> From: linux-hyperv-owner@xxxxxxxxxxxxxxx > > ... > > When a Linux guest runs on Hyper-V (x86_32, x86_64, or ARM64) , we have a > > front-end balloon driver in the guest, which balloons up/down and > > hot adds/removes the guest's memory when the host requests that. The > > problem > > is: the back-end driver on the host can not really save and restore the states > > related to the front-end balloon driver on guest hibernation, so we made the > > decision that balloon up/down and hot-add/remove are not supported when > > we enable hibernation for a guest; BTW, we still want to load the front-end > > driver in the guest, because the driver has a functionality of reporting the > > guest's memory pressure to the host, which we think is useful. > > > > On x86_32 and x86_64, we enable hibernation for a guest by enabling > > the virtual ACPI S4 state for the guest; on ARM64, so far we don't have the > > host side changes required to support guest hibernation, so the details are > > still unclear. > > > > After I discussed with Michael Kelley, it looks we don't really need to > > export drivers/acpi/sleep.c: acpi_sleep_state_supported(), but I think we do > > need to make it non-static. > > > > Now I propose the below changes. I plan to submit a patch first for the > > changes made to drivers/acpi/sleep.c and include/acpi/acpi_bus.h in a few > > days, if there is no objection. > > > > Please let me know how you think of this. Thanks! > > No. > > Hibernation should be always supported, no matter what firmware. If it > can powerdown, it can hibernate. > > That is for x86-32/64, too. > Pavel Hi Pavel, Yes, I totally agree hibernation should be always supported, as long as the system can power down. This is also true for Linux guest running on Hyper-V, when the pava-virtualized drivers are not loaded in the guest However, unluckily the situation is a little different when the pava-virtualized drivers are loaded in the guest: 1) Old Hyper-V hosts are unable to support guest hibernation and these hosts don't support the virtual ACPI S4 state. 2) The recent Hyper-V host is able to support guest hibernation, but when guest hibernation is used, the guest needs to disable some features in the guest balloon driver, as I explained in the previous mail; on the other hand, we also want to keep the ability to enable the features, so the Hyper-V guys decided to use the absence of the virtual ACPI S4 state to signify the ability of enabling the features, and there is a host command tool which can enable or disable the virtual ACPI S4 state for a guest. In summary: 1) On old Hyper-V hosts or a recent host, the virtual ACPI S4 state is unsupported or disabled, respectively, and we don't allow Linux guest hibernation so we can use the full features of the guest balloon driver. 2) On a recent host, after we use the host command tool to enable the virtual ACPI S4 state, we allow Linux guest hibernation and we only use a subset of the full features of the guest balloon driver. This is why I need to know if the virtual ACPI S4 state is supported-and-enabled. Exporting acpi_sleep_state_supported() may be overkill, and now making it non-static seems the only way to help me out. IMO the function is unlikely to change in the future, and it should be stable enough to become non-static. Looking forward to your insights! Thanks, -- Dexuan
