Skip to contentSkip to conversation

x86/intel_rdt: Locking and initialization fixes

PATCH In review Reinette Chatre 3 participants 9 messages Oct 20, 2017 – Oct 21, 2017 x86

lore
Share & tools
lore
#0 lore raw

cover x86/intel_rdt: Locking and initialization fixes

Hi Thomas,

Please consider these three fixes for the RDT subsystem. They are submitted as series for convenience, not because of dependencies among them.

Thank you

Reinette

Reinette Chatre (3):
x86/intel_rdt: Initialize bitmask of shareable resource if CDP enabled x86/intel_rdt: Fix potential deadlock during resctrl unmount x86/intel_rdt: Fix potential deadlock during resctrl mount

 arch/x86/kernel/cpu/intel_rdt.c          |  1 +
 arch/x86/kernel/cpu/intel_rdt_rdtgroup.c | 18 ++++++++++--------
 2 files changed, 11 insertions(+), 8 deletions(-)
Signature · 1 line
2.13.5
#1 lore raw

PATCH 1/3 x86/intel_rdt: Initialize bitmask of shareable resource if CDP enabled

The platform informs us via CPUID.(EAX=0x10, ECX=res#):EBX[31:0] (valid res# are only 1 for L3 and 2 for L2) which unit of the allocation may be used by other entities in the platform. This information is valid whether CDP (Code and Data Prioritization) is enabled or not.

Ensure that the bitmask of shareable resource is initialized when CDP is enabled.

Fixes 0dd2d7494cd8 ("x86/intel_rdt: Show bitmask of shareable resource

with other executing units"

---
 arch/x86/kernel/cpu/intel_rdt.c | 1 +
 1 file changed, 1 insertion(+)
arch/x86/kernel/cpu/intel_rdt.c +1 -0
index cd5fc61ba450..88dcf8479013 100644
@@ -267,6 +267,7 @@ static void rdt_get_cdp_l3_config(int type)
 	r->num_closid = r_l3->num_closid / 2;
 	r->cache.cbm_len = r_l3->cache.cbm_len;
 	r->default_ctrl = r_l3->default_ctrl;
+	r->cache.shareable_bits = r_l3->cache.shareable_bits;
 	r->data_width = (r->cache.cbm_len + 3) / 4;
 	r->alloc_capable = true;
 	/*
Signature · 1 line
2.13.5
#2 lore raw

[tip:x86/cache] x86/intel_rdt: Initialize bitmask of shareable resource if CDP enabled

x86/intel_rdt: Initialize bitmask of shareable resource if CDP enabled
Commit-ID:  95953034fb24c16ad0047a98b16427e5935830c4
Gitweb:     https://git.kernel.org/tip/95953034fb24c16ad0047a98b16427e5935830c4
Author:     Reinette Chatre <[email protected]>
AuthorDate: Fri, 20 Oct 2017 02:16:57 -0700
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Sat, 21 Oct 2017 17:12:21 +0200

x86/intel_rdt: Initialize bitmask of shareable resource if CDP enabled

The platform informs via CPUID.(EAX=0x10, ECX=res#):EBX[31:0] (valid res# are only 1 for L3 and 2 for L2) which unit of the allocation may be used by other entities in the platform. This information is valid whether CDP (Code and Data Prioritization) is enabled or not.

Ensure that the bitmask of shareable resource is initialized when CDP is enabled.

---
 arch/x86/kernel/cpu/intel_rdt.c | 1 +
 1 file changed, 1 insertion(+)
arch/x86/kernel/cpu/intel_rdt.c +1 -0
index cd5fc61..88dcf84 100644
@@ -267,6 +267,7 @@ static void rdt_get_cdp_l3_config(int type)
 	r->num_closid = r_l3->num_closid / 2;
 	r->cache.cbm_len = r_l3->cache.cbm_len;
 	r->default_ctrl = r_l3->default_ctrl;
+	r->cache.shareable_bits = r_l3->cache.shareable_bits;
 	r->data_width = (r->cache.cbm_len + 3) / 4;
 	r->alloc_capable = true;
 	/*
#3 lore raw

PATCH 2/3 x86/intel_rdt: Fix potential deadlock during resctrl unmount

In reply to Reinette Chatre

Lockdep is warning us about a potential deadlock:

[   66.782842] ======================================================
[   66.782888] WARNING: possible circular locking dependency detected
[   66.782937] 4.14.0-rc2-test-test+ #48 Not tainted
[   66.782983] ------------------------------------------------------
[   66.783052] umount/336 is trying to acquire lock:
[   66.783117]  (cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff81032395>] rdt_kill_sb+0x215/0x390
[   66.783193]
               but task is already holding lock:
[   66.783244]  (rdtgroup_mutex){+.+.}, at: [<ffffffff810321b6>] rdt_kill_sb+0x36/0x390
[   66.783305]
               which lock already depends on the new lock.
[   66.783364]
               the existing dependency chain (in reverse order) is:
[   66.783419]
               -> #3 (rdtgroup_mutex){+.+.}:
[   66.783467]        __lock_acquire+0x1293/0x13f0
[   66.783509]        lock_acquire+0xaf/0x220
[   66.783543]        __mutex_lock+0x71/0x9b0
[   66.783575]        mutex_lock_nested+0x1b/0x20
[   66.783610]        intel_rdt_online_cpu+0x3b/0x430
[   66.783649]        cpuhp_invoke_callback+0xab/0x8e0
[   66.783687]        cpuhp_thread_fun+0x7a/0x150
[   66.783722]        smpboot_thread_fn+0x1cc/0x270
[   66.783764]        kthread+0x16e/0x190
[   66.783794]        ret_from_fork+0x27/0x40
[   66.783825]
               -> #2 (cpuhp_state){+.+.}:
[   66.783870]        __lock_acquire+0x1293/0x13f0
[   66.783906]        lock_acquire+0xaf/0x220
[   66.783938]        cpuhp_issue_call+0x102/0x170
[   66.783974]        __cpuhp_setup_state_cpuslocked+0x154/0x2a0
[   66.784023]        __cpuhp_setup_state+0xc7/0x170
[   66.784061]        page_writeback_init+0x43/0x67
[   66.784097]        pagecache_init+0x43/0x4a
[   66.784131]        start_kernel+0x3ad/0x3f7
[   66.784165]        x86_64_start_reservations+0x2a/0x2c
[   66.784204]        x86_64_start_kernel+0x72/0x75
[   66.784241]        verify_cpu+0x0/0xfb
[   66.784270]
               -> #1 (cpuhp_state_mutex){+.+.}:
[   66.784319]        __lock_acquire+0x1293/0x13f0
[   66.784355]        lock_acquire+0xaf/0x220
[   66.784387]        __mutex_lock+0x71/0x9b0
[   66.784419]        mutex_lock_nested+0x1b/0x20
[   66.784454]        __cpuhp_setup_state_cpuslocked+0x52/0x2a0
[   66.784497]        __cpuhp_setup_state+0xc7/0x170
[   66.784535]        page_alloc_init+0x28/0x30
[   66.784569]        start_kernel+0x148/0x3f7
[   66.784602]        x86_64_start_reservations+0x2a/0x2c
[   66.784642]        x86_64_start_kernel+0x72/0x75
[   66.784678]        verify_cpu+0x0/0xfb
[   66.784707]
               -> #0 (cpu_hotplug_lock.rw_sem){++++}:
[   66.784759]        check_prev_add+0x32f/0x6e0
[   66.784794]        __lock_acquire+0x1293/0x13f0
[   66.784830]        lock_acquire+0xaf/0x220
[   66.784863]        cpus_read_lock+0x3d/0xb0
[   66.784896]        rdt_kill_sb+0x215/0x390
[   66.784930]        deactivate_locked_super+0x3e/0x70
[   66.784968]        deactivate_super+0x40/0x60
[   66.785003]        cleanup_mnt+0x3f/0x80
[   66.785034]        __cleanup_mnt+0x12/0x20
[   66.785070]        task_work_run+0x8b/0xc0
[   66.785103]        exit_to_usermode_loop+0x94/0xa0
[   66.786804]        syscall_return_slowpath+0xe8/0x150
[   66.788502]        entry_SYSCALL_64_fastpath+0xab/0xad
[   66.790194]
               other info that might help us debug this:
[   66.795139] Chain exists of:
                 cpu_hotplug_lock.rw_sem --> cpuhp_state --> rdtgroup_mutex
[   66.800035]  Possible unsafe locking scenario:
[   66.803267]        CPU0                    CPU1
[   66.804867]        ----                    ----
[   66.806443]   lock(rdtgroup_mutex);
[   66.808002]                                lock(cpuhp_state);
[   66.809565]                                lock(rdtgroup_mutex);
[   66.811110]   lock(cpu_hotplug_lock.rw_sem);
[   66.812608]
                *** DEADLOCK ***
[   66.816983] 2 locks held by umount/336:
[   66.818418]  #0:  (&type->s_umount_key#35){+.+.}, at: [<ffffffff81229738>] deactivate_super+0x38/0x60
[   66.819922]  #1:  (rdtgroup_mutex){+.+.}, at: [<ffffffff810321b6>] rdt_kill_sb+0x36/0x390

When the resctrl filesystem is unmounted we should obtain the locks in the same order as was done when the cpus came online, cpu_hotplug_lock before rdtgroup_mutex. With cpu_hotplug_lock now obtained earlier we also modify the static_branch_disable() calls to the cpuslocked variant.

---
 arch/x86/kernel/cpu/intel_rdt_rdtgroup.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/intel_rdt_rdtgroup.c +5 -5
index 8a61b20c7e51..fc27f55e90dc 100644
@@ -1364,9 +1364,7 @@ static void rmdir_all_sub(void)
 		kfree(rdtgrp);
 	}
 	/* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
-	get_online_cpus();
 	update_closid_rmid(cpu_online_mask, &rdtgroup_default);
-	put_online_cpus();
 
 	kernfs_remove(kn_info);
 	kernfs_remove(kn_mongrp);
@@ -1377,6 +1375,7 @@ static void rdt_kill_sb(struct super_block *sb)
 {
 	struct rdt_resource *r;
 
+	get_online_cpus();
 	mutex_lock(&rdtgroup_mutex);
 
 	/*Put everything back to default values. */
@@ -1384,11 +1383,12 @@ static void rdt_kill_sb(struct super_block *sb)
 		reset_all_ctrls(r);
 	cdp_disable();
 	rmdir_all_sub();
-	static_branch_disable(&rdt_alloc_enable_key);
-	static_branch_disable(&rdt_mon_enable_key);
-	static_branch_disable(&rdt_enable_key);
+	static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
+	static_branch_disable_cpuslocked(&rdt_mon_enable_key);
+	static_branch_disable_cpuslocked(&rdt_enable_key);
 	kernfs_kill_sb(sb);
 	mutex_unlock(&rdtgroup_mutex);
+	put_online_cpus();
 }
 
 static struct file_system_type rdt_fs_type = {
Signature · 1 line
2.13.5
Acked-by on v1 2/3
#4 lore raw

RE: [PATCH 2/3] x86/intel_rdt: Fix potential deadlock during resctrl unmount

Quoted 4 lines

From: Chatre, Reinette on Friday, October 20, 2017 2:17 AM
Subject: [PATCH 2/3] x86/intel_rdt: Fix potential deadlock during resctrl unmount

#5 lore raw

[tip:x86/cache] x86/intel_rdt: Fix potential deadlock during resctrl unmount

In reply to Reinette Chatre
x86/intel_rdt: Fix potential deadlock during resctrl unmount
Commit-ID:  36b6f9fcb8928c06b6638a4cf91bc9d69bb49aa2
Gitweb:     https://git.kernel.org/tip/36b6f9fcb8928c06b6638a4cf91bc9d69bb49aa2
Author:     Reinette Chatre <[email protected]>
AuthorDate: Fri, 20 Oct 2017 02:16:58 -0700
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Sat, 21 Oct 2017 17:12:21 +0200

x86/intel_rdt: Fix potential deadlock during resctrl unmount

Lockdep warns about a potential deadlock:

[   66.782842] ======================================================
[   66.782888] WARNING: possible circular locking dependency detected
[   66.782937] 4.14.0-rc2-test-test+ #48 Not tainted
[   66.782983] ------------------------------------------------------
[   66.783052] umount/336 is trying to acquire lock:
[   66.783117]  (cpu_hotplug_lock.rw_sem){++++}, at: [<ffffffff81032395>] rdt_kill_sb+0x215/0x390
[   66.783193]
               but task is already holding lock:
[   66.783244]  (rdtgroup_mutex){+.+.}, at: [<ffffffff810321b6>] rdt_kill_sb+0x36/0x390
[   66.783305]
               which lock already depends on the new lock.
[   66.783364]
               the existing dependency chain (in reverse order) is:
[   66.783419]
               -> #3 (rdtgroup_mutex){+.+.}:
[   66.783467]        __lock_acquire+0x1293/0x13f0
[   66.783509]        lock_acquire+0xaf/0x220
[   66.783543]        __mutex_lock+0x71/0x9b0
[   66.783575]        mutex_lock_nested+0x1b/0x20
[   66.783610]        intel_rdt_online_cpu+0x3b/0x430
[   66.783649]        cpuhp_invoke_callback+0xab/0x8e0
[   66.783687]        cpuhp_thread_fun+0x7a/0x150
[   66.783722]        smpboot_thread_fn+0x1cc/0x270
[   66.783764]        kthread+0x16e/0x190
[   66.783794]        ret_from_fork+0x27/0x40
[   66.783825]
               -> #2 (cpuhp_state){+.+.}:
[   66.783870]        __lock_acquire+0x1293/0x13f0
[   66.783906]        lock_acquire+0xaf/0x220
[   66.783938]        cpuhp_issue_call+0x102/0x170
[   66.783974]        __cpuhp_setup_state_cpuslocked+0x154/0x2a0
[   66.784023]        __cpuhp_setup_state+0xc7/0x170
[   66.784061]        page_writeback_init+0x43/0x67
[   66.784097]        pagecache_init+0x43/0x4a
[   66.784131]        start_kernel+0x3ad/0x3f7
[   66.784165]        x86_64_start_reservations+0x2a/0x2c
[   66.784204]        x86_64_start_kernel+0x72/0x75
[   66.784241]        verify_cpu+0x0/0xfb
[   66.784270]
               -> #1 (cpuhp_state_mutex){+.+.}:
[   66.784319]        __lock_acquire+0x1293/0x13f0
[   66.784355]        lock_acquire+0xaf/0x220
[   66.784387]        __mutex_lock+0x71/0x9b0
[   66.784419]        mutex_lock_nested+0x1b/0x20
[   66.784454]        __cpuhp_setup_state_cpuslocked+0x52/0x2a0
[   66.784497]        __cpuhp_setup_state+0xc7/0x170
[   66.784535]        page_alloc_init+0x28/0x30
[   66.784569]        start_kernel+0x148/0x3f7
[   66.784602]        x86_64_start_reservations+0x2a/0x2c
[   66.784642]        x86_64_start_kernel+0x72/0x75
[   66.784678]        verify_cpu+0x0/0xfb
[   66.784707]
               -> #0 (cpu_hotplug_lock.rw_sem){++++}:
[   66.784759]        check_prev_add+0x32f/0x6e0
[   66.784794]        __lock_acquire+0x1293/0x13f0
[   66.784830]        lock_acquire+0xaf/0x220
[   66.784863]        cpus_read_lock+0x3d/0xb0
[   66.784896]        rdt_kill_sb+0x215/0x390
[   66.784930]        deactivate_locked_super+0x3e/0x70
[   66.784968]        deactivate_super+0x40/0x60
[   66.785003]        cleanup_mnt+0x3f/0x80
[   66.785034]        __cleanup_mnt+0x12/0x20
[   66.785070]        task_work_run+0x8b/0xc0
[   66.785103]        exit_to_usermode_loop+0x94/0xa0
[   66.786804]        syscall_return_slowpath+0xe8/0x150
[   66.788502]        entry_SYSCALL_64_fastpath+0xab/0xad
[   66.790194]
               other info that might help us debug this:
[   66.795139] Chain exists of:
                 cpu_hotplug_lock.rw_sem --> cpuhp_state --> rdtgroup_mutex
[   66.800035]  Possible unsafe locking scenario:
[   66.803267]        CPU0                    CPU1
[   66.804867]        ----                    ----
[   66.806443]   lock(rdtgroup_mutex);
[   66.808002]                                lock(cpuhp_state);
[   66.809565]                                lock(rdtgroup_mutex);
[   66.811110]   lock(cpu_hotplug_lock.rw_sem);
[   66.812608]
                *** DEADLOCK ***
[   66.816983] 2 locks held by umount/336:
[   66.818418]  #0:  (&type->s_umount_key#35){+.+.}, at: [<ffffffff81229738>] deactivate_super+0x38/0x60
[   66.819922]  #1:  (rdtgroup_mutex){+.+.}, at: [<ffffffff810321b6>] rdt_kill_sb+0x36/0x390

When the resctrl filesystem is unmounted the locks should be obtain in the locks 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_disable() calls to the _cpulocked variant because now cpu hotplug lock is held already.

[ tglx: Switched to cpus_read_[un]lock ]

---
 arch/x86/kernel/cpu/intel_rdt_rdtgroup.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/intel_rdt_rdtgroup.c +5 -5
index 8a61b20..8ce5d03 100644
@@ -1364,9 +1364,7 @@ static void rmdir_all_sub(void)
 		kfree(rdtgrp);
 	}
 	/* Notify online CPUs to update per cpu storage and PQR_ASSOC MSR */
-	get_online_cpus();
 	update_closid_rmid(cpu_online_mask, &rdtgroup_default);
-	put_online_cpus();
 
 	kernfs_remove(kn_info);
 	kernfs_remove(kn_mongrp);
@@ -1377,6 +1375,7 @@ static void rdt_kill_sb(struct super_block *sb)
 {
 	struct rdt_resource *r;
 
+	cpus_read_lock();
 	mutex_lock(&rdtgroup_mutex);
 
 	/*Put everything back to default values. */
@@ -1384,11 +1383,12 @@ static void rdt_kill_sb(struct super_block *sb)
 		reset_all_ctrls(r);
 	cdp_disable();
 	rmdir_all_sub();
-	static_branch_disable(&rdt_alloc_enable_key);
-	static_branch_disable(&rdt_mon_enable_key);
-	static_branch_disable(&rdt_enable_key);
+	static_branch_disable_cpuslocked(&rdt_alloc_enable_key);
+	static_branch_disable_cpuslocked(&rdt_mon_enable_key);
+	static_branch_disable_cpuslocked(&rdt_enable_key);
 	kernfs_kill_sb(sb);
 	mutex_unlock(&rdtgroup_mutex);
+	cpus_read_unlock();
 }
 
 static struct file_system_type rdt_fs_type = {
#6 lore raw

PATCH 3/3 x86/intel_rdt: Fix potential deadlock during resctrl mount

In reply to Reinette Chatre

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 we should obtain the locks in the same order as was done when the cpus came online, cpu_hotplug_lock before rdtgroup_mutex. Since we now obtain cpu_hotplug_lock earlier we modify the static_branch_enable calls to the cpuslocked variants.

---
 arch/x86/kernel/cpu/intel_rdt_rdtgroup.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/x86/kernel/cpu/intel_rdt_rdtgroup.c +5 -3
index fc27f55e90dc..26c7602512a0 100644
@@ -1149,6 +1149,7 @@ static struct dentry *rdt_mount(struct file_system_type *fs_type,
 	struct dentry *dentry;
 	int ret;
 
+	get_online_cpus();
 	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);
+	put_online_cpus();
 
 	return dentry;
 }
Signature · 1 line
2.13.5
Acked-by on v1 3/3
#7 lore raw

RE: [PATCH 3/3] x86/intel_rdt: Fix potential deadlock during resctrl mount

Quoted 3 lines

From: Chatre, Reinette on Friday, October 20, 2017 2:17 AM
Subject: [PATCH 3/3] x86/intel_rdt: Fix potential deadlock during resctrl mount

#8 lore raw

[tip:x86/cache] x86/intel_rdt: Fix potential deadlock during resctrl mount

In reply to Reinette Chatre
x86/intel_rdt: Fix potential deadlock during resctrl mount
Commit-ID:  87943db7dfb0c5ee5aa74a9ac06346fadd9695c8
Gitweb:     https://git.kernel.org/tip/87943db7dfb0c5ee5aa74a9ac06346fadd9695c8
Author:     Reinette Chatre <[email protected]>
AuthorDate: Fri, 20 Oct 2017 02:16:59 -0700
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Sat, 21 Oct 2017 17:12:22 +0200

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 ]

---
 arch/x86/kernel/cpu/intel_rdt_rdtgroup.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/x86/kernel/cpu/intel_rdt_rdtgroup.c +5 -3
index 8ce5d03..64c5ff9 100644
@@ -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 @@ out_cdp:
 out:
 	rdt_last_cmd_clear();
 	mutex_unlock(&rdtgroup_mutex);
+	cpus_read_unlock();
 
 	return dentry;
 }