On Mon 19 Mar 08:18 PDT 2018, Christ, Austin wrote: > On 3/14/2018 5:40 PM, Bjorn Andersson wrote: > > On Wed 14 Mar 13:05 PDT 2018, Christ, Austin wrote: > > > > > This patch is designed to address a possible issue where an interrupt is > > > fired but not yet handled by the i2c-qup driver and the kernel goes down as > > > kexec prepares to start a secondary kernel. In this case it is possible the > > > interrupt will be left unhandled. > > > > > > > What's the problem with not handling this IRQ? Does the hardware end up > > in some funky state? Are you worried that the kexec kernel will not > > function because there's a pending I2C interrupt? > > > The potential issue is that if an IRQ is left high, the secondary kernel > driver for I2C may enter a bad state trying to handle an interrupt that does > not apply to its context. In the QUP driver case, it may be less of an issue > because the probe does a soft reset of the core. > This is a very common issue with hardware blocks used by the bootloader, so drivers must be implemented in such a way that they can handle this scenario. > > > This is not unique to the i2c-qup driver, similar patches have gone into > > > other drivers such as the following for the arm-smmu-v3 driver. > > > > > > > The only similarity I can see here is that the SMMU is shut down in > > order for any ongoing memory transaction to stop, which could be analog > > to ongoing BAM operations in the QUP driver. But in that case I think > > the shutdown should go in the BAM, to stop any ongoing transactions. > My point wasn't that the shutdown was addressing the same issues between > these drivers, but that registering a shutdown function is the standard code > path for exposing clean tear down for kexec. Kexec does not call remove on > active drivers; it calls shutdown. > We do know that there are devices out there with e.g. backlight drivers on i2c that will be operated in order to display a splash screen in the boot loader. As such the driver must attempt to clear this state and/or be able to handle any spurious interrupts. > Is there any negative impact for adding this function? > This must be handled already and as such this extra code should be totally unnecessary, and we shouldn't merge unnecessary code. And as you say above, the i2c-qup driver does reset the core so this should be "less of an issue", not merging this patch means that if there is any issues left we expose these to more testing. Regards, Bjorn > > > > Regards, > > Bjorn > > > > > commit 7aa8619a66aea52b145e04cbab4f8d6a4e5f3f3b > > > Author: Nate Watterson <nwatters@xxxxxxxxxxxxxx> > > > Date: Thu Jun 29 18:18:15 2017 -0400 > > > > > > iommu/arm-smmu-v3: Implement shutdown method > > > > > > The shutdown method disables the SMMU to avoid corrupting a new kernel > > > started with kexec. > > > > > > Signed-off-by: Nate Watterson <nwatters@xxxxxxxxxxxxxx> > > > Signed-off-by: Will Deacon <will.deacon@xxxxxxx> > > > > > > > > > > > > On 2/2/2018 4:36 PM, Bjorn Andersson wrote: > > > > On Tue 16 Jan 13:35 PST 2018, Austin Christ wrote: > > > > > > > > > This shutdown method disables I2C to avoid corrupting a new kernel > > > > > started with kexec. > > > > > > > > > > > > > Can you elaborate on the issue you're seeing here? In what way is the > > > > i2c-qup driver special, will there be similar patches for all other > > > > drivers in the system? > > > > > > > > Regards, > > > > Bjorn > > > > > > > > > > -- > > > Qualcomm Datacenter Technologies as an affiliate of Qualcomm Technologies, > > > Inc. > > > Qualcomm Technologies, Inc. is a member of the > > > Code Aurora Forum, a Linux Foundation Collaborative Project. > > -- > Qualcomm Datacenter Technologies as an affiliate of Qualcomm Technologies, > Inc. > Qualcomm Technologies, Inc. is a member of the > Code Aurora Forum, a Linux Foundation Collaborative Project.
