Make function "rds_ib_try_reuse_ibmr" return NULL in case
memory region could not be allocated, since callers
simply check if the return value is not NULL.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Waiting for activity on the "clean_list" to quiesce is no substitute
for proper locking.
We can have multiple threads competing for "llist_del_first"
via "rds_ib_reuse_mr", and a single thread competing
for "llist_del_all" and "llist_del_first" via "rds_ib_flush_mr_pool".
Since "llist_del_first" depends on "list->first->next" not to change
in the midst of the operation, simply waiting for all current calls
to "rds_ib_reuse_mr" to quiesce across all CPUs is woefully inadequate:
By the time "wait_clean_list_grace" is done iterating over all CPUs to see
that there is no concurrent caller to "rds_ib_reuse_mr", a new caller may
have just shown up on the first CPU.
Furthermore, <linux/llist.h> explicitly calls out the need for locking:
* Cases where locking is needed:
* If we have multiple consumers with llist_del_first used in one consumer,
* and llist_del_first or llist_del_all used in other consumers,
* then a lock is needed.
Also, while at it, drop the unused "pool" parameter
from "list_to_llist_nodes".
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the following tests last for several hours, the problem will occur.
Server:
rds-stress -r 1.1.1.16 -D 1M
Client:
rds-stress -r 1.1.1.14 -s 1.1.1.16 -D 1M -T 30
The following will occur.
"
Starting up....
tsks tx/s rx/s tx+rx K/s mbi K/s mbo K/s tx us/c rtt us cpu
%
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
"
>From vmcore, we can find that clean_list is NULL.
>From the source code, rds_mr_flushd calls rds_ib_mr_pool_flush_worker.
Then rds_ib_mr_pool_flush_worker calls
"
rds_ib_flush_mr_pool(pool, 0, NULL);
"
Then in function
"
int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
int free_all, struct rds_ib_mr **ibmr_ret)
"
ibmr_ret is NULL.
In the source code,
"
...
list_to_llist_nodes(pool, &unmap_list, &clean_nodes, &clean_tail);
if (ibmr_ret)
*ibmr_ret = llist_entry(clean_nodes, struct rds_ib_mr, llnode);
/* more than one entry in llist nodes */
if (clean_nodes->next)
llist_add_batch(clean_nodes->next, clean_tail, &pool->clean_list);
...
"
When ibmr_ret is NULL, llist_entry is not executed. clean_nodes->next
instead of clean_nodes is added in clean_list.
So clean_nodes is discarded. It can not be used again.
The workqueue is executed periodically. So more and more clean_nodes are
discarded. Finally the clean_list is NULL.
Then this problem will occur.
Fixes: 1bc144b625 ("net, rds, Replace xlist in net/rds/xlist.h with llist")
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Before the commit 490ea5967b ("RDS: IB: move FMR code to its own file"),
when the dirty_count is greater than 9/10 of max_items of 8K pool,
1M pool is used, Vice versa. After the commit 490ea5967b ("RDS: IB: move
FMR code to its own file"), the above is removed. When we make the
following tests.
Server:
rds-stress -r 1.1.1.16 -D 1M
Client:
rds-stress -r 1.1.1.14 -s 1.1.1.16 -D 1M
The following will appear.
"
connecting to 1.1.1.16:4000
negotiated options, tasks will start in 2 seconds
Starting up..header from 1.1.1.166:4001 to id 4001 bogus
..
tsks tx/s rx/s tx+rx K/s mbi K/s mbo K/s tx us/c rtt us
cpu %
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
1 0 0 0.00 0.00 0.00 0.00 0.00 -1.00
...
"
So this exchange between 8K and 1M pool is added back.
Fixes: commit 490ea5967b ("RDS: IB: move FMR code to its own file")
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The BTF conflicts were simple overlapping changes.
The virtio_net conflict was an overlap of a fix of statistics counter,
happening alongisde a move over to a bonafide statistics structure
rather than counting value on the stack.
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the IPv6 dependency from RDS.
Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are many data structures (RDS socket options) used by RDS apps
which use a 32 bit integer to store IP address. To support IPv6,
struct in6_addr needs to be used. To ensure backward compatibility, a
new data structure is introduced for each of those data structures
which use a 32 bit integer to represent an IP address. And new socket
options are introduced to use those new structures. This means that
existing apps should work without a problem with the new RDS module.
For apps which want to use IPv6, those new data structures and socket
options can be used. IPv4 mapped address is used to represent IPv4
address in the new data structures.
v4: Revert changes to SO_RDS_TRANSPORT
Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch changes the internal representation of an IP address to use
struct in6_addr. IPv4 address is stored as an IPv4 mapped address.
All the functions which take an IP address as argument are also
changed to use struct in6_addr. But RDS socket layer is not modified
such that it still does not accept IPv6 address from an application.
And RDS layer does not accept nor initiate IPv6 connections.
v2: Fixed sparse warnings.
Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use max_1m_mrs/max_8k_mrs while setting max_items, as the former
variables are set based on the underlying device attributes.
Signed-off-by: Avinash Repaka <avinash.repaka@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
refcount_t type and corresponding API should be
used instead of atomic_t when the variable is used as
a reference counter. This allows to avoid accidental
refcounter overflows that might lead to use-after-free
situations.
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David Windsor <dwindsor@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for multipath RDS, split the rds_connection
structure into a base structure, and a per-path struct rds_conn_path.
The base structure tracks information and locks common to all
paths. The workqs for send/recv/shutdown etc are tracked per
rds_conn_path. Thus the workq callbacks now work with rds_conn_path.
This commit allows for one rds_conn_path per rds_connection, and will
be extended into multiple conn_paths in subsequent commits.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
alloc_workqueue replaces deprecated create_workqueue().
Since the driver is infiniband which can be used as block device and the
workqueue seems involved in regular operation of the device, so a
dedicated workqueue has been used with WQ_MEM_RECLAIM set to guarantee
forward progress under memory pressure.
Since there are only a fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fastreg MR(FRMR) is another method with which one can
register memory to HCA. Some of the newer HCAs supports only fastreg
mr mode, so we need to add support for it to have RDS functional
on them.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Avinash Repaka <avinash.repaka@oracle.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
No functional change.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Keep fmr related filed in its own struct. Fastreg MR structure
will be added to the union.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
No functional changes. This is in preperation towards adding
fastreg memory resgitration support.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8K message sizes are pretty important usecase for RDS current
workloads so we make provison to have 8K mrs available from the pool.
Based on number of SG's in the RDS message, we pick a pool to use.
Also to make sure that we don't under utlise mrs when say 8k messages
are dominating which could lead to 8k pull being exhausted, we fall-back
to 1m pool till 8k pool recovers for use.
This helps to at least push ~55 kB/s bidirectional data which
is a nice improvement.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Fix below warning by marking rds_ib_fmr_wq static
net/rds/ib_rdma.c:87:25: warning: symbol 'rds_ib_fmr_wq' was not declared. Should it be static?
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
rds_ib_mr already keeps the pool handle which it associates
with. Lets use that instead of round about way of fetching
it from rds_ib_device.
No functional change.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
RDS IB mr pool has its own workqueue 'rds_ib_fmr_wq', so we need
to use queue_delayed_work() to kick the work. This was hurting
the performance since pool maintenance was less often triggered
from other path.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
synchronize_rcu() slowing down un-necessarily the socket shutdown
path. It is used just kfree() the ip addresses in rds_ib_remove_ipaddr()
which is perfect usecase for kfree_rcu();
So lets use that to gain some speedup.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Memory allocated for 'ibmr' uses kzalloc_node() which already
initialises the memory to zero. There is no need to do
memset() 0 on that memory.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
FMR flush is an expensive and time consuming operation. Reduce the
frequency of FMR pool flush by 50% so that more FMR work gets accumulated
for more efficient flushing.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RDS FMR flush operation and also it races with connect/reconect
which happes a lot with RDS. FMR flush being on common rds_wq aggrevates
the problem. Lets push RDS FMR pool flush work to its own worker.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rds_ib_flush_mr_pool(), dirty_count accounts the clean ones
which is wrong. This can lead to a negative dirty count value.
Lets fix it.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the asserion level since its not fatal and can be hit
in normal execution paths. There is no need to take the
system down.
We keep the WARN_ON() to detect the condition if we get
here with bad pages.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the ip address tables hasn't changed, there is no need to remove
them only to be added back again.
Lets fix it.
Reviewed-by: Ajaykumar Hotchandani <ajaykumar.hotchandani@oracle.com>
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes: 3e0249f9c0 ("RDS/IB: add refcount tracking to struct rds_ib_device")
There lacks a dropping on rds_ib_device.refcount in case rds_ib_alloc_fmr
failed(mr pool running out). this lead to the refcount overflow.
A complain in line 117(see following) is seen. From vmcore:
s_ib_rdma_mr_pool_depleted is 2147485544 and rds_ibdev->refcount is -2147475448.
That is the evidence the mr pool is used up. so rds_ib_alloc_fmr is very likely
to return ERR_PTR(-EAGAIN).
115 void rds_ib_dev_put(struct rds_ib_device *rds_ibdev)
116 {
117 BUG_ON(atomic_read(&rds_ibdev->refcount) <= 0);
118 if (atomic_dec_and_test(&rds_ibdev->refcount))
119 queue_work(rds_wq, &rds_ibdev->free_work);
120 }
fix is to drop refcount when rds_ib_alloc_fmr failed.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Reviewed-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Replace uses of get_cpu_var for address calculation through this_cpu_ptr.
Cc: netdev@vger.kernel.org
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The functionality of xlist and llist is almost same. This patch
replace xlist with llist to avoid code duplication.
Known issues: don't know how to test this, need special hardware?
Signed-off-by: Huang Ying <ying.huang@intel.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Andy Grover <andy.grover@oracle.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
With cmwq, there's no reason to use dedicated rds_ib_fmr_wq - it's not
in the memory reclaim path and the maximum number of concurrent work
items is bound by the number of devices. Drop it and use system_wq
instead. This rds_ib_fmr_init/exit() noops. Both removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Grover <andy.grover@oracle.com>
The RDS protocol has lots of functions that should be
declared static. rds_message_get/add_version_extension is
removed since it defined but never used.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is basically just a cleanup. IRQs were disabled on the previous
line so we don't need to do it again here. In the current code IRQs
would get turned on one line earlier than intended.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The RDS IB device list wasn't protected by any locking. Traversal in
both the get_mr and FMR flushing paths could race with additon and
removal.
List manipulation is done with RCU primatives and is protected by the
write side of a rwsem. The list traversal in the get_mr fast path is
protected by a rcu read critical section. The FMR list traversal is
more problematic because it can block while traversing the list. We
protect this with the read side of the rwsem.
Signed-off-by: Zach Brown <zach.brown@oracle.com>
Flushing FMRs is somewhat expensive, and is currently kicked off when
the interrupt handler notices that we are getting low. The result of
this is that FMR flushing only happens from the interrupt cpus.
This spreads the load more effectively by triggering flushes just before
we allocate a new FMR.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
This patch moves the FMR flushing work in to its own mult-threaded work queue.
This is to maintain performance in preparation for returning the main krdsd
work queue back to a single threaded work queue to avoid deep-rooted
concurrency bugs.
This is also good because it further separates FMRs, which might be removed
some day, from the rest of the code base.
Signed-off-by: Zach Brown <zach.brown@oracle.com>
IB connections were not being destroyed during rmmod.
First, recently IB device removal callback was changed to disconnect
connections that used the removing device rather than destroying them. So
connections with devices during rmmod were not being destroyed.
Second, rds_ib_destroy_nodev_conns() was being called before connections are
disassociated with devices. It would almost never find connections in the
nodev list.
We first get rid of rds_ib_destroy_conns(), which is no longer called, and
refactor the existing caller into the main body of the function and get rid of
the list and lock wrappers.
Then we call rds_ib_destroy_nodev_conns() *after* ib_unregister_client() has
removed the IB device from all the conns and put the conns on the nodev list.
The result is that IB connections are destroyed by rmmod.
Signed-off-by: Zach Brown <zach.brown@oracle.com>
Using a delayed work queue helps us make sure a healthy number of FMRs
have queued up over the limit. It makes for a large improvement in RDMA
iops.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
FRM allocation and recycling is performance critical and fairly lock
intensive. The current code has a per connection lock that all
processes bang on and it becomes a major bottleneck on large systems.
This changes things to use a number of cmpxchg based lists instead,
allowing us to go through the whole FMR lifecycle without locking inside
RDS.
Zach Brown pointed out that our usage of cmpxchg for xlist removal is
racey if someone manages to remove and add back an FMR struct into the list
while another CPU can see the FMR's address at the head of the list.
The second CPU might assume the list hasn't changed when in fact any
number of operations might have happened in between the deletion and
reinsertion.
This commit maintains a per cpu count of CPUs that are currently
in xlist removal, and establishes a grace period to make sure that
nobody can see an entry we have just removed from the list.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The RDS IB client .remove callback used to free the rds_ibdev for the given
device unconditionally. This could race other users of the struct. This patch
adds refcounting so that we only free the rds_ibdev once all of its users are
done.
Many rds_ibdev users are tied to connections. We give the connection a
reference and change these users to reference the device in the connection
instead of looking it up in the IB client data. The only user of the IB client
data remaining is the first lookup of the device as connections are built up.
Incrementing the reference count of a device found in the IB client data could
race with final freeing so we use an RCU grace period to make sure that freeing
won't happen until those lookups are done.
MRs need the rds_ibdev to get at the pool that they're freed in to. They exist
outside a connection and many MRs can reference different devices from one
socket, so it was natural to have each MR hold a reference. MR refs can be
dropped from interrupt handlers and final device teardown can block so we push
it off to a work struct. Pool teardown had to be fixed to cancel its pending
work instead of deadlocking waiting for all queued work, including itself, to
finish.
MRs get their reference from the global device list, which gets a reference.
It is left unprotected by locks and remains racy. A simple global lock would
be a significant bottleneck. More scalable (complicated) locking should be
done carefully in a later patch.
Signed-off-by: Zach Brown <zach.brown@oracle.com>
The RDS bind lookups are somewhat expensive in terms of CPU
time and locking overhead. This commit changes them into a
faster RCU based hash tree instead of the rbtrees they were using
before.
On large NUMA systems it is a significant improvement.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Allocate send/recv rings in memory that is node-local to the HCA.
This significantly helps performance.
Signed-off-by: Andy Grover <andy.grover@oracle.com>
rds_ib_get_device is called very often as we turn an
ip address into a corresponding device structure. It currently
take a global spinlock as it walks different lists to find active
devices.
This commit changes the lists over to RCU, which isn't very complex
because they are not updated very often at all.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Implement a CMSG-based interface to do FADD and CSWP ops.
Alter send routines to handle atomic ops.
Add atomic counters to stats.
Add xmit_atomic() to struct rds_transport
Inline rds_ib_send_unmap_rdma into unmap_rm
Signed-off-by: Andy Grover <andy.grover@oracle.com>
