Ranconfig failure between VT-d and HVM

Ranconfig failure between VT-d and HVM

[Andrew Cooper]

https://gitlab.com/xen-project/people/andyhhp/xen/-/jobs/2355562119

ld: prelink.o: in function `vtd_setup':
drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): undefined
reference to `opt_hap_2mb'
drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): relocation
truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_2mb'
ld: drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): undefined
reference to `opt_hap_1gb'
drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): relocation
truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_1gb'
ld: ./.xen-syms.0: hidden symbol `opt_hap_2mb' isn't defined
ld: final link failed: bad value
make[3]: *** [arch/x86/Makefile:137: xen-syms] Error 1

It's really not clear why this has first show up now.  There has been a
huge quantity of noise in Gitlab CI for the past 3 weeks with ARM being
offline, but I can't see anything relevant in the large number of IOMMU
changes in that period.

Anyway, the issue is simple.  It's vtd_ept_page_compatible() in a
!CONFIG_HVM build.

~Andrew

Re: Ranconfig failure between VT-d and HVM

[Andrew Cooper]

On 20/04/2022 18:49, Andrew Cooper wrote:
> https://gitlab.com/xen-project/people/andyhhp/xen/-/jobs/2355562119
>
> ld: prelink.o: in function `vtd_setup':
> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): undefined
> reference to `opt_hap_2mb'
> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): relocation
> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_2mb'
> ld: drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): undefined
> reference to `opt_hap_1gb'
> drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): relocation
> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_1gb'
> ld: ./.xen-syms.0: hidden symbol `opt_hap_2mb' isn't defined
> ld: final link failed: bad value
> make[3]: *** [arch/x86/Makefile:137: xen-syms] Error 1
>
> It's really not clear why this has first show up now.  There has been a
> huge quantity of noise in Gitlab CI for the past 3 weeks with ARM being
> offline, but I can't see anything relevant in the large number of IOMMU
> changes in that period.
>
> Anyway, the issue is simple.  It's vtd_ept_page_compatible() in a
> !CONFIG_HVM build.

And just after sending, I realised it's not an IOMMU change.

commit c479415610f0448bb07d39185e72ef36e08ebec0
Author: Jan Beulich <jbeulich@suse.com>
Date:   Fri Apr 8 13:50:29 2022

    x86/P2M: p2m.c is HVM-only


CI disputes the claim.

~Andrew

Re: Ranconfig failure between VT-d and HVM

[Jan Beulich]

On 20.04.2022 20:01, Andrew Cooper wrote:
> On 20/04/2022 18:49, Andrew Cooper wrote:
>> https://gitlab.com/xen-project/people/andyhhp/xen/-/jobs/2355562119
>>
>> ld: prelink.o: in function `vtd_setup':
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): undefined
>> reference to `opt_hap_2mb'
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): relocation
>> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_2mb'
>> ld: drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): undefined
>> reference to `opt_hap_1gb'
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): relocation
>> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_1gb'
>> ld: ./.xen-syms.0: hidden symbol `opt_hap_2mb' isn't defined
>> ld: final link failed: bad value
>> make[3]: *** [arch/x86/Makefile:137: xen-syms] Error 1
>>
>> It's really not clear why this has first show up now.  There has been a
>> huge quantity of noise in Gitlab CI for the past 3 weeks with ARM being
>> offline, but I can't see anything relevant in the large number of IOMMU
>> changes in that period.
>>
>> Anyway, the issue is simple.  It's vtd_ept_page_compatible() in a
>> !CONFIG_HVM build.
> 
> And just after sending, I realised it's not an IOMMU change.
> 
> commit c479415610f0448bb07d39185e72ef36e08ebec0
> Author: Jan Beulich <jbeulich@suse.com>
> Date:   Fri Apr 8 13:50:29 2022
> 
>     x86/P2M: p2m.c is HVM-only
> 
> 
> CI disputes the claim.

I'm afraid it can't be this simple, as I've been routinely doing !HVM
build for quite some time. There must be some further interaction,
albeit potentially with not-yet-committed further changes of mine. I'll
investigate.

Jan

Re: Ranconfig failure between VT-d and HVM

[Jan Beulich]

On 20.04.2022 20:01, Andrew Cooper wrote:
> On 20/04/2022 18:49, Andrew Cooper wrote:
>> https://gitlab.com/xen-project/people/andyhhp/xen/-/jobs/2355562119
>>
>> ld: prelink.o: in function `vtd_setup':
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): undefined
>> reference to `opt_hap_2mb'
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x219f6): relocation
>> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_2mb'
>> ld: drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): undefined
>> reference to `opt_hap_1gb'
>> drivers/passthrough/vtd/iommu.c:(.init.text+0x21a23): relocation
>> truncated to fit: R_X86_64_PC32 against undefined symbol `opt_hap_1gb'
>> ld: ./.xen-syms.0: hidden symbol `opt_hap_2mb' isn't defined
>> ld: final link failed: bad value
>> make[3]: *** [arch/x86/Makefile:137: xen-syms] Error 1
>>
>> It's really not clear why this has first show up now.  There has been a
>> huge quantity of noise in Gitlab CI for the past 3 weeks with ARM being
>> offline, but I can't see anything relevant in the large number of IOMMU
>> changes in that period.
>>
>> Anyway, the issue is simple.  It's vtd_ept_page_compatible() in a
>> !CONFIG_HVM build.
> 
> And just after sending, I realised it's not an IOMMU change.
> 
> commit c479415610f0448bb07d39185e72ef36e08ebec0
> Author: Jan Beulich <jbeulich@suse.com>
> Date:   Fri Apr 8 13:50:29 2022
> 
>     x86/P2M: p2m.c is HVM-only
> 
> 
> CI disputes the claim.

Just as an intermediate result: In my !HVM build (with gcc11) this

        if ( !vtd_ept_page_compatible(iommu) )
            clear_iommu_hap_pt_share();

simply compiles to nothing, which I assume is the result of
clear_iommu_hap_pt_share() doing nothing in that case. Hence no
reference to opt_hap_{2mb,1gb}.

Jan