Last time I checked[1], SeaBIOS required 1MiB of RAM, and the failure modes were mean.
Back then, I asked whether we should enforce a suitable minimum RAM size[2]. Peter Maydell replied that modelling RAM constraints involves an expedition into the Generality Swamps, and wished me better luck than he had.
Four and a half years later, the failure modes are as mean as ever. For instance,
$ qemu-system-x86_64 --nodefaults -device VGA -m 640k
simply hangs for me, and
$ qemu-system-x86_64 --nodefaults -device VGA -m 16k
crashes with "qemu: fatal: Trying to execute code outside RAM or ROM at 0x0000000000004000" and a register dump with TCG, or the even less helpful "KVM internal error. Suberror: 1" with KVM.
Waiting for "someone" to design and implement the completely general solution has had the predictable result: nothing.
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
[1] Message-ID: 87fw7xwqkq.fsf@blackfin.pond.sub.org https://www.seabios.org/pipermail/seabios/2012-August/004343.html [2] Message-ID: 87wr1921rd.fsf@blackfin.pond.sub.org https://lists.gnu.org/archive/html/qemu-devel/2012-08/msg01319.html
On 22.03.2017 10:08, Markus Armbruster wrote: [...]
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
Thomas
On 22.03.2017 11:03, Thomas Huth wrote:
On 22.03.2017 10:08, Markus Armbruster wrote: [...]
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
Thomas
Or maybe simply warn the user that things may go wrong instead of enforcing it.
On Mi, 2017-03-22 at 11:19 +0100, David Hildenbrand wrote:
On 22.03.2017 11:03, Thomas Huth wrote:
On 22.03.2017 10:08, Markus Armbruster wrote: [...]
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
Thomas
Or maybe simply warn the user that things may go wrong instead of enforcing it.
Why bother? I have my doubts physical i440fx works with less than 1M either, given that this memory is needed to shadow the roms. Possibly you can't even find dimms that are small to plug them into such a system to try ...
I'd say just add a hard limit and be done with it.
Maybe exclude isapc. That one hasn't shadow support so things have at least a chance to work with less than 1M of memory. But honestly I'd rather drop isapc, together with ia64 and sparc. I mean, what is the use case? 'pc' machine type is compatible enough with vga and ide ports being on the standard isa locations so even msdos which has no pci support at all boots happily.
cheers, Gerd
On 23.03.2017 09:19, Gerd Hoffmann wrote:
On Mi, 2017-03-22 at 11:19 +0100, David Hildenbrand wrote:
On 22.03.2017 11:03, Thomas Huth wrote:
On 22.03.2017 10:08, Markus Armbruster wrote: [...]
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
Thomas
Or maybe simply warn the user that things may go wrong instead of enforcing it.
Why bother? I have my doubts physical i440fx works with less than 1M either, given that this memory is needed to shadow the roms. Possibly you can't even find dimms that are small to plug them into such a system to try ...
Because it seems to work if you supply the correct rom. We are trying to catch user errors, don't we?
I'd say just add a hard limit and be done with it.
"640K ought to be enough for anybody". Any limit we set will become out of date.
Maybe exclude isapc. That one hasn't shadow support so things have at least a chance to work with less than 1M of memory. But honestly I'd rather drop isapc, together with ia64 and sparc. I mean, what is the use case? 'pc' machine type is compatible enough with vga and ide ports being on the standard isa locations so even msdos which has no pci support at all boots happily.
I think I like that idea.
cheers, Gerd
On Wed, 22 Mar 2017 11:03:44 +0100 Thomas Huth thuth@redhat.com wrote:
On 22.03.2017 10:08, Markus Armbruster wrote: [...]
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
We can print warning that minimum size will be enforced in 1-2 releases since 2.9/10 and when it's enforced send users that need less to an old qemu version.
Thomas
On Wed, Mar 22, 2017 at 01:25:49PM +0100, Igor Mammedov wrote:
On Wed, 22 Mar 2017 11:03:44 +0100 Thomas Huth thuth@redhat.com wrote:
On 22.03.2017 10:08, Markus Armbruster wrote:
Are we now ready to accept a simple & stupid patch that actually helps users, say letting boards that care declare minimum and maximum RAM size? And make PC reject RAM size less than 1MiB, even though "someone" might conceivably have firmware that works with less?
I'd say enforce a minimum RAM size on the normal "pc" and "q35" machine, but still allow smaller sizes on the "isapc" machine. So if "someone" comes around and claims to have a legacy firmware that wants less memory than 1MiB, just point them to the isapc machine. Just my 0.02 €.
We can print warning that minimum size will be enforced in 1-2 releases since 2.9/10 and when it's enforced send users that need less to an old qemu version.
I did not think that QEMU ever worked on an x86 machine with less than 1MiB of ram. Certainly SeaBIOS has never supported it. I'm pretty sure that Bochs BIOS (qemu < v0.12) did not support it either - its ram detection system assumed a minimum of 1MiB.
My $.02 - I agree with Markus - place a minimum limit of 1MiB on x86 machines. If someone is writing a new firmware, updating QEMU to change the limit is the very least of their concerns.
-Kevin
-
David Hildenbrand -
Gerd Hoffmann -
Igor Mammedov -
Kevin O'Connor -
Markus Armbruster -
Thomas Huth