https://bugzilla.kernel.org/show_bug.cgi?id=220705
--- Comment #14 from Lukas Wunner ([email protected]) --- Hm, L1 is indeed already enabled by BIOS, both on the Root Port and on the wifi card. This all looks fine. And after resume, BIOS has enabled L1 on the Root Port again, which pci_restore_state() overwrites (it clears the ASPM Control bits). Hence the incongruence with the wifi card, which also has L1 enabled when it is re-enumerated after the hotplug event on resume, which causes the message to be printed with the other patch. So this all looks fine and plausible. I went back to the Intel spec and noticed this: "T_power_on_scale value in L1 PM substates control 2 register Current implementation in StP supports only the value of 00 in the T_power_on_scale field (bits 1:0). Other settings are not supported. Details: According to the PCIe standard, the L1 PM Substates Control 2 Register (Offset 0Ch) can accept several scale range values. Stone Peak supports only the default value of 00b which indicates range of 2 uSec. Impact: Most of the customers take the Intel reference design, which is set by default to 00b in this field, so no impact. Other customers need to make sure they set the value to 00b." https://fcc.report/FCC-ID/2AVBM-GK5/5728379.pdf lspci shows this value for the L1 PM SS Control 1 register: LTR1.2_Threshold=106496ns Unfortunately lspci only shows the product of the scale and value fields, but since the value can be at most 1023, it seems the scale was set to 1024 (= 010b) and the value to 104. PCIe r7.0 sec 7.8.3.3 says about the LTR_L1.2_THRESHOLD_Scale field: "Hardware operation is undefined if software writes a Not-Permitted value to this field" Which would explain why the device has an allergic reaction to even being put into D0. So the current theory is that the problems are caused by the kernel writing an unsupported value to LTR_L1.2_THRESHOLD_Scale. I'll see to it that I hack together a patch to verify that theory. We do not have support so far for devices which support only specific values for that field, so we'd have to retrofit that. -- You may reply to this email to add a comment. You are receiving this mail because: You are watching the assignee of the bug. _______________________________________________ acpi-bugzilla mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/acpi-bugzilla
