Re: [PATCH v3 4/4] mshv: Handle insufficient root memory hypervisor statuses

[Wei Liu]

On Fri, Feb 06, 2026 at 06:54:55PM +0000, Michael Kelley wrote:
> From: Stanislav Kinsburskii <skinsburskii@xxxxxxxxxxxxxxxxxxx> Sent: Thursday, February 5, 2026 10:42 AM
> > To: kys@xxxxxxxxxxxxx; haiyangz@xxxxxxxxxxxxx; wei.liu@xxxxxxxxxx;
> > decui@xxxxxxxxxxxxx; longli@xxxxxxxxxxxxx
> > Cc: linux-hyperv@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> > Subject: [PATCH v3 4/4] mshv: Handle insufficient root memory hypervisor statuses
> > 
> > When creating guest partition objects, the hypervisor may fail to
> > allocate root partition pages and return an insufficient memory status.
> > In this case, deposit memory using the root partition ID instead.
> > 
> > Note: This error should never occur in a guest of L1VH partition context.
> > 
> > Signed-off-by: Stanislav Kinsburskii <skinsburskii@xxxxxxxxxxxxxxxxxxx>
> > ---
> >  drivers/hv/hv_common.c      |    2 +
> >  drivers/hv/hv_proc.c        |   14 ++++++++++
> >  include/hyperv/hvgdk_mini.h |   58 ++++++++++++++++++++++---------------------
> >  3 files changed, 46 insertions(+), 28 deletions(-)
> > 
> > diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c
> > index f20596276662..6b67ac616789 100644
> > --- a/drivers/hv/hv_common.c
> > +++ b/drivers/hv/hv_common.c
> > @@ -794,6 +794,8 @@ static const struct hv_status_info hv_status_infos[] = {
> >  	_STATUS_INFO(HV_STATUS_PROPERTY_VALUE_OUT_OF_RANGE,	-EIO),
> >  	_STATUS_INFO(HV_STATUS_INSUFFICIENT_MEMORY,		-ENOMEM),
> >  	_STATUS_INFO(HV_STATUS_INSUFFICIENT_CONTIGUOUS_MEMORY,	-ENOMEM),
> > +	_STATUS_INFO(HV_STATUS_INSUFFICIENT_ROOT_MEMORY,	-ENOMEM),
> > +	_STATUS_INFO(HV_STATUS_INSUFFICIENT_CONTIGUOUS_ROOT_MEMORY, 	-ENOMEM),
> >  	_STATUS_INFO(HV_STATUS_INVALID_PARTITION_ID,		-EINVAL),
> >  	_STATUS_INFO(HV_STATUS_INVALID_VP_INDEX,		-EINVAL),
> >  	_STATUS_INFO(HV_STATUS_NOT_FOUND,			-EIO),
> > diff --git a/drivers/hv/hv_proc.c b/drivers/hv/hv_proc.c
> > index 181f6d02bce3..5f4fd9c3231c 100644
> > --- a/drivers/hv/hv_proc.c
> > +++ b/drivers/hv/hv_proc.c
> > @@ -121,6 +121,18 @@ int hv_deposit_memory_node(int node, u64 partition_id,
> >  	case HV_STATUS_INSUFFICIENT_CONTIGUOUS_MEMORY:
> >  		num_pages = HV_MAX_CONTIGUOUS_ALLOCATION_PAGES;
> >  		break;
> > +
> > +	case HV_STATUS_INSUFFICIENT_CONTIGUOUS_ROOT_MEMORY:
> > +		num_pages = HV_MAX_CONTIGUOUS_ALLOCATION_PAGES;
> > +		fallthrough;
> > +	case HV_STATUS_INSUFFICIENT_ROOT_MEMORY:
> > +		if (!hv_root_partition()) {
> > +			hv_status_err(hv_status, "Unexpected root memory deposit\n");
> > +			return -ENOMEM;
> > +		}
> > +		partition_id = HV_PARTITION_ID_SELF;
> > +		break;
> > +
> 
> Per the discussion in v1 of this patch set, if the number of pages that should be
> deposited in a particular situation is different from what this function provides,
> the fallback is to use hv_call_deposit_pages() directly. From what I see, there's
> only one such fallback case after a hypercall failure -- in hv_do_map_gpa_hcall().
> The other uses of hv_call_deposit_pages() are initial deposits when creating a
> VP or partition.
> 
> But if hv_call_deposit_pages() is used directly, the logic added here to detect
> insufficient root memory and deposit to HV_PARTITION_ID_SELF isn't applied.
> So if the hypercall in hv_do_map_gpa_hcall() fails with insufficient root
> memory, the deposit is done to the wrong partition ID. If that case can
> actually happen, then some additional logic is needed in
> hv_do_map_gpa_hcall() to handle it. Or there needs to be a fallback
> function that contains the logic.

Stanislav, how about this comment? Please submit a follow-up patch if
necessary.

Wei

> 
> Other than that, everything else in this patch set looks good to me.
> 
> Michael
> 
> >  	default:
> >  		hv_status_err(hv_status, "Unexpected!\n");
> >  		return -ENOMEM;