Re: [PATCH] tpm: ibmvtpm: Avoid error message when process gets signal while waiting

[Stefan Berger <stefanb@linux.ibm.com>]

On 7/29/21 8:57 PM, Jarkko Sakkinen wrote:
> On Thu, Jul 29, 2021 at 09:39:18AM -0400, Stefan Berger wrote:
>> On 7/28/21 5:50 PM, Jarkko Sakkinen wrote:
>>> On Mon, Jul 26, 2021 at 11:00:51PM -0400, Stefan Berger wrote:
>>>> On 7/26/21 10:42 PM, Jarkko Sakkinen wrote:
>>>>> On Mon, Jul 12, 2021 at 12:25:05PM -0400, Stefan Berger wrote:
>>>>>> From: Stefan Berger <stefanb@linux.ibm.com>
>>>>>>
>>>>>> When rngd is run as root then lots of these types of message will appear
>>>>>> in the kernel log if the TPM has been configure to provide random bytes:
>>>>>>
>>>>>> [ 7406.275163] tpm tpm0: tpm_transmit: tpm_recv: error -4
>>>>>>
>>>>>> The issue is caused by the following call that is interrupted while
>>>>>> waiting for the TPM's response.
>>>>>>
>>>>>> sig = wait_event_interruptible(ibmvtpm->wq,
>>>>>>                                   !ibmvtpm->tpm_processing_cmd);
>>>>>>
>>>>>> The solution is to use wait_event() instead.
>>>>> Why?
>>>> So it becomes uninterruptible and these error messages go away.
>>> We do not want to make a process uninterruptible. That would prevent
>>> killing it.
>> I guess we'll have to go back to this one then:
>> https://www.spinics.net/lists/linux-integrity/msg16741.html
> Makes a heck lot more sense.
>
> There's a typo in the commit message: PM_STATUS_BUSY
>
> Also the commit message lacks explanation of this change completely:
>
> @@ -690,8 +688,15 @@ static int tpm_ibmvtpm_probe(struct vio_dev *vio_dev,
>   		goto init_irq_cleanup;
>   	}
>   
> -	if (!strcmp(id->compat, "IBM,vtpm20")) {
> +
> +	if (!strcmp(id->compat, "IBM,vtpm20"))
>   		chip->flags |= TPM_CHIP_FLAG_TPM2;
> +
> +	rc = tpm_get_timeouts(chip);
> +	if (rc)
> +		goto init_irq_cleanup;
> +
> +	if (chip->flags & TPM_CHIP_FLAG_TPM2) {
>   		rc = tpm2_get_cc_attrs_tbl(chip);
>
> The last paragraph should be rewritten in imperative form.


will fix.

>
> Finally, you could simplify the fix by simply changing the type of
> tpm_processing_cmd to u8, and just set it to 'true' and 'false',
> which will set the first bit.

Are you sure? It's a bit mask we are using this with. Using 'true' for 
these type of operations doesn't sound right.

         u8 status = chip->ops->status(chip);
         if ((status & chip->ops->req_complete_mask) ==
             chip->ops->req_complete_val)

             goto out_recv;

https://elixir.bootlin.com/linux/latest/source/drivers/char/tpm/tpm-interface.c#L108

@@ -457,7 +455,7 @@ static const struct tpm_class_ops tpm_ibmvtpm = {
  	.send = tpm_ibmvtpm_send,
  	.cancel = tpm_ibmvtpm_cancel,
  	.status = tpm_ibmvtpm_status,
-	.req_complete_mask = 0,
+	.req_complete_mask = TPM_STATUS_BUSY,
  	.req_complete_val = 0,
  	.req_canceled = tpm_ibmvtpm_req_canceled,
  };


    Stefan

>
> /Jarkko