More than 4 network devices with HVM?

[Tom <tom-42@xxxxxx>]

Hi experts,

I tried to start a HVM qcow2 under Xen 4.20.1-pre (packaged for Debian 
13), which uses up to 6 vmxnet3 network devices.
The VM refuses to start:

> libxl: error: libxl_dm.c:3387:device_model_spawn_outcome: Domain 57:domain 57 
> device model: spawn failed (rc=-3)
> libxl: error: libxl_dm.c:3607:device_model_postconfig_done: Domain 57:Post DM 
> startup configs failed, rc=-3
> libxl: error: libxl_create.c:1957:domcreate_devmodel_started: Domain 57:device 
> model did not start: -3
> libxl: error: libxl_aoutils.c:646:libxl__kill_xs_path: Device Model already 
> exited
> libxl: error: libxl_xshelp.c:206:libxl__xs_read_mandatory: xenstore read 
> failed: `/libxl/57/type': No such file or directory
> libxl: warning: libxl_dom.c:49:libxl__domain_type: unable to get domain type 
> for domid=57, assuming HVM
> libxl: error: libxl_domain.c:1616:domain_destroy_domid_cb: Domain 
> 57:xc_domain_destroy failed: No such process
> libxl: error: libxl_domain.c:1133:domain_destroy_callback: Domain 57:Unable to 
> destroy guest
> libxl: error: libxl_domain.c:1060:domain_destroy_cb: Domain 57:Destruction of 
> domain failed
I remember patching qemu 7.2 years ago with a dirty hack found in the 
net, allocating more memory during VM startup, which was working fine 
for Xen version 4.14 (Debian 12) at this time.
Just for the curious, the qemu patch:

> --- a/hw/i386/xen/xen-hvm.c
> +++ b/hw/i386/xen/xen-hvm.c
> @@ -266,6 +266,7 @@
>      unsigned long nr_pfn;
>      xen_pfn_t *pfn_list;
>      int i;
> +    xc_dominfo_t info;
>
>      if (runstate_check(RUN_STATE_INMIGRATE)) {
>          /* RAM already populated in Xen */
> @@ -288,6 +289,17 @@
>          pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
>      }
>
> +    /* increase maxmem before calling xc_domain_populate_physmap */
> +    /* 
> https://patchwork.ozlabs.org/project/qemu-devel/patch/alpine.DEB.2.02.1411251742280.14135@xxxxxxxxxxxxxxxxxxxxxxx/
>  */
> +    if ((xc_domain_getinfo(xen_xc, xen_domid, 1, &info) != 1)
> +         || (info.domid != xen_domid)) {
> +        hw_error("xc_domain_getinfo failed");
> +    }
> +    if (xc_domain_setmaxmem(xen_xc, xen_domid, info.max_memkb +
> +                            (nr_pfn * XC_PAGE_SIZE / 1024)) < 0) {
> +        hw_error("xc_domain_setmaxmem failed");
> +    }
> +
>      if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, 
> pfn_list)) {
>          error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
>                     ram_addr);
That same patch is not working with qemu 10 and Xen 4.20: some old 
interfaces were removed, and the Xen parts in the source were 
reorganized quite a bit. This is probably a good thing, so let's forget 
about the dirty patch above. ;-)

Am I doing something wrong in the first place? Is there a solution to 
the "bug", not allowing more than 4 vmxnet3 per HVM? Any chance to get a 
HVM with 5+ network interfaces started on Xen 4.20?
Kind regards
Tom