[PATCH] tpm: interrupt clear fix

[PATCH] tpm: interrupt clear fix

[Kylene Jo Hall]

Under stress testing I found that the interrupt is not always cleared.
This is a bug and this patch should go into 2.6.18 and 2.6.17.x.

Signed-off-by: Kylene Hall <kjhall@us.ibm.com>
---
--- tcg/tpmdd/drivers/char/tpm/tpm_tis.c	2006-06-07 11:37:08.000000000 -0700
+++ linux-2.6.17/drivers/char/tpm/tpm_tis.c	2006-07-10 12:58:28.000000000 -0700
@@ -424,6 +424,7 @@ static irqreturn_t tis_int_handler(int i
 	iowrite32(interrupt,
 		  chip->vendor.iobase +
 		  TPM_INT_STATUS(chip->vendor.locality));
+	mb();
 	return IRQ_HANDLED;
 }
 

Re: [PATCH] tpm: interrupt clear fix

[Alan Cox]

Ar Mer, 2006-07-12 am 14:01 -0700, ysgrifennodd Kylene Jo Hall:
> Under stress testing I found that the interrupt is not always cleared.
> This is a bug and this patch should go into 2.6.18 and 2.6.17.x.
> 
> Signed-off-by: Kylene Hall <kjhall@us.ibm.com>
> ---
> --- tcg/tpmdd/drivers/char/tpm/tpm_tis.c	2006-06-07 11:37:08.000000000 -0700
> +++ linux-2.6.17/drivers/char/tpm/tpm_tis.c	2006-07-10 12:58:28.000000000 -0700
> @@ -424,6 +424,7 @@ static irqreturn_t tis_int_handler(int i
>  	iowrite32(interrupt,
>  		  chip->vendor.iobase +
>  		  TPM_INT_STATUS(chip->vendor.locality));
> +	mb();

NAK

This looks incorrect.

The iowrite32 will be posted for mmio. mb() may half hide the problem on
some processors but you should read from the same PCI device as per the
PCI spec.

Re: [PATCH] tpm: interrupt clear fix

[linux-os (Dick Johnson)]

On Wed, 12 Jul 2006, Kylene Jo Hall wrote:

> Under stress testing I found that the interrupt is not always cleared.
> This is a bug and this patch should go into 2.6.18 and 2.6.17.x.
>
> Signed-off-by: Kylene Hall <kjhall@us.ibm.com>
> ---
> --- tcg/tpmdd/drivers/char/tpm/tpm_tis.c	2006-06-07 11:37:08.000000000 -0700
> +++ linux-2.6.17/drivers/char/tpm/tpm_tis.c	2006-07-10 12:58:28.000000000 -0700
> @@ -424,6 +424,7 @@ static irqreturn_t tis_int_handler(int i
> 	iowrite32(interrupt,
> 		  chip->vendor.iobase +
> 		  TPM_INT_STATUS(chip->vendor.locality));
> +	mb();
> 	return IRQ_HANDLED;
> }

PCI devices need a final read to flush all pending writes. Whatever
mb() does, just hides the problem.

Cheers,
Dick Johnson
Penguin : Linux version 2.6.16.24 on an i686 machine (5592.63 BogoMips).
New book: http://www.AbominableFirebug.com/
_
\x1a\x04

****************************************************************
The information transmitted in this message is confidential and may be privileged.  Any review, retransmission, dissemination, or other use of this information by persons or entities other than the intended recipient is prohibited.  If you are not the intended recipient, please notify Analogic Corporation immediately - by replying to this message or by sending an email to DeliveryErrors@analogic.com - and destroy all copies of this information, including any attachments, without reading or disclosing them.

Thank you.

Re: [PATCH] tpm: interrupt clear fix

[Kylene Jo Hall]

Under stress testing I found that the interrupt is not always cleared.
This is a bug and this patch should go into 2.6.18 and 2.6.17.x.

On Thu, 2006-07-13 at 07:45 -0400, linux-os (Dick Johnson) wrote:

> PCI devices need a final read to flush all pending writes. Whatever
> mb() does, just hides the problem.


Signed-off-by: Kylene Hall <kjhall@us.ibm.com>
---

--- linux-2.6.18-rc1/drivers/char/tpm/tpm_tis.c	2006-07-13 14:46:39.727500500 -0500
+++ linux-2.6.18-rc1-tpm/drivers/char/tpm/tpm_tis.c	2006-07-13 14:47:33.878884750 -0500
@@ -424,6 +424,7 @@ static irqreturn_t tis_int_handler(int i
 	iowrite32(interrupt,
 		  chip->vendor.iobase +
 		  TPM_INT_STATUS(chip->vendor.locality));
+	ioread32(chip->vendor.iobase + TPM_INT_STATUS(chip->vendor.locality));
 	return IRQ_HANDLED;
 }