & VFPEXC_EX) { /* We have an exception, signal a SIGFPE */ ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGFPE; if (fpexc & VFPEXC_UFC) ksi.ksi_code = FPE_FLTUND; else if (fpexc & VFPEXC_OFC) ksi.ksi_code = FPE_FLTOVF; else if (fpexc & VFPEXC_IOC) ksi.ksi_code = FPE_FLTINV; ksi.ksi_addr = (void *)addr; trapsignal(curthread, &ksi); return 0; } return 1; } /* * If the last time this thread used the VFP it was on this core, and * the last thread to use the VFP on this core was this thread, then the * VFP state is valid, otherwise restore this thread's state to the VFP. */ fmxr(fpexc, fpexc | VFPEXC_EN); curpcb = curthread->td_pcb; cpu = PCPU_GET(cpuid); if (curpcb->pcb_vfpcpu != cpu || curthread != PCPU_GET(fpcurthread)) { vfp_restore(&curpcb->pcb_vfpstate); curpcb->pcb_vfpcpu = cpu; PCPU_SET(fpcurthread, curthread); } critical_exit(); return (0); } /* * Restore the given state to the VFP hardware. */ static void vfp_restore(struct vfp_state *vfpsave) { uint32_t fpexc; /* On vfpv3 we may need to restore FPINST and FPINST2 */ fpexc = vfpsave->fpexec; if (fpexc & VFPEXC_EX) { fmxr(fpinst, vfpsave->fpinst); if (fpexc & VFPEXC_FP2V) fmxr(fpinst2, vfpsave->fpinst2); } fmxr(fpscr, vfpsave->fpscr); __asm __volatile( " .fpu vfpv2\n" " .fpu vfpv3\n" " vldmia %0!, {d0-d15}\n" /* d0-d15 */ " cmp %1, #0\n" /* -D16 or -D32? */ " vldmiane %0!, {d16-d31}\n" /* d16-d31 */ " addeq %0, %0, #128\n" /* skip missing regs */ : "+&r" (vfpsave) : "r" (is_d32) : "cc" ); fmxr(fpexc, fpexc); } /* * If the VFP is on, save its current state and turn it off if requested to do * so. If the VFP is not on, does not change the values at *vfpsave. Caller is * responsible for preventing a context switch while this is running. */ void vfp_store(struct vfp_state *vfpsave, boolean_t disable_vfp) { uint32_t fpexc; fpexc = fmrx(fpexc); /* Is the vfp enabled? */ if (fpexc & VFPEXC_EN) { vfpsave->fpexec = fpexc; vfpsave->fpscr = fmrx(fpscr); /* On vfpv3 we may need to save FPINST and FPINST2 */ if (fpexc & VFPEXC_EX) { vfpsave->fpinst = fmrx(fpinst); if (fpexc & VFPEXC_FP2V) vfpsave->fpinst2 = fmrx(fpinst2); fpexc &= ~VFPEXC_EX; } __asm __volatile( " .fpu vfpv2\n" " .fpu vfpv3\n" " vstmia %0!, {d0-d15}\n" /* d0-d15 */ " cmp %1, #0\n" /* -D16 or -D32? */ " vstmiane r0!, {d16-d31}\n" /* d16-d31 */ " addeq %0, %0, #128\n" /* skip missing regs */ : "+&r" (vfpsave) : "r" (is_d32) : "cc" ); if (disable_vfp) fmxr(fpexc , fpexc & ~VFPEXC_EN); } } /* * The current thread is dying. If the state currently in the hardware belongs * to the current thread, set fpcurthread to NULL to indicate that the VFP * hardware state does not belong to any thread. If the VFP is on, turn it off. * Called only from cpu_throw(), so we don't have to worry about a context * switch here. */ void vfp_discard(struct thread *td) { u_int tmp; if (PCPU_GET(fpcurthread) == td) PCPU_SET(fpcurthread, NULL); tmp = fmrx(fpexc); if (tmp & VFPEXC_EN) fmxr(fpexc, tmp & ~VFPEXC_EN); } #endif Index: head/sys/arm/arm/vm_machdep.c =================================================================== --- head/sys/arm/arm/vm_machdep.c (revision 288982) +++ head/sys/arm/arm/vm_machdep.c (revision 288983) @@ -1,350 +1,352 @@ /*- * Copyright (c) 1982, 1986 The Regents of the University of California. * Copyright (c) 1989, 1990 William Jolitz * Copyright (c) 1994 John Dyson * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department, and William Jolitz. * * Redistribution and use in source and binary :forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * struct switchframe and trapframe must both be a multiple of 8 * for correct stack alignment. */ CTASSERT(sizeof(struct switchframe) == 48); CTASSERT(sizeof(struct trapframe) == 80); +uint32_t initial_fpscr = VFPSCR_DN | VFPSCR_FZ; + /* * Finish a fork operation, with process p2 nearly set up. * Copy and update the pcb, set up the stack so that the child * ready to run and return to user mode. */ void cpu_fork(register struct thread *td1, register struct proc *p2, struct thread *td2, int flags) { struct pcb *pcb2; struct trapframe *tf; struct mdproc *mdp2; if ((flags & RFPROC) == 0) return; /* Point the pcb to the top of the stack */ pcb2 = (struct pcb *) (td2->td_kstack + td2->td_kstack_pages * PAGE_SIZE) - 1; #ifdef __XSCALE__ #ifndef CPU_XSCALE_CORE3 pmap_use_minicache(td2->td_kstack, td2->td_kstack_pages * PAGE_SIZE); #endif #endif td2->td_pcb = pcb2; /* Clone td1's pcb */ bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); /* Point to mdproc and then copy over td1's contents */ mdp2 = &p2->p_md; bcopy(&td1->td_proc->p_md, mdp2, sizeof(*mdp2)); /* Point the frame to the stack in front of pcb and copy td1's frame */ td2->td_frame = (struct trapframe *)pcb2 - 1; *td2->td_frame = *td1->td_frame; /* * Create a new fresh stack for the new process. * Copy the trap frame for the return to user mode as if from a * syscall. This copies most of the user mode register values. */ pmap_set_pcb_pagedir(vmspace_pmap(p2->p_vmspace), pcb2); pcb2->pcb_regs.sf_r4 = (register_t)fork_return; pcb2->pcb_regs.sf_r5 = (register_t)td2; pcb2->pcb_regs.sf_lr = (register_t)fork_trampoline; pcb2->pcb_regs.sf_sp = STACKALIGN(td2->td_frame); pcb2->pcb_vfpcpu = -1; - pcb2->pcb_vfpstate.fpscr = VFPSCR_DN; + pcb2->pcb_vfpstate.fpscr = initial_fpscr; tf = td2->td_frame; tf->tf_spsr &= ~PSR_C; tf->tf_r0 = 0; tf->tf_r1 = 0; /* Setup to release spin count in fork_exit(). */ td2->td_md.md_spinlock_count = 1; td2->td_md.md_saved_cspr = PSR_SVC32_MODE;; #if __ARM_ARCH >= 6 td2->td_md.md_tp = td1->td_md.md_tp; #else td2->td_md.md_tp = *(register_t *)ARM_TP_ADDRESS; #endif } void cpu_thread_swapin(struct thread *td) { } void cpu_thread_swapout(struct thread *td) { } void cpu_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame; int fixup; #ifdef __ARMEB__ u_int call; #endif frame = td->td_frame; fixup = 0; #ifdef __ARMEB__ /* * __syscall returns an off_t while most other syscalls return an * int. As an off_t is 64-bits and an int is 32-bits we need to * place the returned data into r1. As the lseek and frerebsd6_lseek * syscalls also return an off_t they do not need this fixup. */ call = frame->tf_r7; if (call == SYS___syscall) { register_t *ap = &frame->tf_r0; register_t code = ap[_QUAD_LOWWORD]; if (td->td_proc->p_sysent->sv_mask) code &= td->td_proc->p_sysent->sv_mask; fixup = (code != SYS_freebsd6_lseek && code != SYS_lseek) ? 1 : 0; } #endif switch (error) { case 0: if (fixup) { frame->tf_r0 = 0; frame->tf_r1 = td->td_retval[0]; } else { frame->tf_r0 = td->td_retval[0]; frame->tf_r1 = td->td_retval[1]; } frame->tf_spsr &= ~PSR_C; /* carry bit */ break; case ERESTART: /* * Reconstruct the pc to point at the swi. */ #if __ARM_ARCH >= 7 if ((frame->tf_spsr & PSR_T) != 0) frame->tf_pc -= THUMB_INSN_SIZE; else #endif frame->tf_pc -= INSN_SIZE; break; case EJUSTRETURN: /* nothing to do */ break; default: frame->tf_r0 = error; frame->tf_spsr |= PSR_C; /* carry bit */ break; } } /* * Initialize machine state (pcb and trap frame) for a new thread about to * upcall. Put enough state in the new thread's PCB to get it to go back * userret(), where we can intercept it again to set the return (upcall) * Address and stack, along with those from upcals that are from other sources * such as those generated in thread_userret() itself. */ void cpu_set_upcall(struct thread *td, struct thread *td0) { bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); bcopy(td0->td_pcb, td->td_pcb, sizeof(struct pcb)); td->td_pcb->pcb_regs.sf_r4 = (register_t)fork_return; td->td_pcb->pcb_regs.sf_r5 = (register_t)td; td->td_pcb->pcb_regs.sf_lr = (register_t)fork_trampoline; td->td_pcb->pcb_regs.sf_sp = STACKALIGN(td->td_frame); td->td_frame->tf_spsr &= ~PSR_C; td->td_frame->tf_r0 = 0; /* Setup to release spin count in fork_exit(). */ td->td_md.md_spinlock_count = 1; td->td_md.md_saved_cspr = PSR_SVC32_MODE; } /* * Set that machine state for performing an upcall that has to * be done in thread_userret() so that those upcalls generated * in thread_userret() itself can be done as well. */ void cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, stack_t *stack) { struct trapframe *tf = td->td_frame; tf->tf_usr_sp = STACKALIGN((int)stack->ss_sp + stack->ss_size); tf->tf_pc = (int)entry; tf->tf_r0 = (int)arg; tf->tf_spsr = PSR_USR32_MODE; } int cpu_set_user_tls(struct thread *td, void *tls_base) { td->td_md.md_tp = (register_t)tls_base; if (td == curthread) { critical_enter(); #if __ARM_ARCH >= 6 set_tls(tls_base); #else *(register_t *)ARM_TP_ADDRESS = (register_t)tls_base; #endif critical_exit(); } return (0); } void cpu_thread_exit(struct thread *td) { } void cpu_thread_alloc(struct thread *td) { td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_pages * PAGE_SIZE) - 1; /* * Ensure td_frame is aligned to an 8 byte boundary as it will be * placed into the stack pointer which must be 8 byte aligned in * the ARM EABI. */ td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb) - 1; #ifdef __XSCALE__ #ifndef CPU_XSCALE_CORE3 pmap_use_minicache(td->td_kstack, td->td_kstack_pages * PAGE_SIZE); #endif #endif } void cpu_thread_free(struct thread *td) { } void cpu_thread_clean(struct thread *td) { } /* * Intercept the return address from a freshly forked process that has NOT * been scheduled yet. * * This is needed to make kernel threads stay in kernel mode. */ void cpu_set_fork_handler(struct thread *td, void (*func)(void *), void *arg) { td->td_pcb->pcb_regs.sf_r4 = (register_t)func; /* function */ td->td_pcb->pcb_regs.sf_r5 = (register_t)arg; /* first arg */ } /* * Software interrupt handler for queued VM system processing. */ void swi_vm(void *dummy) { if (busdma_swi_pending) busdma_swi(); } void cpu_exit(struct thread *td) { } Index: head/sys/arm/include/md_var.h =================================================================== --- head/sys/arm/include/md_var.h (revision 288982) +++ head/sys/arm/include/md_var.h (revision 288983) @@ -1,74 +1,76 @@ /*- * Copyright (c) 1995 Bruce D. Evans. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the author nor the names of contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: FreeBSD: src/sys/i386/include/md_var.h,v 1.40 2001/07/12 * $FreeBSD$ */ #ifndef _MACHINE_MD_VAR_H_ #define _MACHINE_MD_VAR_H_ extern long Maxmem; extern char sigcode[]; extern int szsigcode; extern uint32_t *vm_page_dump; extern int vm_page_dump_size; extern int (*_arm_memcpy)(void *, void *, int, int); extern int (*_arm_bzero)(void *, int, int); extern int _min_memcpy_size; extern int _min_bzero_size; #define DST_IS_USER 0x1 #define SRC_IS_USER 0x2 #define IS_PHYSICAL 0x4 enum cpu_class { CPU_CLASS_NONE, CPU_CLASS_ARM9TDMI, CPU_CLASS_ARM9ES, CPU_CLASS_ARM9EJS, CPU_CLASS_ARM10E, CPU_CLASS_ARM10EJ, CPU_CLASS_CORTEXA, CPU_CLASS_KRAIT, CPU_CLASS_XSCALE, CPU_CLASS_ARM11J, CPU_CLASS_MARVELL }; extern enum cpu_class cpu_class; struct dumperinfo; extern int busdma_swi_pending; void busdma_swi(void); void dump_add_page(vm_paddr_t); void dump_drop_page(vm_paddr_t); int minidumpsys(struct dumperinfo *); +extern uint32_t initial_fpscr; + #endif /* !_MACHINE_MD_VAR_H_ */