missing boot rom: is it really a fatal error?

missing boot rom: is it really a fatal error?

[Michael Tokarev]

Hi!

In old good world ;), there was qemu which didn't require boot roms to be present
for all devices for which bootrom file is defined, missing rom was just a warning.
But this changed in 2014, 9 years ago, with this commit:

commit 178e785fb4507ec3462dc772bbe08303416ece47
From: Marcel Apfelbaum <marcel.a@redhat.com>
Date: Mon, 27 Oct 2014 19:34:41 +0200
Subject: [PATCH] hw/pci: fixed error flow in pci_qdev_init

   Verify return code for pci_add_option_rom.

where inability to load rom file started being treated as an error.
Up until now I didn't even know about this change, until today when someone bugged
me about non-working qemu on debian, due to missing network boot roms (this a
packaging issue due to me being unaware of the above change).

What is the reason to require boot roms to be present and throw an error if not?

I'm about to revert that old change on debian, to make it just a warning instead
of an error (the code is different now, but the same principle applies), - because
I dislike dependencies which are useless 99.9% of the time and are trivial to
install when actually needed.

Thanks,

/mjt

Re: missing boot rom: is it really a fatal error?

[Thomas Huth]

On 07/05/2023 19.56, Michael Tokarev wrote:
> Hi!
> 
> In old good world ;), there was qemu which didn't require boot roms to be 
> present
> for all devices for which bootrom file is defined, missing rom was just a 
> warning.
> But this changed in 2014, 9 years ago, with this commit:
> 
> commit 178e785fb4507ec3462dc772bbe08303416ece47
> From: Marcel Apfelbaum <marcel.a@redhat.com>

Marcel left Red Hat a while ago ... CC:-ing him with his current address - 
maybe he remembers why that fix was needed...

  Thomas


> Date: Mon, 27 Oct 2014 19:34:41 +0200
> Subject: [PATCH] hw/pci: fixed error flow in pci_qdev_init
> 
>    Verify return code for pci_add_option_rom.
> 
> where inability to load rom file started being treated as an error.
> Up until now I didn't even know about this change, until today when someone 
> bugged
> me about non-working qemu on debian, due to missing network boot roms (this a
> packaging issue due to me being unaware of the above change).
> 
> What is the reason to require boot roms to be present and throw an error if 
> not?
> 
> I'm about to revert that old change on debian, to make it just a warning 
> instead
> of an error (the code is different now, but the same principle applies), - 
> because
> I dislike dependencies which are useless 99.9% of the time and are trivial to
> install when actually needed.
> 
> Thanks,
> 
> /mjt
> 

Re: missing boot rom: is it really a fatal error?

[Michael S. Tsirkin]

On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
> Hi!
> 
> In old good world ;), there was qemu which didn't require boot roms to be present
> for all devices for which bootrom file is defined, missing rom was just a warning.
> But this changed in 2014, 9 years ago, with this commit:
> 
> commit 178e785fb4507ec3462dc772bbe08303416ece47
> From: Marcel Apfelbaum <marcel.a@redhat.com>
> Date: Mon, 27 Oct 2014 19:34:41 +0200
> Subject: [PATCH] hw/pci: fixed error flow in pci_qdev_init
> 
>   Verify return code for pci_add_option_rom.
> 
> where inability to load rom file started being treated as an error.
> Up until now I didn't even know about this change, until today when someone bugged
> me about non-working qemu on debian, due to missing network boot roms (this a
> packaging issue due to me being unaware of the above change).
> 
> What is the reason to require boot roms to be present and throw an error if not?
> 
> I'm about to revert that old change on debian, to make it just a warning instead
> of an error (the code is different now, but the same principle applies), - because
> I dislike dependencies which are useless 99.9% of the time and are trivial to
> install when actually needed.
> 
> Thanks,
> 
> /mjt
> 

I advise against it.
If you boot guest on a system with boot rom not installed you will not
be able to migrate to a system with boot rom installed.
why not? because we don't know how big to make the rom BAR.
And users will not discover until much much later after they have
painted themselves into a corner.

-- 
MST

Re: missing boot rom: is it really a fatal error?

[Markus Armbruster]

"Michael S. Tsirkin" <mst@redhat.com> writes:

> On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
>> Hi!
>> 
>> In old good world ;), there was qemu which didn't require boot roms to be present
>> for all devices for which bootrom file is defined, missing rom was just a warning.
>> But this changed in 2014, 9 years ago, with this commit:
>> 
>> commit 178e785fb4507ec3462dc772bbe08303416ece47
>> From: Marcel Apfelbaum <marcel.a@redhat.com>
>> Date: Mon, 27 Oct 2014 19:34:41 +0200
>> Subject: [PATCH] hw/pci: fixed error flow in pci_qdev_init
>> 
>>   Verify return code for pci_add_option_rom.
>> 
>> where inability to load rom file started being treated as an error.
>> Up until now I didn't even know about this change, until today when someone bugged
>> me about non-working qemu on debian, due to missing network boot roms (this a
>> packaging issue due to me being unaware of the above change).
>> 
>> What is the reason to require boot roms to be present and throw an error if not?
>> 
>> I'm about to revert that old change on debian, to make it just a warning instead
>> of an error (the code is different now, but the same principle applies), - because
>> I dislike dependencies which are useless 99.9% of the time and are trivial to
>> install when actually needed.
>> 
>> Thanks,
>> 
>> /mjt
>> 
>
> I advise against it.
> If you boot guest on a system with boot rom not installed you will not
> be able to migrate to a system with boot rom installed.
> why not? because we don't know how big to make the rom BAR.

For what it's worth: we know when property "romsize" is set.

> And users will not discover until much much later after they have
> painted themselves into a corner.

Re: missing boot rom: is it really a fatal error?

[Michael Tokarev]

08.05.2023 13:28, Michael S. Tsirkin wrote:
> On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
..

>> I'm about to revert that old change on debian, to make it just a warning instead
>> of an error (the code is different now, but the same principle applies), - because
>> I dislike dependencies which are useless 99.9% of the time and are trivial to
>> install when actually needed.
..
> I advise against it.
> If you boot guest on a system with boot rom not installed you will not
> be able to migrate to a system with boot rom installed.
> why not? because we don't know how big to make the rom BAR.
> And users will not discover until much much later after they have
> painted themselves into a corner.

Yes, I know about the migration. Actually there's an old bug report open
against debian qemu package, - the context is similar to the old bios128
vs bios256 thing in qemu upstream, - boot roms might change in size too.

In this context though, the talk is not about migration at all. The missing
dep is in Xen HVM qemu package, a xen-only build of qemu-system-i386. And
this one fails to start unless the boot roms are provided. It is not even
capable of migration to begin with :)

Thank you for the reminder, - very useful.

/mjt

Re: missing boot rom: is it really a fatal error?

[Michael S. Tsirkin]

On Mon, May 08, 2023 at 01:42:04PM +0300, Michael Tokarev wrote:
> 08.05.2023 13:28, Michael S. Tsirkin wrote:
> > On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
> ..
> 
> > > I'm about to revert that old change on debian, to make it just a warning instead
> > > of an error (the code is different now, but the same principle applies), - because
> > > I dislike dependencies which are useless 99.9% of the time and are trivial to
> > > install when actually needed.
> ..
> > I advise against it.
> > If you boot guest on a system with boot rom not installed you will not
> > be able to migrate to a system with boot rom installed.
> > why not? because we don't know how big to make the rom BAR.
> > And users will not discover until much much later after they have
> > painted themselves into a corner.
> 
> Yes, I know about the migration. Actually there's an old bug report open
> against debian qemu package, - the context is similar to the old bios128
> vs bios256 thing in qemu upstream, - boot roms might change in size too.
> 
> In this context though, the talk is not about migration at all. The missing
> dep is in Xen HVM qemu package, a xen-only build of qemu-system-i386. And
> this one fails to start unless the boot roms are provided. It is not even
> capable of migration to begin with :)
> 
> Thank you for the reminder, - very useful.
> 
> /mjt

I guess we decided we'd rather not handle reports from users about net
boot not working. It's true most users don't need net boot but then
that's true for most qemu functionality - 99% of users
probably need 1% of the functionality. It's just a different 1% for each
user...

-- 
MST

Re: missing boot rom: is it really a fatal error?

[Warner Losh]

[-- Attachment #1: Type: text/plain, Size: 1960 bytes --]

On Mon, May 8, 2023 at 5:48 AM Michael S. Tsirkin <mst@redhat.com> wrote:

> On Mon, May 08, 2023 at 01:42:04PM +0300, Michael Tokarev wrote:
> > 08.05.2023 13:28, Michael S. Tsirkin wrote:
> > > On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
> > ..
> >
> > > > I'm about to revert that old change on debian, to make it just a
> warning instead
> > > > of an error (the code is different now, but the same principle
> applies), - because
> > > > I dislike dependencies which are useless 99.9% of the time and are
> trivial to
> > > > install when actually needed.
> > ..
> > > I advise against it.
> > > If you boot guest on a system with boot rom not installed you will not
> > > be able to migrate to a system with boot rom installed.
> > > why not? because we don't know how big to make the rom BAR.
> > > And users will not discover until much much later after they have
> > > painted themselves into a corner.
> >
> > Yes, I know about the migration. Actually there's an old bug report open
> > against debian qemu package, - the context is similar to the old bios128
> > vs bios256 thing in qemu upstream, - boot roms might change in size too.
> >
> > In this context though, the talk is not about migration at all. The
> missing
> > dep is in Xen HVM qemu package, a xen-only build of qemu-system-i386. And
> > this one fails to start unless the boot roms are provided. It is not even
> > capable of migration to begin with :)
> >
> > Thank you for the reminder, - very useful.
> >
> > /mjt
>
> I guess we decided we'd rather not handle reports from users about net
> boot not working. It's true most users don't need net boot but then
> that's true for most qemu functionality - 99% of users
> probably need 1% of the functionality. It's just a different 1% for each
> user...
>

Yea, but to every user, their 1% is the most important thing ever...
Or so it seems some days :)

Warner

[-- Attachment #2: Type: text/html, Size: 2537 bytes --]

Re: missing boot rom: is it really a fatal error?

[Michael S. Tsirkin]

On Mon, May 15, 2023 at 10:02:43AM -0600, Warner Losh wrote:
> 
> 
> On Mon, May 8, 2023 at 5:48 AM Michael S. Tsirkin <mst@redhat.com> wrote:
> 
>     On Mon, May 08, 2023 at 01:42:04PM +0300, Michael Tokarev wrote:
>     > 08.05.2023 13:28, Michael S. Tsirkin wrote:
>     > > On Sun, May 07, 2023 at 08:56:23PM +0300, Michael Tokarev wrote:
>     > ..
>     >
>     > > > I'm about to revert that old change on debian, to make it just a
>     warning instead
>     > > > of an error (the code is different now, but the same principle
>     applies), - because
>     > > > I dislike dependencies which are useless 99.9% of the time and are
>     trivial to
>     > > > install when actually needed.
>     > ..
>     > > I advise against it.
>     > > If you boot guest on a system with boot rom not installed you will not
>     > > be able to migrate to a system with boot rom installed.
>     > > why not? because we don't know how big to make the rom BAR.
>     > > And users will not discover until much much later after they have
>     > > painted themselves into a corner.
>     >
>     > Yes, I know about the migration. Actually there's an old bug report open
>     > against debian qemu package, - the context is similar to the old bios128
>     > vs bios256 thing in qemu upstream, - boot roms might change in size too.
>     >
>     > In this context though, the talk is not about migration at all. The
>     missing
>     > dep is in Xen HVM qemu package, a xen-only build of qemu-system-i386. And
>     > this one fails to start unless the boot roms are provided. It is not even
>     > capable of migration to begin with :)
>     >
>     > Thank you for the reminder, - very useful.
>     >
>     > /mjt
> 
>     I guess we decided we'd rather not handle reports from users about net
>     boot not working. It's true most users don't need net boot but then
>     that's true for most qemu functionality - 99% of users
>     probably need 1% of the functionality. It's just a different 1% for each
>     user...
> 
> 
> Yea, but to every user, their 1% is the most important thing ever...
> Or so it seems some days :)
> 
> Warner

Yep. So I do not think we will be reverting this change in QEMU.

-- 
MST