> On 20 Jan 2020, at 16:06, Leon Romanovsky <leon@xxxxxxxxxx> wrote:
>
> On Fri, Jan 17, 2020 at 02:56:22PM +0100, Håkon Bugge wrote:
>> Using CX-3 virtual functions, either from a bare-metal machine or
>> pass-through from a VM, MAD packets are proxied through the PF driver.
>>
>> Since the VF drivers have separate name spaces for MAD Transaction Ids
>> (TIDs), the PF driver has to re-map the TIDs and keep the book keeping
>> in a cache.
>>
>> Following the RDMA Connection Manager (CM) protocol, it is clear when
>> an entry has to evicted from the cache. When a DREP is sent from
>> mlx4_ib_multiplex_cm_handler(), id_map_find_del() is called. Similar
>> when a REJ is received by the mlx4_ib_demux_cm_handler(),
>> id_map_find_del() is called.
>>
>> This function wipes out the TID in use from the IDR or XArray and
>> removes the id_map_entry from the table.
>>
>> In short, it does everything except the topping of the cake, which is
>> to remove the entry from the list and free it. In other words, for the
>> DREP and REJ cases enumerated above, both will leak one id_map_entry.
>>
>> Signed-off-by: Håkon Bugge <haakon.bugge@xxxxxxxxxx>
>> ---
>> drivers/infiniband/hw/mlx4/cm.c | 7 ++++++-
>> 1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
>> index ecd6cadd529a..1df6d3ccfc62 100644
>> --- a/drivers/infiniband/hw/mlx4/cm.c
>> +++ b/drivers/infiniband/hw/mlx4/cm.c
>> @@ -197,8 +197,13 @@ static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
>> if (!ent)
>> goto out;
>> found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id);
>> - if (found_ent && found_ent == ent)
>> + if (found_ent && found_ent == ent) {
>> rb_erase(&found_ent->node, sl_id_map);
>> + if (!ent->scheduled_delete) {
>
> Why do we need to check scheduled_delete?
1. Node receives a DREQ and mlx4_ib_demux_cm_handler() is called, which again calls schedule_delayed(), which sets scheduled_delete.
2. DREQ is proxied over to the VM, which replies with a DREP.
3. The DREP is proxied over to the PF driver, mlx4_ib_multiplex_cm_handler() is called, id_map_find_del() is called. If it is freed now (without checking scheduled_delete), it will be a double free when the delayed work kicks in.
But this documents that the commit message is not accurate, it is only the REJ case that has a leak.
> Isn't this to mark call to timeout cleanup (id_map_ent_timeout), which
> can't race with id_map_find_del()? They both hold the same spinlock.
No race, but it can be set as per the above.
If you agree, I will send a v2 with corrected commit message.
Thxs, Håkon
>
> Thanks
>
>> + list_del(&ent->list);
>> + kfree(ent);
>> + }
>> + }
>> out:
>> spin_unlock(&sriov->id_map_lock);
>> }
>> --
>> 2.20.1
>>
