From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by mx.groups.io with SMTP id smtpd.web11.19743.1590174216035109220 for ; Fri, 22 May 2020 12:03:36 -0700 Authentication-Results: mx.groups.io; dkim=missing; spf=pass (domain: arm.com, ip: 217.140.110.172, mailfrom: ard.biesheuvel@arm.com) Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id EE50A55D; Fri, 22 May 2020 12:03:33 -0700 (PDT) Received: from [192.168.1.81] (unknown [10.37.8.117]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 42FC33F68F; Fri, 22 May 2020 12:03:32 -0700 (PDT) Subject: Re: [edk2-devel] [PATCH v2] ArmPkg/PlatformBootManagerLib: connect non-discoverable USB hosts To: devel@edk2.groups.io, lersek@redhat.com Cc: liming.gao@intel.com, leif@nuviainc.com, philmd@redhat.com References: <20200522084006.32006-1-ard.biesheuvel@arm.com> From: "Ard Biesheuvel" Message-ID: Date: Fri, 22 May 2020 21:03:29 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.7.0 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8; format=flowed Content-Language: en-US Content-Transfer-Encoding: 7bit On 5/22/20 8:57 PM, Laszlo Ersek via groups.io wrote: > On 05/22/20 10:40, Ard Biesheuvel wrote: >> The way the BDS handles the short-form USB device path of the console >> keyboard relies on USB host controllers to be locatable via their PCI >> metadata, which implies that these controllers already have a PCI I/O >> protocol installed on their handle. >> >> This is not the case for non-discoverable USB host controllers that are >> supported by the NonDiscoverable PCI device driver. These controllers >> must be connected first, or the BDS will never notice their existence, >> and will not enable any USB keyboards connected through them. >> >> Let's work around this by connecting these handles explicitly. This is >> a bit of a stopgap, but it is the cleanest way of dealing with this >> without violating the UEFI driver model entirely. This ensures that >> platforms that do not rely on ConnectAll() will keep working as >> expected. >> >> Signed-off-by: Ard Biesheuvel >> --- >> >> This is tagged as v2 since it is a followup to >> >> 'MdeModulePkg/NonDiscoverablePciDeviceDxe: connect handles on registration' >> >> that I sent out yesterday. Laslzo made the point that it would be >> unfortunate to add yet another hack that defeats the UEFI driver model >> entirely, so this time, the change is in the BDS where it belongs. >> >> Note that only USB console devices are affected by this: short form >> block IO device paths used for booting are expanded and cached in a >> "HDDP" variable, and so they work as expected with fast boot without >> connect all. >> >> ArmPkg/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf | 4 ++ >> ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c | 42 ++++++++++++++++++++ >> 2 files changed, 46 insertions(+) >> >> diff --git a/ArmPkg/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf b/ArmPkg/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf >> index 87c9b1150c54..2f726d117d7d 100644 >> --- a/ArmPkg/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf >> +++ b/ArmPkg/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf >> @@ -66,6 +66,9 @@ [Pcd] >> gEfiMdePkgTokenSpaceGuid.PcdPlatformBootTimeOut >> >> [Guids] >> + gEdkiiNonDiscoverableEhciDeviceGuid >> + gEdkiiNonDiscoverableUhciDeviceGuid >> + gEdkiiNonDiscoverableXhciDeviceGuid >> gEfiFileInfoGuid >> gEfiFileSystemInfoGuid >> gEfiFileSystemVolumeLabelInfoIdGuid >> @@ -74,6 +77,7 @@ [Guids] >> gUefiShellFileGuid >> >> [Protocols] >> + gEdkiiNonDiscoverableDeviceProtocolGuid >> gEfiDevicePathProtocolGuid >> gEfiGraphicsOutputProtocolGuid >> gEfiLoadedImageProtocolGuid >> diff --git a/ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c b/ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c >> index 3411219fbfdb..4aca1382b042 100644 >> --- a/ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c >> +++ b/ArmPkg/Library/PlatformBootManagerLib/PlatformBm.c >> @@ -23,10 +23,12 @@ >> #include >> #include >> #include >> +#include >> #include >> #include >> #include >> #include >> +#include >> #include >> #include >> >> @@ -254,6 +256,37 @@ IsPciDisplay ( >> } >> >> >> +/** >> + This FILTER_FUNCTION checks if a handle corresponds to a non-discoverable >> + USB host controller. >> +**/ >> +STATIC >> +BOOLEAN >> +EFIAPI >> +IsUsbHost ( >> + IN EFI_HANDLE Handle, >> + IN CONST CHAR16 *ReportText >> + ) >> +{ >> + NON_DISCOVERABLE_DEVICE *Device; >> + EFI_STATUS Status; >> + >> + Status = gBS->HandleProtocol (Handle, >> + &gEdkiiNonDiscoverableDeviceProtocolGuid, >> + (VOID **)&Device); >> + if (EFI_ERROR (Status)) { >> + return FALSE; >> + } >> + >> + if (CompareGuid (Device->Type, &gEdkiiNonDiscoverableUhciDeviceGuid) || >> + CompareGuid (Device->Type, &gEdkiiNonDiscoverableEhciDeviceGuid) || >> + CompareGuid (Device->Type, &gEdkiiNonDiscoverableXhciDeviceGuid)) { >> + return TRUE; >> + } >> + return FALSE; >> +} >> + >> + >> /** >> This CALLBACK_FUNCTION attempts to connect a handle non-recursively, asking >> the matching driver to produce all first-level child handles. >> @@ -574,6 +607,15 @@ PlatformBootManagerBeforeConsole ( >> // >> FilterAndProcess (&gEfiGraphicsOutputProtocolGuid, NULL, AddOutput); >> >> + // >> + // The core BDS code connects short-form USB device paths by explicitly >> + // looking for handles with PCI I/O installed, and checking the PCI class >> + // code whether it matches the one for a USB host controller. This means >> + // non-discoverable USB host controllers need to have the non-discoverable >> + // PCI driver attached first. >> + // >> + FilterAndProcess (&gEdkiiNonDiscoverableDeviceProtocolGuid, IsUsbHost, Connect); >> + >> // >> // Add the hardcoded short-form USB keyboard device path to ConIn. >> // >> > > Awesome! So we don't mind the full proto stack on these handles, as long > as "these handles" means exactly the small set we want. > > Reviewed-by: Laszlo Ersek > Indeed. One thing I could not figure out is whether we should pass the 'end device path' as remaining device path to ConnectController, to prevent the USB bus driver from instantiating any new handles, and only attach to the one that has the PCI I/O and USB host controller protocols installed. However, in order for the BDS to be able to connect any USB keyboards, it will have to enumerate the entire subordinate USB hierarchy anyway, so that does not really matter in practice.