8:14 a.m.
Hello Dave,
This patch series for gcore extension module fix a bug and improves
register restoration for active tasks. Could you review these? I'm
going to release these after your review.
In summary,
- The first patch fixes a bug that a target task context is wrongly
specified in the process of getting NT_PRSTATUS on vmcore in
ELF. Without this patch, all the register values but the one of
the task user mannually specifies are unintensionally equal each
other.
- The remainig patches implements register restorement for active
tasks from exception frame and KVMDUMP notes. By this, register
values for active tasks can be correctly restored for KDUMP,
NETDUMP, DISKDUMP and KVMDUMP formats.
Thanks.
HATAYAMA Daisuke
7:41 a.m.
----- Original Message -----
Hello Dave,
This patch series for gcore extension module fix a bug and improves
register restoration for active tasks. Could you review these? I'm
going to release these after your review.
The patches look fine to me -- although it is not necessary for you
to get my review/approval to update the gcore source code. You are the
maintainer, so you can send me an updated gcore.tar.bz2 file whenever
you like, and I will post it on the crash extensions web page.
Thanks,
Dave
In summary,
- The first patch fixes a bug that a target task context is wrongly
specified in the process of getting NT_PRSTATUS on vmcore in
ELF. Without this patch, all the register values but the one of
the task user mannually specifies are unintensionally equal each
other.
- The remainig patches implements register restorement for active
tasks from exception frame and KVMDUMP notes. By this, register
values for active tasks can be correctly restored for KDUMP,
NETDUMP, DISKDUMP and KVMDUMP formats.
Thanks.
HATAYAMA Daisuke
8:14 a.m.
From: Dave Anderson <anderson(a)redhat.com>
Subject: Re: [Crash-utility] [PATCH 0/6] gcore: a bug fix and improvements on register
restoration for active tasks
Date: Tue, 8 Feb 2011 08:41:24 -0500 (EST)
----- Original Message -----
> Hello Dave,
>
> This patch series for gcore extension module fix a bug and improves
> register restoration for active tasks. Could you review these? I'm
> going to release these after your review.
The patches look fine to me -- although it is not necessary for you
to get my review/approval to update the gcore source code. You are the
maintainer, so you can send me an updated gcore.tar.bz2 file whenever
you like, and I will post it on the crash extensions web page.
Thanks for your review.
I've understood the procedure. You told me that some times.
The review was to confirm that the usage of crash library is correct.
Well, I'll release a new version soon.
Thanks.
HATAYAMA Daisuke
8:14 a.m.
New subject: [PATCH 1/6] x86_64: Choose note information corredponding to proper task context
Each NT_PRSTATUS corresponds to each CPU, but here CURRENT_CONTEXT()
is wrongly specified, so all the register values for active tasks
except for the task of the current context is wrongly equal to that of
CURRENT_CONTEXT().
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 1e98b48..781ee38 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -1369,7 +1369,7 @@ static int genregs_get(struct task_context *target,
if (active && KDUMP_DUMPFILE()) {
struct user_regs_struct *note_regs_p;
- note_regs_p = get_regs_from_elf_notes(CURRENT_CONTEXT());
+ note_regs_p = get_regs_from_elf_notes(target);
memcpy(¬e_regs, note_regs_p, sizeof(struct user_regs_struct));
/*
--
1.7.4
8:14 a.m.
New subject: [PATCH 2/6] x86: Introduce get_active_regs()
There are several methods to get register values for active tasks such as note
information and exception frames on kernel stacks. This is the preparation
patch to support such features abstracting them as the feature collecting
active registers.
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 41 ++++++++++++++++++++---------------------
1 files changed, 20 insertions(+), 21 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 781ee38..377df6d 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -1354,30 +1354,29 @@ static ulong gcore_x86_64_get_cpu__pda_oldrsp(int cpu)
return oldrsp;
}
+static int get_active_regs(struct task_context *target,
+ struct user_regs_struct *regs)
+{
+ if (get_netdump_arch() != EM_NONE) {
+ struct user_regs_struct *note = get_regs_from_elf_notes(target);
+ memcpy(regs, note, sizeof(struct user_regs_struct));
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
static int genregs_get(struct task_context *target,
const struct user_regset *regset,
unsigned int size, void *buf)
{
struct user_regs_struct *regs = (struct user_regs_struct *)buf;
- struct user_regs_struct note_regs;
+ struct user_regs_struct active_regs;
const int active = is_task_active(target->task);
- /*
- * vmcore generated by kdump contains NT_PRSTATUS including
- * general register values for active tasks.
- */
- if (active && KDUMP_DUMPFILE()) {
- struct user_regs_struct *note_regs_p;
-
- note_regs_p = get_regs_from_elf_notes(target);
- memcpy(¬e_regs, note_regs_p, sizeof(struct user_regs_struct));
-
- /*
- * If the task was in kernel-mode at the kernel crash, note
- * information is not what we would like.
- */
- if (user_mode(¬e_regs)) {
- memcpy(regs, ¬e_regs, sizeof(struct user_regs_struct));
+ if (active && get_active_regs(target, &active_regs)) {
+ if (user_mode(&active_regs)) {
+ memcpy(regs, &active_regs, sizeof(*regs));
return 0;
}
}
@@ -1415,7 +1414,7 @@ static int genregs_get(struct task_context *target,
* entire registers are saved for special system calls.
*/
if (!gxt->is_special_syscall(nr_syscall))
- restore_rest(target->task, (struct pt_regs *)regs, ¬e_regs);
+ restore_rest(target->task, (struct pt_regs *)regs, &active_regs);
/*
* See FIXUP_TOP_OF_STACK in arch/x86/kernel/entry_64.S.
@@ -1437,7 +1436,7 @@ static int genregs_get(struct task_context *target,
/* Exceptions and NMI */
else if (vector < 20) {
restore_rest(target->task, (struct pt_regs *)regs,
- ¬e_regs);
+ &active_regs);
restore_segment_registers(target->task, regs);
}
@@ -1454,14 +1453,14 @@ static int genregs_get(struct task_context *target,
if (!gxt->is_special_ia32_syscall(nr_syscall))
restore_rest(target->task,
(struct pt_regs *)regs,
- ¬e_regs);
+ &active_regs);
restore_segment_registers(target->task, regs);
}
/* Muskable Interrupts */
else if (vector < 256) {
restore_rest(target->task, (struct pt_regs *)regs,
- ¬e_regs);
+ &active_regs);
restore_segment_registers(target->task, regs);
}
--
1.7.4
8:14 a.m.
New subject: [PATCH 3/6] x86_64: Collect register values on exception frame
KDUMP, NETDUMP and DISKDUMP all saves register values for active tasks onto
exception frame at kernel crash. This patch implements the logic to collect the
register values by retrieving EFRAME information from back_trace()'s output.
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 67 ++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 67 insertions(+), 0 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 377df6d..29fe808 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -1354,6 +1354,64 @@ static ulong gcore_x86_64_get_cpu__pda_oldrsp(int cpu)
return oldrsp;
}
+static int
+gcore_find_regs_from_bt_output(FILE *output, char *buf, size_t bufsize)
+{
+ while (fgets(buf, bufsize, output))
+ if (strncmp(buf, " RIP:", 8) == 0)
+ return TRUE;
+
+ return FALSE;
+}
+
+static int
+gcore_get_regs_from_bt_output(FILE *output, struct pt_regs *regs)
+{
+ char buf[BUFSIZE];
+
+ if (gcore_find_regs_from_bt_output(output, buf, BUFSIZE) == FALSE)
+ return FALSE;
+
+ sscanf(buf, " RIP: %016lx RSP: %016lx RFLAGS: %08lx\n",
+ ®s->rip, ®s->rsp, ®s->eflags);
+ fscanf(output, " RAX: %016lx RBX: %016lx RCX: %016lx\n",
+ ®s->rax, ®s->rbx, ®s->rcx);
+ fscanf(output, " RDX: %016lx RSI: %016lx RDI: %016lx\n",
+ ®s->rdx, ®s->rsi, ®s->rdi);
+ fscanf(output, " RBP: %016lx R8: %016lx R9: %016lx\n",
+ ®s->rbp, ®s->r8, ®s->r9);
+ fscanf(output, " R10: %016lx R11: %016lx R12: %016lx\n",
+ ®s->r10, ®s->r11, ®s->r12);
+ fscanf(output, " R13: %016lx R14: %016lx R15: %016lx\n",
+ ®s->r13, ®s->r14, ®s->r15);
+ fscanf(output, " ORIG_RAX: %016lx CS: %04lx SS: %04lx\n",
+ ®s->orig_rax, ®s->cs, ®s->ss);
+
+ return TRUE;
+}
+
+static int
+gcore_get_regs_from_eframe(struct task_context *tc, struct pt_regs *regs)
+{
+ int ret;
+ struct bt_info bt;
+
+ BZERO(&bt, sizeof(struct bt_info));
+ bt.stackbuf = NULL;
+ bt.tc = tc;
+ bt.task = tc->task;
+ bt.stackbase = GET_STACKBASE(tc->task);
+ bt.stacktop = GET_STACKTOP(tc->task);
+
+ open_tmpfile();
+ back_trace(&bt);
+ rewind(pc->tmpfile);
+ ret = gcore_get_regs_from_bt_output(pc->tmpfile, regs);
+ close_tmpfile();
+
+ return ret;
+}
+
static int get_active_regs(struct task_context *target,
struct user_regs_struct *regs)
{
@@ -1363,6 +1421,15 @@ static int get_active_regs(struct task_context *target,
return TRUE;
}
+ if (gcore_get_regs_from_eframe(target, (struct pt_regs *)regs)) {
+ /*
+ * EFRAME contains CS and SS only. Here collects the
+ * remaining part of segment registers.
+ */
+ restore_segment_registers(target->task, regs);
+ return TRUE;
+ }
+
return FALSE;
}
--
1.7.4
8:14 a.m.
New subject: [PATCH 4/6] x86_64: Collect register values for active tasks from KVMDUMP notes
KVMDUMP format contains register values for active tasks. So, get the
register values in case of KVMDUMP format.
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 38 ++++++++++++++++++++++++++++++++++++++
1 files changed, 38 insertions(+), 0 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 29fe808..03c3579 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -15,6 +15,7 @@
#if defined(X86) || defined(X86_64)
#include "defs.h"
+#include "kvmdump.h"
#ifdef X86_64
#include "unwind_x86_64.h"
#endif
@@ -1412,9 +1413,46 @@ gcore_get_regs_from_eframe(struct task_context *tc, struct pt_regs
*regs)
return ret;
}
+static void
+get_regs_from_kvmdump_notes(struct task_context *target,
+ struct user_regs_struct *regs)
+{
+ struct register_set *r = &kvm->registers[target->processor];
+
+ regs->cs = r->cs;
+ regs->ss = r->ss;
+ regs->ds = r->ds;
+ regs->es = r->es;
+ regs->fs = r->fs;
+ regs->gs = r->gs;
+ regs->ip = r->ip;
+ regs->flags = r->flags;
+ regs->ax = r->regs[0];
+ regs->cx = r->regs[1];
+ regs->dx = r->regs[2];
+ regs->bx = r->regs[3];
+ regs->sp = r->regs[4];
+ regs->bp = r->regs[5];
+ regs->si = r->regs[6];
+ regs->di = r->regs[7];
+ regs->r8 = r->regs[8];
+ regs->r9 = r->regs[9];
+ regs->r10 = r->regs[10];
+ regs->r11 = r->regs[11];
+ regs->r12 = r->regs[12];
+ regs->r13 = r->regs[13];
+ regs->r14 = r->regs[14];
+ regs->r15 = r->regs[15];
+}
+
static int get_active_regs(struct task_context *target,
struct user_regs_struct *regs)
{
+ if (KVMDUMP_DUMPFILE()) {
+ get_regs_from_kvmdump_notes(target, regs);
+ return TRUE;
+ }
+
if (get_netdump_arch() != EM_NONE) {
struct user_regs_struct *note = get_regs_from_elf_notes(target);
memcpy(regs, note, sizeof(struct user_regs_struct));
--
1.7.4
8:14 a.m.
New subject: [PATCH 5/6] x86: Collect register values for active tasks from KVMDUMP notes
KVMDUMP format contains register values for active tasks. So, get the
register values in case of KVMDUMP format.
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 31 +++++++++++++++++++++++++++++++
1 files changed, 31 insertions(+), 0 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 03c3579..8c129d3 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -1742,6 +1742,30 @@ void gcore_x86_64_regsets_init(void)
#endif /* X86_64 */
#ifdef X86
+static void
+get_regs_from_kvmdump_notes(struct task_context *target,
+ struct user_regs_struct *regs)
+{
+ struct register_set *r = &kvm->registers[target->processor];
+
+ regs->ax = r->regs[0];
+ regs->cx = r->regs[1];
+ regs->dx = r->regs[2];
+ regs->bx = r->regs[3];
+ regs->sp = r->regs[4];
+ regs->bp = r->regs[5];
+ regs->si = r->regs[6];
+ regs->di = r->regs[7];
+ regs->cs = r->cs;
+ regs->ss = r->ss;
+ regs->ds = r->ds;
+ regs->es = r->es;
+ regs->fs = r->fs;
+ regs->gs = r->gs;
+ regs->ip = r->ip;
+ regs->flags = r->flags;
+}
+
static int genregs_get32(struct task_context *target,
const struct user_regset *regset,
unsigned int size, void *buf)
@@ -1750,6 +1774,13 @@ static int genregs_get32(struct task_context *target,
char *pt_regs_buf;
ulonglong pt_regs_addr;
+ if (is_task_active(target->task) && KVMDUMP_DUMPFILE()) {
+ get_regs_from_kvmdump_notes(target, regs);
+ if (user_mode(regs)) {
+ return TRUE;
+ }
+ }
+
pt_regs_buf = GETBUF(SIZE(pt_regs));
pt_regs_addr = machdep->get_stacktop(target->task) - SIZE(pt_regs);
--
1.7.4
8:14 a.m.
New subject: [PATCH 6/6] x86_64: Trace RBP chain on kernel stack only if unwind() fails
- unwind() logic attempts to collect all the callee-saved registers including
RBP. So, RBP chain tracing is necessary only if this unwind() logic doesn't
succeed.
- Also, verify whether a given task is not active. This is because for active
tasks, RBP saved on the corresponding kernel stacks could be modified.
Signed-off-by: HATAYAMA Daisuke <d.hatayama(a)jp.fujitsu.com>
---
src/libgcore/gcore_x86.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/libgcore/gcore_x86.c b/src/libgcore/gcore_x86.c
index 8c129d3..871a9db 100644
--- a/src/libgcore/gcore_x86.c
+++ b/src/libgcore/gcore_x86.c
@@ -1258,7 +1258,7 @@ static inline void restore_rest(ulong task, struct pt_regs *regs,
* could trace the value of bp until its value became a
* user-space address. See comments of restore_frame_pointer.
*/
- if (machdep->flags & FRAMEPOINTER) {
+ else if ((machdep->flags & FRAMEPOINTER) && !is_task_active(task)) {
regs->rbp = restore_frame_pointer(task);
}
}
--
1.7.4
231
days inactive
231
days old
devel@lists.crash-utility.osci.io
8 comments
2 participants
participants (2)
-
Dave Anderson -
HATAYAMA Daisuke