[v2] eal/linux: enable the hugepage mem dump
Checks
Commit Message
These hugepages include important structures. we should dump these
hugepages into a coredump file for debugging when generating a coredump.
Signed-off-by: Li Feng <fengli@smartx.com>
---
lib/eal/linux/eal_memalloc.c | 2 ++
1 file changed, 2 insertions(+)
Comments
On Fri, 1 Apr 2022 17:10:04 +0800
Li Feng <fengli@smartx.com> wrote:
> These hugepages include important structures. we should dump these
> hugepages into a coredump file for debugging when generating a coredump.
>
> Signed-off-by: Li Feng <fengli@smartx.com>
> ---
> lib/eal/linux/eal_memalloc.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/lib/eal/linux/eal_memalloc.c b/lib/eal/linux/eal_memalloc.c
> index f8b1588cae..93c4f396cf 100644
> --- a/lib/eal/linux/eal_memalloc.c
> +++ b/lib/eal/linux/eal_memalloc.c
> @@ -677,6 +677,8 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
> __func__);
> #endif
>
> + eal_mem_set_dump(addr, alloc_sz, true);
> +
> huge_recover_sigbus();
>
> ms->addr = addr;
Don't merge this patch as is please; it would cause a lot of pain
in a cloud environment.
In our environment core dumps are collected (via systemd) and uploaded
to a central server. With this kind of change the processing would get
overloaded with multi-gigabyte core dump size. Probably couldn't even
save a core dump on these kind of smart nics.
This needs to be optional (from command line) and default to the current
behavior (not dumping huge pages).
2022-04-05 15:46 (UTC-0700), Stephen Hemminger:
> On Fri, 1 Apr 2022 17:10:04 +0800
> Li Feng <fengli@smartx.com> wrote:
>
> > These hugepages include important structures. we should dump these
> > hugepages into a coredump file for debugging when generating a coredump.
> >
> > Signed-off-by: Li Feng <fengli@smartx.com>
> > ---
> > lib/eal/linux/eal_memalloc.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/lib/eal/linux/eal_memalloc.c b/lib/eal/linux/eal_memalloc.c
> > index f8b1588cae..93c4f396cf 100644
> > --- a/lib/eal/linux/eal_memalloc.c
> > +++ b/lib/eal/linux/eal_memalloc.c
> > @@ -677,6 +677,8 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
> > __func__);
> > #endif
> >
> > + eal_mem_set_dump(addr, alloc_sz, true);
> > +
> > huge_recover_sigbus();
> >
> > ms->addr = addr;
>
>
> Don't merge this patch as is please; it would cause a lot of pain
> in a cloud environment.
>
> In our environment core dumps are collected (via systemd) and uploaded
> to a central server. With this kind of change the processing would get
> overloaded with multi-gigabyte core dump size. Probably couldn't even
> save a core dump on these kind of smart nics.
>
>
> This needs to be optional (from command line) and default to the current
> behavior (not dumping huge pages).
Maybe expose eal_mem_set_dump() as rte_mem_set_dump()?
This would allow to implement the feature easily using memory callbacks.
Better, one can enable hugepages to dump selectively:
for example, dump some interesting hash tables but skip rings and mempools.
On Wed, Apr 6, 2022 at 6:46 AM Stephen Hemminger
<stephen@networkplumber.org> wrote:
>
> On Fri, 1 Apr 2022 17:10:04 +0800
> Li Feng <fengli@smartx.com> wrote:
>
> > These hugepages include important structures. we should dump these
> > hugepages into a coredump file for debugging when generating a coredump.
> >
> > Signed-off-by: Li Feng <fengli@smartx.com>
> > ---
> > lib/eal/linux/eal_memalloc.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/lib/eal/linux/eal_memalloc.c b/lib/eal/linux/eal_memalloc.c
> > index f8b1588cae..93c4f396cf 100644
> > --- a/lib/eal/linux/eal_memalloc.c
> > +++ b/lib/eal/linux/eal_memalloc.c
> > @@ -677,6 +677,8 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
> > __func__);
> > #endif
> >
> > + eal_mem_set_dump(addr, alloc_sz, true);
> > +
> > huge_recover_sigbus();
> >
> > ms->addr = addr;
>
>
> Don't merge this patch as is please; it would cause a lot of pain
> in a cloud environment.
>
> In our environment core dumps are collected (via systemd) and uploaded
> to a central server. With this kind of change the processing would get
> overloaded with multi-gigabyte core dump size. Probably couldn't even
> save a core dump on these kind of smart nics.
>
>
> This needs to be optional (from command line) and default to the current
> behavior (not dumping huge pages).
On Linux, just with this patch, the coredump will not include these
hugepages which are shared,
we should write 0x73 to /proc/self/coredump_filter.
This is the coredump_filter explanation:
Since kernel 2.6.23, the Linux-specific
/proc/[pid]/coredump_filter file can be used to control which memory
segments are written to the core dump
file in the event that a core dump is performed for the process
with the corresponding process ID.
The value in the file is a bit mask of memory mapping types
(see mmap(2)). If a bit is set in the mask, then memory mappings of
the corresponding
type are dumped; otherwise they are not dumped. The bits in
this file have the following meanings:
bit 0 Dump anonymous private mappings.
bit 1 Dump anonymous shared mappings.
bit 2 Dump file-backed private mappings.
bit 3 Dump file-backed shared mappings.
bit 4 (since Linux 2.6.24)
Dump ELF headers.
bit 5 (since Linux 2.6.28)
Dump private huge pages.
bit 6 (since Linux 2.6.28)
Dump shared huge pages.
bit 7 (since Linux 4.4)
Dump private DAX pages.
bit 8 (since Linux 4.4)
Dump shared DAX pages.
By default, the following bits are set: 0, 1, 4 (if the
CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS kernel configuration option is
enabled), and 5. This
default can be modified at boot time using the coredump_filter
boot option.
On Wed, 6 Apr 2022 02:14:46 +0300
Dmitry Kozlyuk <dmitry.kozliuk@gmail.com> wrote:
> >
> > Don't merge this patch as is please; it would cause a lot of pain
> > in a cloud environment.
> >
> > In our environment core dumps are collected (via systemd) and uploaded
> > to a central server. With this kind of change the processing would get
> > overloaded with multi-gigabyte core dump size. Probably couldn't even
> > save a core dump on these kind of smart nics.
> >
> >
> > This needs to be optional (from command line) and default to the current
> > behavior (not dumping huge pages).
>
> Maybe expose eal_mem_set_dump() as rte_mem_set_dump()?
> This would allow to implement the feature easily using memory callbacks.
> Better, one can enable hugepages to dump selectively:
> for example, dump some interesting hash tables but skip rings and mempools.
As was mentioned in thread core_dump_filter will also control these.
So it won't impact users who do not enable it.
Since the granularity is a the page level, it doesn't make sense
to try and be selective for hash tables, rings, mempools etc.
Looks good as is, though it might need a rebase.
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
On Fri, 1 Apr 2022 17:10:04 +0800
Li Feng <fengli@smartx.com> wrote:
> These hugepages include important structures. we should dump these
> hugepages into a coredump file for debugging when generating a coredump.
>
> Signed-off-by: Li Feng <fengli@smartx.com>
This needs to be config option to avoid surprising existing users.
@@ -677,6 +677,8 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
__func__);
#endif
+ eal_mem_set_dump(addr, alloc_sz, true);
+
huge_recover_sigbus();
ms->addr = addr;