Skip to contentSkip to conversation

4.0.2-stable-queue build failures due to 'clockevents: Fix cpu_down() race for hrtimer based broadcasting'

DISCUSSION Guenter Roeck 3 participants 8 messages May 11, 2015 – May 12, 2015

Subsystems (2)
Stable branch Nohz, dynticks support
lore
Share & tools
lore

Guenter Roeck

author
#0 lore raw

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

---

Bisect log:

# bad: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting # good: [072cab659c9368586d6417cfd6ec2d2c68469c67] Linux 4.0.2
git bisect start 'HEAD' 'v4.0.2'
# good: [7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d] cdc-acm: prevent infinite loop when parsing CDC headers. git bisect good 7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d
# good: [8b55f847f1e9964ac2ab657cc0e1334ed8be214c] ext4: fix data corruption caused by unwritten and delayed extents git bisect good 8b55f847f1e9964ac2ab657cc0e1334ed8be214c
# good: [7c0e04a5e08bec7f9101d20214f92b5f466a556b] drm/radeon: only mark audio as connected if the monitor supports it (v3) git bisect good 7c0e04a5e08bec7f9101d20214f92b5f466a556b
# good: [da9778d808f0629fd804421e97294f25b81b1885] drm/radeon: fix lockup when BOs aren't part of the VM on release git bisect good da9778d808f0629fd804421e97294f25b81b1885
# good: [68cd28ab24ce16ac7f3978a7627eae5b599c3d8b] drm/radeon: check new address before removing old one git bisect good 68cd28ab24ce16ac7f3978a7627eae5b599c3d8b
# good: [c20b6545f649724a733e008e7124fa9d1e114c71] Drivers: hv: vmbus: Don't wait after requesting offers git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Guenter

#1 lore raw
Quoted 39 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

---

Bisect log:

# bad: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting
# good: [072cab659c9368586d6417cfd6ec2d2c68469c67] Linux 4.0.2 git bisect start 'HEAD' 'v4.0.2'
# good: [7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d] cdc-acm: prevent infinite loop when parsing CDC headers.
git bisect good 7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d
# good: [8b55f847f1e9964ac2ab657cc0e1334ed8be214c] ext4: fix data corruption caused by unwritten and delayed extents
git bisect good 8b55f847f1e9964ac2ab657cc0e1334ed8be214c
# good: [7c0e04a5e08bec7f9101d20214f92b5f466a556b] drm/radeon: only mark audio as connected if the monitor supports it (v3)
git bisect good 7c0e04a5e08bec7f9101d20214f92b5f466a556b
# good: [da9778d808f0629fd804421e97294f25b81b1885] drm/radeon: fix lockup when BOs aren't part of the VM on release
git bisect good da9778d808f0629fd804421e97294f25b81b1885
# good: [68cd28ab24ce16ac7f3978a7627eae5b599c3d8b] drm/radeon: check new address before removing old one
git bisect good 68cd28ab24ce16ac7f3978a7627eae5b599c3d8b
# good: [c20b6545f649724a733e008e7124fa9d1e114c71] Drivers: hv: vmbus:
Don't wait after requesting offers
git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

Quoted 6 lines
# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

Signed-off-by Preeti U. Murthy <[email protected]>
include/linux/tick.h +1 -0
index e3e9d20..b62ff06 100644
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }
 #endif /* !CONFIG_GENERIC_CLOCKEVENTS */
 
 # ifdef CONFIG_NO_HZ_COMMON

Thanks

Regards
Preeti U Murthy

Quoted 3 lines

Guenter

maintainer
#2 lore raw
Quoted 62 lines · Preeti U Murthy
Quoted 39 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

---

Bisect log:

# bad: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting
# good: [072cab659c9368586d6417cfd6ec2d2c68469c67] Linux 4.0.2 git bisect start 'HEAD' 'v4.0.2'
# good: [7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d] cdc-acm: prevent infinite loop when parsing CDC headers.
git bisect good 7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d
# good: [8b55f847f1e9964ac2ab657cc0e1334ed8be214c] ext4: fix data corruption caused by unwritten and delayed extents
git bisect good 8b55f847f1e9964ac2ab657cc0e1334ed8be214c
# good: [7c0e04a5e08bec7f9101d20214f92b5f466a556b] drm/radeon: only mark audio as connected if the monitor supports it (v3)
git bisect good 7c0e04a5e08bec7f9101d20214f92b5f466a556b
# good: [da9778d808f0629fd804421e97294f25b81b1885] drm/radeon: fix lockup when BOs aren't part of the VM on release
git bisect good da9778d808f0629fd804421e97294f25b81b1885
# good: [68cd28ab24ce16ac7f3978a7627eae5b599c3d8b] drm/radeon: check new address before removing old one
git bisect good 68cd28ab24ce16ac7f3978a7627eae5b599c3d8b
# good: [c20b6545f649724a733e008e7124fa9d1e114c71] Drivers: hv: vmbus:
Don't wait after requesting offers
git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

Quoted 6 lines
# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

Signed-off-by Preeti U. Murthy <[email protected]>
include/linux/tick.h +0 -0
index e3e9d20..b62ff06 100644

include/linux/tick.h static inline void tick_clock_notify(void) { } · line 140

Quoted 7 lines
include/linux/tick.h +1 -0
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }
 #endif /* !CONFIG_GENERIC_CLOCKEVENTS */
 
 # ifdef CONFIG_NO_HZ_COMMON

I'm just going to drop this patch, as it's causing way too many problems and something major is wrong with the backport, again. It's obvious you never tested this both times it was sent to me :(

greg k-h

author
#3 lore raw
In reply to Preeti U Murthy
Quoted 14 lines · Preeti U Murthy
Quoted 13 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

[ ... ]

Quoted 18 lines
Quoted 7 lines

git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

+++ b/include/linux/tick.h

include/linux/tick.h static inline void tick_clock_notify(void) { } · line 137

Quoted 4 lines
include/linux/tick.h +1 -0
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }

Possibly it does, but why is the back-port so much different from the original patch ? This would be the third declaration of this dummy function, which just seems wrong. The original patch only needs one.

Guenter

#4 lore raw
Quoted 33 lines · Guenter Roeck
Quoted 14 lines · Preeti U Murthy
Quoted 13 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

[ ... ]

Quoted 16 lines
Quoted 7 lines

git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

Quoted 10 lines

include/linux/tick.h static inline void tick_clock_notify(void) { } · line 137

Quoted 6 lines

+++ b/include/linux/tick.h

include/linux/tick.h +1 -0
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }

Possibly it does, but why is the back-port so much different from the original patch ? This would be the third declaration of this dummy function, which just seems wrong. The original patch only needs one.

That is because these two patches simplified the layout of the related files.

c1797baf6880174: tick: Move core only declarations and functions to core
b7475eb599ddb2e: tick: Simplify tick-internal.h

These patches are not yet in stable.

Regards
Preeti U Murthy

Quoted 3 lines

Guenter

maintainer
#5 lore raw
Quoted 45 lines · Preeti U Murthy
Quoted 33 lines · Guenter Roeck
Quoted 14 lines · Preeti U Murthy
Quoted 13 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

[ ... ]

Quoted 16 lines
Quoted 7 lines

git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

Quoted 10 lines

include/linux/tick.h static inline void tick_clock_notify(void) { } · line 137

Quoted 6 lines

+++ b/include/linux/tick.h

include/linux/tick.h +1 -0
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }

Possibly it does, but why is the back-port so much different from the original patch ? This would be the third declaration of this dummy function, which just seems wrong. The original patch only needs one.

Quoted 7 lines

That is because these two patches simplified the layout of the related files.

c1797baf6880174: tick: Move core only declarations and functions to core
b7475eb599ddb2e: tick: Simplify tick-internal.h

These patches are not yet in stable.

Then you should have asked for them to be merged as well. I don't want to deviate from what is in Linus's tree if at all possible. This long email exchange is the exact reason why (patch errors, build errors, wasted time, confusion, etc.)

Please resend me the exact git commit ids that you wish to have added to the stable tree, in the order you wish to have them added. If something needs to be changed/backported, be very explicit as to why it needs to be changed, and that you have done so.

thanks,

greg k-h

#6 lore raw
Quoted 59 lines · Greg Kroah-Hartman
Quoted 45 lines · Preeti U Murthy
Quoted 33 lines · Guenter Roeck
Quoted 14 lines · Preeti U Murthy
Quoted 13 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

[ ... ]

Quoted 16 lines
Quoted 7 lines

git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f]
clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Please let me know if the following patch fixes the issue:

clockevents: Fix build error caused by fix for hotplug race with hrtimer broadcast

commit: c71309a3658d ("clockevents: Fix cpu_down() race for hrtimer based broadcasting") causes a build error on certain archs where CONFIG_GENERIC_CLOCKEVENTS
is not set. This patch fixes this.

Quoted 10 lines

include/linux/tick.h static inline void tick_clock_notify(void) { } · line 137

Quoted 6 lines

+++ b/include/linux/tick.h

include/linux/tick.h +1 -0
@@ -134,6 +134,7 @@ static inline void tick_clock_notify(void) { }
 static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 static inline void tick_irq_enter(void) { }
 static inline int tick_oneshot_mode_active(void) { return 0; }
+static inline void hotplug_cpu__broadcast_tick_pull(int dead_cpu) { }

Possibly it does, but why is the back-port so much different from the original patch ? This would be the third declaration of this dummy function, which just seems wrong. The original patch only needs one.

Quoted 7 lines

That is because these two patches simplified the layout of the related files.

c1797baf6880174: tick: Move core only declarations and functions to core
b7475eb599ddb2e: tick: Simplify tick-internal.h

These patches are not yet in stable.

Then you should have asked for them to be merged as well. I don't want to deviate from what is in Linus's tree if at all possible. This long email exchange is the exact reason why (patch errors, build errors, wasted time, confusion, etc.)

Quoted 4 lines

Please resend me the exact git commit ids that you wish to have added to the stable tree, in the order you wish to have them added. If something needs to be changed/backported, be very explicit as to why it needs to be changed, and that you have done so.

Ok Greg, I will do so. I apologize for the confusion created.

Regards
Preeti U Murthy

Quoted 5 lines

thanks,

greg k-h

maintainer
#7 lore raw
In reply to Guenter Roeck
Quoted 31 lines · Guenter Roeck

Build is still going on, so there may be other affected architectures.

---

Building ia64:defconfig ... failed
Building parisc:generic-32bit_defconfig ... failed
--------------
Error log:
kernel/cpu.c: In function '_cpu_down':
kernel/cpu.c:415:2: error: implicit declaration of function 'hotplug_cpu__broadcast_tick_pull' [-Werror=implicit-function-declaration]
cc1: some warnings being treated as errors

---

Bisect log:

# bad: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting # good: [072cab659c9368586d6417cfd6ec2d2c68469c67] Linux 4.0.2
git bisect start 'HEAD' 'v4.0.2'
# good: [7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d] cdc-acm: prevent infinite loop when parsing CDC headers. git bisect good 7e62c9a30d01ac14be2bc3e048b737e1b0c8d08d
# good: [8b55f847f1e9964ac2ab657cc0e1334ed8be214c] ext4: fix data corruption caused by unwritten and delayed extents git bisect good 8b55f847f1e9964ac2ab657cc0e1334ed8be214c
# good: [7c0e04a5e08bec7f9101d20214f92b5f466a556b] drm/radeon: only mark audio as connected if the monitor supports it (v3) git bisect good 7c0e04a5e08bec7f9101d20214f92b5f466a556b
# good: [da9778d808f0629fd804421e97294f25b81b1885] drm/radeon: fix lockup when BOs aren't part of the VM on release git bisect good da9778d808f0629fd804421e97294f25b81b1885
# good: [68cd28ab24ce16ac7f3978a7627eae5b599c3d8b] drm/radeon: check new address before removing old one git bisect good 68cd28ab24ce16ac7f3978a7627eae5b599c3d8b
# good: [c20b6545f649724a733e008e7124fa9d1e114c71] Drivers: hv: vmbus: Don't wait after requesting offers git bisect good c20b6545f649724a733e008e7124fa9d1e114c71

Quoted 5 lines
# first bad commit: [c71309a3658d323d56261d4590b2c5214e61b05f] clockevents: Fix cpu_down() race for hrtimer based broadcasting

---

Reverting the bad commit fixes the problem.

Thanks for letting me know, that commit is now removed.

greg k-h