x86/intel_rdt: Fix potential deadlock during resctrl mount

Sai reported a warning during some MBA tests:

[  236.755559] ======================================================
[  236.762443] WARNING: possible circular locking dependency detected
[  236.769328] 4.14.0-rc4-yocto-standard #8 Not tainted
[  236.774857] ------------------------------------------------------
[  236.781738] mount/10091 is trying to acquire lock:
[  236.787071]  (cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff8117f892>] static_key_enable+0x12/0x30
[  236.797058]
               but task is already holding lock:
[  236.803552]  (&type->s_umount_key#37/1){+.+.}, at: [<ffffffff81208b2f>] sget_userns+0x32f/0x520
[  236.813247]
               which lock already depends on the new lock.

[  236.822353]
               the existing dependency chain (in reverse order) is:
[  236.830686]
               -> #4 (&type->s_umount_key#37/1){+.+.}:
[  236.837756]        __lock_acquire+0x1100/0x11a0
[  236.842799]        lock_acquire+0xdf/0x1d0
[  236.847363]        down_write_nested+0x46/0x80
[  236.852310]        sget_userns+0x32f/0x520
[  236.856873]        kernfs_mount_ns+0x7e/0x1f0
[  236.861728]        rdt_mount+0x30c/0x440
[  236.866096]        mount_fs+0x38/0x150
[  236.870262]        vfs_kern_mount+0x67/0x150
[  236.875015]        do_mount+0x1df/0xd50
[  236.879286]        SyS_mount+0x95/0xe0
[  236.883464]        entry_SYSCALL_64_fastpath+0x18/0xad
[  236.889183]
               -> #3 (rdtgroup_mutex){+.+.}:
[  236.895292]        __lock_acquire+0x1100/0x11a0
[  236.900337]        lock_acquire+0xdf/0x1d0
[  236.904899]        __mutex_lock+0x80/0x8f0
[  236.909459]        mutex_lock_nested+0x1b/0x20
[  236.914407]        intel_rdt_online_cpu+0x3b/0x4a0
[  236.919745]        cpuhp_invoke_callback+0xce/0xb80
[  236.925177]        cpuhp_thread_fun+0x1c5/0x230
[  236.930222]        smpboot_thread_fn+0x11a/0x1e0
[  236.935362]        kthread+0x152/0x190
[  236.939536]        ret_from_fork+0x27/0x40
[  236.944097]
               -> #2 (cpuhp_state-up){+.+.}:
[  236.950199]        __lock_acquire+0x1100/0x11a0
[  236.955241]        lock_acquire+0xdf/0x1d0
[  236.959800]        cpuhp_issue_call+0x12e/0x1c0
[  236.964845]        __cpuhp_setup_state_cpuslocked+0x13b/0x2f0
[  236.971242]        __cpuhp_setup_state+0xa7/0x120
[  236.976483]        page_writeback_init+0x43/0x67
[  236.981623]        pagecache_init+0x38/0x3b
[  236.986281]        start_kernel+0x3c6/0x41a
[  236.990931]        x86_64_start_reservations+0x2a/0x2c
[  236.996650]        x86_64_start_kernel+0x72/0x75
[  237.001793]        verify_cpu+0x0/0xfb
[  237.005966]
               -> #1 (cpuhp_state_mutex){+.+.}:
[  237.012364]        __lock_acquire+0x1100/0x11a0
[  237.017408]        lock_acquire+0xdf/0x1d0
[  237.021969]        __mutex_lock+0x80/0x8f0
[  237.026527]        mutex_lock_nested+0x1b/0x20
[  237.031475]        __cpuhp_setup_state_cpuslocked+0x54/0x2f0
[  237.037777]        __cpuhp_setup_state+0xa7/0x120
[  237.043013]        page_alloc_init+0x28/0x30
[  237.047769]        start_kernel+0x148/0x41a
[  237.052425]        x86_64_start_reservations+0x2a/0x2c
[  237.058145]        x86_64_start_kernel+0x72/0x75
[  237.063284]        verify_cpu+0x0/0xfb
[  237.067456]
               -> #0 (cpu_hotplug_lock.rw_sem){++++}:
[  237.074436]        check_prev_add+0x401/0x800
[  237.079286]        __lock_acquire+0x1100/0x11a0
[  237.084330]        lock_acquire+0xdf/0x1d0
[  237.088890]        cpus_read_lock+0x42/0x90
[  237.093548]        static_key_enable+0x12/0x30
[  237.098496]        rdt_mount+0x406/0x440
[  237.102862]        mount_fs+0x38/0x150
[  237.107035]        vfs_kern_mount+0x67/0x150
[  237.111787]        do_mount+0x1df/0xd50
[  237.116058]        SyS_mount+0x95/0xe0
[  237.120233]        entry_SYSCALL_64_fastpath+0x18/0xad
[  237.125952]
               other info that might help us debug this:

[  237.134867] Chain exists of:
                 cpu_hotplug_lock.rw_sem --> rdtgroup_mutex --> &type->s_umount_key#37/1

[  237.148425]  Possible unsafe locking scenario:

[  237.155015]        CPU0                    CPU1
[  237.160057]        ----                    ----
[  237.165100]   lock(&type->s_umount_key#37/1);
[  237.169952]                                lock(rdtgroup_mutex);
[  237.176641]
lock(&type->s_umount_key#37/1);
[  237.184287]   lock(cpu_hotplug_lock.rw_sem);
[  237.189041]
                *** DEADLOCK ***

When the resctrl filesystem is mounted the locks must be acquired in the
same order as was done when the cpus came online:

     cpu_hotplug_lock before rdtgroup_mutex.

This also requires to switch the static_branch_enable() calls to the
_cpulocked variant because now cpu hotplug lock is held already.

[ tglx: Switched to cpus_read_[un]lock ]

Reported-by: Sai Praneeth Prakhya <[email protected]>
Signed-off-by: Reinette Chatre <[email protected]>
Tested-by: Sai Praneeth Prakhya <[email protected]>
Acked-by: Vikas Shivappa <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: https://lkml.kernel.org/r/9c41b91bc2f47d9e95b62b213ecdb45623c47a9f.1508490116.git.reinette.chatre@intel.com
Signed-off-by: Thomas Gleixner <[email protected]>

arch/x86/kernel/cpu/intel_rdt_rdtgroup.c

@@ -1149,6 +1149,7 @@ static struct dentry *rdt_mount(struct file_system_type *fs_type,
 	struct dentry *dentry;
 	int ret;
 
+	cpus_read_lock();
 	mutex_lock(&rdtgroup_mutex);
 	/*
 	 * resctrl file system can only be mounted once.
@@ -1198,12 +1199,12 @@ static struct dentry *rdt_mount(struct file_system_type *fs_type,
 		goto out_mondata;
 
 	if (rdt_alloc_capable)
-		static_branch_enable(&rdt_alloc_enable_key);
+		static_branch_enable_cpuslocked(&rdt_alloc_enable_key);
 	if (rdt_mon_capable)
-		static_branch_enable(&rdt_mon_enable_key);
+		static_branch_enable_cpuslocked(&rdt_mon_enable_key);
 
 	if (rdt_alloc_capable || rdt_mon_capable)
-		static_branch_enable(&rdt_enable_key);
+		static_branch_enable_cpuslocked(&rdt_enable_key);
 
 	if (is_mbm_enabled()) {
 		r = &rdt_resources_all[RDT_RESOURCE_L3];
@@ -1226,6 +1227,7 @@ static struct dentry *rdt_mount(struct file_system_type *fs_type,
 out:
 	rdt_last_cmd_clear();
 	mutex_unlock(&rdtgroup_mutex);
+	cpus_read_unlock();
 
 	return dentry;
 }