Re: [PATCH] tpm: Fix the timeout & use ktime

[Orlov, Ivan]

On 11/06/2025 18:02, Jarkko Sakkinen wrote:
> > Instead, perform the check in the following way:
> >
> > 1. Read the current timestamp
> > 2. Read the completion status. If completed, return the result
> > 3. Sleep
> > 4. Check if the timestamp read at step 1 exceeds the timeout. Return
> >     an error if it does
> > 5. Goto 1
> >
> > Also, use ktime instead of jiffes as a more reliable and precise timing
> > source.
>
> "also", i.e. a logically separate change which should be split up to
> a separate patch.

Got it, will send this enhancement as a separate patch.

> > +		curr_time = ktime_get();
> >   		u8 status = tpm_chip_status(chip);
> >   		if ((status & chip->ops->req_complete_mask) ==
> >   		    chip->ops->req_complete_val)
> > @@ -140,7 +149,7 @@ static ssize_t tpm_try_transmit(struct tpm_chip *chip, void *buf, size_t bufsiz)
> >   
> >   		tpm_msleep(TPM_TIMEOUT_POLL);
> >   		rmb();
> > -	} while (time_before(jiffies, stop));
> > +	} while (ktime_before(curr_time, timeout));
>
>
> Wouldn't it be simpler fix to just check completion after dropping out
> of the loop?

yes, this should also solve the problem without taking additional space 
on the stack with new vars. Will update in V2, thanks!

> And declare this before tpm_try_transmit():
>
> static bool tpm_transmit_completed(struct tpm_chip *chip)
> {
> 	u8 status = tpm_chip_status(chip);
> 	
> 	return (status & chip->ops->req_complete_mask) == chip->ops->req_complete_val;
> }

Cool, will do.

Thanks for the review and sorry for the late reply.

--
Kind regards,
Ivan Orlov