From: Bjorn H. <bjo...@hp...> - 2006-07-26 12:20:38
|
> Yes ideally that would be the case. But many hardware vendors are not > supporting ACPI BIOS entries for the TPM and this was not a requirement > for 1.1 devices. ACPI should describe all motherboard devices that consume resources or need a driver. Are you saying there's some exception to this for TPM 1.1 devices? Why would TPM be different from other motherboard devices in this regard? > As you will notice the tpm_tis driver now has away > around the PNP stuff for machines that don't have the ACPI entry becaus= e > there are just too many cases that it doesn't work for. Yes, I noticed this (and even mentioned it below). > If you want to > add similar support to atmel and nsc where the pnp is attempted first > that might work (though for legacy reasons I don't want to make the non= - > PNP version for those drivers now require a module parameter as is the > case in tis. We can probably get away with blindly probing on x86. But these drivers shouldn't be x86-specific, and it's not safe to blindly probe on some other architectures (ia64, for instance). Do you know what PNP IDs the atmel and nsc devices would use? > On Tue, 2006-07-25 at 16:49 -0600, Bjorn Helgaas wrote: >> tpm_infineon.c and tpm_tis.c use pnp_register_driver() to bind to >> devices with matching PNP IDs. That seems like the right way to >> do it. >> >> But tpm_atmel.c and tpm_nsc.c just blindly poke around looking for >> their device. That seems wrong. Is there some reason they can't >> use pnp_register_driver() also? I'd be glad to send a patch to >> fix this if somebody knows the appropriate list of PNP IDs that >> these drivers should claim. >> >> I like the recent tpm_tis.c changes to give a hook for use when >> ACPI doesn't correctly describe the TPM device. That will give >> manufacturers a hint that they're doing something wrong. >> >> Bjorn > > |