On Tue, Mar 18, 2025 at 11:29:26AM -0700, Breno Leitao wrote: > On Tue, Mar 18, 2025 at 05:34:55PM +0000, Mark Brown wrote: > > On Tue, Mar 18, 2025 at 10:02:47AM -0700, Breno Leitao wrote: > > > > > Makes sense. Another question, for platforms like this one that doesn't > > > have the device reset methods, what can we do to stop the bleed? > > > > > Basically every message that is sent to the SPI controller will fail, > > > which will trigger the device_reet() which is a no-op, but the device > > > will continue to be online. Should we disable the device after some > > > point? > > > > The SPI controller is only going to be doing something because some > > driver for an attached SPI device is trying to do something. Presumably > > whatever driver that is won't be having a good time and can hopefully > > figure something out, though given that SPI is simple and not > > hotpluggable this isn't really something that comes up a lot in > > production so I'd be unsurprised to see things just keep on retrying. > > I'd expect to see any substantial error handling in the driver for the > > device rather than in the controller. > > Good point. In my specific case, this is coming from tpm_tis, > which is not aware that the device is totally dead, and continues to ask > for random numbers: > > tegra_qspi_transfer_one_message > __spi_pump_transfer_message > __spi_sync > spi_sync > tpm_tis_spi_transfer > tpm_tis_spi_read_bytes > tpm_tis_request_locality > tpm_chip_start > tpm_try_get_ops > tpm_find_get_ops > tpm_get_random > tpm_hwrng_read > hwrng_fillfn > kthread > ret_from_fork > > Looking at tpm_tis, it seems it doesn't care if the the SPI is dead, and > just forward through the requests, which never complete. Adding Arnd to > see if he has any idea about this. > > Arnd, > > Summary of the proiblem: tpm_tis is trying to read random numbers > through a dead SPI controller. That causes infinite amounts of warnings > on the kernel, given that the controller is WARNing on time outs (which > is being fixed in one of the patches in this patchset). > > Question: Should tpm_tis be aware that the underneath SPI controller is > dead, and eventually get unplugged? Adding Arnd to the email.
