| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
FQ rate limiting suffers from two problems, reported
by Steinar :
1) FQ enforces a delay when flow quantum is exhausted in order
to reduce cpu overhead. But if packets are small, current
delay computation is slightly wrong, and observed rates can
be too high.
Steinar had this problem because he disabled TSO and GSO,
and default FQ quantum is 2*1514.
(Of course, I wish recent TSO auto sizing changes will help
to not having to disable TSO in the first place)
2) maxrate was not used for forwarded flows (skbs not attached
to a socket)
Tested:
tc qdisc add dev eth0 root est 1sec 4sec fq maxrate 8Mbit
netperf -H lpq84 -l 1000 &
sleep 10 ; tc -s qdisc show dev eth0
qdisc fq 8003: root refcnt 32 limit 10000p flow_limit 100p buckets 1024
quantum 3028 initial_quantum 15140 maxrate 8000Kbit
Sent 16819357 bytes 11258 pkt (dropped 0, overlimits 0 requeues 0)
rate 7831Kbit 653pps backlog 7570b 5p requeues 0
44 flows (43 inactive, 1 throttled), next packet delay 2977352 ns
0 gc, 0 highprio, 5545 throttled
lpq83:~# tcpdump -p -i eth0 host lpq84 -c 12
09:02:52.079484 IP lpq83 > lpq84: . 1389536928:1389538376(1448) ack 3808678021 win 457 <nop,nop,timestamp 961812 572609068>
09:02:52.079499 IP lpq83 > lpq84: . 1448:2896(1448) ack 1 win 457 <nop,nop,timestamp 961812 572609068>
09:02:52.079906 IP lpq84 > lpq83: . ack 2896 win 16384 <nop,nop,timestamp 572609080 961812>
09:02:52.082568 IP lpq83 > lpq84: . 2896:4344(1448) ack 1 win 457 <nop,nop,timestamp 961815 572609071>
09:02:52.082581 IP lpq83 > lpq84: . 4344:5792(1448) ack 1 win 457 <nop,nop,timestamp 961815 572609071>
09:02:52.083017 IP lpq84 > lpq83: . ack 5792 win 16384 <nop,nop,timestamp 572609083 961815>
09:02:52.085678 IP lpq83 > lpq84: . 5792:7240(1448) ack 1 win 457 <nop,nop,timestamp 961818 572609074>
09:02:52.085693 IP lpq83 > lpq84: . 7240:8688(1448) ack 1 win 457 <nop,nop,timestamp 961818 572609074>
09:02:52.086117 IP lpq84 > lpq83: . ack 8688 win 16384 <nop,nop,timestamp 572609086 961818>
09:02:52.088792 IP lpq83 > lpq84: . 8688:10136(1448) ack 1 win 457 <nop,nop,timestamp 961821 572609077>
09:02:52.088806 IP lpq83 > lpq84: . 10136:11584(1448) ack 1 win 457 <nop,nop,timestamp 961821 572609077>
09:02:52.089217 IP lpq84 > lpq83: . ack 11584 win 16384 <nop,nop,timestamp 572609090 961821>
Reported-by: Steinar H. Gunderson <sesse@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
fq_reset() should drops all packets in queue, including
throttled flows.
This patch moves code from fq_destroy() to fq_reset()
to do the cleaning.
fq_change() must stop calling fq_dequeue() if all remaining
packets are from throttled flows.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
kbuild bot reported following m68k build error :
net/sched/sch_fq.c: In function 'fq_dequeue':
>> net/sched/sch_fq.c:491:2: error: implicit declaration of function
'prefetch' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors
While we are fixing this, move this prefetch() call a bit earlier.
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Uses perfect flow match (not stochastic hash like SFQ/FQ_codel)
- Uses the new_flow/old_flow separation from FQ_codel
- New flows get an initial credit allowing IW10 without added delay.
- Special FIFO queue for high prio packets (no need for PRIO + FQ)
- Uses a hash table of RB trees to locate the flows at enqueue() time
- Smart on demand gc (at enqueue() time, RB tree lookup evicts old
unused flows)
- Dynamic memory allocations.
- Designed to allow millions of concurrent flows per Qdisc.
- Small memory footprint : ~8K per Qdisc, and 104 bytes per flow.
- Single high resolution timer for throttled flows (if any).
- One RB tree to link throttled flows.
- Ability to have a max rate per flow. We might add a socket option
to add per socket limitation.
Attempts have been made to add TCP pacing in TCP stack, but this
seems to add complex code to an already complex stack.
TCP pacing is welcomed for flows having idle times, as the cwnd
permits TCP stack to queue a possibly large number of packets.
This removes the 'slow start after idle' choice, hitting badly
large BDP flows, and applications delivering chunks of data
as video streams.
Nicely spaced packets :
Here interface is 10Gbit, but flow bottleneck is ~20Mbit
cwin is big, yet FQ avoids the typical bursts generated by TCP
(as in netperf TCP_RR -- -r 100000,100000)
15:01:23.545279 IP A > B: . 78193:81089(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.545394 IP B > A: . ack 81089 win 3668 <nop,nop,timestamp 11597985 1115>
15:01:23.546488 IP A > B: . 81089:83985(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.546565 IP B > A: . ack 83985 win 3668 <nop,nop,timestamp 11597986 1115>
15:01:23.547713 IP A > B: . 83985:86881(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.547778 IP B > A: . ack 86881 win 3668 <nop,nop,timestamp 11597987 1115>
15:01:23.548911 IP A > B: . 86881:89777(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.548949 IP B > A: . ack 89777 win 3668 <nop,nop,timestamp 11597988 1115>
15:01:23.550116 IP A > B: . 89777:92673(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.550182 IP B > A: . ack 92673 win 3668 <nop,nop,timestamp 11597989 1115>
15:01:23.551333 IP A > B: . 92673:95569(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.551406 IP B > A: . ack 95569 win 3668 <nop,nop,timestamp 11597991 1115>
15:01:23.552539 IP A > B: . 95569:98465(2896) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.552576 IP B > A: . ack 98465 win 3668 <nop,nop,timestamp 11597992 1115>
15:01:23.553756 IP A > B: . 98465:99913(1448) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.554138 IP A > B: P 99913:100001(88) ack 65248 win 3125 <nop,nop,timestamp 1115 11597805>
15:01:23.554204 IP B > A: . ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.554234 IP B > A: . 65248:68144(2896) ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.555620 IP B > A: . 68144:71040(2896) ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.557005 IP B > A: . 71040:73936(2896) ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.558390 IP B > A: . 73936:76832(2896) ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.559773 IP B > A: . 76832:79728(2896) ack 100001 win 3668 <nop,nop,timestamp 11597993 1115>
15:01:23.561158 IP B > A: . 79728:82624(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.562543 IP B > A: . 82624:85520(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.563928 IP B > A: . 85520:88416(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.565313 IP B > A: . 88416:91312(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.566698 IP B > A: . 91312:94208(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.568083 IP B > A: . 94208:97104(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.569467 IP B > A: . 97104:100000(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.570852 IP B > A: . 100000:102896(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.572237 IP B > A: . 102896:105792(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.573639 IP B > A: . 105792:108688(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.575024 IP B > A: . 108688:111584(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.576408 IP B > A: . 111584:114480(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
15:01:23.577793 IP B > A: . 114480:117376(2896) ack 100001 win 3668 <nop,nop,timestamp 11597994 1115>
TCP timestamps show that most packets from B were queued in the same ms
timeframe (TSval 1159799{3,4}), but FQ managed to send them right
in time to avoid a big burst.
In slow start or steady state, very few packets are throttled [1]
FQ gets a bunch of tunables as :
limit : max number of packets on whole Qdisc (default 10000)
flow_limit : max number of packets per flow (default 100)
quantum : the credit per RR round (default is 2 MTU)
initial_quantum : initial credit for new flows (default is 10 MTU)
maxrate : max per flow rate (default : unlimited)
buckets : number of RB trees (default : 1024) in hash table.
(consumes 8 bytes per bucket)
[no]pacing : disable/enable pacing (default is enable)
All of them can be changed on a live qdisc.
$ tc qd add dev eth0 root fq help
Usage: ... fq [ limit PACKETS ] [ flow_limit PACKETS ]
[ quantum BYTES ] [ initial_quantum BYTES ]
[ maxrate RATE ] [ buckets NUMBER ]
[ [no]pacing ]
$ tc -s -d qd
qdisc fq 8002: dev eth0 root refcnt 32 limit 10000p flow_limit 100p buckets 256 quantum 3028 initial_quantum 15140
Sent 216532416 bytes 148395 pkt (dropped 0, overlimits 0 requeues 14)
backlog 0b 0p requeues 14
511 flows, 511 inactive, 0 throttled
110 gc, 0 highprio, 0 retrans, 1143 throttled, 0 flows_plimit
[1] Except if initial srtt is overestimated, as if using
cached srtt in tcp metrics. We'll provide a fix for this issue.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
| |
The next patch will use it to determine if a cgroup is newly created
while we're iterating the cgroup hierarchy.
tj: Rephrased the comment on top of cgroup_serial_nr_cursor.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
|
| |
|
|
|
|
|
|
|
| |
The memory allocated in cgroup_add_cftypes() should be freed. The
effect of this bug is we leak a bit memory everytime we unload
cfq-iosched module if blkio cgroup is enabled.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, there's no easy way to find out the next sibling cgroup
unless it's known that the current cgroup is accessed from the
parent's children list in a single RCU critical section. This in turn
forces all iterators to require whole iteration to be enclosed in a
single RCU critical section, which sometimes is too restrictive. This
patch implements cgroup_next_sibling() which can reliably determine
the next sibling regardless of the state of the current cgroup as long
as it's accessible.
It currently is impossible to determine the next sibling after
dropping RCU read lock because the cgroup being iterated could be
removed anytime and if RCU read lock is dropped, nothing guarantess
its ->sibling.next pointer is accessible. A removed cgroup would
continue to point to its next sibling for RCU accesses but stop
receiving updates from the sibling. IOW, the next sibling could be
removed and then complete its grace period while RCU read lock is
dropped, making it unsafe to dereference ->sibling.next after dropping
and re-acquiring RCU read lock.
This can be solved by adding a way to traverse to the next sibling
without dereferencing ->sibling.next. This patch adds a monotonically
increasing cgroup serial number, cgroup->serial_nr, which guarantees
that all cgroup->children lists are kept in increasing serial_nr
order. A new function, cgroup_next_sibling(), is implemented, which,
if CGRP_REMOVED is not set on the current cgroup, follows
->sibling.next; otherwise, traverses the parent's ->children list
until it sees a sibling with higher ->serial_nr.
This allows the function to always return the next sibling regardless
of the state of the current cgroup without adding overhead in the fast
path.
Further patches will update the iterators to use cgroup_next_sibling()
so that they allow dropping RCU read lock and blocking while iteration
is in progress which in turn will be used to simplify controllers.
v2: Typo fix as per Serge.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
|
| |
|
|
|
|
|
|
|
|
| |
cgroup_is_removed() no longer has external users and it shouldn't grow
any - controllers should deal with cgroup_subsys_state on/offline
state instead of cgroup removal state. Make it static.
While at it, make it return bool.
Signed-off-by: Tejun Heo <tj@kernel.org>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
kdbus folks want a sane way to determine the cgroup path that a given
task belongs to on a given hierarchy, which is a reasonble thing to
expect from cgroup core.
Implement task_cgroup_path_from_hierarchy().
v2: Dropped unnecessary NULL check on the return value of
task_cgroup_from_root() as suggested by Li Zefan.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Greg Kroah-Hartman <greg@kroah.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Kay Sievers <kay@vrfy.org>
Cc: Lennart Poettering <lennart@poettering.net>
Cc: Daniel Mack <daniel@zonque.org>
|
| |
|
|
|
|
|
|
|
| |
We want to be able to lookup a hierarchy from its id and cyclic
allocation is a whole lot simpler with idr. Convert to idr and use
idr_alloc_cyclc().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
|
| |
|
|
|
|
|
|
| |
Now that hierarchy_id alloc / free are protected by the cgroup
mutexes, there's no need for this separate lock. Drop it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We're planning to converting hierarchy_ida to an idr and use it to
look up hierarchy from its id. As we want the mapping to happen
atomically with cgroupfs_root registration, this patch refactors
hierarchy_id init / exit so that ida operations happen inside
cgroup_[root_]mutex.
* s/init_root_id()/cgroup_init_root_id()/ and make it return 0 or
-errno like a normal function.
* Move hierarchy_id initialization from cgroup_root_from_opts() into
cgroup_mount() block where the root is confirmed to be used and
being registered while holding both mutexes.
* Split cgroup_drop_id() into cgroup_exit_root_id() and
cgroup_free_root(), so that ID release can happen before dropping
the mutexes in cgroup_kill_sb(). The latter expects hierarchy_id to
be exited before being invoked.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, when the last reference of a blkcg_gq is put, all then
release operations sans the actual freeing happen directly in
blkg_put(). As blkg_put() may be called under queue_lock, all
pd_exit_fn()s may be too. This makes it impossible for pd_exit_fn()s
to use del_timer_sync() on timers which grab the queue_lock which is
an irq-safe lock due to the deadlock possibility described in the
comment on top of del_timer_sync().
This can be easily avoided by perfoming the release operations in the
RCU callback instead of directly from blkg_put(). This patch moves
the blkcg_gq release operations to the RCU callback.
As this leaves __blkg_release() with only call_rcu() invocation,
blkg_rcu_free() is renamed to __blkg_release_rcu(), exported and
call_rcu() invocation is now done directly from blkg_put() instead of
going through __blkg_release() which is removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, when creating a new blkcg_gq, each policy's pd_init_fn() is
invoked in blkg_alloc() before the parent is linked. This makes it
difficult for policies to perform initializations which are dependent
on the parent.
This patch moves pd_init_fn() invocations to blkg_create() after the
parent blkg is linked where the new blkg is fully initialized. As
this means that blkg_free() can't assume that pd's are initialized,
pd_exit_fn() invocations are moved to __blkg_release(). This
guarantees that pd_exit_fn() is also invoked with fully initialized
blkgs with valid parent pointers.
This will help implementing hierarchy support in blk-throttle.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
| |
This will be used by blk-throttle hierarchy support.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
| |
blk-throttle hierarchy support will make use of it. Move
blkg_for_each_descendant_pre() from block/blk-cgroup.c to
block/blk-cgroup.h.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
In blkg_create(), after lookup of parent fails, the control jumps to
error path with the error code encoded into @blkg. The error path
doesn't use @blkg for the return value. It returns ERR_PTR(ret).
Make lookup fail path set @ret instead of @blkg.
Note that the parent lookup is guaranteed to succeed at that point and
the condition check is purely for sanity and triggers WARN when fails.
As such, I don't think it's necessary to mark it for -stable.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With the recent updates, blk-throttle is finally ready for proper
hierarchy support. Dispatching now honors service_queue->parent_sq
and propagates correctly. The only thing missing is setting
->parent_sq correctly so that throtl_grp hierarchy matches the cgroup
hierarchy.
This patch updates throtl_pd_init() such that service_queues form the
same hierarchy as the cgroup hierarchy if sane_behavior is enabled.
As this concludes proper hierarchy support for blkcg, the shameful
.broken_hierarchy tag is removed from blkio_subsys.
v2: Updated blkio-controller.txt as suggested by Vivek.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Cc: Li Zefan <lizefan@huawei.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
blk_throtl_bio() has a quick exit path for throtl_grps without limits
configured. It looks at the bps and iops limits and if both are not
configured, the bio is issued immediately. While this is correct in
the current flat hierarchy as each throtl_grp behaves completely
independently, it would become wrong in proper hierarchy mode. A
group without any limits could still be limited by one of its
ancestors and bio's queued for such group should not bypass
blk-throtl.
As having a quick bypass mechanism is beneficial, this patch
reimplements the mechanism such that it's correct even with proper
hierarchy. throtl_grp->has_rules[] is added. These booleans are
updated for the whole subtree whenever a config is updated so that
has_rules[] of the whole subtree stays synchronized. They're also
updated when a new throtl_grp comes online so that it can't escape the
limits of its ancestors.
As no throtl_grp has another throtl_grp as parent now, this patch
doesn't yet make any behavior differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With the planned proper hierarchy support, a bio will climb up the
tree before actually being dispatched. This makes sure bio is also
subjected to parent's throttling limits, if any.
It might happen that parent is idle and when bio is transferred to
parent, a new slice starts fresh. But that is incorrect as parents
wait time should have started when bio was queued in child group and
causes IOs to be throttled more than configured as they climb the
hierarchy.
Given the fact that we have not written hierarchical algorithm in a
way where child's and parents time slices are synchronized, we
transfer the child's start time to parent if parent was idling. If
parent was busy doing dispatch of other bios all this while, this is
not an issue.
Child's slice start time is passed to parent. Parent looks at its
last expired slice start time. If child's start time is after parents
old start time, that means parent had been idle and after parent
went idle, child had an IO queued. So use child's start time as
parent start time.
If parent's start time is after child's start time, that means,
when IO got queued in child group, parent was not idle. But later
it dispatched some IO, its slice got trimmed and then it went idle.
After a while child's request got shifted in parent group. In this
case use parent's old start time as new start time as that's the
duration of slice we did not use.
This logic is far from perfect as if there are multiple childs
then first child transferring the bio decides the start time while
a bio might have queued up even earlier in other child, which is
yet to be transferred up to parent. In that case we will lose
time and bandwidth in parent. This patch is just an approximation
to make situation somewhat better.
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With flat hierarchy, there's only single level of dispatching
happening and fairness beyond that point is the responsibility of the
rest of the block layer and driver, which usually works out okay;
however, with the planned hierarchy support,
service_queue->bio_lists[] can be filled up by bios from a single
source. While the limits would still be honored, it'd be very easy to
starve IOs from siblings or children.
To avoid such starvation, this patch implements throtl_qnode and
converts service_queue->bio_lists[] to lists of per-source qnodes
which in turn contains the bio's. For example, when a bio is
dispatched from a child group, the bio doesn't get queued on
->bio_lists[] directly but it first gets queued on the group's qnode
which in turn gets queued on service_queue->queued[]. When
dispatching for the upper level, the ->queued[] list is consumed in
round-robing order so that the dispatch windows is consumed fairly by
all IO sources.
There are two ways a bio can come to a throtl_grp - directly queued to
the group or dispatched from a child. For the former
throtl_grp->qnode_on_self[rw] is used. For the latter, the child's
->qnode_on_parent[rw].
Note that this means that the child which is contributing a bio to its
parent should stay pinned until all its bios are dispatched to its
grand-parent. This patch moves blkg refcnting from bio add/remove
spots to qnode activation/deactivation so that the blkg containing an
active qnode is always pinned. As child pins the parent, this is
sufficient for keeping the relevant sub-tree pinned while bios are in
flight.
The starvation issue was spotted by Vivek Goyal.
v2: The original patch used the same throtl_grp->qnode_on_self/parent
for reads and writes causing RWs to be queued incorrectly if there
already are outstanding IOs in the other direction. They should
be throtl_grp->qnode_on_self/parent[2] so that READs and WRITEs
can use different qnodes. Spotted by Vivek Goyal.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_pending_timer_fn() currently assumes that the parent_sq is the
top level one and the bio's dispatched are ready to be issued;
however, this assumption will be wrong with proper hierarchy support.
This patch makes the following changes to make
throtl_pending_timer_fn() ready for hiearchy.
* If the parent_sq isn't the top-level one, update the parent
throtl_grp's dispatch time and schedule the next dispatch as
necessary. If the parent's dispatch time is now, repeat the
function for the parent throtl_grp.
* If the parent_sq is the top-level one, kick issue work_item as
before.
* The debug message printed by throtl_log() now prints out the
service_queue's nr_queued[] instead of the total nr_queued as the
latter becomes uninteresting and misleading with hierarchical
dispatch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
tg_dispatch_one_bio() currently assumes that the parent_sq is the top
level one and the bio being dispatched is ready to be issued; however,
this assumption will be wrong with proper hierarchy support. This
patch makes the following changes to make tg_dispatch_on_bio() ready
for hiearchy.
* throtl_data->nr_queued[] is incremented in blk_throtl_bio() instead
of throtl_add_bio_tg() so that throtl_add_bio_tg() can be used to
transfer a bio from a child tg to its parent.
* tg_dispatch_one_bio() is updated to distinguish whether its parent
is another throtl_grp or the throtl_data. If former, the bio is
transferred to the parent throtl_grp using throtl_add_bio_tg(). If
latter, the bio is ready to be issued and put on the top-level
service_queue's bio_lists[] and throtl_data->nr_queued is
decremented.
As all throtl_grps currently have the top level service_queue as their
->parent_sq, this patch in itself doesn't make any behavior
difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, blk_throtl_bio() issues the passed in bio directly if it's
within limits of its associated tg (throtl_grp). This behavior
becomes incorrect with hierarchy support as the bio should be
accounted to and throttled by the ancestor throtl_grps too.
This patch makes the direct issue path of blk_throtl_bio() to loop
until it reaches the top-level service_queue or gets throttled. If
the former, the bio can be issued directly; otherwise, it gets queued
at the first layer it was above limits.
As tg->parent_sq is always the top-level service queue currently, this
patch in itself doesn't make any behavior differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current blk_throtl_drain() assumes that all active throtl_grps are
queued on throtl_data->service_queue, which won't be true once
hierarchy support is implemented.
This patch makes blk_throtl_drain() perform post-order walk of the
blkg hierarchy draining each associated throtl_grp, which guarantees
that all bios will eventually be pushed to the top-level service_queue
in throtl_data.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, blk_throtl_dispatch_work_fn() is responsible for both
dispatching bio's from throtl_grp's according to their limits and then
issuing the dispatched bios.
This patch moves the dispatch part to throtl_pending_timer_fn() so
that the work item is kicked iff there are bio's to issue. This is to
avoid work item execution at each step when hierarchy support is
enabled. bio's will be dispatched towards the top-level service_queue
from the timers at each layer and the work item will only be used to
issue the bio's which reached the top-level service_queue.
While fetching bio's to issue from bio_lists[],
blk_throtl_dispatch_work_fn() fetches all READs before WRITEs. While
the original code also dispatched READs first, if multiple throtl_grps
are dispatched on the same run, WRITEs from throtl_grp which is
dispatched first would precede READs from throtl_grps which are
dispatched later. While this is a behavior change, given that the
previous code already prioritized READs and block layer generally
prioritizes and segregates READs from WRITEs, this isn't likely to
make any noticeable differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_select_dispatch() only dispatches throtl_quantum bios on each
invocation. blk_throtl_dispatch_work_fn() in turn depends on
throtl_schedule_next_dispatch() scheduling the next dispatch window
immediately so that undue delays aren't incurred. This effectively
chains multiple dispatch work item executions back-to-back when there
are more than throtl_quantum bios to dispatch on a given tick.
There is no reason to finish the current work item just to repeat it
immediately. This patch makes throtl_schedule_next_dispatch() return
%false without doing anything if the current dispatch window is still
open and updates blk_throtl_dispatch_work_fn() repeat dispatching
after cpu_relax() on %false return.
This change will help implementing hierarchy support as dispatching
will be done from pending_timer and immediate reschedule of timer
function isn't supported and doesn't make much sense.
While this patch changes how dispatch behaves when there are more than
throtl_quantum bios to dispatch on a single tick, the behavior change
is immaterial.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_data->dispatch_work
Currently, throtl_data->dispatch_work is a delayed_work item which
handles both delayed dispatch and issuing bios. The two tasks will be
separated to support proper hierarchy. To prepare for that, this
patch separates out the timer into throtl_service_queue->pending_timer
from throtl_data->dispatch_work and make the latter a work_struct.
* As the timer is now per-service_queue, it's initialized and
del_sync'd as its corresponding service_queue is created and
destroyed. The timer, when triggered, simply schedules
throtl_data->dispathc_work for execution.
* throtl_schedule_delayed_work() is renamed to
throtl_schedule_pending_timer() and takes @sq and @expires now.
* Simiarly, throtl_schedule_next_dispatch() now takes @sq, which
should be the parent_sq of the service_queue which just got a new
bio or updated. As the parent_sq is always the top-level
service_queue now, this doesn't change anything at this point.
This patch doesn't introduce any behavior differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
update with it
With proper hierarchy support, a bio can be dispatched multiple times
until it reaches the top-level service_queue and we don't want to
update dispatch stats at each step. They are local stats and will be
kept local. If recursive stats are necessary, they should be
implemented separately and definitely not by updating counters
recursively on each dispatch.
This patch moves REQ_THROTTLED setting to throtl_charge_bio() and gate
stats update with it so that dispatch stats are updated only on the
first time the bio is charged to a throtl_grp, which will always be
the throtl_grp the bio was originally queued to.
This means that REQ_THROTTLED would be set even for bios which don't
get throttled. As we don't want bios to leave blk-throtl with the
flag set, move REQ_THROTLLED clearing to the end of blk_throtl_bio()
and clear if the bio is being issued directly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now that both throtl_data and throtl_grp embed throtl_service_queue,
we can unify throtl_log() and throtl_log_tg().
* sq_to_tg() is added. This returns the throtl_grp a service_queue is
embedded in. If the service_queue is the top-level one embedded in
throtl_data, NULL is returned.
* sq_to_td() is added. A service_queue is always associated with a
throtl_data. This function finds the associated td and returns it.
* throtl_log() is updated to take throtl_service_queue instead of
throtl_data. If the service_queue is one embedded in throtl_grp, it
prints the same header as throtl_log_tg() did. If it's one embedded
in throtl_data, it behaves the same as before. This renders
throtl_log_tg() unnecessary. Removed.
This change is necessary for hierarchy support as we're gonna be using
the same code paths to dispatch bios to intermediate service_queues
embedded in throtl_grps and the top-level service_queue embedded in
throtl_data.
This patch doesn't make any behavior changes.
v2: throtl_log() didn't print a space after blkg path. Updated so
that it prints a space after throtl_grp path. Spotted by Vivek.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
To prepare for hierarchy support, this patch adds
throtl_service_queue->service_sq which points to the arent
service_queue. Currently, for all service_queues embedded in
throtl_grps, it points to throtl_data->service_queue. As
throtl_data->service_queue doesn't have a parent its parent_sq is set
to NULL.
There are a number of functions which take both throtl_grp *tg and
throtl_service_queue *parent_sq. With this patch, the parent
service_queue can be determined from @tg and the @parent_sq arguments
are removed.
This patch doesn't make any behavior differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When blk_throtl_bio() wants to queue a bio to a tg (throtl_grp), it
avoids invoking tg_update_disptime() and
throtl_schedule_next_dispatch() if the tg already has bios queued in
that direction. As a new bio is appeneded after the existing ones, it
can't change the tg's next dispatch time or the parent's dispatch
schedule.
This optimization is currently open coded in blk_throtl_bio().
Whether the target biolist was occupied was recorded in a local
variable and later used to skip disptime update. This patch moves
generalizes it so that throtl_add_bio_tg() sets a new flag
THROTL_TG_WAS_EMPTY if the biolist was empty before the new bio was
added. tg_update_disptime() clears the flag automatically.
blk_throtl_bio() is updated to simply test the flag before updating
disptime.
This patch doesn't make any functional differences now but will enable
using the same optimization for recursive dispatch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_service_queues will eventually form a tree which is anchored at
throtl_data->service_queue and queue bios will climb the tree to the
top service_queue to be executed.
This patch makes the dispatch paths in blk_throtl_dispatch_work_fn()
and blk_throtl_drain() to dispatch bios to
throtl_data->service_queue.bio_lists[] instead of the on-stack
bio_lists. This will keep the final dispatch to the top level
service_queue share the same mechanism as dispatches through the rest
of the hierarchy.
As bio's should be issued in a sleepable context,
blk_throtl_dispatch_work_fn() transfers all dispatched bio's from the
service_queue bio_lists[] into an onstack one before dropping
queue_lock and issuing the bio's.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_service_queues will eventually form a tree which is anchored at
throtl_data->service_queue and queue bios will climb the tree to the
top service_queue to be executed.
This patch moves bio_lists[] and nr_queued[] from throtl_grp to its
service_queue to prepare for that. As currently only the
throtl_data->service_queue is in use, this patch just ends up moving
throtl_grp->bio_lists[] and ->nr_queued[] to
throtl_grp->service_queue.bio_lists[] and ->nr_queued[] without making
any functional differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, there's single service_queue per queue -
throtl_data->service_queue. All active throtl_grp's are queued on the
queue and dispatched according to their limits. To support hierarchy,
this will be expanded such that active throtl_grp's form a tree
anchored at throtl_data->service_queue and chained through each
intermediate throtl_grp's service_queue.
This patch adds throtl_grp->service_queue to prepare for hierarchy
support. The initialization function - throtl_service_queue_init() -
is added and replaces the macro initializer. The newly added
tg->service_queue isn't used yet. Following patches will do.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_service_queue will be the building block of hierarchy support
and will form a tree. This patch updates its usages as arguments to
reduce confusion.
* When a service queue is used as the parent role - the host of the
rbtree - use @parent_sq instead of @sq.
* For functions taking both @tg and @parent_sq, reorder them so that
the order is (@tg, @parent_sq) not the other way around. This makes
the code follow the usual convention of specifying the primary
target of the operation as the first argument.
This patch doesn't make any functional differences.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_service_queue will be used as the basic block to implement
hierarchy support. Pass around throtl_service_queue *sq instead of
throtl_data *td in the following functions which will be used across
multiple levels of hierarchy.
* [__]throtl_enqueue/dequeue_tg()
* throtl_add_bio_tg()
* tg_update_disptime()
* throtl_select_dispatch()
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add throtl_grp->td so that the td (throtl_data) a given tg
(throtl_grp) belongs to can be determined, and remove @td argument
from functions which take both @td and @tg as the former now can be
determined from the latter.
This generally simplifies the code and removes a number of cases where
@td is passed as an argument without being actually used. This will
also help hierarchy support implementation.
While at it, in multi-line conditions, move the logical operators
leading broken lines to the end of the previous line.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
| |
blk-throttle is still using function-defining macros to define flag
handling functions, which went out style at least a decade ago.
Just define the flag as bitmask and use direct bit operations.
This patch doesn't make any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_rb_root will be expanded to cover more roles for hierarchy
support. Rename it to throtl_service_queue and make its fields more
descriptive.
* rb -> pending_tree
* left -> first_pending
* count -> nr_pending
* min_disptime -> first_pending_disptime
This patch is purely cosmetic.
Signed-off-by: Tejun Heo <tj@kernel.org
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_nr_queued() is used in several places to avoid performing
certain operations when the throtl_data is empty. This usually is
useless as those paths usually aren't traveled if there's no bio
queued.
* throtl_schedule_delayed_work() skips scheduling dispatch work item
if @td doesn't have any bios queued; however, the only case it can
be called when @td is empty is from tg_set_conf() which isn't
something we should be optimizing for.
* throtl_schedule_next_dispatch() takes a quick exit if @td is empty;
however, right after that it triggers BUG if the service tree is
empty. The two conditions are equivalent and it can just test
@st->count for the quick exit.
* blk_throtl_dispatch_work_fn() skips dispatch if @td is empty. This
work function isn't usually invoked when @td is empty. The only
possibility is from tg_set_conf() and when it happens the normal
dispatching path can handle empty @td fine. No need to add special
skip path.
This patch removes the above three unnecessary optimizations, which
leave throtl_log() call in blk_throtl_dispatch_work_fn() the only user
of throtl_nr_queued(). Remove throtl_nr_queued() and open code it in
throtl_log(). I don't think we need td->nr_queued[] at all. Maybe we
can remove it later.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
| |
Move throtl_schedule_delayed_work() above its first user so that the
forward declaration can be removed.
This patch is pure relocaiton.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
blk-throttle is about to go through major restructuring to support
hierarchy. Do cosmetic updates in preparation.
* s/throtl_data->throtl_work/throtl_data->dispatch_work/
* s/blk_throtl_work()/blk_throtl_dispatch_work_fn()/
* Collapse throtl_dispatch() into blk_throtl_dispatch_work_fn()
This patch is purely cosmetic.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When bps or iops configuration changes, blk-throttle records the new
configuration and sets a flag indicating that the config has changed.
The flag is checked in the bio dispatch path and applied. This
deferred config application was necessary due to limitations in blkcg
framework, which haven't existed for quite a while now.
This patch removes the deferred config application mechanism and
applies new configurations directly from tg_set_conf(), which is
simpler.
v2: Dropped unnecessary throtl_schedule_delayed_work() call from
tg_set_conf() as suggested by Vivek Goyal.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
throtl_select_dispatch()
throtl_select_dispatch() calls throtl_enqueue_tg() right after
tg_update_disptime(), which always calls the function anyway. The
call is, while harmless, unnecessary. Remove it.
This patch doesn't introduce any behavior difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
This reverts commit 207f5d933450512d44043ee9436a95f175598137.
|
