On Fri, 2019-09-06 at 14:37 +0200, Jan Lübbe wrote:

This is due to the SPI accesses performed by tis_int_handler (which
will sleep). Switching to devm_request_threaded_irq fixes this and
leads to a successful IRQ probe.

Aah, right through tpm_tis_read32/write32(). This is definitely a new
regression. Thanks for reporting this! This was completely missed when
the support for other than TCG MMIO was implemented for the TIS driver.

This should have a patch of it own with your reported-by unless you
care to send bug fix for it.

But: It seems that the IRQ is not acked correctly, as the interrupt
line stays low. I suspect this is because the tpm_chip_stop from
http://git.infradead.org/users/jjs/linux-tpmdd.git/commitdiff/9b558deab2c5d7dc23d5f7a4064892ede482ad32
happens before the threaded handler runs. I'm currently unable to
verify that though, as my build machine's disk just died. :/

I cannot recall all the nyances related to interrupt probing but with a
quick look I wonder why it does not utilize int_queue. Then
tpm_chip_stop() could be done synchronously.

Regards,
Jan

/Jarkko