| Message ID | 1419776431-26217-1-git-send-email-christoffer.dall@linaro.org |
|---|---|
| State | Accepted |
| Commit | d6ad36913083d683aad4e02e53580c995f1a6ede |
| Headers | show |
On Sun, Dec 28, 2014 at 09:46:20PM +0000, Marc Zyngier wrote: > On 2014-12-28 14:20, Christoffer Dall wrote: > >Commit 0b46b8a718c6e ("clocksource: arch_timer: Fix code to...") > >fixes > >timer issues with certain ARMv7 platforms, but unfortunately breaks > >arm64 platforms with hyp mode (EL2) enabled. > > > >The commit only sets arch_timer_use_virtual to false under > >CONFIG_ARM, > >but forgets that the config variable is also set in other code paths > >(actually, right underneath the check in the patch) with detrimental > >consequences as we've now introduced a direct early call to BUG() on > >practically all arm64 platforms. > > > >One could argue that this code could be refactored to use different > >variables for checking which *timer* to use and which *counter* to > >use, > >which seems to be the desired difference between ARM and arm64 in > >this > >case, but this approach fixes an urgent issue for now. > > > >Cc: Sonny Rao <sonnyrao@chromium.org> > >Cc: Catalin Marinas <catalin.marinas@arm.com> > >Cc: Daniel Lezcano <daniel.lezcano@linaro.org> > >Cc: Olof Johansson <olof@lixom.net> > >Cc: Mark Rutland <mark.rutland@arm.com> > >Cc: Catalin Marinas <catalin.marinas@arm.com> > >Cc: Marc Zyngier <marc.zyngier@arm.com> > >Cc: Yingjoe Chen <yingjoe.chen@mediatek.com> > >Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org> > >--- > >This was apparently already discovered by Yingjoe Chen in this thread > >https://lkml.org/lkml/2014/11/24/41 and Catalin recommended a similar > >fix. > > I'm increasingly worried about the time it takes to get such critical > fixes into the tree (arm64 is *dead* without it). > > What is holding this patch which, as far as I remember, has been posted > by Catalin almost three weeks ago? > I didn't find that since I didn't think I'd have to go back that long on lakml for something that breaks boot of an entire architecture. Sorry for the confusion of a second patch, but fwiw, I now spent another few hours bisecting this, so I would really like to see this fix go into mainline ASAP as well to save others the trouble. Thanks, -Christoffer -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On 12/29/2014 08:13 PM, Mark Rutland wrote: > [Adding Arnd] > > On Mon, Dec 29, 2014 at 08:35:09AM +0000, Christoffer Dall wrote: >> On Sun, Dec 28, 2014 at 09:46:20PM +0000, Marc Zyngier wrote: >>> On 2014-12-28 14:20, Christoffer Dall wrote: >>>> Commit 0b46b8a718c6e ("clocksource: arch_timer: Fix code to...") >>>> fixes >>>> timer issues with certain ARMv7 platforms, but unfortunately breaks >>>> arm64 platforms with hyp mode (EL2) enabled. >>>> >>>> The commit only sets arch_timer_use_virtual to false under >>>> CONFIG_ARM, >>>> but forgets that the config variable is also set in other code paths >>>> (actually, right underneath the check in the patch) with detrimental >>>> consequences as we've now introduced a direct early call to BUG() on >>>> practically all arm64 platforms. >>>> >>>> One could argue that this code could be refactored to use different >>>> variables for checking which *timer* to use and which *counter* to >>>> use, >>>> which seems to be the desired difference between ARM and arm64 in >>>> this >>>> case, but this approach fixes an urgent issue for now. >>>> >>>> Cc: Sonny Rao <sonnyrao@chromium.org> >>>> Cc: Catalin Marinas <catalin.marinas@arm.com> >>>> Cc: Daniel Lezcano <daniel.lezcano@linaro.org> >>>> Cc: Olof Johansson <olof@lixom.net> >>>> Cc: Mark Rutland <mark.rutland@arm.com> >>>> Cc: Catalin Marinas <catalin.marinas@arm.com> >>>> Cc: Marc Zyngier <marc.zyngier@arm.com> >>>> Cc: Yingjoe Chen <yingjoe.chen@mediatek.com> >>>> Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org> >>>> --- >>>> This was apparently already discovered by Yingjoe Chen in this thread >>>> https://lkml.org/lkml/2014/11/24/41 and Catalin recommended a similar >>>> fix. >>> >>> I'm increasingly worried about the time it takes to get such critical >>> fixes into the tree (arm64 is *dead* without it). >>> >>> What is holding this patch which, as far as I remember, has been posted >>> by Catalin almost three weeks ago? >>> >> I didn't find that since I didn't think I'd have to go back that long on >> lakml for something that breaks boot of an entire architecture. Sorry >> for the confusion of a second patch, but fwiw, I now spent another few >> hours bisecting this, so I would really like to see this fix go into >> mainline ASAP as well to save others the trouble. > > Last I knew, Arnd was going to take the fix [1], which has been in > arm-soc's for-next and fixes branches for almost two weeks now. It > didn't make it into the last pull req due to some confusion over who was > going to take it. I asked arm-soc team to take the fix because it was depending on a patch which was already in their tree [1]. Merry Christmas -- Daniel [1] http://www.spinics.net/lists/arm-kernel/msg385704.html > Arnd, what's the plan for getting this into mainline ASAP? > > Mark. > > [1] https://git.kernel.org/cgit/linux/kernel/git/arm/arm-soc.git/commit/?h=fixes&id=d6ad36913083d683aad4e02e53580c995f1a6ede >
diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c index 6a79fc4..095c177 100644 --- a/drivers/clocksource/arm_arch_timer.c +++ b/drivers/clocksource/arm_arch_timer.c @@ -462,7 +462,7 @@ static void __init arch_counter_register(unsigned type) /* Register the CP15 based counter if we have one */ if (type & ARCH_CP15_TIMER) { - if (arch_timer_use_virtual) + if (IS_ENABLED(CONFIG_ARM64) || arch_timer_use_virtual) arch_timer_read_counter = arch_counter_get_cntvct; else arch_timer_read_counter = arch_counter_get_cntpct;
Commit 0b46b8a718c6e ("clocksource: arch_timer: Fix code to...") fixes timer issues with certain ARMv7 platforms, but unfortunately breaks arm64 platforms with hyp mode (EL2) enabled. The commit only sets arch_timer_use_virtual to false under CONFIG_ARM, but forgets that the config variable is also set in other code paths (actually, right underneath the check in the patch) with detrimental consequences as we've now introduced a direct early call to BUG() on practically all arm64 platforms. One could argue that this code could be refactored to use different variables for checking which *timer* to use and which *counter* to use, which seems to be the desired difference between ARM and arm64 in this case, but this approach fixes an urgent issue for now. Cc: Sonny Rao <sonnyrao@chromium.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Daniel Lezcano <daniel.lezcano@linaro.org> Cc: Olof Johansson <olof@lixom.net> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Marc Zyngier <marc.zyngier@arm.com> Cc: Yingjoe Chen <yingjoe.chen@mediatek.com> Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org> --- This was apparently already discovered by Yingjoe Chen in this thread https://lkml.org/lkml/2014/11/24/41 and Catalin recommended a similar fix. drivers/clocksource/arm_arch_timer.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)