Hi, Am Dienstag, den 22.01.2019, 13:17 +0000 schrieb Aisheng Dong: > > > > From: Lucas Stach [mailto:l.stach@xxxxxxxxxxxxxx] > > Sent: Tuesday, January 22, 2019 8:52 PM > > > > To: Aisheng Dong <aisheng.dong@xxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx > > > > > > Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; shawnguo@xxxxxxxxxx; dl-linux-imx > > > > > > > > <linux-imx@xxxxxxx>; robh+dt@xxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; > > > > tglx@xxxxxxxxxxxxx; Marc Zyngier <marc.zyngier@xxxxxxx> > > Subject: Re: [PATCH 4/4] irq: imx: irqsteer: add multi output interrupts support > > > > Am Dienstag, den 22.01.2019, 12:03 +0000 schrieb Aisheng Dong: > > > > > > From: Lucas Stach [mailto:l.stach@xxxxxxxxxxxxxx] > > > > > > > > Sent: Tuesday, January 22, 2019 6:59 PM > > > > > > > > Am Dienstag, den 22.01.2019, 10:39 +0000 schrieb Aisheng Dong: > > > > > > > > From: Lucas Stach [mailto:l.stach@xxxxxxxxxxxxxx] > > > > > > > > > > > > Sent: Friday, January 18, 2019 6:23 PM > > > > > > > > > > [...] > > > > > > > > This has been discussed when upstreaming the driver. The > > > > > > > > controller may support multiple output IRQs, but only one > > > > > > > > them is actually used depending on the CHANCTRL config. > > > > > > > > There is no use in hooking up all the output IRQs in DT, if > > > > > > > > only one of them is actually used. Some of the outputs may > > > > > > > > not even be visible to the Linux system, but may belong to a > > > > > > > > Cortex M4 subsystem. All of those configurations can be > > > > > > > > described in DT by changing the upstream interrupt and > > > > > > > > "fsl,channel" in a > > > > > > > > > > > > coherent way. > > > > > > > > > > > > > > > > Please correct me if my understanding is totally wrong. > > > > > > > > > > > > > > I'm afraid your understanding of CHAN seems wrong. > > > > > > > (Binding doc of that property needs change as well). > > > > > > > > > > > > > > On QXP DC SS, the IRQSTEER supports 512 interrupts with 8 > > > > > > > interrupt output Conntected to GIC. > > > > > > > The current driver does not support it as it assumes only one > > > > > > > interrupt > > > > > > > > > > > > output used. > > > > > > > > > > > > Okay, so let's take a step back. The description in the QXP RM > > > > > > is actually better than what I've seen until now. Still it's > > > > > > totally confusing that > > > > > > > > the "channel" > > > > > > terminology used with different meanings in docs. Let's try to > > > > > > avoid this as much as possible. > > > > > > > > > > > > So to get things straight: Each irqsteer controller has a number > > > > > > of IRQ > > > > > > > > groups. > > > > > > All the input IRQs of one group are ORed together to form on output > > > > IRQ. > > > > > > Depending on the SoC integration, a group can contain 32 or > > > > > > 64 IRQs, where DCSS irqsteer on MX8M and the big 512 input > > > > > > controllers on QXP and QM both use 64 IRQs per group. You are > > > > > > claiming that the smaller controllers on both QXP am QM have > > > > > > only 32 > > > > > > > > IRQs per group, right? > > > > > > > > > > > > So the only change that is needed is that the driver needs to > > > > > > know the number of input IRQs per group, with a default of 64 to > > > > > > not break DT > > > > > > > > compatibility. > > > > > > > > > > > > > > > > Not exactly. > > > > > from HW point of view , there're two parameters during IRQSTEER > > > > > > > > integration. > > > > > For example, > > > > > DC in QXP: > > > > > > > > > > > > > > > > > parameter IRQCHAN = 1; > > > > //Number of IRQ Channels/Slots > > > > > > > parameter NINT32 = 8; //Number of interrupts in > > > > multiple > > > > > > > > of 32 > > > > > > > > If this is always in multiples of 32, the only change we need to > > > > make to the driver is to fix DT binding and interpretation of the > > > > "fsl,irq-groups" property to be in multiples of 32. > > > > > > > > This means i.MX8MQ DCSS irqsteer would need to change to 2 > > > > irq-groups, but as this isn't used upstream yet we can still do this > > > > change without breaking too much stuff and I would rather correct > > > > this now than keeping a DT binding around that doesn't match the HW. > > > > > > > > > > We want to avoid using of irq-groups as it's wrong. > > > Stick to HW parameters, only channel number and interrupts number should > > > > be used. > > > > The fsl,irq-groups property is exactly your NINT32 parameter above. I just > > wrongly assumed that it's always in multiples of 64, as that's what the > > i.MX8MQ DCSS irqsteer module looks like. We should fix this and be done with > > it. > > > > No, not exactly the same thing. Using group will confuse people that the group is 32. > However, internally Group is fixed 64 interrupts although it may not use all the > 64 interrupts. E.g. 32 interrupts. > See CHn_MINTDIS register which is also defined fixed to 64. > > The two HW parameter for integration is already very clear. We should use interrupts > Number for the channel. Not group. Okay, I see that the name irq-groups is confusing for you. But then I find the -per-chan naming confusing. So given that we seem to agree to split each channel into it's own DT node, there is no need to name the property "something-per-chan", as it's implied by the split DT nodes that all properties in one node are referencing a channel. May I suggest to name the property "fsl,num-irqs"? Regards, Lucas
