[v3] eal: fix create user mem map repeatedly when it exists
Checks
Commit Message
From: Yunjian Wang <wangyunjian@huawei.com>
Currently, a issue that a container has many devices and the
application will map the same memory many times. The kernel
driver returns EEXIST as long as there are overlapping memory
areas. As a result, we repeatedly create new user mem map entry
for the same memory segment and this will lead to no more space
for other user mem maps.
To resolve the issue, add support to remove the same entry in
the function compact_user_maps().
Fixes: 0cbce3a167f1 ("vfio: skip DMA map failure if already mapped")
Cc: stable@dpdk.org
Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
v3:
Only update commit log and title
---
lib/librte_eal/linux/eal_vfio.c | 5 +++++
1 file changed, 5 insertions(+)
Comments
16/10/2020 11:28, wangyunjian:
> From: Yunjian Wang <wangyunjian@huawei.com>
>
> Currently, a issue that a container has many devices and the
> application will map the same memory many times. The kernel
> driver returns EEXIST as long as there are overlapping memory
> areas. As a result, we repeatedly create new user mem map entry
> for the same memory segment and this will lead to no more space
> for other user mem maps.
>
> To resolve the issue, add support to remove the same entry in
> the function compact_user_maps().
Sorry I don't understand the explanations above.
Anatoly, please could you help in rewording?
20/10/2020 16:09, Thomas Monjalon:
> 16/10/2020 11:28, wangyunjian:
> > From: Yunjian Wang <wangyunjian@huawei.com>
> >
> > Currently, a issue that a container has many devices and the
> > application will map the same memory many times. The kernel
> > driver returns EEXIST as long as there are overlapping memory
> > areas. As a result, we repeatedly create new user mem map entry
> > for the same memory segment and this will lead to no more space
> > for other user mem maps.
> >
> > To resolve the issue, add support to remove the same entry in
> > the function compact_user_maps().
>
> Sorry I don't understand the explanations above.
> Anatoly, please could you help in rewording?
Ping for rewording please.
15/11/2020 15:23, Thomas Monjalon:
> 20/10/2020 16:09, Thomas Monjalon:
> > 16/10/2020 11:28, wangyunjian:
> > > From: Yunjian Wang <wangyunjian@huawei.com>
> > >
> > > Currently, a issue that a container has many devices and the
> > > application will map the same memory many times. The kernel
> > > driver returns EEXIST as long as there are overlapping memory
> > > areas. As a result, we repeatedly create new user mem map entry
> > > for the same memory segment and this will lead to no more space
> > > for other user mem maps.
> > >
> > > To resolve the issue, add support to remove the same entry in
> > > the function compact_user_maps().
> >
> > Sorry I don't understand the explanations above.
> > Anatoly, please could you help in rewording?
>
> Ping for rewording please.
What is the conclusion? This fix is not worth the effort?
> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Monday, November 23, 2020 2:20 AM
> To: anatoly.burakov@intel.com; wangyunjian <wangyunjian@huawei.com>
> Cc: stable@dpdk.org; dev@dpdk.org; david.marchand@redhat.com; Lilijun
> (Jerry) <jerry.lilijun@huawei.com>; xudingke <xudingke@huawei.com>;
> bruce.richardson@intel.com; john.mcnamara@intel.com; asafp@nvidia.com
> Subject: Re: [dpdk-stable] [dpdk-dev] [PATCH v3] eal: fix create user mem map
> repeatedly when it exists
>
> 15/11/2020 15:23, Thomas Monjalon:
> > 20/10/2020 16:09, Thomas Monjalon:
> > > 16/10/2020 11:28, wangyunjian:
> > > > From: Yunjian Wang <wangyunjian@huawei.com>
> > > >
> > > > Currently, a issue that a container has many devices and the
> > > > application will map the same memory many times. The kernel driver
> > > > returns EEXIST as long as there are overlapping memory areas. As a
> > > > result, we repeatedly create new user mem map entry for the same
> > > > memory segment and this will lead to no more space for other user
> > > > mem maps.
> > > >
> > > > To resolve the issue, add support to remove the same entry in the
> > > > function compact_user_maps().
> > >
> > > Sorry I don't understand the explanations above.
> > > Anatoly, please could you help in rewording?
> >
> > Ping for rewording please.
>
> What is the conclusion? This fix is not worth the effort?
>
In my opinion, this issue needs to be fixed. Currently, the 'user_mem_maps->maps[]'
may store many same user mem maps but the maps array is limited, which will lead
to other mem maps cannot be created because of no more space left.
Thanks,
Yunjian
On 20-Oct-20 3:09 PM, Thomas Monjalon wrote:
> 16/10/2020 11:28, wangyunjian:
>> From: Yunjian Wang <wangyunjian@huawei.com>
>>
>> Currently, a issue that a container has many devices and the
>> application will map the same memory many times. The kernel
>> driver returns EEXIST as long as there are overlapping memory
>> areas. As a result, we repeatedly create new user mem map entry
>> for the same memory segment and this will lead to no more space
>> for other user mem maps.
>>
>> To resolve the issue, add support to remove the same entry in
>> the function compact_user_maps().
>
> Sorry I don't understand the explanations above.
> Anatoly, please could you help in rewording?
>
>
>
Apologies for delay, fell through the cracks.
Suggested rewording:
Currently, user mem maps will check if the newly mapped area is adjacent
to any existing mapping, but will not check if the mapping is identical
because it assumes that the API will never get called with the same
mapping twice. This will result in duplicate entries in the user mem
maps list.
Fix it by also checking for duplicate mappings, and skipping them if
they are found.
@@ -167,6 +167,10 @@ adjust_map(struct user_mem_map *src, struct user_mem_map *end,
static int
merge_map(struct user_mem_map *left, struct user_mem_map *right)
{
+ /* merge the same maps into one */
+ if (memcmp(left, right, sizeof(struct user_mem_map)) == 0)
+ goto out;
+
if (left->addr + left->len != right->addr)
return 0;
if (left->iova + left->len != right->iova)
@@ -174,6 +178,7 @@ merge_map(struct user_mem_map *left, struct user_mem_map *right)
left->len += right->len;
+out:
memset(right, 0, sizeof(*right));
return 1;