Dear Timothy,
Thank you for your reply.
Am Sonntag, den 11.12.2016, 15:54 -0600 schrieb Timothy Pearson:
Several devices using the Intel 945 chipset copied code for PCI reset, costing 200 ms of boot time.
/* Force PCIRST# */ pci_write_config16(PCI_DEV(0, 0x1e, 0), BCTRL, SBR); udelay(200 * 1000); pci_write_config16(PCI_DEV(0, 0x1e, 0), BCTRL, 0);The change-set Ia37d9f0ecf5655531616edb20b53757d5d47b42f [1] removes that code from the Lenovo X60.
That code was added for some crypto card on a Roda device.
My question is, if removing that code is fine, or if it should be left in and be made configurable (Kconfig/NVRAM)?
Are there often cases where there are extensions card with problems, that need such a PCI reset?
We have run into this issue on the KGPE-D16 and LSI SAS controllers. However, in this case it's not so much the reset itself as it is the time it takes for the card to start up and become ready for PCI scan.
We handled this with a devicetree.cb option to set a delay between reset and PCI scan. Perhaps something similar could be used in this instance?
I guess you are referring to the following line in `src/mainboard/asus/kgpe-d16/devicetree.cb`?
``` register "pcie_settling_time" = "1000000" # Allow PIKE to be detected / configured ```
This option is only available for that particular southbridge, AMD SR5650, right?
In the Lenovo X60 case, it’d probably would be better to have it as a run time option in NVRAM, so that it can be configured depending if such a problematic card is used or not.
Would it be feasible to implement something like that in `src/device`, so that all devices can use it?
Or is that overkill, and such pluggable(?), that means external, cards are rare, and the code should just be removed for the Lenovo X60?
Thanks,
Paul