Skip to contentSkip to conversation

tpm: Fix TPM 1.2 Shutdown sequence to prevent future TPM operations

PATCH In review Douglas Anderson 4 participants 21 messages Jul 11, 2019 – Aug 5, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tpm: Fix TPM 1.2 Shutdown sequence to prevent future TPM operations

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

---
This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

[1] https://lkml.kernel.org/r/CAD=FV=UoSV9LKOTMuXKRfgFir+7_qPkuhSLN6XJEKPiRPuJJwg@mail.gmail.com

 drivers/char/tpm/tpm-chip.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/char/tpm/tpm-chip.c +4 -5
index 46caadca916a..f784b6fd93b4 100644
@@ -187,12 +187,11 @@ static int tpm_class_shutdown(struct device *dev)
 {
 	struct tpm_chip *chip = container_of(dev, struct tpm_chip, dev);
 
-	if (chip->flags & TPM_CHIP_FLAG_TPM2) {
-		down_write(&chip->ops_sem);
+	down_write(&chip->ops_sem);
+	if (chip->flags & TPM_CHIP_FLAG_TPM2)
 		tpm2_shutdown(chip, TPM2_SU_CLEAR);
-		chip->ops = NULL;
-		up_write(&chip->ops_sem);
-	}
+	chip->ops = NULL;
+	up_write(&chip->ops_sem);
 
 	return 0;
 }
Signature · 1 line
2.22.0.410.gd8fdbe21b5-goog
#1 lore raw
Quoted 23 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Quoted 4 lines

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Jason

#2 lore raw

Hi,

Quoted 25 lines · Jason Gunthorpe
Quoted 23 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Quoted 7 lines
Quoted 4 lines

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Ah, got it. Thanks for catching! Should we just give up on trying to get this to stable then, or are the sysfs ops locking patches also easy to queue up?

-Doug

maintainer
#3 lore raw
In reply to Jason Gunthorpe
Quoted 24 lines · Jason Gunthorpe
Quoted 23 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Quoted 7 lines
Quoted 4 lines

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

thanks,

greg k-h

#4 lore raw
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

The last sentence suggests there are other patches needed too though..

Jason

maintainer
#5 lore raw
Quoted 40 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200
Quoted 8 lines

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

thanks,

greg k-h

#6 lore raw

Hi,

Quoted 49 lines · Greg KH
Quoted 47 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

Quoted 4 lines
Quoted 1 line

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

The patch ("Fix TPM 1.2 Shutdown sequence to prevent future TPM operations")? It's already done. It just got merge conflicts when going back to 4.19 which is why I sent the backport.

-Doug

maintainer
#7 lore raw
Quoted 52 lines · Doug Anderson

Hi,

Quoted 49 lines · Greg KH
Quoted 47 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

The patch ("Fix TPM 1.2 Shutdown sequence to prevent future TPM operations")? It's already done. It just got merge conflicts when going back to 4.19 which is why I sent the backport.

But the sysfs comment means I should not apply this backport then?

Totally confused by this long thread, sorry.

What am I supposed to do for the stable trees here?

thanks,

greg k-h

#8 lore raw

Hi,

Quoted 63 lines · Greg KH
Quoted 52 lines · Doug Anderson

Hi,

Quoted 49 lines · Greg KH
Quoted 47 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

The patch ("Fix TPM 1.2 Shutdown sequence to prevent future TPM operations")? It's already done. It just got merge conflicts when going back to 4.19 which is why I sent the backport.

Quoted 5 lines

But the sysfs comment means I should not apply this backport then?

Totally confused by this long thread, sorry.

What am I supposed to do for the stable trees here?

I think the answer is to drop my backport for now and Jarkko says he'll take a fresh look at it in 2 weeks when he's back from his leave. Thus my understanding:

* On mainline: fixed

* On 5.2 / 5.1: you've already got this picked to stable. Good

* On 4.14 / 4.19: Jarkko will look at in 2 weeks.

* On 4.9 and older: I'd propose skipping unless someone is known to need a solution here.

-Doug

maintainer
#9 lore raw
Quoted 76 lines · Doug Anderson

Hi,

Quoted 63 lines · Greg KH
Quoted 52 lines · Doug Anderson

Hi,

Quoted 49 lines · Greg KH
Quoted 47 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

The patch ("Fix TPM 1.2 Shutdown sequence to prevent future TPM operations")? It's already done. It just got merge conflicts when going back to 4.19 which is why I sent the backport.

Quoted 5 lines

But the sysfs comment means I should not apply this backport then?

Totally confused by this long thread, sorry.

What am I supposed to do for the stable trees here?

I think the answer is to drop my backport for now and Jarkko says he'll take a fresh look at it in 2 weeks when he's back from his leave. Thus my understanding:

Quoted 8 lines

* On mainline: fixed

* On 5.2 / 5.1: you've already got this picked to stable. Good

* On 4.14 / 4.19: Jarkko will look at in 2 weeks.

* On 4.9 and older: I'd propose skipping unless someone is known to need a solution here.

Thanks, that makes sense now.

greg k-h

#10 lore raw
Quoted 81 lines · Greg KH
Quoted 80 lines · Doug Anderson

Hi,

Quoted 63 lines · Greg KH
Quoted 52 lines · Doug Anderson

Hi,

Quoted 49 lines · Greg KH
Quoted 47 lines · Jason Gunthorpe
Quoted 30 lines · Greg KH
Quoted 29 lines · Jason Gunthorpe
Quoted 27 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Quoted 4 lines
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

And what commit added that?

commit 2677ca98ae377517930c183248221f69f771c921

Author: Jarkko Sakkinen <[email protected]>
Date:   Sun Nov 4 11:38:27 2018 +0200

tpm: use tpm_try_get_ops() in tpm-sysfs.c.

Use tpm_try_get_ops() in tpm-sysfs.c so that we can consider moving other decorations (locking, localities, power management for example) inside it. This direction can be of course taken only after other call sites for tpm_transmit() have been treated in the same way.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

The last sentence suggests there are other patches needed too though..

So 5.1. So does this original patch need to go into the 5.2 and 5.1 kernels?

The patch ("Fix TPM 1.2 Shutdown sequence to prevent future TPM operations")? It's already done. It just got merge conflicts when going back to 4.19 which is why I sent the backport.

Quoted 5 lines

But the sysfs comment means I should not apply this backport then?

Totally confused by this long thread, sorry.

What am I supposed to do for the stable trees here?

I think the answer is to drop my backport for now and Jarkko says he'll take a fresh look at it in 2 weeks when he's back from his leave. Thus my understanding:

* On mainline: fixed

* On 5.2 / 5.1: you've already got this picked to stable. Good

Quoted 8 lines
Quoted 4 lines

* On 4.14 / 4.19: Jarkko will look at in 2 weeks.

* On 4.9 and older: I'd propose skipping unless someone is known to need a solution here.

Thanks, that makes sense now.

greg k-h

I have not forgotten this but might have to postpone the backport after Linux Plumbers. Just have lots of stuff in my queue ATM but right after the conference I have good slot to do the backports.

/Jarkko

#11 lore raw
In reply to Doug Anderson
Quoted 6 lines · Doug Anderson

* On 5.2 / 5.1: you've already got this picked to stable. Good

* On 4.14 / 4.19: Jarkko will look at in 2 weeks.

* On 4.9 and older: I'd propose skipping unless someone is known to need a solution here.

I'll prioritize 4.14 and 4.19.

If it doesn't become a too big struggle, I'll try to fix also older but no final word on that at this point.

/Jarkko

#12 lore raw
In reply to Greg KH
Quoted 5 lines · Greg KH

But the sysfs comment means I should not apply this backport then?

Totally confused by this long thread, sorry.

What am I supposed to do for the stable trees here?

I'll work out a proper patch set for stable kernels with necessary patches and cover letter with a brief summary in the week starting 29th of this month when I come back from vacation.

/Jarkko

#13 lore raw
In reply to Jason Gunthorpe
Quoted 24 lines · Jason Gunthorpe
Quoted 23 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Quoted 7 lines
Quoted 4 lines

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

/Jarkko

#14 lore raw
Quoted 4 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active. Secondly, why the crash would not realize with TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

/Jarkko

#15 lore raw
Quoted 10 lines · Jarkko Sakkinen
Quoted 4 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active.

Not strictly, AFAIK

Quoted 3 lines

Secondly, why the crash would not realize with
TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

TPM2 doesn't use the unlocked sysfs path

Jason

#16 lore raw
Quoted 12 lines · Jason Gunthorpe
Quoted 10 lines · Jarkko Sakkinen
Quoted 4 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active.

Quoted 7 lines

Not strictly, AFAIK

Quoted 3 lines

Secondly, why the crash would not realize with
TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

TPM2 doesn't use the unlocked sysfs path

Gah, sorry :-) I should have known that.

I can go through the patches needed when I come back from my leave after two weeks.

/Jarkko

#17 lore raw
Quoted 19 lines · Jarkko Sakkinen
Quoted 18 lines · Jason Gunthorpe
Quoted 10 lines · Jarkko Sakkinen
Quoted 4 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active.

Not strictly, AFAIK

Quoted 3 lines

Secondly, why the crash would not realize with
TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

Quoted 6 lines
Quoted 1 line

TPM2 doesn't use the unlocked sysfs path

Gah, sorry :-) I should have known that.

I can go through the patches needed when I come back from my leave after two weeks.

It might require a number of patches but maybe it makes also overally sense to fix the racy sysfs code in stable kernels.

/Jarkko

#18 lore raw
Quoted 22 lines · Jarkko Sakkinen
Quoted 21 lines · Jarkko Sakkinen
Quoted 19 lines · Jason Gunthorpe
Quoted 10 lines · Jarkko Sakkinen
Quoted 4 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active.

Not strictly, AFAIK

Quoted 3 lines

Secondly, why the crash would not realize with
TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

TPM2 doesn't use the unlocked sysfs path

Quoted 7 lines
Quoted 4 lines

Gah, sorry :-) I should have known that.

I can go through the patches needed when I come back from my leave after two weeks.

It might require a number of patches but maybe it makes also overally sense to fix the racy sysfs code in stable kernels.

The sysfs isn't racy, it justs used a different locking scheme

Jason

#19 lore raw
In reply to Jarkko Sakkinen

Hi,

Quoted 5 lines · Jarkko Sakkinen
Quoted 3 lines · Jarkko Sakkinen
Quoted 2 lines

Careful with this, you can't backport this to any kernels that don't have the sysfs ops locking changes or they will crash in sysfs code.

Quoted 8 lines
Quoted 1 line

Oops, I was way too fast! Thanks Jason.

Hmm... hold on a second.

How would the crash realize? I mean this is at the point when user space should not be active. Secondly, why the crash would not realize with TPM2? The only thing the fix is doing is to do the same thing with TPM1 essentially.

I will continue to remind that I'm pretty TPM-clueless (mostly I just took someone else's patch and posted it), but I will note that people on the Chrome OS team seemed concerned by the sysfs locking too. After seeing Jason's message this morning I dug a little bit and found
<https://crbug.com/819265>

-Doug

Acked-by on v1 1/1
#20 lore raw
In reply to Douglas Anderson
Quoted 24 lines · Douglas Anderson

From: Vadim Sukhomlinov <[email protected]>

commit db4d8cb9c9f2af71c4d087817160d866ed572cc9 upstream.

TPM 2.0 Shutdown involve sending TPM2_Shutdown to TPM chip and disabling future TPM operations. TPM 1.2 behavior was different, future TPM operations weren't disabled, causing rare issues. This patch ensures that future TPM operations are disabled.

Fixes d1bd4a792d39 ("tpm: Issue a TPM2_Shutdown for TPM2 devices.")
[dianders: resolved merge conflicts with mainline]

---
This is the backport of the patch referenced above to 4.19 as was done in Chrome OS. See <https://crrev.com/c/1495114> for details. It presumably applies to some older kernels. NOTE that the problem itself has existed for a long time, but continuing to backport this exact solution to super old kernels is out of scope for me. For those truly interested feel free to reference the past discussion [1].

Quoted 4 lines

Reason for backport: mainline has commit a3fbfae82b4c ("tpm: take TPM chip power gating out of tpm_transmit()") and commit 719b7d81f204 ("tpm: introduce tpm_chip_start() and tpm_chip_stop()") and it didn't seem like a good idea to backport 17 patches to avoid the conflict.

/Jarkko