2:09 a.m.
This patchset deals with the issue when CONFIG_VMAP_STACK
enabled, kmem fails to print task context.
patch 1 deals with "kmem <virtual_address>"
patch 3 deals with "kmem <physical_address>"
patch 2 is needed by patch 3, which will convert the correct
physical address to virtual address.
v2 -> v3: Add the kmem <physical_address> case.
v1: https://listman.redhat.com/archives/crash-utility/2022-August/010061.html
v2: https://listman.redhat.com/archives/crash-utility/2022-August/010074.html
Tao Liu (3):
Fix kmem failing to print task context when address is vmalloced stack
Fix page offset issue when converting physical to virtual address
Let kmem print task context when physical address is vmalloced stack
memory.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
--
2.33.1
2:09 a.m.
New subject: [PATCH v3 1/3] Fix kmem failing to print task context when address is vmalloced stack
When kernel enabled CONFIG_VMAP_STACK, stack can be allocated to
vmalloced area. Currently crash didn't handle the case, as a result,
kmem will not print the task context as expected. This patch fix the
bug by checking if the address is a vmalloced stack first.
Before:
crash> kmem ffffb7efce9bbe28
VMAP_AREA VM_STRUCT ADDRESS RANGE SIZE
ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000 20480
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffdd28220dc000 1883700000 0 0 1 50000000000000
After:
crash> kmem ffffb7efce9bbe28
PID: 847
COMMAND: "khungtaskd"
TASK: ffff94f8038f4000 [THREAD_INFO: ffff94f8038f4000]
CPU: 72
STATE: TASK_RUNNING (PANIC)
VMAP_AREA VM_STRUCT ADDRESS RANGE SIZE
ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000 20480
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffdd28220dc000 1883700000 0 0 1 50000000000000
Signed-off-by: Tao Liu <ltao(a)redhat.com>
---
memory.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/memory.c b/memory.c
index 7339f0c..9ab5781 100644
--- a/memory.c
+++ b/memory.c
@@ -13477,6 +13477,10 @@ kmem_search(struct meminfo *mi)
* Check for a valid mapped address.
*/
if ((mi->memtype == KVADDR) && IS_VMALLOC_ADDR(mi->spec_addr)) {
+ if ((task = stkptr_to_task(vaddr)) && (tc = task_to_context(task))) {
+ show_context(tc);
+ fprintf(fp, "\n");
+ }
if (kvtop(NULL, mi->spec_addr, &paddr, 0)) {
mi->flags = orig_flags | VMLIST_VERIFY;
dump_vmlist(mi);
--
2.33.1
2:09 a.m.
New subject: [PATCH v3 2/3] Fix page offset issue when converting physical to virtual address
When trying to convert a physical address to its virtual
address in dump_vmap_area() and dump_vmlist(), the vi->retval
is added by 2 values: the page aligned address "pcheck"
and page offset address "PAGEOFFSET(paddr)".
However "paddr" is given by "pcheck", is also page aligned,
so "PAGEOFFSET(paddr)" is always 0.
In this patch, we will use PAGEOFFSET(vi->spec_addr) to give the
page offset, vi->spec_addr is the physical address we'd like
to convert, which contains the correct page offset.
Signed-off-by: Tao Liu <ltao(a)redhat.com>
---
memory.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/memory.c b/memory.c
index 9ab5781..1b6f9ba 100644
--- a/memory.c
+++ b/memory.c
@@ -8861,7 +8861,7 @@ dump_vmlist(struct meminfo *vi)
(vi->spec_addr < (paddr+PAGESIZE()))) {
if (vi->flags & GET_PHYS_TO_VMALLOC) {
vi->retval = pcheck +
- PAGEOFFSET(paddr);
+ PAGEOFFSET(vi->spec_addr);
return;
} else
fprintf(fp,
@@ -9010,7 +9010,7 @@ dump_vmap_area(struct meminfo *vi)
(vi->spec_addr < (paddr+PAGESIZE()))) {
if (vi->flags & GET_PHYS_TO_VMALLOC) {
vi->retval = pcheck +
- PAGEOFFSET(paddr);
+ PAGEOFFSET(vi->spec_addr);
FREEBUF(ld->list_ptr);
return;
} else
--
2.33.1
2:09 a.m.
New subject: [PATCH v3 3/3] Let kmem print task context when physical address is vmalloced stack
Patch [1] enables kmem to print task context if the given virtual
address is a vmalloced stack.
This patch handles the same issue except the given address is
physical address.
Before:
crash> kmem 1883700e28
VMAP_AREA VM_STRUCT ADDRESS RANGE SIZE
ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000 20480
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffdd28220dc000 1883700000 0 0 1 50000000000000
After:
crash> kmem 1883700e28
PID: 847
COMMAND: "khungtaskd"
TASK: ffff94f8038f4000 [THREAD_INFO: ffff94f8038f4000]
CPU: 72
STATE: TASK_RUNNING (PANIC)
VMAP_AREA VM_STRUCT ADDRESS RANGE SIZE
ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000 20480
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffdd28220dc000 1883700000 0 0 1 50000000000000
[1]: https://listman.redhat.com/archives/crash-utility/2022-August/010074.html
Signed-off-by: Tao Liu <ltao(a)redhat.com>
---
memory.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/memory.c b/memory.c
index 1b6f9ba..eac8262 100644
--- a/memory.c
+++ b/memory.c
@@ -13506,6 +13506,10 @@ kmem_search(struct meminfo *mi)
mi->flags &= ~GET_PHYS_TO_VMALLOC;
if (mi->retval) {
+ if ((task = stkptr_to_task(mi->retval)) && (tc = task_to_context(task))) {
+ show_context(tc);
+ fprintf(fp, "\n");
+ }
if ((sp = value_search(mi->retval, &offset))) {
show_symbol(sp, offset,
SHOW_LINENUM | SHOW_RADIX());
--
2.33.1
12:32 a.m.
New subject: [PATCH v3 3/3] Let kmem print task context when physical address is vmalloced stack
On 2022/09/05 16:09, Tao Liu wrote:
Patch [1] enables kmem to print task context if the given virtual
address is a vmalloced stack.
This patch handles the same issue except the given address is
physical address.
If we do this, it would be better to support a kernel without
CONFIG_VMAP_STACK, too. Any reason not to support?
crash> sys
KERNEL: vmlinux-3.10.0-957.el7.x86_64
...
crash> kmem ffff88007cfdbf80
PID: 1
COMMAND: "systemd"
TASK: ffff88007cfd0000 [THREAD_INFO: ffff88007cfd8000]
CPU: 0
STATE: TASK_INTERRUPTIBLE
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffea0001f3f6c0 7cfdb000 0 0 0 1fffff00000000
crash> kmem 7cfdbf80
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffea0001f3f6c0 7cfdb000 0 0 0 1fffff00000000
Thanks,
Kazu
>
> Before:
> crash> kmem 1883700e28
> VMAP_AREA VM_STRUCT ADDRESS RANGE
SIZE
> ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000
20480
>
> PAGE PHYSICAL MAPPING INDEX CNT FLAGS
> ffffdd28220dc000 1883700000 0 0 1 50000000000000
>
> After:
> crash> kmem 1883700e28
> PID: 847
> COMMAND: "khungtaskd"
> TASK: ffff94f8038f4000 [THREAD_INFO: ffff94f8038f4000]
> CPU: 72
> STATE: TASK_RUNNING (PANIC)
>
> VMAP_AREA VM_STRUCT ADDRESS RANGE
SIZE
> ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000
20480
>
> PAGE PHYSICAL MAPPING INDEX CNT FLAGS
> ffffdd28220dc000 1883700000 0 0 1 50000000000000
>
> [1]: https://listman.redhat.com/archives/crash-utility/2022-August/010074.html
>
> Signed-off-by: Tao Liu <ltao(a)redhat.com>
> ---
> memory.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/memory.c b/memory.c
> index 1b6f9ba..eac8262 100644
> --- a/memory.c
> +++ b/memory.c
> @@ -13506,6 +13506,10 @@ kmem_search(struct meminfo *mi)
> mi->flags &= ~GET_PHYS_TO_VMALLOC;
>
> if (mi->retval) {
> + if ((task = stkptr_to_task(mi->retval)) && (tc =
task_to_context(task))) {
> + show_context(tc);
> + fprintf(fp, "\n");
> + }
> if ((sp = value_search(mi->retval, &offset))) {
> show_symbol(sp, offset,
> SHOW_LINENUM | SHOW_RADIX());
7:55 p.m.
New subject: [PATCH v3 3/3] Let kmem print task context when physical address is vmalloced stack
Hi Kazu,
On Tue, Sep 6, 2022 at 1:32 PM HAGIO KAZUHITO(萩尾 一仁) <k-hagio-ab(a)nec.com> wrote:
On 2022/09/05 16:09, Tao Liu wrote:
> Patch [1] enables kmem to print task context if the given virtual
> address is a vmalloced stack.
>
> This patch handles the same issue except the given address is
> physical address.
If we do this, it would be better to support a kernel without
CONFIG_VMAP_STACK, too. Any reason not to support?
I will support it in the v4 patch. Sorry for the late reply, I was
busy working on the maple tree support for crash, which is in a final
stage and code refactoring/testing. Maybe it will take a few days
before I work on the v4.
Thanks,
Tao Liu
crash> sys
KERNEL: vmlinux-3.10.0-957.el7.x86_64
...
crash> kmem ffff88007cfdbf80
PID: 1
COMMAND: "systemd"
TASK: ffff88007cfd0000 [THREAD_INFO: ffff88007cfd8000]
CPU: 0
STATE: TASK_INTERRUPTIBLE
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffea0001f3f6c0 7cfdb000 0 0 0 1fffff00000000
crash> kmem 7cfdbf80
PAGE PHYSICAL MAPPING INDEX CNT FLAGS
ffffea0001f3f6c0 7cfdb000 0 0 0 1fffff00000000
Thanks,
Kazu
>
> Before:
> crash> kmem 1883700e28
> VMAP_AREA VM_STRUCT ADDRESS RANGE
SIZE
> ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000
20480
>
> PAGE PHYSICAL MAPPING INDEX CNT FLAGS
> ffffdd28220dc000 1883700000 0 0 1 50000000000000
>
> After:
> crash> kmem 1883700e28
> PID: 847
> COMMAND: "khungtaskd"
> TASK: ffff94f8038f4000 [THREAD_INFO: ffff94f8038f4000]
> CPU: 72
> STATE: TASK_RUNNING (PANIC)
>
> VMAP_AREA VM_STRUCT ADDRESS RANGE
SIZE
> ffff94eb9102c640 ffff94eb9102b140 ffffb7efce9b8000 - ffffb7efce9bd000
20480
>
> PAGE PHYSICAL MAPPING INDEX CNT FLAGS
> ffffdd28220dc000 1883700000 0 0 1 50000000000000
>
> [1]: https://listman.redhat.com/archives/crash-utility/2022-August/010074.html
>
> Signed-off-by: Tao Liu <ltao(a)redhat.com>
> ---
> memory.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/memory.c b/memory.c
> index 1b6f9ba..eac8262 100644
> --- a/memory.c
> +++ b/memory.c
> @@ -13506,6 +13506,10 @@ kmem_search(struct meminfo *mi)
> mi->flags &= ~GET_PHYS_TO_VMALLOC;
>
> if (mi->retval) {
> + if ((task = stkptr_to_task(mi->retval)) && (tc =
task_to_context(task))) {
> + show_context(tc);
> + fprintf(fp, "\n");
> + }
> if ((sp = value_search(mi->retval, &offset))) {
> show_symbol(sp, offset,
> SHOW_LINENUM | SHOW_RADIX());
8:30 p.m.
New subject: [PATCH v3 3/3] Let kmem print task context when physical address is vmalloced stack
On 2022/09/08 9:55, Tao Liu wrote:
Hi Kazu,
On Tue, Sep 6, 2022 at 1:32 PM HAGIO KAZUHITO(萩尾 一仁) <k-hagio-ab(a)nec.com> wrote:
>
>
> On 2022/09/05 16:09, Tao Liu wrote:
>> Patch [1] enables kmem to print task context if the given virtual
>> address is a vmalloced stack.
>>
>> This patch handles the same issue except the given address is
>> physical address.
>
> If we do this, it would be better to support a kernel without
> CONFIG_VMAP_STACK, too. Any reason not to support?
>
I will support it in the v4 patch. Sorry for the late reply, I was
I would like it to be consistent in behavior for the different kernel
configs if it can be implemented easily. But if a big change is needed,
no need to work hard for this.. I can also ack this patch as it is.
busy working on the maple tree support for crash, which is in a
final
stage and code refactoring/testing. Maybe it will take a few days
before I work on the v4.
Good, thank you for the huge work!
Thanks,
Kazu
738
days inactive
742
days old
devel@lists.crash-utility.osci.io
6 comments
2 participants
participants (2)
-
HAGIO KAZUHITO(萩尾 一仁) -
Tao Liu