[PATCH V5 0/4] blk-mq: implement queue quiesce via percpu_ref for BLK_MQ_F_BLOCKING

[Ming Lei]

On Fri, Sep 11, 2020 at 11:34:45AM -0700, Sagi Grimberg wrote:
> 
> > > Hi Jens,
> > > 
> > > The 1st patch add .mq_quiesce_mutex for serializing quiesce/unquiesce,
> > > and prepares for replacing srcu with percpu_ref.
> > > 
> > > The 2nd patch replaces srcu with percpu_ref.
> > > 
> > > The 3rd patch adds tagset quiesce interface.
> > > 
> > > The 4th patch applies tagset quiesce interface for NVMe subsystem.
> > 
> > Tested some reset storms and target restarts during traffic with
> > nvme-tcp.
> > 
> > Seems that no apparent breakage.
> > 
> > So:
> > 
> > Tested-by: Sagi Grimberg <sagi at grimberg.me>
> 
> Probably unrelated to this patches, but I do see new
> kmemleak complaints in the form of:
> --
> unreferenced object 0xffff9440dbf3c240 (size 64):
>   comm "systemd", pid 1, jiffies 4306444056 (age 25034.440s)
>   hex dump (first 32 bytes):
>     00 00 00 00 00 00 00 00 b0 fe 13 99 ff ff ff ff  ................
>     00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
>   backtrace:
>     [<00000000f1d0b20e>] percpu_ref_init+0x5f/0xf0
>     [<000000009598103f>] cgroup_mkdir+0xe9/0x440
>     [<0000000001b93c19>] kernfs_iop_mkdir+0x57/0x80
>     [<000000001ed0f985>] vfs_mkdir+0x10e/0x1d0
>     [<00000000cac65f7e>] do_mkdirat+0xec/0x120
>     [<00000000956db630>] do_syscall_64+0x33/0x80
>     [<000000001c2b0e1a>] entry_SYSCALL_64_after_hwframe+0x44/0xa9

Hello Sagi,

Could you test the following patch and see if the leak is fixed?

diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index dd247747ec14..e8b7a8b66415 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -4962,6 +4962,7 @@ static void css_free_rwork_fn(struct work_struct *work)
 			psi_cgroup_free(cgrp);
 			if (cgroup_on_dfl(cgrp))
 				cgroup_rstat_exit(cgrp);
+			percpu_ref_exit(&cgrp->self.refcnt);
 			kfree(cgrp);
 		} else {
 			/*

Thanks, 
Ming