[PATCH 1/1] efi_loader: stop watchdogs in ExitBootServices()

[Tom Rini]

On Tue, Jan 31, 2023 at 02:03:10PM +0200, Ilias Apalodimas wrote:
> Hi all,
> 
> On Mon, Jan 30, 2023 at 01:30:49PM -0500, Tom Rini wrote:
> > On Mon, Jan 30, 2023 at 01:13:55PM -0500, Tom Rini wrote:
> > > On Sat, Jan 28, 2023 at 09:57:45AM +0100, Heinrich Schuchardt wrote:
> > >
> > > > The UEFI specification requires for ExitBootServices() that "the boot
> > > > services watchdog timer is disabled". We already disable the software
> > > > watchdog. We should additionally disable the hardware watchdogs.
> > > >
> > > > Reported-by: Andre Przywara <andre.przywara at arm.com>
> > > > Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>
> > > > ---
> > > >  lib/efi_loader/efi_boottime.c | 10 ++++++----
> > > >  1 file changed, 6 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/lib/efi_loader/efi_boottime.c b/lib/efi_loader/efi_boottime.c
> > > > index ba28989f36..71215af9d2 100644
> > > > --- a/lib/efi_loader/efi_boottime.c
> > > > +++ b/lib/efi_loader/efi_boottime.c
> > > > @@ -19,6 +19,7 @@
> > > >  #include <u-boot/crc.h>
> > > >  #include <usb.h>
> > > >  #include <watchdog.h>
> > > > +#include <wdt.h>
> > > >  #include <asm/global_data.h>
> > > >  #include <asm/setjmp.h>
> > > >  #include <linux/libfdt_env.h>
> > > > @@ -2171,6 +2172,11 @@ static efi_status_t EFIAPI efi_exit_boot_services(efi_handle_t image_handle,
> > > >  			list_del(&evt->link);
> > > >  	}
> > > >
> > > > +	/* Disable watchdogs */
> > > > +	efi_set_watchdog(0);
> > > > +	if IS_ENABLED(CONFIG_WDT)
> > > > +		wdt_stop_all();
> > > > +
> > > >  	if (!efi_st_keep_devices) {
> > > >  		bootm_disable_interrupts();
> > > >  		if (IS_ENABLED(CONFIG_USB_DEVICE))
> > > > @@ -2196,10 +2202,6 @@ static efi_status_t EFIAPI efi_exit_boot_services(efi_handle_t image_handle,
> > > >
> > > >  	/* Recalculate CRC32 */
> > > >  	efi_update_table_header_crc32(&systab.hdr);
> > > > -
> > > > -	/* Give the payload some time to boot */
> > > > -	efi_set_watchdog(0);
> > > > -	schedule();
> > > >  out:
> > > >  	if (IS_ENABLED(CONFIG_EFI_TCG2_PROTOCOL)) {
> > > >  		if (ret != EFI_SUCCESS)
> > >
> > > I thought we had rejected going down this path since the UEFI spec is
> > > unhelpfully wrong if it insists this?
> >
> > Because, to be clear, stopping hardware watchdogs is not to be done. The
> > one in-tree caller of wdt_stop_all is very questionable. You cannot
> > seriously stop a watchdog until someone else can hopefully resume it as
> > that violates the function of a hardware watchdog. A pure software
> > watchdog is one thing, and a hardware watchdog is another. I feel like
> > the most likely answer here is that someone needs to, still, push back
> > to the UEFI specification to get hardware watchdogs better understood
> > and handled, as it must never be stopped once started and if you cannot
> > reach the next stage in time, that's an engineering issue to resolve. My
> > first guess is that ExitBootServices should service the watchdog one
> > last time to ensure the largest window of time for the OS to take over
> > servicing of the watchdog.
> >
> 
> There's two scenarios I can think of
> 1. After U-Boot is done it can disable the hardware watchdog.
>    The kernel will go through the EFI-stub -> kernel proper -> watchdog
>    gets re-initialized.  In that case you are *hoping* that device won't
>    hang in the efi-stub or until the wd is up again.
> 2. EFI makes sure the hardware wd gets configured with the highest allowed
>    value.  The efi-stub doesn't have any driver to refresh the wd, so we
>    will again rely on the wd driver coming up and refreshing the timers.

You cannot stop the hardware watchdog, period. I think in the previous
thread about this it was noted that some hardware watchdogs cannot be
disabled, it's not function that the watchdog supports. Someone needs to
go and talk with the UEFI specification people and get this addressed.
There is no sane path for "disable the hardware watchdog".

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20230131/5365f88b/attachment.sig>