On Thu, Mar 13, 2025 at 11:25:19AM +0530, Siddharth Vadapalli wrote: > On Wed, Mar 12, 2025 at 11:16:00AM -0500, Bjorn Helgaas wrote: > > On Wed, Mar 05, 2025 at 06:50:18PM +0530, Siddharth Vadapalli wrote: > > > Commit under Fixes assigned the value of 'linkdown_irq_regfield' for the > > > J784S4 SoC as 'LINK_DOWN' which corresponds to BIT(1). However, according > > > to the Technical Reference Manual and Register Documentation for the J784S4 > > > SoC [0], BIT(1) corresponds to "ENABLE_SYS_EN_PCIE_DPA_1" which is __NOT__ > > > the field for the link-state interrupt. Instead, it is BIT(10) of the > > > "PCIE_INTD_ENABLE_REG_SYS_2" register that corresponds to the link-state > > > field named as "ENABLE_SYS_EN_PCIE_LINK_STATE". > > > > I guess the reason we want this is that on J784S4, we ignore actual > > link-down interrupts (and we don't write STATUS_CLR_REG_SYS_2 to clear > > the interrupt indication, so maybe there's an interrupt storm), and we > > think some other interrupt (DPA_1, whatever that is) is actually a > > link-down interrupt? > > While it is true that actual link-down interrupts are ignored, it is not > the case that there's an interrupt storm because the same incorrect macro > is used to enable the interrupt line. Since the enables an interrupt for > DPA_1 which never fires, we don't run into the situation where we are not > clearing the interrupt (the interrupt handler will look for the same > incorrect field to clear the interrupt if it does fire for DPA_1, but that > doesn't happen). The 'linkdown_irq_regfield' corresponds to the > "link-state" field not just in the J784S4 SoC, but in all SoCs supported by > the pci-j721e.c driver. It is only in J721E that it is BIT(1) > [LINK_DOWN macro], while in all other SoCs (J784S4 included), it is BIT(10) > [J7200_LINK_DOWN macro since it was first added for J7200 SoC]. Matt > probably referred to J721E's Technical Reference Manual and ended up > incorrectly assigning "LINK_DOWN", due to which the driver is enabling > the DPA_1 interrupt and the interrupt handler is also going to look for > the field corresponding to receiving an interrupt for DPA_1. So I guess without this patch, we incorrectly ignore link-down interrupts on J784S4. It's good to have a one-sentence motivation like that somewhere in the commit log that we can pull out and include in the merge commit log and the pull request. > I can only hope that the URL will redirect to the latest version of > the User Guide if at all it becomes invalid. OK, thanks, I guess there's nothing more to do ;) I guess that manual is not really designed for collaborative development. Thanks for the patient hand holding! Bjorn
