[v4] eal: detach memsegs on cleanup
Checks
Commit Message
Currently, we don't detach the shared memory on EAL cleanup, which
leaves the page table descriptors still holding on to the file
descriptors as well as memory space occupied by them. Fix it by adding
another detach stage that closes the internal memory allocator resource
references, detaches shared fbarrays and unmaps the shared mem config.
Bugzilla ID: 380
Bugzilla ID: 381
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
Notes:
v4:
- Fix Windows build
v3:
- Added missing Bugzilla ID for similar bug
- Fixed deadlock on exit
v2:
- Fixed checkpatch warnings
Not backporting to stable because this fix isn't critical but is rather
"nice to have".
lib/librte_eal/common/eal_common_memory.c | 54 ++++++++++++++
lib/librte_eal/common/eal_memalloc.h | 3 +
lib/librte_eal/common/eal_private.h | 7 ++
lib/librte_eal/freebsd/eal.c | 2 +
lib/librte_eal/freebsd/eal_memalloc.c | 5 ++
lib/librte_eal/linux/eal.c | 2 +
lib/librte_eal/linux/eal_memalloc.c | 85 +++++++++++++++++++++++
lib/librte_eal/windows/eal.c | 3 +-
lib/librte_eal/windows/eal_memalloc.c | 7 ++
9 files changed, 167 insertions(+), 1 deletion(-)
Comments
On 14-Sep-20 2:04 PM, Anatoly Burakov wrote:
> Currently, we don't detach the shared memory on EAL cleanup, which
> leaves the page table descriptors still holding on to the file
> descriptors as well as memory space occupied by them. Fix it by adding
> another detach stage that closes the internal memory allocator resource
> references, detaches shared fbarrays and unmaps the shared mem config.
>
> Bugzilla ID: 380
> Bugzilla ID: 381
>
> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
> ---
Hi Stephen,
You were the original submitter for the above bugzilla issues. Could you
please review the patch?
Stephen, ping.
15/10/2020 11:54, Burakov, Anatoly:
> On 14-Sep-20 2:04 PM, Anatoly Burakov wrote:
> > Currently, we don't detach the shared memory on EAL cleanup, which
> > leaves the page table descriptors still holding on to the file
> > descriptors as well as memory space occupied by them. Fix it by adding
> > another detach stage that closes the internal memory allocator resource
> > references, detaches shared fbarrays and unmaps the shared mem config.
> >
> > Bugzilla ID: 380
> > Bugzilla ID: 381
> >
> > Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
> > ---
>
> Hi Stephen,
>
> You were the original submitter for the above bugzilla issues. Could you
> please review the patch?
>
>
That's a pity we didn't get this patch in DPDK 20.11.
Anatoly, Stephen, what happened? It is not interesting anymore?
Anyone else to review?
20/10/2020 13:53, Thomas Monjalon:
> Stephen, ping.
>
> 15/10/2020 11:54, Burakov, Anatoly:
> > On 14-Sep-20 2:04 PM, Anatoly Burakov wrote:
> > > Currently, we don't detach the shared memory on EAL cleanup, which
> > > leaves the page table descriptors still holding on to the file
> > > descriptors as well as memory space occupied by them. Fix it by adding
> > > another detach stage that closes the internal memory allocator resource
> > > references, detaches shared fbarrays and unmaps the shared mem config.
> > >
> > > Bugzilla ID: 380
> > > Bugzilla ID: 381
> > >
> > > Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
> > > ---
> >
> > Hi Stephen,
> >
> > You were the original submitter for the above bugzilla issues. Could you
> > please review the patch?
On 22-Nov-20 6:16 PM, Thomas Monjalon wrote:
> That's a pity we didn't get this patch in DPDK 20.11.
>
> Anatoly, Stephen, what happened? It is not interesting anymore?
>
> Anyone else to review?
>
It is a good patch and should be merged. I wanted to get it into LTS but
no such luck apparently.
>
> 20/10/2020 13:53, Thomas Monjalon:
>> Stephen, ping.
>>
>> 15/10/2020 11:54, Burakov, Anatoly:
>>> On 14-Sep-20 2:04 PM, Anatoly Burakov wrote:
>>>> Currently, we don't detach the shared memory on EAL cleanup, which
>>>> leaves the page table descriptors still holding on to the file
>>>> descriptors as well as memory space occupied by them. Fix it by adding
>>>> another detach stage that closes the internal memory allocator resource
>>>> references, detaches shared fbarrays and unmaps the shared mem config.
>>>>
>>>> Bugzilla ID: 380
>>>> Bugzilla ID: 381
>>>>
>>>> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
>>>> ---
>>>
>>> Hi Stephen,
>>>
>>> You were the original submitter for the above bugzilla issues. Could you
>>> please review the patch?
>
>
>
27/11/2020 13:56, Burakov, Anatoly:
> On 22-Nov-20 6:16 PM, Thomas Monjalon wrote:
> > That's a pity we didn't get this patch in DPDK 20.11.
> >
> > Anatoly, Stephen, what happened? It is not interesting anymore?
> >
> > Anyone else to review?
> >
>
> It is a good patch and should be merged. I wanted to get it into LTS but
> no such luck apparently.
Anatoly, you must follow-up more closely,
not waiting the last day.
I know it is hard to get any reply from Stephen,
but it is your responsibility.
And I don't understand why nobody else found time to look at it.
> > 20/10/2020 13:53, Thomas Monjalon:
> >> Stephen, ping.
> >>
> >> 15/10/2020 11:54, Burakov, Anatoly:
> >>> On 14-Sep-20 2:04 PM, Anatoly Burakov wrote:
> >>>> Currently, we don't detach the shared memory on EAL cleanup, which
> >>>> leaves the page table descriptors still holding on to the file
> >>>> descriptors as well as memory space occupied by them. Fix it by adding
> >>>> another detach stage that closes the internal memory allocator resource
> >>>> references, detaches shared fbarrays and unmaps the shared mem config.
> >>>>
> >>>> Bugzilla ID: 380
> >>>> Bugzilla ID: 381
> >>>>
> >>>> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
> >>>> ---
> >>>
> >>> Hi Stephen,
> >>>
> >>> You were the original submitter for the above bugzilla issues. Could you
> >>> please review the patch?
On 27-Nov-20 1:21 PM, Thomas Monjalon wrote:
> 27/11/2020 13:56, Burakov, Anatoly:
>> On 22-Nov-20 6:16 PM, Thomas Monjalon wrote:
>>> That's a pity we didn't get this patch in DPDK 20.11.
>>>
>>> Anatoly, Stephen, what happened? It is not interesting anymore?
>>>
>>> Anyone else to review?
>>>
>>
>> It is a good patch and should be merged. I wanted to get it into LTS but
>> no such luck apparently.
>
> Anatoly, you must follow-up more closely,
> not waiting the last day.
>
> I know it is hard to get any reply from Stephen,
> but it is your responsibility.
> And I don't understand why nobody else found time to look at it.
>
Apologies for late reply. Let's try this for 21.05 :)
Stephen, please review as time permits.
On Mon, 14 Sep 2020 14:04:05 +0100
Anatoly Burakov <anatoly.burakov@intel.com> wrote:
> Currently, we don't detach the shared memory on EAL cleanup, which
> leaves the page table descriptors still holding on to the file
> descriptors as well as memory space occupied by them. Fix it by adding
> another detach stage that closes the internal memory allocator resource
> references, detaches shared fbarrays and unmaps the shared mem config.
>
> Bugzilla ID: 380
> Bugzilla ID: 381
>
> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Sure looks good. We should put more tests of cleanup
in the test suite.
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Sorry for the late response been stuck in internal project stuff.
On Mon, Sep 14, 2020 at 3:06 PM Anatoly Burakov
<anatoly.burakov@intel.com> wrote:
>
> Currently, we don't detach the shared memory on EAL cleanup, which
> leaves the page table descriptors still holding on to the file
> descriptors as well as memory space occupied by them. Fix it by adding
> another detach stage that closes the internal memory allocator resource
> references, detaches shared fbarrays and unmaps the shared mem config.
>
> Bugzilla ID: 380
> Bugzilla ID: 381
>
> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Applied, thanks.
@@ -1002,6 +1002,60 @@ rte_extmem_detach(void *va_addr, size_t len)
return sync_memory(va_addr, len, false);
}
+/* detach all EAL memory */
+int
+rte_eal_memory_detach(void)
+{
+ struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+ size_t page_sz = rte_mem_page_size();
+ unsigned int i;
+
+ rte_rwlock_write_lock(&mcfg->memory_hotplug_lock);
+
+ /* detach internal memory subsystem data first */
+ if (eal_memalloc_cleanup())
+ RTE_LOG(ERR, EAL, "Could not release memory subsystem data\n");
+
+ for (i = 0; i < RTE_DIM(mcfg->memsegs); i++) {
+ struct rte_memseg_list *msl = &mcfg->memsegs[i];
+
+ /* skip uninitialized segments */
+ if (msl->base_va == NULL)
+ continue;
+ /*
+ * external segments are supposed to be detached at this point,
+ * but if they aren't, we can't really do anything about it,
+ * because if we skip them here, they'll become invalid after
+ * we unmap the memconfig anyway. however, if this is externally
+ * referenced memory, we have no business unmapping it.
+ */
+ if (!msl->external)
+ if (rte_mem_unmap(msl->base_va, msl->len) != 0)
+ RTE_LOG(ERR, EAL, "Could not unmap memory: %s\n",
+ strerror(errno));
+
+ /*
+ * we are detaching the fbarray rather than destroying because
+ * other processes might still reference this fbarray, and we
+ * have no way of knowing if they still do.
+ */
+ if (rte_fbarray_detach(&msl->memseg_arr))
+ RTE_LOG(ERR, EAL, "Could not detach fbarray: %s\n",
+ rte_strerror(rte_errno));
+ }
+ rte_rwlock_write_unlock(&mcfg->memory_hotplug_lock);
+
+ /*
+ * we've detached the memseg lists, so we can unmap the shared mem
+ * config - we can't zero it out because it might still be referenced
+ * by other processes.
+ */
+ rte_mem_unmap(mcfg, RTE_ALIGN(sizeof(*mcfg), page_sz));
+ rte_eal_get_configuration()->mem_config = NULL;
+
+ return 0;
+}
+
/* init memory subsystem */
int
rte_eal_memory_init(void)
@@ -93,4 +93,7 @@ eal_memalloc_get_seg_fd_offset(int list_idx, int seg_idx, size_t *offset);
int
eal_memalloc_init(void);
+int
+eal_memalloc_cleanup(void);
+
#endif /* EAL_MEMALLOC_H */
@@ -433,6 +433,13 @@ int rte_eal_hugepage_init(void);
*/
int rte_eal_hugepage_attach(void);
+/**
+ * Detaches all memory mappings from a process.
+ *
+ * This function is private to the EAL.
+ */
+int rte_eal_memory_detach(void);
+
/**
* Find a bus capable of identifying a device.
*
@@ -963,6 +963,8 @@ rte_eal_cleanup(void)
eal_get_internal_configuration();
rte_service_finalize();
rte_mp_channel_cleanup();
+ /* after this point, any DPDK pointers will become dangling */
+ rte_eal_memory_detach();
rte_trace_save();
eal_trace_fini();
eal_cleanup_config(internal_conf);
@@ -74,6 +74,11 @@ eal_memalloc_get_seg_fd_offset(int list_idx __rte_unused,
return -ENOTSUP;
}
+int eal_memalloc_cleanup(void)
+{
+ return 0;
+}
+
int
eal_memalloc_init(void)
{
@@ -1359,6 +1359,8 @@ rte_eal_cleanup(void)
rte_memseg_walk(mark_freeable, NULL);
rte_service_finalize();
rte_mp_channel_cleanup();
+ /* after this point, any DPDK pointers will become dangling */
+ rte_eal_memory_detach();
rte_trace_save();
eal_trace_fini();
eal_cleanup_config(internal_conf);
@@ -1418,6 +1418,31 @@ secondary_msl_create_walk(const struct rte_memseg_list *msl,
return 0;
}
+static int
+secondary_msl_destroy_walk(const struct rte_memseg_list *msl,
+ void *arg __rte_unused)
+{
+ struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+ struct rte_memseg_list *local_msl;
+ int msl_idx, ret;
+
+ if (msl->external)
+ return 0;
+
+ msl_idx = msl - mcfg->memsegs;
+ local_msl = &local_memsegs[msl_idx];
+
+ ret = rte_fbarray_destroy(&local_msl->memseg_arr);
+ if (ret < 0) {
+ RTE_LOG(ERR, EAL, "Cannot destroy local memory map\n");
+ return -1;
+ }
+ local_msl->base_va = NULL;
+ local_msl->len = 0;
+
+ return 0;
+}
+
static int
alloc_list(int list_idx, int len)
{
@@ -1450,6 +1475,34 @@ alloc_list(int list_idx, int len)
return 0;
}
+static int
+destroy_list(int list_idx)
+{
+ const struct internal_config *internal_conf =
+ eal_get_internal_configuration();
+
+ /* single-file segments mode does not need fd list */
+ if (!internal_conf->single_file_segments) {
+ int *fds = fd_list[list_idx].fds;
+ int i;
+ /* go through each fd and ensure it's closed */
+ for (i = 0; i < fd_list[list_idx].len; i++) {
+ if (fds[i] >= 0) {
+ close(fds[i]);
+ fds[i] = -1;
+ }
+ }
+ free(fds);
+ fd_list[list_idx].fds = NULL;
+ fd_list[list_idx].len = 0;
+ } else if (fd_list[list_idx].memseg_list_fd >= 0) {
+ close(fd_list[list_idx].memseg_list_fd);
+ fd_list[list_idx].count = 0;
+ fd_list[list_idx].memseg_list_fd = -1;
+ }
+ return 0;
+}
+
static int
fd_list_create_walk(const struct rte_memseg_list *msl,
void *arg __rte_unused)
@@ -1467,6 +1520,20 @@ fd_list_create_walk(const struct rte_memseg_list *msl,
return alloc_list(msl_idx, len);
}
+static int
+fd_list_destroy_walk(const struct rte_memseg_list *msl, void *arg __rte_unused)
+{
+ struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+ int msl_idx;
+
+ if (msl->external)
+ return 0;
+
+ msl_idx = msl - mcfg->memsegs;
+
+ return destroy_list(msl_idx);
+}
+
int
eal_memalloc_set_seg_fd(int list_idx, int seg_idx, int fd)
{
@@ -1603,6 +1670,24 @@ eal_memalloc_get_seg_fd_offset(int list_idx, int seg_idx, size_t *offset)
return 0;
}
+int
+eal_memalloc_cleanup(void)
+{
+ /* close all remaining fd's - these are per-process, so it's safe */
+ if (rte_memseg_list_walk_thread_unsafe(fd_list_destroy_walk, NULL))
+ return -1;
+
+ /* destroy the shadow page table if we're a secondary process */
+ if (rte_eal_process_type() == RTE_PROC_PRIMARY)
+ return 0;
+
+ if (rte_memseg_list_walk_thread_unsafe(secondary_msl_destroy_walk,
+ NULL))
+ return -1;
+
+ return 0;
+}
+
int
eal_memalloc_init(void)
{
@@ -250,7 +250,8 @@ rte_eal_cleanup(void)
{
struct internal_config *internal_conf =
eal_get_internal_configuration();
-
+ /* after this point, any DPDK pointers will become dangling */
+ rte_eal_memory_detach();
eal_cleanup_config(internal_conf);
return 0;
}
@@ -437,6 +437,13 @@ eal_memalloc_sync_with_primary(void)
return -1;
}
+int
+eal_memalloc_cleanup(void)
+{
+ /* not implemented */
+ return 0;
+}
+
int
eal_memalloc_init(void)
{