Date: Sat, 2 May 2009 08:44:31 GMT From: Marko Zec <zec@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 161467 for review Message-ID: <200905020844.n428iVBU010553@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=161467 Change 161467 by zec@zec_amdx2 on 2009/05/02 08:44:24 IFC @ 161464 Affected files ... .. //depot/projects/vimage/src/sys/amd64/amd64/local_apic.c#17 integrate .. //depot/projects/vimage/src/sys/amd64/amd64/mp_machdep.c#21 integrate .. //depot/projects/vimage/src/sys/amd64/conf/GENERIC#29 integrate .. //depot/projects/vimage/src/sys/amd64/isa/clock.c#10 integrate .. //depot/projects/vimage/src/sys/amd64/linux32/linux32_sysvec.c#15 integrate .. //depot/projects/vimage/src/sys/compat/linux/linux_futex.c#10 integrate .. //depot/projects/vimage/src/sys/conf/NOTES#48 integrate .. //depot/projects/vimage/src/sys/conf/files#65 integrate .. //depot/projects/vimage/src/sys/conf/options#47 integrate .. //depot/projects/vimage/src/sys/dev/acpica/acpi.c#19 integrate .. //depot/projects/vimage/src/sys/dev/ata/ata-all.h#14 integrate .. //depot/projects/vimage/src/sys/dev/ata/ata-disk.c#12 integrate .. //depot/projects/vimage/src/sys/dev/ata/ata-queue.c#12 integrate .. //depot/projects/vimage/src/sys/dev/sk/if_sk.c#9 integrate .. //depot/projects/vimage/src/sys/dev/usb/wlan/if_ural.c#7 integrate .. //depot/projects/vimage/src/sys/i386/conf/GENERIC#32 integrate .. //depot/projects/vimage/src/sys/i386/i386/local_apic.c#17 integrate .. //depot/projects/vimage/src/sys/i386/i386/mp_machdep.c#20 integrate .. //depot/projects/vimage/src/sys/i386/linux/linux_sysvec.c#11 integrate .. //depot/projects/vimage/src/sys/isa/atrtc.c#2 integrate .. //depot/projects/vimage/src/sys/kern/kern_osd.c#4 integrate .. //depot/projects/vimage/src/sys/mips/include/pmap.h#4 integrate .. //depot/projects/vimage/src/sys/mips/mips/pmap.c#11 integrate .. //depot/projects/vimage/src/sys/modules/Makefile#46 integrate .. //depot/projects/vimage/src/sys/net/ieee8023ad_lacp.c#11 integrate .. //depot/projects/vimage/src/sys/net/if_bridge.c#29 integrate .. //depot/projects/vimage/src/sys/net/if_gif.h#13 integrate .. //depot/projects/vimage/src/sys/net/if_lagg.c#22 integrate .. //depot/projects/vimage/src/sys/net/route.c#49 integrate .. //depot/projects/vimage/src/sys/net/vnet.h#24 integrate .. //depot/projects/vimage/src/sys/net80211/ieee80211_ddb.c#22 integrate .. //depot/projects/vimage/src/sys/netinet/igmp.c#37 integrate .. //depot/projects/vimage/src/sys/netinet/ip_fw.h#30 integrate .. //depot/projects/vimage/src/sys/netinet/tcp_timewait.c#32 integrate .. //depot/projects/vimage/src/sys/netinet/vinet.h#57 integrate .. //depot/projects/vimage/src/sys/netinet6/udp6_usrreq.c#42 integrate .. //depot/projects/vimage/src/sys/netinet6/vinet6.h#36 integrate .. //depot/projects/vimage/src/sys/netipsec/vipsec.h#26 integrate .. //depot/projects/vimage/src/sys/pc98/conf/GENERIC#21 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_atalk.c#3 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_audit.c#6 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_cred.c#3 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_inet.c#13 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_inet6.c#5 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_internal.h#10 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_net.c#8 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_pipe.c#8 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_posix_sem.c#9 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_posix_shm.c#4 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_priv.c#5 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_process.c#11 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_socket.c#7 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_system.c#7 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_sysv_msg.c#7 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_sysv_sem.c#7 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_sysv_shm.c#7 integrate .. //depot/projects/vimage/src/sys/security/mac/mac_vfs.c#11 integrate Differences ... ==== //depot/projects/vimage/src/sys/amd64/amd64/local_apic.c#17 (text+ko) ==== @@ -32,7 +32,7 @@ */ #include <sys/cdefs.h> -__FBSDID("$FreeBSD: src/sys/amd64/amd64/local_apic.c,v 1.52 2009/02/21 23:15:34 jeff Exp $"); +__FBSDID("$FreeBSD: src/sys/amd64/amd64/local_apic.c,v 1.54 2009/05/01 20:53:37 mav Exp $"); #include "opt_hwpmc_hooks.h" #include "opt_kdtrace.h" @@ -112,7 +112,7 @@ u_long la_stat_ticks; u_long la_prof_ticks; /* Include IDT_SYSCALL to make indexing easier. */ - u_int la_ioint_irqs[APIC_NUM_IOINTS + 1]; + int la_ioint_irqs[APIC_NUM_IOINTS + 1]; } static lapics[MAX_APIC_ID + 1]; /* XXX: should thermal be an NMI? */ @@ -254,6 +254,8 @@ lapics[apic_id].la_lvts[i] = lvts[i]; lapics[apic_id].la_lvts[i].lvt_active = 0; } + for (i = 0; i <= APIC_NUM_IOINTS; i++) + lapics[apic_id].la_ioint_irqs[i] = -1; lapics[apic_id].la_ioint_irqs[IDT_SYSCALL - APIC_IO_INTS] = IRQ_SYSCALL; lapics[apic_id].la_ioint_irqs[APIC_TIMER_INT - APIC_IO_INTS] = IRQ_TIMER; @@ -363,11 +365,15 @@ lapic_setup_clock(void) { u_long value; + int i; /* Can't drive the timer without a local APIC. */ if (lapic == NULL) return (0); + if (resource_int_value("apic", 0, "clock", &i) == 0 && i == 0) + return (0); + /* Start off with a divisor of 2 (power on reset default). */ lapic_timer_divisor = 2; @@ -807,7 +813,7 @@ */ mtx_lock_spin(&icu_lock); for (vector = 0; vector < APIC_NUM_IOINTS; vector++) { - if (lapics[apic_id].la_ioint_irqs[vector] != 0) + if (lapics[apic_id].la_ioint_irqs[vector] != -1) continue; lapics[apic_id].la_ioint_irqs[vector] = irq; mtx_unlock_spin(&icu_lock); @@ -847,7 +853,7 @@ for (vector = 0; vector < APIC_NUM_IOINTS; vector++) { /* Vector is in use, end run. */ - if (lapics[apic_id].la_ioint_irqs[vector] != 0) { + if (lapics[apic_id].la_ioint_irqs[vector] != -1) { run = 0; first = 0; continue; @@ -932,7 +938,7 @@ sched_bind(td, apic_cpuid(apic_id)); thread_unlock(td); mtx_lock_spin(&icu_lock); - lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = 0; + lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS] = -1; mtx_unlock_spin(&icu_lock); thread_lock(td); sched_unbind(td); @@ -944,11 +950,15 @@ u_int apic_idt_to_irq(u_int apic_id, u_int vector) { + int irq; KASSERT(vector >= APIC_IO_INTS && vector != IDT_SYSCALL && vector <= APIC_IO_INTS + APIC_NUM_IOINTS, ("Vector %u does not map to an IRQ line", vector)); - return (lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS]); + irq = lapics[apic_id].la_ioint_irqs[vector - APIC_IO_INTS]; + if (irq < 0) + irq = 0; + return (irq); } #ifdef DDB @@ -974,7 +984,7 @@ db_printf("Interrupts bound to lapic %u\n", apic_id); for (i = 0; i < APIC_NUM_IOINTS + 1 && !db_pager_quit; i++) { irq = lapics[apic_id].la_ioint_irqs[i]; - if (irq == 0 || irq == IRQ_SYSCALL) + if (irq == -1 || irq == IRQ_SYSCALL) continue; db_printf("vec 0x%2x -> ", i + APIC_IO_INTS); if (irq == IRQ_TIMER) ==== //depot/projects/vimage/src/sys/amd64/amd64/mp_machdep.c#21 (text+ko) ==== @@ -25,7 +25,7 @@ */ #include <sys/cdefs.h> -__FBSDID("$FreeBSD: src/sys/amd64/amd64/mp_machdep.c,v 1.304 2009/04/29 06:54:40 jeff Exp $"); +__FBSDID("$FreeBSD: src/sys/amd64/amd64/mp_machdep.c,v 1.305 2009/04/30 22:10:04 jkim Exp $"); #include "opt_cpu.h" #include "opt_kstack_pages.h" @@ -292,6 +292,10 @@ static void topo_probe(void) { + static int cpu_topo_probed = 0; + + if (cpu_topo_probed) + return; logical_cpus = logical_cpus_mask = 0; if (cpu_high >= 0xb) @@ -299,9 +303,10 @@ else if (cpu_high) topo_probe_0x4(); if (cpu_cores == 0) - cpu_cores = mp_ncpus; + cpu_cores = mp_ncpus > 0 ? mp_ncpus : 1; if (cpu_logical == 0) cpu_logical = 1; + cpu_topo_probed = 1; } struct cpu_group * @@ -313,6 +318,7 @@ * Determine whether any threading flags are * necessry. */ + topo_probe(); if (cpu_logical > 1 && hyperthreading_cpus) cg_flags = CG_FLAG_HTT; else if (cpu_logical > 1) ==== //depot/projects/vimage/src/sys/amd64/conf/GENERIC#29 (text+ko) ==== @@ -16,7 +16,7 @@ # If you are in doubt as to the purpose or necessity of a line, check first # in NOTES. # -# $FreeBSD: src/sys/amd64/conf/GENERIC,v 1.523 2009/04/10 00:40:48 jfv Exp $ +# $FreeBSD: src/sys/amd64/conf/GENERIC,v 1.524 2009/05/01 17:20:16 sam Exp $ cpu HAMMER ident GENERIC @@ -292,8 +292,10 @@ device ulpt # Printer device umass # Disks/Mass storage - Requires scbus and da device ums # Mouse +device rum # Ralink Technology RT2501USB wireless NICs +device uath # Atheros AR5523 wireless NICs device ural # Ralink Technology RT2500USB wireless NICs -device rum # Ralink Technology RT2501USB wireless NICs +device zyd # ZyDAS zb1211/zb1211b wireless NICs device urio # Diamond Rio 500 MP3 player # USB Serial devices device uark # Technologies ARK3116 based serial adapters ==== //depot/projects/vimage/src/sys/amd64/isa/clock.c#10 (text+ko) ==== @@ -33,7 +33,7 @@ */ #include <sys/cdefs.h> -__FBSDID("$FreeBSD: src/sys/amd64/isa/clock.c,v 1.243 2008/04/22 19:38:27 phk Exp $"); +__FBSDID("$FreeBSD: src/sys/amd64/isa/clock.c,v 1.244 2009/05/01 21:43:04 mav Exp $"); /* * Routines to handle clock hardware. @@ -376,6 +376,17 @@ mtx_unlock_spin(&clock_lock); } +static void +i8254_restore(void) +{ + + mtx_lock_spin(&clock_lock); + outb(TIMER_MODE, TIMER_SEL0 | TIMER_RATEGEN | TIMER_16BIT); + outb(TIMER_CNTR0, i8254_real_max_count & 0xff); + outb(TIMER_CNTR0, i8254_real_max_count >> 8); + mtx_unlock_spin(&clock_lock); +} + /* This is separate from startrtclock() so that it can be called early. */ void i8254_init(void) @@ -558,6 +569,14 @@ return(0); } +static int +attimer_resume(device_t dev) +{ + + i8254_restore(); + return(0); +} + static device_method_t attimer_methods[] = { /* Device interface */ DEVMETHOD(device_probe, attimer_probe), @@ -565,7 +584,7 @@ DEVMETHOD(device_detach, bus_generic_detach), DEVMETHOD(device_shutdown, bus_generic_shutdown), DEVMETHOD(device_suspend, bus_generic_suspend), - DEVMETHOD(device_resume, bus_generic_resume), + DEVMETHOD(device_resume, attimer_resume), { 0, 0 } }; ==== //depot/projects/vimage/src/sys/amd64/linux32/linux32_sysvec.c#15 (text+ko) ==== @@ -31,7 +31,7 @@ */ #include <sys/cdefs.h> -__FBSDID("$FreeBSD: src/sys/amd64/linux32/linux32_sysvec.c,v 1.46 2009/04/05 09:27:19 dchagin Exp $"); +__FBSDID("$FreeBSD: src/sys/amd64/linux32/linux32_sysvec.c,v 1.47 2009/05/01 15:36:02 dchagin Exp $"); #include "opt_compat.h" #ifndef COMPAT_IA32 @@ -128,7 +128,7 @@ static void linux32_fixlimit(struct rlimit *rl, int which); extern LIST_HEAD(futex_list, futex) futex_list; -extern struct sx futex_sx; +extern struct mtx futex_mtx; static eventhandler_tag linux_exit_tag; static eventhandler_tag linux_schedtail_tag; @@ -1117,7 +1117,7 @@ mtx_init(&emul_lock, "emuldata lock", NULL, MTX_DEF); sx_init(&emul_shared_lock, "emuldata->shared lock"); LIST_INIT(&futex_list); - sx_init(&futex_sx, "futex protection lock"); + mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit, NULL, 1000); linux_schedtail_tag = EVENTHANDLER_REGISTER(schedtail, @@ -1149,7 +1149,7 @@ linux_device_unregister_handler(*ldhp); mtx_destroy(&emul_lock); sx_destroy(&emul_shared_lock); - sx_destroy(&futex_sx); + mtx_destroy(&futex_mtx); EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag); EVENTHANDLER_DEREGISTER(schedtail, linux_schedtail_tag); EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag); ==== //depot/projects/vimage/src/sys/compat/linux/linux_futex.c#10 (text+ko) ==== @@ -32,7 +32,7 @@ */ #include <sys/cdefs.h> -__FBSDID("$FreeBSD: src/sys/compat/linux/linux_futex.c,v 1.19 2009/04/19 13:48:42 dchagin Exp $"); +__FBSDID("$FreeBSD: src/sys/compat/linux/linux_futex.c,v 1.20 2009/05/01 15:36:02 dchagin Exp $"); #if 0 __KERNEL_RCSID(1, "$NetBSD: linux_futex.c,v 1.7 2006/07/24 19:01:49 manu Exp $"); #endif @@ -62,419 +62,284 @@ #include <compat/linux/linux_futex.h> #include <compat/linux/linux_emul.h> +MALLOC_DEFINE(M_FUTEX, "futex", "Linux futexes"); +MALLOC_DEFINE(M_FUTEX_WP, "futex wp", "Linux futexes wp"); + struct futex; struct waiting_proc { - struct thread *wp_t; - struct futex *wp_new_futex; + uint32_t wp_flags; + struct futex *wp_futex; TAILQ_ENTRY(waiting_proc) wp_list; }; + struct futex { - void *f_uaddr; - int f_refcount; + struct sx f_lck; + uint32_t *f_uaddr; + uint32_t f_refcount; LIST_ENTRY(futex) f_list; TAILQ_HEAD(lf_waiting_proc, waiting_proc) f_waiting_proc; }; LIST_HEAD(futex_list, futex) futex_list; -struct sx futex_sx; /* this protects the LIST of futexes */ -#define FUTEX_LOCK sx_xlock(&futex_sx) -#define FUTEX_UNLOCK sx_xunlock(&futex_sx) +#define FUTEX_LOCK(f) sx_xlock(&(f)->f_lck) +#define FUTEX_UNLOCK(f) sx_xunlock(&(f)->f_lck) +#define FUTEX_INIT(f) sx_init_flags(&(f)->f_lck, "ftlk", 0) +#define FUTEX_DESTROY(f) sx_destroy(&(f)->f_lck) +#define FUTEX_ASSERT_LOCKED(f) sx_assert(&(f)->f_lck, SA_XLOCKED) -#define FUTEX_LOCKED 1 -#define FUTEX_UNLOCKED 0 +struct mtx futex_mtx; /* protects the futex list */ +#define FUTEXES_LOCK mtx_lock(&futex_mtx) +#define FUTEXES_UNLOCK mtx_unlock(&futex_mtx) -#define FUTEX_SYSTEM_LOCK mtx_lock(&Giant) -#define FUTEX_SYSTEM_UNLOCK mtx_unlock(&Giant) +/* flags for futex_get() */ +#define FUTEX_CREATE_WP 0x1 /* create waiting_proc */ +#define FUTEX_DONTCREATE 0x2 /* don't create futex if not exists */ +#define FUTEX_DONTEXISTS 0x4 /* return EINVAL if futex exists */ -static struct futex *futex_get(void *, int); -static void futex_put(struct futex *); -static int futex_sleep(struct futex *, struct thread *, unsigned long); -static int futex_wake(struct futex *, int, struct futex *, int); -static int futex_atomic_op(struct thread *td, int encoded_op, caddr_t uaddr); +/* wp_flags */ +#define FUTEX_WP_REQUEUED 0x1 /* wp requeued - wp moved from wp_list + * of futex where thread sleep to wp_list + * of another futex. + */ +#define FUTEX_WP_REMOVED 0x2 /* wp is woken up and removed from futex + * wp_list to prevent double wakeup. + */ /* support.s */ -int futex_xchgl(int oparg, caddr_t uaddr, int *oldval); -int futex_addl(int oparg, caddr_t uaddr, int *oldval); -int futex_orl(int oparg, caddr_t uaddr, int *oldval); -int futex_andl(int oparg, caddr_t uaddr, int *oldval); -int futex_xorl(int oparg, caddr_t uaddr, int *oldval); +int futex_xchgl(int oparg, uint32_t *uaddr, int *oldval); +int futex_addl(int oparg, uint32_t *uaddr, int *oldval); +int futex_orl(int oparg, uint32_t *uaddr, int *oldval); +int futex_andl(int oparg, uint32_t *uaddr, int *oldval); +int futex_xorl(int oparg, uint32_t *uaddr, int *oldval); -int -linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args) +static void +futex_put(struct futex *f, struct waiting_proc *wp) { - int val; - int ret; - struct l_timespec timeout = {0, 0}; - int error = 0; - struct futex *f; - struct futex *newf; - int timeout_hz; - struct timeval tv = {0, 0}; - struct futex *f2; - int op_ret; - struct linux_emuldata *em; + + FUTEX_ASSERT_LOCKED(f); + if (wp != NULL) { + if ((wp->wp_flags & FUTEX_WP_REMOVED) == 0) + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + free(wp, M_FUTEX_WP); + } -#ifdef DEBUG - if (ldebug(sys_futex)) - printf(ARGS(futex, "%p, %i, %i, *, %p, %i"), args->uaddr, args->op, - args->val, args->uaddr2, args->val3); -#endif + FUTEXES_LOCK; + if (--f->f_refcount == 0) { + LIST_REMOVE(f, f_list); + FUTEXES_UNLOCK; + FUTEX_UNLOCK(f); - /* - * Our implementation provides only privates futexes. Most of the apps - * should use private futexes but don't claim so. Therefore we treat - * all futexes as private by clearing the FUTEX_PRIVATE_FLAG. It works - * in most cases (ie. when futexes are not shared on file descriptor - * or between different processes.). - */ - args->op = (args->op & ~LINUX_FUTEX_PRIVATE_FLAG); + FUTEX_DESTROY(f); + free(f, M_FUTEX); + return; + } - switch (args->op) { - case LINUX_FUTEX_WAIT: - FUTEX_SYSTEM_LOCK; + FUTEXES_UNLOCK; + FUTEX_UNLOCK(f); +} - if ((error = copyin(args->uaddr, - &val, sizeof(val))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; - } +static int +futex_get0(uint32_t *uaddr, struct futex **newf, uint32_t flags) +{ + struct futex *f, *tmpf; - if (val != args->val) { - FUTEX_SYSTEM_UNLOCK; - return EWOULDBLOCK; - } + *newf = tmpf = NULL; - if (args->timeout != NULL) { - if ((error = copyin(args->timeout, - &timeout, sizeof(timeout))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; +retry: + FUTEXES_LOCK; + LIST_FOREACH(f, &futex_list, f_list) { + if (f->f_uaddr == uaddr) { + if (tmpf != NULL) { + FUTEX_UNLOCK(tmpf); + FUTEX_DESTROY(tmpf); + free(tmpf, M_FUTEX); + } + if (flags & FUTEX_DONTEXISTS) { + FUTEXES_UNLOCK; + return (EINVAL); } - } -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: val = %d, uaddr = %p, " - "*uaddr = %d, timeout = %d.%09lu\n", - td->td_proc->p_pid, args->val, - args->uaddr, val, timeout.tv_sec, - (unsigned long)timeout.tv_nsec); -#endif - tv.tv_usec = timeout.tv_sec * 1000000 + timeout.tv_nsec / 1000; - timeout_hz = tvtohz(&tv); + /* + * Increment refcount of the found futex to + * prevent it from deallocation before FUTEX_LOCK() + */ + ++f->f_refcount; + FUTEXES_UNLOCK; - if (timeout.tv_sec == 0 && timeout.tv_nsec == 0) - timeout_hz = 0; - /* - * If the user process requests a non null timeout, - * make sure we do not turn it into an infinite - * timeout because timeout_hz gets null. - * - * We use a minimal timeout of 1/hz. Maybe it would - * make sense to just return ETIMEDOUT without sleeping. - */ - if (((timeout.tv_sec != 0) || (timeout.tv_nsec != 0)) && - (timeout_hz == 0)) - timeout_hz = 1; - - - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - ret = futex_sleep(f, td, timeout_hz); - futex_put(f); - -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: uaddr = %p, " - "ret = %d\n", td->td_proc->p_pid, args->uaddr, ret); -#endif - - FUTEX_SYSTEM_UNLOCK; - switch (ret) { - case EWOULDBLOCK: /* timeout */ - return ETIMEDOUT; - break; - case EINTR: /* signal */ - return EINTR; - break; - case 0: /* FUTEX_WAKE received */ -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT %d: uaddr = %p, " - "got FUTEX_WAKE\n", - td->td_proc->p_pid, args->uaddr); -#endif - return 0; - break; - default: -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAIT: unexpected ret = %d\n", - ret); -#endif - break; + FUTEX_LOCK(f); + *newf = f; + return (0); } + } - /* NOTREACHED */ - break; + if (flags & FUTEX_DONTCREATE) { + FUTEXES_UNLOCK; + return (0); + } - case LINUX_FUTEX_WAKE: - FUTEX_SYSTEM_LOCK; + if (tmpf == NULL) { + FUTEXES_UNLOCK; + tmpf = malloc(sizeof(*tmpf), M_FUTEX, M_WAITOK | M_ZERO); + tmpf->f_uaddr = uaddr; + tmpf->f_refcount = 1; + FUTEX_INIT(tmpf); + TAILQ_INIT(&tmpf->f_waiting_proc); /* - * XXX: Linux is able to cope with different addresses - * corresponding to the same mapped memory in the sleeping - * and waker process(es). + * Lock the new futex before an insert into the futex_list + * to prevent futex usage by other. */ -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAKE %d: uaddr = %p, val = %d\n", - td->td_proc->p_pid, args->uaddr, args->val); -#endif - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - td->td_retval[0] = futex_wake(f, args->val, NULL, 0); - futex_put(f); + FUTEX_LOCK(tmpf); + goto retry; + } - FUTEX_SYSTEM_UNLOCK; - break; + LIST_INSERT_HEAD(&futex_list, tmpf, f_list); + FUTEXES_UNLOCK; - case LINUX_FUTEX_CMP_REQUEUE: - FUTEX_SYSTEM_LOCK; + *newf = tmpf; + return (0); +} - if ((error = copyin(args->uaddr, - &val, sizeof(val))) != 0) { - FUTEX_SYSTEM_UNLOCK; - return error; - } +static int +futex_get(uint32_t *uaddr, struct waiting_proc **wp, struct futex **f, + uint32_t flags) +{ + int error; - if (val != args->val3) { - FUTEX_SYSTEM_UNLOCK; - return EAGAIN; - } + if (flags & FUTEX_CREATE_WP) { + *wp = malloc(sizeof(struct waiting_proc), M_FUTEX_WP, M_WAITOK); + (*wp)->wp_flags = 0; + } + error = futex_get0(uaddr, f, flags); + if (error) { + if (flags & FUTEX_CREATE_WP) + free(*wp, M_FUTEX_WP); + return (error); + } + if (flags & FUTEX_CREATE_WP) { + TAILQ_INSERT_HEAD(&(*f)->f_waiting_proc, *wp, wp_list); + (*wp)->wp_futex = *f; + } - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - newf = futex_get(args->uaddr2, FUTEX_UNLOCKED); - td->td_retval[0] = futex_wake(f, args->val, newf, - (int)(unsigned long)args->timeout); - futex_put(f); - futex_put(newf); - - FUTEX_SYSTEM_UNLOCK; - break; - - case LINUX_FUTEX_WAKE_OP: - FUTEX_SYSTEM_LOCK; -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX_WAKE_OP: %d: uaddr = %p, op = %d, " - "val = %x, uaddr2 = %p, val3 = %x\n", - td->td_proc->p_pid, args->uaddr, args->op, - args->val, args->uaddr2, args->val3); -#endif - f = futex_get(args->uaddr, FUTEX_UNLOCKED); - f2 = futex_get(args->uaddr2, FUTEX_UNLOCKED); - - /* - * This function returns positive number as results and - * negative as errors - */ - op_ret = futex_atomic_op(td, args->val3, args->uaddr2); -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("futex_atomic_op ret %d\n", op_ret); -#endif - if (op_ret < 0) { - /* XXX: We don't handle the EFAULT yet. */ - if (op_ret != -EFAULT) { - futex_put(f); - futex_put(f2); - FUTEX_SYSTEM_UNLOCK; - return (-op_ret); - } - - futex_put(f); - futex_put(f2); - - FUTEX_SYSTEM_UNLOCK; - return (EFAULT); - } - - ret = futex_wake(f, args->val, NULL, 0); - futex_put(f); - if (op_ret > 0) { - op_ret = 0; - /* - * Linux abuses the address of the timespec parameter - * as the number of retries. - */ - op_ret += futex_wake(f2, - (int)(unsigned long)args->timeout, NULL, 0); - ret += op_ret; - } - futex_put(f2); - td->td_retval[0] = ret; - - FUTEX_SYSTEM_UNLOCK; - break; - - case LINUX_FUTEX_LOCK_PI: - /* not yet implemented */ - return (ENOSYS); - - case LINUX_FUTEX_UNLOCK_PI: - /* not yet implemented */ - return (ENOSYS); - - case LINUX_FUTEX_TRYLOCK_PI: - /* not yet implemented */ - return (ENOSYS); - - case LINUX_FUTEX_REQUEUE: - - /* - * Glibc does not use this operation since Jun 2004 (2.3.3), - * as it is racy and replaced by FUTEX_CMP_REQUEUE operation. - * Glibc versions prior to 2.3.3 fall back to FUTEX_WAKE when - * FUTEX_REQUEUE returned EINVAL. - */ - em = em_find(td->td_proc, EMUL_DONTLOCK); - if (em->used_requeue == 0) { - printf("linux(%s (%d)) sys_futex: " - "unsupported futex_requeue op\n", - td->td_proc->p_comm, td->td_proc->p_pid); - em->used_requeue = 1; - } - return (EINVAL); - - default: - printf("linux_sys_futex: unknown op %d\n", - args->op); - return (ENOSYS); - } - return (0); + return (error); } -static struct futex * -futex_get(void *uaddr, int locked) +static int +futex_sleep(struct futex *f, struct waiting_proc *wp, unsigned long timeout) { - struct futex *f; + int error; - if (locked == FUTEX_UNLOCKED) - FUTEX_LOCK; - LIST_FOREACH(f, &futex_list, f_list) { - if (f->f_uaddr == uaddr) { - f->f_refcount++; - if (locked == FUTEX_UNLOCKED) - FUTEX_UNLOCK; - return f; - } + FUTEX_ASSERT_LOCKED(f); + error = sx_sleep(wp, &f->f_lck, PCATCH, "futex", timeout); + if (wp->wp_flags & FUTEX_WP_REQUEUED) { + KASSERT(f != wp->wp_futex, ("futex != wp_futex")); + futex_put(f, NULL); + f = wp->wp_futex; + FUTEX_LOCK(f); } - f = malloc(sizeof(*f), M_LINUX, M_WAITOK); - f->f_uaddr = uaddr; - f->f_refcount = 1; - TAILQ_INIT(&f->f_waiting_proc); - LIST_INSERT_HEAD(&futex_list, f, f_list); - if (locked == FUTEX_UNLOCKED) - FUTEX_UNLOCK; - - return f; + futex_put(f, wp); + return (error); } -static void -futex_put(f) - struct futex *f; +static int +futex_wake(struct futex *f, int n) { - FUTEX_LOCK; - f->f_refcount--; - if (f->f_refcount == 0) { - LIST_REMOVE(f, f_list); - free(f, M_LINUX); + struct waiting_proc *wp, *wpt; + int count = 0; + + FUTEX_ASSERT_LOCKED(f); + TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { + wp->wp_flags |= FUTEX_WP_REMOVED; + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + wakeup_one(wp); + if (++count == n) + break; } - FUTEX_UNLOCK; - return; + return (count); } static int -futex_sleep(struct futex *f, struct thread *td, unsigned long timeout) +futex_requeue(struct futex *f, int n, struct futex *f2, int n2) { - struct waiting_proc *wp; - int ret; + struct waiting_proc *wp, *wpt; + int count = 0; - wp = malloc(sizeof(*wp), M_LINUX, M_WAITOK); - wp->wp_t = td; - wp->wp_new_futex = NULL; - FUTEX_LOCK; - TAILQ_INSERT_TAIL(&f->f_waiting_proc, wp, wp_list); - FUTEX_UNLOCK; + FUTEX_ASSERT_LOCKED(f); + FUTEX_ASSERT_LOCKED(f2); -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX --> %d tlseep timeout = %ld\n", - td->td_proc->p_pid, timeout); -#endif - ret = tsleep(wp, PCATCH | PZERO, "linuxfutex", timeout); -#ifdef DEBUG - if (ldebug(sys_futex)) - printf("FUTEX -> %d tsleep returns %d\n", - td->td_proc->p_pid, ret); -#endif + TAILQ_FOREACH_SAFE(wp, &f->f_waiting_proc, wp_list, wpt) { + if (++count <= n) { + wp->wp_flags |= FUTEX_WP_REMOVED; + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + wakeup_one(wp); + } else { + wp->wp_flags |= FUTEX_WP_REQUEUED; + /* Move wp to wp_list of f2 futex */ + TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); + TAILQ_INSERT_HEAD(&f2->f_waiting_proc, wp, wp_list); - FUTEX_LOCK; - TAILQ_REMOVE(&f->f_waiting_proc, wp, wp_list); - FUTEX_UNLOCK; - - /* if we got woken up in futex_wake */ - if ((ret == 0) && (wp->wp_new_futex != NULL)) { - /* suspend us on the new futex */ - ret = futex_sleep(wp->wp_new_futex, td, timeout); - /* and release the old one */ - futex_put(wp->wp_new_futex); + /* + * Thread which sleeps on wp after waking should + * acquire f2 lock, so increment refcount of f2 to + * prevent it from premature deallocation. + */ + wp->wp_futex = f2; + FUTEXES_LOCK; + ++f2->f_refcount; + FUTEXES_UNLOCK; + if (count - n >= n2) + break; + } } - free(wp, M_LINUX); - - return ret; + return (count); } static int -futex_wake(struct futex *f, int n, struct futex *newf, int n2) +futex_wait(struct futex *f, struct waiting_proc *wp, struct l_timespec *ts) { - struct waiting_proc *wp; - int count; + struct l_timespec timeout = {0, 0}; + struct timeval tv = {0, 0}; + int timeout_hz; + int error; + + if (ts != NULL) { + error = copyin(ts, &timeout, sizeof(timeout)); + if (error) + return (error); + } + + tv.tv_usec = timeout.tv_sec * 1000000 + timeout.tv_nsec / 1000; + timeout_hz = tvtohz(&tv); + + if (timeout.tv_sec == 0 && timeout.tv_nsec == 0) + timeout_hz = 0; /* - * Linux is very strange it wakes up N threads for - * all operations BUT requeue ones where its N+1 - * mimic this. + * If the user process requests a non null timeout, + * make sure we do not turn it into an infinite + * timeout because timeout_hz gets null. + * + * We use a minimal timeout of 1/hz. Maybe it would + * make sense to just return ETIMEDOUT without sleeping. */ - count = newf ? 0 : 1; + if (((timeout.tv_sec != 0) || (timeout.tv_nsec != 0)) && + (timeout_hz == 0)) + timeout_hz = 1; - FUTEX_LOCK; - TAILQ_FOREACH(wp, &f->f_waiting_proc, wp_list) { - if (count <= n) { - wakeup_one(wp); - count++; - } else { - if (newf != NULL) { - /* futex_put called after tsleep */ - wp->wp_new_futex = futex_get(newf->f_uaddr, - FUTEX_LOCKED); - wakeup_one(wp); - if (count - n >= n2) - break; - } - } - } - FUTEX_UNLOCK; + error = futex_sleep(f, wp, timeout_hz); + if (error == EWOULDBLOCK) + error = ETIMEDOUT; - return count; + return (error); } static int -futex_atomic_op(struct thread *td, int encoded_op, caddr_t uaddr) +futex_atomic_op(struct thread *td, int encoded_op, uint32_t *uaddr) { int op = (encoded_op >> 28) & 7; int cmp = (encoded_op >> 24) & 15; @@ -536,14 +401,237 @@ } int +linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args) +{ + int op_ret, val, ret, nrwake; + struct linux_emuldata *em; + struct waiting_proc *wp; + struct futex *f, *f2; + int error = 0; + + /* + * Our implementation provides only privates futexes. Most of the apps + * should use private futexes but don't claim so. Therefore we treat + * all futexes as private by clearing the FUTEX_PRIVATE_FLAG. It works + * in most cases (ie. when futexes are not shared on file descriptor + * or between different processes.). + */ + args->op = (args->op & ~LINUX_FUTEX_PRIVATE_FLAG); + + switch (args->op) { + case LINUX_FUTEX_WAIT: + +#ifdef DEBUG + if (ldebug(sys_futex)) + printf(ARGS(sys_futex, "futex_wait val %d uaddr %p"), + args->val, args->uaddr); +#endif + error = futex_get(args->uaddr, &wp, &f, FUTEX_CREATE_WP); + if (error) + return (error); + error = copyin(args->uaddr, &val, sizeof(val)); + if (error) { + futex_put(f, wp); + return (error); + } >>> TRUNCATED FOR MAIL (1000 lines) <<<
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200905020844.n428iVBU010553>