From: "Andrew Fish" <afish@apple.com>
To: edk2-devel-groups-io <devel@edk2.groups.io>,
Ard Biesheuvel <ard.biesheuvel@linaro.org>,
Mike Kinney <michael.d.kinney@intel.com>
Cc: leif@nuviainc.com, Ard Biesheuvel <ard.biesheuvel@arm.com>
Subject: Re: [edk2-devel] [PATCH edk2-platforms v2] Silicon/ChaosKeyDxe: don't rely on connect all controllers
Date: Tue, 02 Jun 2020 15:28:40 -0700 [thread overview]
Message-ID: <77023D58-E75D-48E8-B5CE-A51FA08FE9C7@apple.com> (raw)
In-Reply-To: <20200602133849.5422-1-ard.biesheuvel@linaro.org>
> On Jun 2, 2020, at 6:38 AM, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
>
> From: Ard Biesheuvel <ard.biesheuvel@arm.com>
>
> The ChaosKey driver implements the UEFI driver model, and so it is
> not guaranteed that any controllers will be attached to this driver
> unless it is connected explicitly. On many platforms today, this is
> taken care of by the ConnectAll() call that occurs in the BDS, but
> this is not something we should rely on.
>
> So add a protocol notification event that will attempt to connect
> the ChaosKey on each USB I/O protocol instance that carries the
> expected vendor and product IDs. This still relies on the USB host
> controllers to be connected by the platform, which is typically the
> case (given that a USB keyboard is required to interrupt the boot)
>
> On platforms where USB is not connected at all by default, it is really
> not up to a third party driver to meddle with this, and so relying on
> the USB host controllers to have been connected non-reursively is the
> best we can do. Note that third party drivers registered via Driver####
> can set a 'reconnect all' flag if needed to mitigate this.
>
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@arm.com>
> ---
>
> Unfortunately, the previous approach of connecting a short-form USB device
> path containing the ChaosKey's VID/PID turned out not to be reliable in
> practice (it doesn't work at all on ThunderX2 with AMI firmware)
>
> So instead, we have to rely on USB I/O protocols having been instantiated
> by the DXE core. This is what typically happens, given that USB keyboards
> are also exposed via USB I/O protocol instances. The only difference with
> a non-fast [ConnectAll()] boot is that those USB I/O protocol instances
> are not connected further automatically, but it is reasonable to expect
> that the handles themselves have been instantiated.
>
> Platforms that do not produce those USB I/O handles would not be able to
> offer the ability to interrupt the boot or enter the menu using a USB
> keyboard, so this is rather rare in practice. Also, if such platforms do
> exist, it is not up to this 3rd party driver to override this policy and
> enumerate the entire USB hierarchy.
>
> So this means we need to call UsbHwrngDriverBindingSupported() directly
> to check whether the USB I/O protocol in question has the right VID/PID.
> If that succeeds, there is really no point in using ConnectController(),
> so just call UsbHwrngDriverBindingStart() directly to connect the driver
> to the controller.
>
Actually there is a point in using gBS->ConnectController().
1) You make it impossible for the platform to override the drivers choice. The philosophy of EFI has alway been that the drivers should not cary policy and the platforms should cary policy. This inverts that.
2) I’m not sure it is well defined behavior by UEFI to NOT call gBS->ConnectController(). I took a quick look at the DXE Core and the gBS->ConnectController() is not doing any book keeping, other than managing the precedence rules, but it seems like it would be legal to add something in the future. + Mike in case I’m off on this one.
Also It would like be a good idea to have a PCD to turn on/off this feature.
Thanks,
Andrew Fish
> Tested on ThunderX2 using a Driver0000 option.
>
> Silicon/Openmoko/ChaosKeyDxe/DriverBinding.c | 66 ++++++++++++++++++++
> 1 file changed, 66 insertions(+)
>
> diff --git a/Silicon/Openmoko/ChaosKeyDxe/DriverBinding.c b/Silicon/Openmoko/ChaosKeyDxe/DriverBinding.c
> index e7d0d3fe563e..611803c6c339 100644
> --- a/Silicon/Openmoko/ChaosKeyDxe/DriverBinding.c
> +++ b/Silicon/Openmoko/ChaosKeyDxe/DriverBinding.c
> @@ -11,6 +11,9 @@
>
> #include "ChaosKeyDriver.h"
>
> +STATIC VOID *mProtocolNotifyRegistration;
> +STATIC EFI_EVENT mProtocolNotifyRegistrationEvent;
> +
> /**
> Tests to see if this driver supports a given controller.
>
> @@ -157,6 +160,55 @@ EFI_DRIVER_BINDING_PROTOCOL gUsbDriverBinding = {
> };
>
>
> +STATIC
> +VOID
> +EFIAPI
> +UsbHwrngOnProtocolNotify (
> + IN EFI_EVENT Event,
> + IN VOID *Context
> + )
> +{
> + EFI_STATUS Status;
> + EFI_HANDLE *Handles;
> + UINTN HandleCount;
> + UINTN Index;
> +
> + do {
> + Status = gBS->LocateHandleBuffer (ByRegisterNotify, NULL,
> + mProtocolNotifyRegistration, &HandleCount, &Handles);
> + if (EFI_ERROR (Status)) {
> + if (Status != EFI_NOT_FOUND) {
> + DEBUG ((DEBUG_WARN, "%a: LocateHandleBuffer() failed - %r\n",
> + __FUNCTION__, Status));
> + }
> + return;
> + }
> +
> + if (HandleCount == 0) {
> + return;
> + }
> +
> + for (Index = 0; Index < HandleCount; Index++) {
> + Status = UsbHwrngDriverBindingSupported (&gUsbDriverBinding,
> + Handles[Index], NULL);
> + if (EFI_ERROR (Status)) {
> + continue;
> + }
> +
> + //
> + // Attempt to connect the USB device path describing the ChaosKey
> + // hardware via the handle describing a USB host controller.
> + //
> + Status = UsbHwrngDriverBindingStart (&gUsbDriverBinding,
> + Handles[Index], NULL);
> + DEBUG ((DEBUG_VERBOSE, "%a: UsbHwrngDriverBindingStart () returned %r\n",
> + __FUNCTION__, Status));
> + }
> + gBS->FreePool (Handles);
> + } while (1);
> +}
> +
> +
> /**
> The entry point of ChaosKey UEFI Driver.
>
> @@ -185,6 +237,18 @@ EntryPoint (
> NULL, &gChaosKeyDriverComponentName2);
> ASSERT_EFI_ERROR (Status);
>
> + //
> + // This driver produces the EFI Random Number Generator protocol on
> + // compatible USB I/O handles, which is not a protocol that can provide
> + // a boot target. This means that it will not get connected on an ordinary
> + // 'fast' boot (which only connects the console and boot entry device paths)
> + // unless we take extra measures.
> + //
> + mProtocolNotifyRegistrationEvent = EfiCreateProtocolNotifyEvent (
> + &gEfiUsbIoProtocolGuid, TPL_CALLBACK,
> + UsbHwrngOnProtocolNotify, NULL,
> + &mProtocolNotifyRegistration);
> +
> DEBUG ((DEBUG_INIT | DEBUG_INFO, "*** Installed ChaosKey driver! ***\n"));
>
> return EFI_SUCCESS;
> @@ -211,6 +275,8 @@ UnloadImage (
> UINTN HandleCount;
> UINTN Index;
>
> + gBS->CloseEvent (mProtocolNotifyRegistrationEvent);
> +
> //
> // Retrieve all USB I/O handles in the handle database
> //
> --
> 2.20.1
>
>
>
>
next prev parent reply other threads:[~2020-06-02 22:28 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-06-02 13:38 [PATCH edk2-platforms v2] Silicon/ChaosKeyDxe: don't rely on connect all controllers Ard Biesheuvel
2020-06-02 14:49 ` Leif Lindholm
2020-06-02 14:58 ` Ard Biesheuvel
2020-06-02 21:39 ` Leif Lindholm
2020-06-02 22:28 ` Andrew Fish [this message]
2020-06-03 6:32 ` [edk2-devel] " Ard Biesheuvel
2020-06-03 15:47 ` Andrew Fish
2020-06-03 16:04 ` Ard Biesheuvel
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-list from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=77023D58-E75D-48E8-B5CE-A51FA08FE9C7@apple.com \
--to=devel@edk2.groups.io \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox