| Message ID | 20200409050508.96753-1-jaedon.shin@gmail.com (mailing list archive) |
|---|---|
| State | Mainlined |
| Commit | 4405bdf3c57ec28d606bdf5325f1167505bfdcd4 |
| Headers | show |
| Series | ARM: VDSO: Fix incorrect clock_gettime64 | expand |
On 2020-04-09 06:05, Jaedon Shin wrote: > __vdso_*() should be removed and fallback used if CNTCVT is not > available by cntvct_functional(). __vdso_clock_gettime64 when added > previous commit is using the incorrect CNTCVT value in that state. > __vdso_clock_gettime64 is also added to remove it's symbol. Tested-by: Robin Murphy <robin.murphy@arm.com> Turns out this is what was making my RK3288 box virtually unusable since its last couple of updates - it would be nice to get the fix merged and backported ASAP. I suspect it's also the same underlying thing for the Raspberry Pi thread over here: https://lore.kernel.org/linux-arm-kernel/CAK8P3a2Tw2w73ZkK-W6AA9veMK4-miLUx-TL1EuOdP7EdW-AmQ@mail.gmail.com/ Robin. > Fixes: 74d06efb9c2f ("ARM: 8932/1: Add clock_gettime64 entry point") > Signed-off-by: Jaedon Shin <jaedon.shin@gmail.com> > --- > arch/arm/kernel/vdso.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/arch/arm/kernel/vdso.c b/arch/arm/kernel/vdso.c > index e0330a25e1c6..28cfe7bad1bf 100644 > --- a/arch/arm/kernel/vdso.c > +++ b/arch/arm/kernel/vdso.c > @@ -184,6 +184,7 @@ static void __init patch_vdso(void *ehdr) > if (!cntvct_ok) { > vdso_nullpatch_one(&einfo, "__vdso_gettimeofday"); > vdso_nullpatch_one(&einfo, "__vdso_clock_gettime"); > + vdso_nullpatch_one(&einfo, "__vdso_clock_gettime64"); > } > } > >
On Sat, May 30, 2020 at 03:51:35PM +0100, Robin Murphy wrote: > On 2020-04-09 06:05, Jaedon Shin wrote: > > __vdso_*() should be removed and fallback used if CNTCVT is not > > available by cntvct_functional(). __vdso_clock_gettime64 when added > > previous commit is using the incorrect CNTCVT value in that state. > > __vdso_clock_gettime64 is also added to remove it's symbol. > > Tested-by: Robin Murphy <robin.murphy@arm.com> > > Turns out this is what was making my RK3288 box virtually unusable since its > last couple of updates - it would be nice to get the fix merged and > backported ASAP. I suspect it's also the same underlying thing for the > Raspberry Pi thread over here: > > https://lore.kernel.org/linux-arm-kernel/CAK8P3a2Tw2w73ZkK-W6AA9veMK4-miLUx-TL1EuOdP7EdW-AmQ@mail.gmail.com/ Does this mean I need to do something with this patch? It's not been put in the patch system, so I'm afraid it's missed the boat for 5.7.
On 2020-05-30 15:59, Russell King - ARM Linux admin wrote: > On Sat, May 30, 2020 at 03:51:35PM +0100, Robin Murphy wrote: >> On 2020-04-09 06:05, Jaedon Shin wrote: >>> __vdso_*() should be removed and fallback used if CNTCVT is not >>> available by cntvct_functional(). __vdso_clock_gettime64 when added >>> previous commit is using the incorrect CNTCVT value in that state. >>> __vdso_clock_gettime64 is also added to remove it's symbol. >> >> Tested-by: Robin Murphy <robin.murphy@arm.com> >> >> Turns out this is what was making my RK3288 box virtually unusable since its >> last couple of updates - it would be nice to get the fix merged and >> backported ASAP. I suspect it's also the same underlying thing for the >> Raspberry Pi thread over here: >> >> https://lore.kernel.org/linux-arm-kernel/CAK8P3a2Tw2w73ZkK-W6AA9veMK4-miLUx-TL1EuOdP7EdW-AmQ@mail.gmail.com/ > > Does this mean I need to do something with this patch? It's not been > put in the patch system, so I'm afraid it's missed the boat for 5.7. Well, there still seems to be a little uncertainty over whether this is the best fix, so I guess we may as well wait until the end of the merge window just in case somebody comes up with something more solid. In this particular context I'm merely a user who wants my thing to work, but I'll try to keep an eye out and follow up if I don't see any other movement. Robin.
On 2020-06-08 14:54, Robin Murphy wrote: > On 2020-05-30 15:59, Russell King - ARM Linux admin wrote: >> On Sat, May 30, 2020 at 03:51:35PM +0100, Robin Murphy wrote: >>> On 2020-04-09 06:05, Jaedon Shin wrote: >>>> __vdso_*() should be removed and fallback used if CNTCVT is not >>>> available by cntvct_functional(). __vdso_clock_gettime64 when added >>>> previous commit is using the incorrect CNTCVT value in that state. >>>> __vdso_clock_gettime64 is also added to remove it's symbol. >>> >>> Tested-by: Robin Murphy <robin.murphy@arm.com> >>> >>> Turns out this is what was making my RK3288 box virtually unusable >>> since its >>> last couple of updates - it would be nice to get the fix merged and >>> backported ASAP. I suspect it's also the same underlying thing for the >>> Raspberry Pi thread over here: >>> >>> https://lore.kernel.org/linux-arm-kernel/CAK8P3a2Tw2w73ZkK-W6AA9veMK4-miLUx-TL1EuOdP7EdW-AmQ@mail.gmail.com/ >>> >> >> Does this mean I need to do something with this patch? It's not been >> put in the patch system, so I'm afraid it's missed the boat for 5.7. > > Well, there still seems to be a little uncertainty over whether this is > the best fix, so I guess we may as well wait until the end of the merge > window just in case somebody comes up with something more solid. > > In this particular context I'm merely a user who wants my thing to work, > but I'll try to keep an eye out and follow up if I don't see any other > movement. Since the discussion about why the expected fallback behaviour didn't happen seems to have died down, and the Raspberry Pi folks have gone down the commendable path of actually fixing their firmware (!), I've taken the liberty of chucking this into the patch system for the sake of the remaining minority of affected machines. Robin.
diff --git a/arch/arm/kernel/vdso.c b/arch/arm/kernel/vdso.c index e0330a25e1c6..28cfe7bad1bf 100644 --- a/arch/arm/kernel/vdso.c +++ b/arch/arm/kernel/vdso.c @@ -184,6 +184,7 @@ static void __init patch_vdso(void *ehdr) if (!cntvct_ok) { vdso_nullpatch_one(&einfo, "__vdso_gettimeofday"); vdso_nullpatch_one(&einfo, "__vdso_clock_gettime"); + vdso_nullpatch_one(&einfo, "__vdso_clock_gettime64"); } }
__vdso_*() should be removed and fallback used if CNTCVT is not available by cntvct_functional(). __vdso_clock_gettime64 when added previous commit is using the incorrect CNTCVT value in that state. __vdso_clock_gettime64 is also added to remove it's symbol. Fixes: 74d06efb9c2f ("ARM: 8932/1: Add clock_gettime64 entry point") Signed-off-by: Jaedon Shin <jaedon.shin@gmail.com> --- arch/arm/kernel/vdso.c | 1 + 1 file changed, 1 insertion(+)