Re: Aw: Re: Re: [help] Xen 4.14.5 on Devuan 4.0 Chimaera, regression from Xen 4.0.1

[Jan Beulich <jbeulich@xxxxxxxx>]

On 14.03.2023 16:11, Andrew Cooper wrote:
> On 14/03/2023 2:53 pm, Denis wrote:
>> On 14.03.2023 07:37; Jan Beulich wrote:
>>> On 14.03.2023 02:15, Denis wrote:
>>>> On 13.03.2023 10:36, Jan wrote
>>>>> On 10.03.2023 21:50, Denis wrote:
>>>>>> Should I test something else?
>>>>> ... there was no request for any further testing here, for the moment.
>>>> ah...sorry, going by "Would be nice to have this confirmed forthe system
>>>> in question, i.e. without Xen underneath Linux" I thought I could test
>>>> something which might help shed some light on all of this.
>>> Well, yes, that Linux-without-Xen test would still be useful to have
>>> results from. I didn't account for this in my earlier reply because
>>> I had asked for it before already, and I did take "something else"
>>> for meaning anything that might have turned up as useful from the new
>>> data you had provided.
>> What tests could I do or what info should I provide to help?
> 
> Can you please rebuild Xen with this patch:
> 
> diff --git a/xen/drivers/passthrough/amd/iommu_acpi.c
> b/xen/drivers/passthrough/amd/iommu_acpi.c
> index 2fdebd2d74c9..747eae25f56c 100644
> --- a/xen/drivers/passthrough/amd/iommu_acpi.c
> +++ b/xen/drivers/passthrough/amd/iommu_acpi.c
> @@ -1033,7 +1033,7 @@ static int __init parse_ivrs_table(struct
> acpi_table_header *table)
>      const struct acpi_ivrs_header *ivrs_block;
>      unsigned long length;
>      unsigned int apic;
> -    bool_t sb_ioapic = !iommu_intremap;
> +    bool_t sb_ioapic = 1;
>      int error = 0;
>  
>      BUG_ON(!table);
> 
> which should cause the behaviour to revert back to that of Xen 4.0.1 
> (i.e. it will fully ignore the checks relating to the southbridge ioapic).

Alternatively you may want to try the change below (I think I have now
convinced myself that the state change is still possible at this point
in time), with the intended effect of ...

> Confirm that with this, and booting Xen simply with `iommu=1` that full
> DMA remapping and interrupt remapping is considered active.

... DMA remapping active, but interrupt mapping off (i.e. matching
Linux behavior), without any overriding command line options.

Jan

AMD/IOMMU: allow DMA remapping to remain enabled when there's no southbridge 
IO-APIC

The original Linux commit that our respective code was derived from
isn't as heavyhanded as our cloned code: It only disables interrupt
remapping in such a case. Follow that model, noting that it is still
early enough to turn interrupt remapping off on its own.

Fixes: 06bbcaf48d09 ("AMD IOMMU: fail if there is no southbridge IO-APIC")
Reported-by: Denis <tachyon_gun@xxxxxx>
Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
Note that the alternative of disabling per-device interrupt remapping
is undesirable as per XSA-36, yet then again it may still be better than
turning off interrupt remapping altogether. Thoughts?

--- unstable.orig/xen/drivers/passthrough/amd/iommu_acpi.c
+++ unstable/xen/drivers/passthrough/amd/iommu_acpi.c
@@ -1183,7 +1183,7 @@ static int __init cf_check parse_ivrs_ta
     if ( !error && !sb_ioapic )
     {
         if ( amd_iommu_perdev_intremap )
-            error = -ENXIO;
+            iommu_intremap = iommu_intremap_off;
         printk("%sNo southbridge IO-APIC found in IVRS table\n",
                amd_iommu_perdev_intremap ? XENLOG_ERR : XENLOG_WARNING);
     }