From: "Burakov, Anatoly" <anatoly.burakov@intel.com>
To: Hideyuki Yamashita <yamashita.hideyuki@po.ntt-tx.co.jp>, dev@dpdk.org
Cc: Yasufumi Ogawa <ogawa.yasufumi@lab.ntt.co.jp>,
Nakamura Hioryuki <nakamura.hiroyuki@po.ntt-tx.co.jp>
Subject: Re: [dpdk-dev] rte_eal_hotplug_remove() generates error message
Date: Mon, 17 Dec 2018 10:38:51 +0000 [thread overview]
Message-ID: <f0f6d42e-e850-c3d1-1235-222ff2647b39@intel.com> (raw)
In-Reply-To: <0eb59f12-0192-a234-a773-7f3061abea3a@intel.com>
On 17-Dec-18 10:23 AM, Burakov, Anatoly wrote:
> On 17-Dec-18 10:02 AM, Hideyuki Yamashita wrote:
>> Dear Thomas and all,
>>
>> I took a look on dpdk code.
>> I firstly write qustions and my analisys
>> on the current dpdk code follows after that.
>>
>> [1.Questions]
>> I have several questions to ask again.
>> Is my understanding correct about followings
>>
>> Q1: "EAL:ERROR, Invalid memory" is ignorable
>>
>> Q2: there is no big difference between calling
>> rte_eal_hotplug_remove(bus->name, dev->name)
>> and
>> rte_dev_remove(dev) because anyway it
>> reaches to rte_pmd_vhost_remove and encounter
>> the same error.
>>
>> [2.snip from my code]
>> .....
>> rte_eth_dev_close(port_id);
>> ret = rte_dev_remove(dev);
>> if (ret < 0)
>> return ret;
>> rte_eth_dev_release_port(&rte_eth_devices[port_id]);
>>
>> [3. My analysis on dpdk code]
>> static int
>> rte_pmd_vhost_remove(struct rte_vdev_device *dev)
>> {
>> ...........
>> eth_dev_close(eth_dev);
>>
>> rte_free(vring_states[eth_dev->data->port_id]);
>> vring_states[eth_dev->data->port_id] = NULL;
>>
>> rte_eth_dev_release_port(eth_dev);
>>
>> As you can see in rte_eth_vhost.c
>> It calls both eth_dev_close and rte_eth_dev_release_port.
>> And inside both functions, it tries to free mac_addrs.
>> rte_free(dev->data->mac_addrs); //in rth_dev_close
>> rte_free(eth_dev->data->mac_addrs); //in
>> rte_eth_dev_release_port
>>
>> I understand that is the reason why
>> /* Free the memory space back to heap */
>> void rte_free(void *addr)
>> {
>> if (addr == NULL) return;
>> if (malloc_heap_free(malloc_elem_from_data(addr)) < 0)
>> RTE_LOG(ERR, EAL, "Error: Invalid memory\n");
>> }
>> encounter the error.
>> As an experiment, I commented out one of them, "ERR, Invalid memory"
>> disappered.
>>
>> Thanks and BR,
>> Hideyuki Yamashita
>> NTT TechnoCross
>>
>>> Adding my colleague Yasufumi and Hiroyuki as CC.
>>>
>>> We are waiting valuable advice from you.
>>>
>>> Thanks in advance,
>>> Hideyuki Yamashita
>>> NTT TechnoCross
>>>
>>>>
>>>> Dear Thomas and all,
>>>>
>>>> I hope you all get safely back home after DPDK summit.
>>>> (When I get back Japan, it is chilling. (start of winter))
>>>>
>>>> On DPDK 18.11.0, we tried to remove vhost device by using
>>>> rte_eal_hotplug_remove().
>>>> However, following syslog message is printed.
>>>> “EAL: Error: Invalid memory”
>>>>
>>>> At DPDK summit San Jose, we had chance to ask Thomas how to handle
>>>> the error message, and he gave us following advice:
>>>> Replace “rte_eal_hotplug_add()” to “rte_dev_probe(devargs)” and
>>>> “rte_eal_hotplug_remove()” to “rte_eth_dev_close() and
>>>> rte_dev_remove(rte_dev)”
>>>>
>>>> We tested above changes, but the result is the same (i.e., same
>>>> error message is printed).
>>>> The debug log message says:
>>>> ---
>>>> [primary]
>>>> VHOST_CONFIG: vhost-user server: socket created, fd: 17
>>>> VHOST_CONFIG: bind to /tmp/sock0
>>>> EAL: Error: Invalid memory
>>>> VHOST_CONFIG: vhost-user server: socket created, fd: 17
>>>> VHOST_CONFIG: bind to /tmp/sock0
>>>>
>>>> [secondary]
>>>> APP: devargs=eth_vhost0,iface=/tmp/sock0,queues=1
>>>> EAL: request: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> EAL: request: bus_vdev_mp
>>>> EAL: msg: bus_vdev_mp
>>>> EAL: msg: bus_vdev_mp
>>>> EAL: reply: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> rte_eth_promiscuous_disable: Function not supported
>>>> rte_eth_allmulticast_disable: Function not supported
>>>> APP: To Server: add
>>>> EAL: request: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> EAL: reply: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> APP: To Server: del
>>>> APP: devargs=eth_vhost0,iface=/tmp/sock0,queues=1
>>>> EAL: request: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> EAL: request: bus_vdev_mp
>>>> EAL: msg: bus_vdev_mp
>>>> EAL: msg: bus_vdev_mp
>>>> EAL: reply: eal_dev_mp_request
>>>> EAL: msg: eal_dev_mp_request
>>>> rte_eth_promiscuous_disable: Function not supported
>>>> rte_eth_allmulticast_disable: Function not supported
>>>> APP: To Server: add
>>>> ---
>>>>
>>>> We would like to ask:
>>>> 1) Is the message “EAL: Error: Invalid memory” ignorable or not?
>>>> There is no obstacle except this message to re-add the vhost device.
>>>> 2) Which is the better(best?) way to add/del vhost device
>>>> “rte_eal_hotplug_add/remove()” or the way Thomas suggested?
>>>>
>>>> Thanks in advance and have a nice day.
>>>>
>>>> BR,
>>>> Hideyuki Yamashita
>>>> NTT TechnoCross
>>>
>>
>>
>>
>
> Hi Hideyuki,
>
> The error you're referring to (about invalid memory) means that you're
> trying to free a pointer that points to invalid memory. Meaning, either
> the pointer itself is not pointing to an allocated area, or it points to
> memory that has already been freed.
>
> If dev->data->mac_addrs and eth_dev->data->mac_addrs point to the same
> area, this is a bug, because this would lead to double free, and
> rte_malloc will rightly complain about invalid memory. Now, malloc won't
> try to do anything with the invalid memory, so the error itself is
> harmless *as far as malloc is concerned*, but these attempts to free the
> memory twice should be fixed whereever they happen.
>
> I'm not well-versed in dev infrastructure, so i wouldn't be able to say
> which one of the rte_free calls is an extra, unneeded one. This is
> something e.g. Thomas could help with, or the driver maintainer.
>
...with that said, to me, this looks suspicious:
> rte_eth_dev_close(port_id);
> ret = rte_dev_remove(dev);
> if (ret < 0)
> return ret;
> rte_eth_dev_release_port(&rte_eth_devices[port_id]);
So, you're closing the port, then you remove the device, then you
release port. From your own analysis, what dev_remove does is the following:
> rte_pmd_vhost_remove(struct rte_vdev_device *dev)
> {
> ...........
> eth_dev_close(eth_dev);
>
> rte_free(vring_states[eth_dev->data->port_id]);
> vring_states[eth_dev->data->port_id] = NULL;
>
> rte_eth_dev_release_port(eth_dev);
So, you have duplicate call to eth_dev_close() (which is likely not a
problem), and a duplicate call to rte_eth_dev_release_port(). So, if
rte_dev_remove() already calls rte_eth_dev_release_port(), why are you
calling it in your code?
Now, as i said above, i'm not well-versed enough in the dev
infrastructure to know which of the calls should go where, but i'm
pretty sure you shouldn't call rte_eth_dev_release_port twice - this is
likely what's causing your double-free.
--
Thanks,
Anatoly
next prev parent reply other threads:[~2018-12-17 10:38 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-12-09 4:30 [dpdk-dev] dpdk vlan strip offload bug for I350 nic? 1534898891
2018-12-09 5:23 ` =?gb18030?B?MTUzNDg5ODg5MQ==?=
2018-12-18 2:26 ` Zhao1, Wei
2018-12-10 7:53 ` [dpdk-dev] rte_eal_hotplug_remove() generates error message Hideyuki Yamashita
2018-12-11 0:54 ` Hideyuki Yamashita
2018-12-17 10:02 ` Hideyuki Yamashita
2018-12-17 10:23 ` Burakov, Anatoly
2018-12-17 10:38 ` Burakov, Anatoly [this message]
2018-12-17 11:03 ` Hideyuki Yamashita
2018-12-17 10:45 ` Hideyuki Yamashita
2018-12-17 11:21 ` Burakov, Anatoly
2018-12-17 12:10 ` Hideyuki Yamashita
2018-12-18 2:39 ` Tiwei Bie
2018-12-18 3:11 ` Hideyuki Yamashita
2018-12-18 5:12 ` Tiwei Bie
2018-12-18 5:52 ` Hideyuki Yamashita
2018-12-18 6:25 ` Tiwei Bie
2018-12-18 11:30 ` Hideyuki Yamashita
2018-12-18 13:09 ` Tiwei Bie
2018-12-27 5:52 ` Hideyuki Yamashita
2019-01-02 9:19 ` Tiwei Bie
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=f0f6d42e-e850-c3d1-1235-222ff2647b39@intel.com \
--to=anatoly.burakov@intel.com \
--cc=dev@dpdk.org \
--cc=nakamura.hiroyuki@po.ntt-tx.co.jp \
--cc=ogawa.yasufumi@lab.ntt.co.jp \
--cc=yamashita.hideyuki@po.ntt-tx.co.jp \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).