On Mon, Jun 14, 2010 at 02:25:21PM -0400, Kevin O'Connor wrote:
On Mon, Jun 14, 2010 at 05:51:27PM +0300, Avi Kivity wrote:
On 06/14/2010 05:09 PM, Gleb Natapov wrote:
Could we just have qemu build the hpet tables and pass them through to seabios? Perhaps using the qemu_cfg_acpi_additional_tables() method.
Possible, and I considered that. I personally prefer to pass minimum information required for seabios to discover underlying HW and leave ACPI table creation to seabios. That is how things done for HW that seabios can actually detect. If we will go your way pretty soon we will move creation of ACPI/SMBIOS/MP tables into qemu and IMHO this will be step backworkds.
I agree. ACPI is a firmware/OS interface. If we move ACPI table generation into qemu, it becomes a mixed hardware/firmware/OS interface.
This seems to be a philosophical distinction. Lets go over the practical implications.
It seems there was a change in qemu to the hpet functionality. Although the change is solely between qemu and the OS, it's necessary to patch both qemu and seabios for the OS to see the change. This means creating and reviewing patches for two separate repos. This also requires release coordination - the seabios change has to be committed and released, and then qemu needs to be released with the new seabios. Additional changes in seabios tip will get merged into qemu, which could complicate testing.
My patch is completely unrelated to functionality change in qemu. In fact I wrote it before the change and had to rebase. Seabios/qemu have a bug that HPET is always advertise through ACPI table even when qemu haven't created one. So your description above is not accurate. But even if it was accurate propagation features through software stack is common operation everywhere. Think about adding system call to the kernel and updating libc, or adding feature to kvm kernel module and adding patch to use it in qemu.
Better keep those interfaces separate: hardware/firmware (fwcfg) and firmware/OS (acpi).
One could look at the current hpet patch as implementing: qemu -> struct hpet_fw_entry -> seabios -> struct acpi_20_hpet -> OS.
I'm suggesting that we do the following instead: qemu -> struct acpi_20_hpet -> seabios -> struct acpi_20_hpet -> OS.
I'm not suggesting a radical rethink of fwcfg, but I fail to see the advantage in introducing the arbitrary "struct hpet_fw_entry" when there is a perfectly good, well defined, "struct acpi_20_hpet" that already exists. This new arbitrary intermediate format just introduces "make work" for all of us.
Then qemu will have to create ACPI header too and will have to fill details like oem_id/oem_table_id and so on. Now if I want to change them in my bios version it is not enough to edit CONFIG_APPNAME in seabios. If seabios will decide to move to more resent version of ACPI spec it will not be able to do so since it will not fully control table creation.
-- Gleb.