rds: ib: Fix cleanup of rds_ib_cache_gc_worker

rds_ib_free_caches() calls cancel_delayed_work() in an attempt to clean up
the rds_ib_cache_gc_worker(). But, rds_ib_cache_gc_worker() may still be
running after cancel_delayed_work() is called. And since
rds_ib_cache_gc_worker() re-queues itself, a disaster may happen:

BUG: unable to handle page fault for address: 00007ffeaa9bf6cc
[snip]
Call Trace:
 <IRQ>
 run_timer_softirq+0x19/0x2d
 __do_softirq+0xd0/0x2a5
 ? sched_clock_cpu+0x9/0xb6
 __irq_exit_rcu+0xc7/0xf1
 sysvec_apic_timer_interrupt+0x72/0x89
 </IRQ>
 <TASK>
 asm_sysvec_apic_timer_interrupt+0x16/0x1b
RIP: 0010:cpuidle_enter_state+0xc7/0x35d
[snip]
 cpuidle_enter+0x29/0x40
 cpuidle_idle_call+0x143/0x1de
 do_idle+0x81/0xd2
 cpu_startup_entry+0x19/0x1b
 secondary_startup_64_no_verify+0xc2/0x0
 </TASK>

or

general protection fault, probably for non-canonical address
0xffff11de4ab26c00: 0000 [#1] SMP PTI
[snip]
RIP: 0010:__queue_work+0xde/0x40a
[snip]
Call Trace:
 <IRQ>
 ? queue_work_node+0x110/0x105
 call_timer_fn+0x27/0xff
 __run_timers+0x1bd/0x299
 run_timer_softirq+0x19/0x2d
 __do_softirq+0xd0/0x2a5
 ? sched_clock_cpu+0x9/0xb6
 __irq_exit_rcu+0xc7/0xf1
 sysvec_apic_timer_interrupt+0x72/0x89
 </IRQ>
 <TASK>
 asm_sysvec_apic_timer_interrupt+0x16/0x1b
RIP: 0010:cpuidle_enter_state+0xc7/0x35d
[snip]
 ? cpuidle_enter_state+0xb7/0x35d
 cpuidle_enter+0x29/0x40
 cpuidle_idle_call+0x143/0x1de
 do_idle+0x81/0xd2
 cpu_startup_entry+0x19/0x1b
 secondary_startup_64_no_verify+0xc2/0x0
 </TASK>

or

BUG: kernel NULL pointer dereference, address: 0000000000000000
[snip]
RIP: 0010:_raw_spin_lock+0xc/0x51
[snip]
Call Trace:
 <IRQ>
 __queue_work+0x13f/0x40a
 call_timer_fn+0x24/0xff
 __run_timers+0x1bd/0x299
 run_timer_softirq+0x19/0x2d
 __do_softirq+0xcd/0x2a5
 __irq_exit_rcu+0xc7/0xf1
 sysvec_apic_timer_interrupt+0x72/0x89
 </IRQ>
 <TASK>
 asm_sysvec_apic_timer_interrupt+0x16/0x1b
RIP: 0010:cpuidle_enter_state+0xc7/0x35d
[snip]
 cpuidle_enter+0x29/0x40
 cpuidle_idle_call+0x143/0x1de
 do_idle+0x81/0xd2
 cpu_startup_entry+0x19/0x1b
 secondary_startup_64_no_verify+0xc2/0x0
 </TASK>

Fixed by calling cancel_delayed_work_sync() instead.

Orabug: 34806050

Fixes: 8450b32 ("RDS-IB: Add garbage-collection to cache")
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>
Reviewed-by: Sharath Srinivasan <sharath.srinivasan@oracle.com>

Author: Hakon-Bugge

net/rds/ib.c

@@ -421,7 +421,7 @@ static void rds_ib_free_caches(struct rds_ib_device *rds_ibdev)
 {
 	int i;
 
-	cancel_delayed_work(&rds_ibdev->i_cache_gc_work);
+	cancel_delayed_work_sync(&rds_ibdev->i_cache_gc_work);
 	rds_ib_free_inc_cache(&rds_ibdev->i_cache_incs);
 	for (i = 0; i < RDS_FRAG_CACHE_ENTRIES; i++)
 		rds_ib_free_frag_cache(rds_ibdev->i_cache_frags + i, PAGE_SIZE << i);