On 01/29/19 11:54, Tomas Pilar (tpilar) wrote: >> Why TFTP client just pull one packet per second? I think it’s not >> correct and it could use the poll() function to accelerate the >> receive. Why you trying to solve a TFTP layer problem in SNP layer? >> It breaks the design principle of network layer model.
> Yes, I appreciate the principle. However, in practice we don't get to > sell adapters that do not PXE boot and it's pointless to argue with > customers that they have a rubbish TFTP client implementation. So we > put in workarounds into the driver. Actually, I think this is the core principle behind the UEFI forum, and the UEFI spec. You shouldn't have to implement bug compat hacks. The era when an add-on card would work on some platform's BIOS but not on another's should be left behind. You have a spec to point at, and the platform in question was likely certified against that spec in some form (possibly self-certified). Sp I think it would be reasonable to contact the platform vendor, and to direct your own customers to that ticket too. If you have a representative on the USWG, it might make sense to raise the issue there as well, especially if the issue is wide-spread and affects multiple platform vendors. The UEFI spec targets practical, common use cases, and this looks like one. (When a RH customer or partner reports e.g. a RHEL kernel issue to us, and we determine it is a problem in the firmware, we absolutely talk to the platform vendor, and sometimes to standards bodies too. We also advise customers on the applications that they run on RHEL, if they ask and/or care to listen. Plus, some high-profile applications and RHEL are explicitly certified against each other.) ... I don't mean to intrude of course; I'm sorry if I came through like that. Thanks Laszlo _______________________________________________ edk2-devel mailing list edk2-devel@lists.01.org https://lists.01.org/mailman/listinfo/edk2-devel