Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, 23 Feb 2024 at 16:55, Neil Armstrong wrote: > > On 23/02/2024 15:52, Johan Hovold wrote: > > On Fri, Feb 23, 2024 at 03:38:13PM +0100, Neil Armstrong wrote: > >> On 23/02/2024 15:21, Johan Hovold wrote: > > > >>> But it is *not* standalone as I tried to explain above. > >>> > >>> So you have to drop it again as the later patches depend on it and > >>> cannot be merged (through a different tree) without it. > >> > >> drm-misc branches cannot be rebased, it must be reverted, but it can still > >> be applied > >> on drm-misc-next and I'll send a revert patch for drm-misc-fixes if > >> needed, not a big deal. > >> > >>> I thought you had all the acks you needed to take this through drm-misc, > >>> but we can wait a bit more if necessary (and there's no rush to get the > >>> first one in). > >> > >> If you want it to be in v6.9, it's too late since the last drm-misc-next > >> PR has been sent > >> yesterday > >> (https://cgit.freedesktop.org/drm/drm-misc/tag/?h=drm-misc-next-2024-02-22) > >> > >> Please ping Thomas or Maxime, perhaps it's not too late since the > >> drm-misc-next tree > >> really closes on sunday. > > > > I don't want this in 6.9, this is needed for *6.8* as this fixes a DRM > > regression in 6.8-rc1 that breaks the display on machines like the X13s. > > > > If you guys can't sort this out in time, then perhaps Bjorn can take > > this through the Qualcomm tree instead (with DRM acks). > > > > But again, this is fixing a severe *regression* in 6.8-rc1. It can not > > wait for 6.9. > > Right, I can't apply them right now, I send a patchset ack so it can be > applied ASAP, Applied and pushed patches 2-4. Patches 5 and 6 can go through the phy/fixes. There is no need for them to go through drm-misc tree. -- With best wishes Dmitry
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On 23/02/2024 15:52, Johan Hovold wrote: On Fri, Feb 23, 2024 at 03:38:13PM +0100, Neil Armstrong wrote: On 23/02/2024 15:21, Johan Hovold wrote: But it is *not* standalone as I tried to explain above. So you have to drop it again as the later patches depend on it and cannot be merged (through a different tree) without it. drm-misc branches cannot be rebased, it must be reverted, but it can still be applied on drm-misc-next and I'll send a revert patch for drm-misc-fixes if needed, not a big deal. I thought you had all the acks you needed to take this through drm-misc, but we can wait a bit more if necessary (and there's no rush to get the first one in). If you want it to be in v6.9, it's too late since the last drm-misc-next PR has been sent yesterday (https://cgit.freedesktop.org/drm/drm-misc/tag/?h=drm-misc-next-2024-02-22) Please ping Thomas or Maxime, perhaps it's not too late since the drm-misc-next tree really closes on sunday. I don't want this in 6.9, this is needed for *6.8* as this fixes a DRM regression in 6.8-rc1 that breaks the display on machines like the X13s. If you guys can't sort this out in time, then perhaps Bjorn can take this through the Qualcomm tree instead (with DRM acks). But again, this is fixing a severe *regression* in 6.8-rc1. It can not wait for 6.9. Right, I can't apply them right now, I send a patchset ack so it can be applied ASAP, Thanks, Neil Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On 17/02/2024 16:02, Johan Hovold wrote: Starting with 6.8-rc1 the internal display sometimes fails to come up on machines like the Lenovo ThinkPad X13s and the logs indicate that this is due to a regression in the DRM subsystem [1]. This series fixes a race in the pmic_glink_altmode driver which was exposed / triggered by the transparent DRM bridges rework that went into 6.8-rc1 and that manifested itself as a bridge failing to attach and sometimes triggering a NULL-pointer dereference. The intermittent hard resets that have also been reported since 6.8-rc1 unfortunately still remains and suggests that we are dealing with two separate regressions. There is some indication that also the hard resets (e.g. due to register accesses to unclocked hardware) are also due to changes in the DRM subsystem as it happens around the time that the eDP panel and display controller would be initialised during boot (the runtime PM rework?). This remains to be verified, however. Included is also a fix for a related OF node reference leak in the aux-hpd driver found through inspection when reworking the driver. The use-after-free bug is triggered by a probe deferral and highlighted some further bugs in the involved drivers, which were registering child devices before deferring probe. This behaviour is not correct and can both trigger probe deferral loops and potentially also further issues with the DRM bridge implementation. This series can either go through the Qualcomm SoC tree (pmic_glink) or the DRM tree. The PHY patches do not depend on the rest of the series and could possibly be merged separately through the PHY tree. Whichever gets this to mainline the fastest. Johan [1] https://lore.kernel.org/lkml/zctvmlk4ztwcp...@hovoldconsulting.com/ Johan Hovold (5): drm/bridge: aux-hpd: fix OF node leaks drm/bridge: aux-hpd: separate allocation and registration soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free phy: qcom-qmp-combo: fix drm bridge registration phy: qcom-qmp-combo: fix type-c switch registration Rob Clark (1): soc: qcom: pmic_glink: Fix boot when QRTR=m drivers/gpu/drm/bridge/aux-hpd-bridge.c | 70 ++- drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 16 +++--- drivers/soc/qcom/pmic_glink.c | 21 +++ drivers/soc/qcom/pmic_glink_altmode.c | 16 +- include/drm/bridge/aux-bridge.h | 15 + 5 files changed, 102 insertions(+), 36 deletions(-) For the serie: Acked-by: Neil Armstrong After an offline discussion, Dmitry, it's ok to push the remaining patches to drm-misc-fixes. Thanks, Neil
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, Feb 23, 2024 at 03:38:13PM +0100, Neil Armstrong wrote: > On 23/02/2024 15:21, Johan Hovold wrote: > > But it is *not* standalone as I tried to explain above. > > > > So you have to drop it again as the later patches depend on it and > > cannot be merged (through a different tree) without it. > > drm-misc branches cannot be rebased, it must be reverted, but it can still be > applied > on drm-misc-next and I'll send a revert patch for drm-misc-fixes if needed, > not a big deal. > > > I thought you had all the acks you needed to take this through drm-misc, > > but we can wait a bit more if necessary (and there's no rush to get the > > first one in). > > If you want it to be in v6.9, it's too late since the last drm-misc-next PR > has been sent > yesterday > (https://cgit.freedesktop.org/drm/drm-misc/tag/?h=drm-misc-next-2024-02-22) > > Please ping Thomas or Maxime, perhaps it's not too late since the > drm-misc-next tree > really closes on sunday. I don't want this in 6.9, this is needed for *6.8* as this fixes a DRM regression in 6.8-rc1 that breaks the display on machines like the X13s. If you guys can't sort this out in time, then perhaps Bjorn can take this through the Qualcomm tree instead (with DRM acks). But again, this is fixing a severe *regression* in 6.8-rc1. It can not wait for 6.9. Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On 23/02/2024 15:21, Johan Hovold wrote: On Fri, Feb 23, 2024 at 02:52:28PM +0100, Neil Armstrong wrote: On 23/02/2024 13:51, Johan Hovold wrote: On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: On 23/02/2024 12:02, Neil Armstrong wrote: Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git (drm-misc-fixes) [1/6] drm/bridge: aux-hpd: fix OF node leaks https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b [2/6] drm/bridge: aux-hpd: separate allocation and registration (no commit info) [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free (no commit info) [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m (no commit info) [5/6] phy: qcom-qmp-combo: fix drm bridge registration (no commit info) [6/6] phy: qcom-qmp-combo: fix type-c switch registration (no commit info) To clarify, I only applied patch 1 to drm-misc-fixes Ok, but can you please not do that? :) These patches should go in through the same tree to avoid conflicts. I discussed this with Bjorn and Dmitry the other day and the conclusion was that it was easiest to take all of these through DRM. I only applied patch 1, which is a standalone fix and goes into a separate tree, for the next patches it would be indeed simpler for them to go via drm-misc when they are properly acked. But it is *not* standalone as I tried to explain above. So you have to drop it again as the later patches depend on it and cannot be merged (through a different tree) without it. drm-misc branches cannot be rebased, it must be reverted, but it can still be applied on drm-misc-next and I'll send a revert patch for drm-misc-fixes if needed, not a big deal. I thought you had all the acks you needed to take this through drm-misc, but we can wait a bit more if necessary (and there's no rush to get the first one in). If you want it to be in v6.9, it's too late since the last drm-misc-next PR has been sent yesterday (https://cgit.freedesktop.org/drm/drm-misc/tag/?h=drm-misc-next-2024-02-22) Please ping Thomas or Maxime, perhaps it's not too late since the drm-misc-next tree really closes on sunday. Neil Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, Feb 23, 2024 at 04:18:08PM +0200, Dmitry Baryshkov wrote: > On Fri, 23 Feb 2024 at 15:52, Neil Armstrong > wrote: > > On 23/02/2024 13:51, Johan Hovold wrote: > > > On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: > > >> On 23/02/2024 12:02, Neil Armstrong wrote: > > >>> Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git > > >>> (drm-misc-fixes) > > >>> > > >>> [1/6] drm/bridge: aux-hpd: fix OF node leaks > > >>> > > >>> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b > > >>> [2/6] drm/bridge: aux-hpd: separate allocation and registration > > >>> (no commit info) > > >>> [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free > > >>> (no commit info) > > >>> [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m > > >>> (no commit info) > > >>> [5/6] phy: qcom-qmp-combo: fix drm bridge registration > > >>> (no commit info) > > >>> [6/6] phy: qcom-qmp-combo: fix type-c switch registration > > >>> (no commit info) > > >>> > > >> > > >> To clarify, I only applied patch 1 to drm-misc-fixes > > > > > > Ok, but can you please not do that? :) > > > > > > These patches should go in through the same tree to avoid conflicts. > > > > > > I discussed this with Bjorn and Dmitry the other day and the conclusion > > > was that it was easiest to take all of these through DRM. > > > > I only applied patch 1, which is a standalone fix and goes into a separate > > tree, > > for the next patches it would be indeed simpler for them to go via drm-misc > > when > > they are properly acked. > > I think PHY patches can go through a usual route (phy/next or > phy/fixes). They can, but I've explicitly asked Vinod to ack them so that they can go in with the rest of the series through DRM. They also fix a regression that came in through DRM in 6.8-rc1 (the bridge rework which started registering child devices) so it makes sense to also route the fix the same way. And to do it for this cycle. > For patches 3 and 4 I'd need an ack from Bjorn to merge > them through drm-misc-next or drm-misc-fixes. You have Bjorn's ack. He's reviewed all the patches for this purpose and we discussed this in person two days ago. And, again, this has to go in for *this* cycle. You broke the display on the X13s and other machines so this cannot wait for 6.9. Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, Feb 23, 2024 at 02:52:28PM +0100, Neil Armstrong wrote: > On 23/02/2024 13:51, Johan Hovold wrote: > > On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: > >> On 23/02/2024 12:02, Neil Armstrong wrote: > >>> Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git > >>> (drm-misc-fixes) > >>> > >>> [1/6] drm/bridge: aux-hpd: fix OF node leaks > >>> > >>> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b > >>> [2/6] drm/bridge: aux-hpd: separate allocation and registration > >>> (no commit info) > >>> [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free > >>> (no commit info) > >>> [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m > >>> (no commit info) > >>> [5/6] phy: qcom-qmp-combo: fix drm bridge registration > >>> (no commit info) > >>> [6/6] phy: qcom-qmp-combo: fix type-c switch registration > >>> (no commit info) > >>> > >> > >> To clarify, I only applied patch 1 to drm-misc-fixes > > > > Ok, but can you please not do that? :) > > > > These patches should go in through the same tree to avoid conflicts. > > > > I discussed this with Bjorn and Dmitry the other day and the conclusion > > was that it was easiest to take all of these through DRM. > > I only applied patch 1, which is a standalone fix and goes into a separate > tree, > for the next patches it would be indeed simpler for them to go via drm-misc > when > they are properly acked. But it is *not* standalone as I tried to explain above. So you have to drop it again as the later patches depend on it and cannot be merged (through a different tree) without it. I thought you had all the acks you needed to take this through drm-misc, but we can wait a bit more if necessary (and there's no rush to get the first one in). Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, 23 Feb 2024 at 15:52, Neil Armstrong wrote: > > On 23/02/2024 13:51, Johan Hovold wrote: > > On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: > >> On 23/02/2024 12:02, Neil Armstrong wrote: > >>> Hi, > >>> > >>> On Sat, 17 Feb 2024 16:02:22 +0100, Johan Hovold wrote: > Starting with 6.8-rc1 the internal display sometimes fails to come up on > machines like the Lenovo ThinkPad X13s and the logs indicate that this > is due to a regression in the DRM subsystem [1]. > > This series fixes a race in the pmic_glink_altmode driver which was > exposed / triggered by the transparent DRM bridges rework that went into > 6.8-rc1 and that manifested itself as a bridge failing to attach and > sometimes triggering a NULL-pointer dereference. > > [...] > >>> > >>> Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git > >>> (drm-misc-fixes) > >>> > >>> [1/6] drm/bridge: aux-hpd: fix OF node leaks > >>> > >>> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b > >>> [2/6] drm/bridge: aux-hpd: separate allocation and registration > >>> (no commit info) > >>> [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free > >>> (no commit info) > >>> [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m > >>> (no commit info) > >>> [5/6] phy: qcom-qmp-combo: fix drm bridge registration > >>> (no commit info) > >>> [6/6] phy: qcom-qmp-combo: fix type-c switch registration > >>> (no commit info) > >>> > >> > >> To clarify, I only applied patch 1 to drm-misc-fixes > > > > Ok, but can you please not do that? :) > > > > These patches should go in through the same tree to avoid conflicts. > > > > I discussed this with Bjorn and Dmitry the other day and the conclusion > > was that it was easiest to take all of these through DRM. > > I only applied patch 1, which is a standalone fix and goes into a separate > tree, > for the next patches it would be indeed simpler for them to go via drm-misc > when > they are properly acked. I think PHY patches can go through a usual route (phy/next or phy/fixes). For patches 3 and 4 I'd need an ack from Bjorn to merge them through drm-misc-next or drm-misc-fixes. -- With best wishes Dmitry
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On 23/02/2024 13:51, Johan Hovold wrote: On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: On 23/02/2024 12:02, Neil Armstrong wrote: Hi, On Sat, 17 Feb 2024 16:02:22 +0100, Johan Hovold wrote: Starting with 6.8-rc1 the internal display sometimes fails to come up on machines like the Lenovo ThinkPad X13s and the logs indicate that this is due to a regression in the DRM subsystem [1]. This series fixes a race in the pmic_glink_altmode driver which was exposed / triggered by the transparent DRM bridges rework that went into 6.8-rc1 and that manifested itself as a bridge failing to attach and sometimes triggering a NULL-pointer dereference. [...] Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git (drm-misc-fixes) [1/6] drm/bridge: aux-hpd: fix OF node leaks https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b [2/6] drm/bridge: aux-hpd: separate allocation and registration (no commit info) [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free (no commit info) [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m (no commit info) [5/6] phy: qcom-qmp-combo: fix drm bridge registration (no commit info) [6/6] phy: qcom-qmp-combo: fix type-c switch registration (no commit info) To clarify, I only applied patch 1 to drm-misc-fixes Ok, but can you please not do that? :) These patches should go in through the same tree to avoid conflicts. I discussed this with Bjorn and Dmitry the other day and the conclusion was that it was easiest to take all of these through DRM. I only applied patch 1, which is a standalone fix and goes into a separate tree, for the next patches it would be indeed simpler for them to go via drm-misc when they are properly acked. Neil With Vinod acking the PHY patches, I believe you have what you need to merge the whole series now? Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On Fri, Feb 23, 2024 at 12:03:10PM +0100, Neil Armstrong wrote: > On 23/02/2024 12:02, Neil Armstrong wrote: > > Hi, > > > > On Sat, 17 Feb 2024 16:02:22 +0100, Johan Hovold wrote: > >> Starting with 6.8-rc1 the internal display sometimes fails to come up on > >> machines like the Lenovo ThinkPad X13s and the logs indicate that this > >> is due to a regression in the DRM subsystem [1]. > >> > >> This series fixes a race in the pmic_glink_altmode driver which was > >> exposed / triggered by the transparent DRM bridges rework that went into > >> 6.8-rc1 and that manifested itself as a bridge failing to attach and > >> sometimes triggering a NULL-pointer dereference. > >> > >> [...] > > > > Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git > > (drm-misc-fixes) > > > > [1/6] drm/bridge: aux-hpd: fix OF node leaks > > > > https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b > > [2/6] drm/bridge: aux-hpd: separate allocation and registration > >(no commit info) > > [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free > >(no commit info) > > [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m > >(no commit info) > > [5/6] phy: qcom-qmp-combo: fix drm bridge registration > >(no commit info) > > [6/6] phy: qcom-qmp-combo: fix type-c switch registration > >(no commit info) > > > > To clarify, I only applied patch 1 to drm-misc-fixes Ok, but can you please not do that? :) These patches should go in through the same tree to avoid conflicts. I discussed this with Bjorn and Dmitry the other day and the conclusion was that it was easiest to take all of these through DRM. With Vinod acking the PHY patches, I believe you have what you need to merge the whole series now? Johan
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
On 23/02/2024 12:02, Neil Armstrong wrote: Hi, On Sat, 17 Feb 2024 16:02:22 +0100, Johan Hovold wrote: Starting with 6.8-rc1 the internal display sometimes fails to come up on machines like the Lenovo ThinkPad X13s and the logs indicate that this is due to a regression in the DRM subsystem [1]. This series fixes a race in the pmic_glink_altmode driver which was exposed / triggered by the transparent DRM bridges rework that went into 6.8-rc1 and that manifested itself as a bridge failing to attach and sometimes triggering a NULL-pointer dereference. [...] Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git (drm-misc-fixes) [1/6] drm/bridge: aux-hpd: fix OF node leaks https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b [2/6] drm/bridge: aux-hpd: separate allocation and registration (no commit info) [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free (no commit info) [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m (no commit info) [5/6] phy: qcom-qmp-combo: fix drm bridge registration (no commit info) [6/6] phy: qcom-qmp-combo: fix type-c switch registration (no commit info) To clarify, I only applied patch 1 to drm-misc-fixes Thanks, Neil
Re: [PATCH 0/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free
Hi, On Sat, 17 Feb 2024 16:02:22 +0100, Johan Hovold wrote: > Starting with 6.8-rc1 the internal display sometimes fails to come up on > machines like the Lenovo ThinkPad X13s and the logs indicate that this > is due to a regression in the DRM subsystem [1]. > > This series fixes a race in the pmic_glink_altmode driver which was > exposed / triggered by the transparent DRM bridges rework that went into > 6.8-rc1 and that manifested itself as a bridge failing to attach and > sometimes triggering a NULL-pointer dereference. > > [...] Thanks, Applied to https://anongit.freedesktop.org/git/drm/drm-misc.git (drm-misc-fixes) [1/6] drm/bridge: aux-hpd: fix OF node leaks https://cgit.freedesktop.org/drm/drm-misc/commit/?id=9ee485bdda68d6d3f5728cbe3150eb9013d7d22b [2/6] drm/bridge: aux-hpd: separate allocation and registration (no commit info) [3/6] soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free (no commit info) [4/6] soc: qcom: pmic_glink: Fix boot when QRTR=m (no commit info) [5/6] phy: qcom-qmp-combo: fix drm bridge registration (no commit info) [6/6] phy: qcom-qmp-combo: fix type-c switch registration (no commit info) -- Neil
