|Subject:||Re: USB3.0 XHCI support on GRUB2|
|Date:||Sun, 5 Mar 2017 20:41:53 +0530|
I am not using EFI. I am using coreboot with Intel FSP. Intel confirmed that XHCI controller is initialized by FSP and any standard XHCI driver should work on top of that.Problem Details:Requirement:My requirement is to load GRUB2 as payload from coreboot and Load Linux using GRUB2 (from USB pen drive and Hard Disk) based on grub menu selectionProblem I am facing:1. GRUB2 does not recognize USB pen drive, so can not boot linux from USB pen drive2. GRUB2 does not recognize USB keyboard so not able to select menu entryBoard: Intel Oxbohill CRB (Apollo lake)On Thu, Mar 2, 2017 at 9:29 AM, Andrei Borzenkov <address@hidden> wrote:02.03.2017 06:00, Gailu Singh пишет:
> Thanks for the update. I will try your patches on Apollo lake and report
> back the results. Intel strategy seems to be XHCI only option on newer
> boards so sooner or later xhci support is required in Grub. Thanks for your
> effort to take initiative.
On physical platforms grub normally relies on firmware to access
devices. What exact problem do you solve? Why do you need direct xHCI
access? What firmware this board has (I presume, EFI)?
> On Wed, Mar 1, 2017 at 4:07 PM, Bjørn Forsman <address@hidden>
>> On 1 March 2017 at 07:57, Andrei Borzenkov <address@hidden> wrote:
>>> Yes, Bjørn Forsman intended to work on implementation.
>> So now is apparently a good time to tell you about the status of that
>> implementation :-)
>> The good:
>> I ported the xHCI driver from Coreboot / libpayload to GRUB. In basic
>> tests, it works. (I tested a Lenovo G50-80 laptop with Intel Wildcat
>> Point-LP USB xHCI controller [8086:9cb1].)
>> The bad:
>> The code doesn't support HUBs and completely hangs on one xHCI
>> controller I tested (Intel NUC6i5SYH Skylake with Sunrise Point-LP
>> xHCI USB controller [8086:9d2f]).
>> The HUB problem is due to GRUB using a different code path for HUBs
>> than normal devices. I implemented hooks for "control_transfer" and
>> "bulk_transfer" in GRUB USB stack, because that required little
>> changes and it made it easy to hook it up with the xHCI driver. But
>> when GRUB sees a HUB, it uses the "old" setup_transfer code path which
>> isn't implemented in the xHCI driver.
>> I'm thinking the "hanging" issue (the controller never responds to a
>> basic "NOP" command) is a problem with the Coreboot driver itself, or
>> more specifically, the lack of a "quirk" handling for this controller.
>> But blaming that driver doesn't fix it.
>> About the implementation. I tried to change as little code as
>> possible, both in GRUB and the Coreboot driver, thinking that it'd
>> make maintenance easier. My reasoning was (perhaps wrongly?) that
>> there is little help to get in the community, and if I reuse enough of
>> the Coreboot driver, I can easiliy re-import the driver if/when they
>> fix some bugs. This "change as little as possible" policy means that
>> now the xHCI driver initializes and tracks some state that also GRUB
>> USB stack does. It works but it's not something I think is good enough
>> to integrate in mainline GRUB. (That's one of the reasons why I didn't
>> tell you about the status earlier. I don't see how to make it "clean"
>> enough for mainline. Also, the above issues are sort of blockers in
>> The ugly :-)
>> (I just cleaned up history and rebased on master, so it's less ugly
>> than it was a few hours ago.)
>> Best regards,
>> Bjørn Forsman
|[Prev in Thread]||Current Thread||[Next in Thread]|