d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add stack_save_td_running(), a function to trace the kernel stack of a

Browse Source 
running thread.

It is currently implemented only on amd64 and i386; on these
architectures, it is implemented by raising an NMI on the CPU on which
the target thread is currently running. Unlike stack_save_td(), it may
fail, for example if the thread is running in user mode.

This change also modifies the kern.proc.kstack sysctl to use this function,
so that stacks of running threads are shown in the output of "procstat -kk".
This is handy for debugging threads that are stuck in a busy loop.

Reviewed by:	bdrewery, jhb, kib
Sponsored by:	EMC / Isilon Storage Division
Differential Revision:	https://reviews.freebsd.org/D3256
This commit is contained in:
Mark Johnston 2015-09-11 03:54:37 +00:00
parent 1e954a7c63
commit 610141cebb
14 changed files with 165 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
sys/amd64/amd64/trap.c
View File

@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$");
#include "opt_hwpmc_hooks.h"
#include "opt_isa.h"
#include "opt_kdb.h"
#include "opt_stack.h"
#include <sys/param.h>
#include <sys/bus.h>
@ -91,6 +92,7 @@ PMC_SOFT_DEFINE( , , page_fault, write);
#ifdef SMP
#include <machine/smp.h>
#endif
#include <machine/stack.h>
#include <machine/tss.h>
#ifdef KDTRACE_HOOKS
@ -202,18 +204,25 @@ trap(struct trapframe *frame)
goto out;
}
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI. If the PMC module is
* active, pass the 'rip' value to the PMC module's interrupt
* handler. A return value of '1' from the handler means that
* the NMI was handled by it and we can return immediately.
*/
if (type == T_NMI && pmc_intr &&
(*pmc_intr)(PCPU_GET(cpuid), frame))
goto out;
if (type == T_NMI) {
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI. If the PMC module is
* active, pass the 'rip' value to the PMC module's interrupt
* handler. A non-zero return value from the handler means that
* the NMI was consumed by it and we can return immediately.
*/
if (pmc_intr != NULL &&
(*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
goto out;
#endif
#ifdef STACK
if (stack_nmi_handler(frame) != 0)
goto out;
#endif
}
if (type == T_MCHK) {
mca_intr();
goto out;

7
sys/arm/arm/stack_machdep.c
View File

@ -71,6 +71,13 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(st, frame);
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
return (EOPNOTSUPP);
}
void
stack_save(struct stack *st)
{

7
sys/arm64/arm64/stack_machdep.c
View File

@ -72,6 +72,13 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(st, &frame);
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
return (EOPNOTSUPP);
}
void
stack_save(struct stack *st)
{

31
sys/i386/i386/trap.c
View File

@ -50,6 +50,7 @@ __FBSDID("$FreeBSD$");
#include "opt_isa.h"
#include "opt_kdb.h"
#include "opt_npx.h"
#include "opt_stack.h"
#include "opt_trap.h"
#include <sys/param.h>
@ -94,6 +95,7 @@ PMC_SOFT_DEFINE( , , page_fault, write);
#ifdef SMP
#include <machine/smp.h>
#endif
#include <machine/stack.h>
#include <machine/tss.h>
#include <machine/vm86.h>
@ -219,19 +221,26 @@ trap(struct trapframe *frame)
goto out;
}
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI so we check for that first.
* If the HWPMC module is active, 'pmc_hook' will point to
* the function to be called. A return value of '1' from the
* hook means that the NMI was handled by it and that we can
* return immediately.
*/
if (type == T_NMI && pmc_intr &&
(*pmc_intr)(PCPU_GET(cpuid), frame))
goto out;
if (type == T_NMI) {
#ifdef HWPMC_HOOKS
/*
* CPU PMCs interrupt using an NMI so we check for that first.
* If the HWPMC module is active, 'pmc_hook' will point to
* the function to be called. A non-zero return value from the
* hook means that the NMI was consumed by it and that we can
* return immediately.
*/
if (pmc_intr != NULL &&
(*pmc_intr)(PCPU_GET(cpuid), frame) != 0)
goto out;
#endif
#ifdef STACK
if (stack_nmi_handler(frame) != 0)
goto out;
#endif
}
if (type == T_MCHK) {
mca_intr();
goto out;

11
sys/kern/kern_proc.c
View File

@ -2517,11 +2517,14 @@ sysctl_kern_proc_kstack(SYSCTL_HANDLER_ARGS)
sizeof(kkstp->kkst_trace), SBUF_FIXEDLEN);
thread_lock(td);
kkstp->kkst_tid = td->td_tid;
if (TD_IS_SWAPPED(td))
if (TD_IS_SWAPPED(td)) {
kkstp->kkst_state = KKST_STATE_SWAPPED;
else if (TD_IS_RUNNING(td))
kkstp->kkst_state = KKST_STATE_RUNNING;
else {
} else if (TD_IS_RUNNING(td)) {
if (stack_save_td_running(st, td) == 0)
kkstp->kkst_state = KKST_STATE_STACKOK;
else
kkstp->kkst_state = KKST_STATE_RUNNING;
} else {
kkstp->kkst_state = KKST_STATE_STACKOK;
stack_save_td(st, td);
}

7
sys/mips/mips/stack_machdep.c
View File

@ -142,6 +142,13 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(st, pc, sp);
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
return (EOPNOTSUPP);
}
void
stack_save(struct stack *st)
{

7
sys/powerpc/powerpc/stack_machdep.c
View File

@ -98,6 +98,13 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(st, frame);
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
return (EOPNOTSUPP);
}
void
stack_save(struct stack *st)
{

7
sys/sparc64/sparc64/stack_machdep.c
View File

@ -82,6 +82,13 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(st, (struct frame *)(td->td_pcb->pcb_sp + SPOFF));
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
return (EOPNOTSUPP);
}
void
stack_save(struct stack *st)
{

3
sys/sys/stack.h
View File

@ -56,9 +56,10 @@ void stack_ktr(u_int, const char *, int, const struct stack *,
#define CTRSTACK(m, st, depth, cheap)
#endif
/* MD Routine. */
/* MD Routines. */
struct thread;
void stack_save(struct stack *);
void stack_save_td(struct stack *, struct thread *);
int stack_save_td_running(struct stack *, struct thread *);
#endif

4
sys/x86/include/apicvar.h
View File

@ -129,12 +129,14 @@
#else
#define IPI_DYN_FIRST (APIC_IPI_INTS + 8)
#endif
#define IPI_DYN_LAST (254) /* IPIs allocated at runtime */
#define IPI_DYN_LAST (253) /* IPIs allocated at runtime */
/*
* IPI_STOP_HARD does not need to occupy a slot in the IPI vector space since
* it is delivered using an NMI anyways.
*/
#define IPI_NMI_FIRST 254
#define IPI_TRACE 254 /* Interrupt for tracing. */
#define IPI_STOP_HARD 255 /* Stop CPU with a NMI. */
/*

4
sys/x86/include/stack.h
View File

@ -54,4 +54,8 @@ struct i386_frame {
};
#endif /* __amd64__ */
#ifdef _KERNEL
int stack_nmi_handler(struct trapframe *);
#endif
#endif /* !_X86_STACK_H */

7
sys/x86/x86/local_apic.c
View File

@ -1703,11 +1703,10 @@ native_lapic_ipi_vectored(u_int vector, int dest)
icrlo = APIC_DESTMODE_PHY | APIC_TRIGMOD_EDGE | APIC_LEVEL_ASSERT;
/*
* IPI_STOP_HARD is just a "fake" vector used to send a NMI.
* Use special rules regard NMI if passed, otherwise specify
* the vector.
* NMI IPIs are just fake vectors used to send a NMI. Use special rules
* regarding NMIs if passed, otherwise specify the vector.
*/
if (vector == IPI_STOP_HARD)
if (vector >= IPI_NMI_FIRST)
icrlo |= APIC_DELMODE_NMI;
else
icrlo |= vector | APIC_DELMODE_FIXED;

12
sys/x86/x86/mp_x86.c
View File

@ -120,7 +120,7 @@ struct cpu_ops cpu_ops;
* Local data and functions.
*/
static volatile cpuset_t ipi_nmi_pending;
static volatile cpuset_t ipi_stop_nmi_pending;
/* used to hold the AP's until we are ready to release them */
struct mtx ap_boot_mtx;
@ -894,7 +894,7 @@ ipi_selected(cpuset_t cpus, u_int ipi)
* Set the mask of receiving CPUs for this purpose.
*/
if (ipi == IPI_STOP_HARD)
CPU_OR_ATOMIC(&ipi_nmi_pending, &cpus);
CPU_OR_ATOMIC(&ipi_stop_nmi_pending, &cpus);
while ((cpu = CPU_FFS(&cpus)) != 0) {
cpu--;
@ -917,7 +917,7 @@ ipi_cpu(int cpu, u_int ipi)
* Set the mask of receiving CPUs for this purpose.
*/
if (ipi == IPI_STOP_HARD)
CPU_SET_ATOMIC(cpu, &ipi_nmi_pending);
CPU_SET_ATOMIC(cpu, &ipi_stop_nmi_pending);
CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi);
ipi_send_cpu(cpu, ipi);
@ -944,7 +944,7 @@ ipi_all_but_self(u_int ipi)
* Set the mask of receiving CPUs for this purpose.
*/
if (ipi == IPI_STOP_HARD)
CPU_OR_ATOMIC(&ipi_nmi_pending, &other_cpus);
CPU_OR_ATOMIC(&ipi_stop_nmi_pending, &other_cpus);
CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi);
lapic_ipi_vectored(ipi, APIC_IPI_DEST_OTHERS);
@ -962,10 +962,10 @@ ipi_nmi_handler()
* and should be handled.
*/
cpuid = PCPU_GET(cpuid);
if (!CPU_ISSET(cpuid, &ipi_nmi_pending))
if (!CPU_ISSET(cpuid, &ipi_stop_nmi_pending))
return (1);
CPU_CLR_ATOMIC(cpuid, &ipi_nmi_pending);
CPU_CLR_ATOMIC(cpuid, &ipi_stop_nmi_pending);
cpustop_handler();
return (0);
}

68
sys/x86/x86/stack_machdep.c
View File

@ -1,4 +1,5 @@
/*-
* Copyright (c) 2015 EMC Corporation
* Copyright (c) 2005 Antoine Brodin
* All rights reserved.
*
@ -29,17 +30,21 @@ __FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/stack.h>
#include <x86/stack.h>
#include <machine/pcb.h>
#include <machine/smp.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <x86/stack.h>
#ifdef __i386__
#define PCB_FP(pcb) ((pcb)->pcb_ebp)
#define TF_FP(tf) ((tf)->tf_ebp)
@ -54,6 +59,14 @@ typedef struct i386_frame *x86_frame_t;
typedef struct amd64_frame *x86_frame_t;
#endif
static struct stack *nmi_stack;
static volatile struct thread *nmi_pending;
#ifdef SMP
static struct mtx nmi_lock;
MTX_SYSINIT(nmi_lock, &nmi_lock, "stack_nmi", MTX_SPIN);
#endif
static void
stack_capture(struct thread *td, struct stack *st, register_t fp)
{
@ -78,6 +91,24 @@ stack_capture(struct thread *td, struct stack *st, register_t fp)
}
}
int
stack_nmi_handler(struct trapframe *tf)
{
/* Don't consume an NMI that wasn't meant for us. */
if (nmi_stack == NULL || curthread != nmi_pending)
return (0);
if (INKERNEL(TF_PC(tf)))
stack_capture(curthread, nmi_stack, TF_FP(tf));
else
/* We interrupted a thread in user mode. */
nmi_stack->depth = 0;
atomic_store_rel_ptr((long *)&nmi_pending, (long)NULL);
return (1);
}
void
stack_save_td(struct stack *st, struct thread *td)
{
@ -90,6 +121,39 @@ stack_save_td(struct stack *st, struct thread *td)
stack_capture(td, st, PCB_FP(td->td_pcb));
}
int
stack_save_td_running(struct stack *st, struct thread *td)
{
THREAD_LOCK_ASSERT(td, MA_OWNED);
MPASS(TD_IS_RUNNING(td));
if (td == curthread) {
stack_save(st);
return (0);
}
#ifdef SMP
mtx_lock_spin(&nmi_lock);
nmi_stack = st;
nmi_pending = td;
ipi_cpu(td->td_oncpu, IPI_TRACE);
while ((void *)atomic_load_acq_ptr((long *)&nmi_pending) != NULL)
cpu_spinwait();
nmi_stack = NULL;
mtx_unlock_spin(&nmi_lock);
if (st->depth == 0)
/* We interrupted a thread in user mode. */
return (EAGAIN);
#else
KASSERT(0, ("curthread isn't running"));
#endif
return (0);
}
void
stack_save(struct stack *st)
{