For some use-cases, it is helpful to have more than one UART available to the guest, but the Arm 'virt' board only creates one. If the second UART slot is not already used for a TrustZone Secure-World-only UART, create it as a NonSecure UART if the user provides a serial backend for it (e.g. via a second -serial command line option).
We've wanted this for literally years; my first attempt at it was this series in 2017: https://lore.kernel.org/all/1512745328-5109-1-git-send-email-peter.mayd...@linaro.org/ More recently Axel Heider revived the idea with a patchset last year: https://lore.kernel.org/qemu-devel/166990501232.22022.1658256124453401108...@git.sr.ht/ However it has previously foundered on the problem that EDK2 does odd things when presented with multiple UARTs in the DTB. (Specifically, it will send the guest GRUB bootloader output to UART1, debug output to both UARTs 0 and 1 depending on how far through boot it is, and the guest kernel will use UART0 since that's what the ACPI tables say.) Several things here I think mean we can finally get over this issue: * I learnt about the device tree "aliases" node; this allows us to explicitly say "this node is the first UART and this node is the second UART". So guests like Linux that follow this part of the DTB spec will always get the UART order correct; and if there are obscure guests that turn out to misbehave, we can point at the spec and say "this is how you should fix this on your end"... * This patch, like Axel's patch, only creates the second UART if the user asks for it on the command line, so any pre-existing command lines will not change behaviour. * Laszlo Ersek has kindly written some EDK2 patches that rationalise what it does when it finds more than one UART. This means that we can tell any users who do want to use two UARTs with EDK2 "you should upgrade your EDK2 blobs to version NNN if you want to do that". I haven't tagged this patchset RFC because I think it is good for review as it stands. However I think we should ideally wait for the EDK2 changes to go upstream first, so that we can fill in the TODO comment in patch 3 and mention in our release notes what the new EDK2 that correctly handles multiple UARTs would be. So I'm not targeting this at 8.2. Review from ACPI experts to confirm that I got the ACPI table changes right in patch 3 would be appreciated :-) thanks -- PMM Peter Maydell (3): hw/arm/virt: Add serial aliases in DTB hw/arm/virt: Rename VIRT_UART and VIRT_SECURE_UART to VIRT_UART[01] hw/arm/virt: allow creation of a second NonSecure UART docs/system/arm/virt.rst | 6 ++++- include/hw/arm/virt.h | 5 ++-- hw/arm/virt-acpi-build.c | 22 ++++++++++------- hw/arm/virt.c | 52 +++++++++++++++++++++++++++++++++------- 4 files changed, 65 insertions(+), 20 deletions(-) -- 2.34.1
