Re: [libvirt] [Qemu-devel] [PATCH] cirrus_vga: adding sanity check for vram size
[adding libvirt] On 05/09/2014 05:54 AM, Gerd Hoffmann wrote:
Hi,
virt-manager/libvirt seems to default to 9 MByte of Vram for cirrus, so this would break a lot of setups.
It wouldn't. libvirt sticks that into the xml, but it doesn't set any qemu parameters. The libvirt parameter actually predates the qemu property for setting the size.
Then we should probably re-evaluate what libvirt does with the parameters, which avoids breaking any guest that happens to be pre-existing with the odd 9MB sizing in the XML.
Looking at datasheets on the web seems to say the chips actually went down to 1 MB or less.
I have my doubts we emulate that correctly (register telling the guest how much memory is actually there etc.). Also it is pretty much useless these days, even the 4MB imply serious constrains when FullHD displays are commonplace. Newer cirrus drivers such as the kernel's drm driver are specifically written to qemu's cirrus cards, I have my doubs that they are prepared to handle 1MB cirrus cards correctly.
Bottom line: Allowing less than 4MB is asking for trouble for no good reason ;)
cheers, Gerd
-- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Am 12.05.2014 19:05, schrieb Eric Blake:
[adding libvirt]
On 05/09/2014 05:54 AM, Gerd Hoffmann wrote:
Hi,
virt-manager/libvirt seems to default to 9 MByte of Vram for cirrus, so this would break a lot of setups.
It wouldn't. libvirt sticks that into the xml, but it doesn't set any qemu parameters. The libvirt parameter actually predates the qemu property for setting the size.
Déjà vu... we had the same problem with pvpanic.
Then we should probably re-evaluate what libvirt does with the parameters, which avoids breaking any guest that happens to be pre-existing with the odd 9MB sizing in the XML.
If libvirt cannot or doesn't want to pass a user-supplied tag to QEMU, the only sane and future-proof solution is for libvirt to signal an error to the user for *any* flags it cannot handle rather than throwing up on a case-by-case basis the question of how libvirt should behave when some previously ignored feature gets implemented. That would prevent us from running into this situation again and again. As further examples, apic, acpi and other x86-centric features got auto-added for s390x guests last time I checked. Apart from anything the user might have specified herself, there's really no reason to add stuff the user did not ask for and that are not applicable. Ignoring a value "9" specified by the user would be very odd; and I guess you have no way of telling who added what to the XML for choosing between dropping and error'ing out. Regards, Andreas - -- SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (GNU/Linux) iQIcBAEBAgAGBQJTcQqiAAoJEPou0S0+fgE/CY8P/jRgyfqNtv2MLbilNXKKXe5A CBftFQv1dlvL4VFVGD9W5ortlyOpPfxSmC6fgnLKNbTsTCtDy1k3zgWp8uSjYNrp /W5oMpOstYxUBA7LtizCnCyPisffV6FlWQBLt8BGnRBPjEyVQXIIDHQuwvYv6Byr EKjCGWMjw76Rfe84Hdl7xDQvfd+qTFjIDbKEfSa3fciJ/svCYKlnTqsPzvtNWbUg 907r5tI2LVT17xaclhex12GQ+uxV4hXcpcUnY3W/lCHvH98NdgFZRf1P8xTdjSmB MbQ6WKOtGK9SqoyPiPzm8NwPduBbbmC69vumEmaG6QCqvTmcEKPNB5WMcXSEo4YB SBJjZrkeNgS7XlUDwL4i45PYSbxi60CByrIrvYzWEielPCUQWMKzckfs+Zxmu0FT nO51q334YmHU1X/oEPDnF8hakwzu9figWJ43uXSP/dm3ifjhFh35/Owmz+pd84Zi cauzlygF0JAHGKurStveIqOZ0o6AJLxhWZP695PbKHAeHvOXo6dSsnLgG6POylP1 N/pTCcvQZV6oZ/NjscHcrmYxwrA4XWSzY8JdzS4iPSqzvCgxFNR/wZ1VTjCZ5Cbx zs+bv+B7JKhjrYs/7dQt8kPv5ZXwh/kEkxkhXX7xpBfdPixx173Gdg3XvS86G6GI q/dszTpb4kJNcsaxNp3y =RZ98 -----END PGP SIGNATURE-----
participants (2)
-
Andreas Färber -
Eric Blake