From: Ross Lagerwall <[email protected]>

When the PCI functions are created, Xen is informed about them and
caches the number of MSI-X entries each function has.  However, the
number of MSI-X entries is not set until after the hardware has been
configured and the VFs have been started. This prevents
PCI-passthrough from working because Xen rejects mapping MSI-X
interrupts to domains because it thinks the MSI-X interrupts don't
exist.

Fix this by moving the call to pci_enable_sriov() later so that the
number of MSI-X entries is set correctly in hardware by the time Xen
reads it.

Cc: Jesse Brandeburg <[email protected]>
Cc: Tony Nguyen <[email protected]>
Signed-off-by: Ross Lagerwall <[email protected]>
Signed-off-by: Javi Merino <[email protected]>
---

I'm unsure about the error path if `pci_enable_sriov()` fails.  Do we
have to undo what `ice_start_vfs()` has started?  I can see that
`ice_start_vfs()` has a teardown at the end, so maybe we need the same
code if `pci_enable_sriov()` fails?

 drivers/net/ethernet/intel/ice/ice_sriov.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_sriov.c 
b/drivers/net/ethernet/intel/ice/ice_sriov.c
index a94a1c48c3de..8a9c8a2fe834 100644
--- a/drivers/net/ethernet/intel/ice/ice_sriov.c
+++ b/drivers/net/ethernet/intel/ice/ice_sriov.c
@@ -878,24 +878,20 @@ static int ice_ena_vfs(struct ice_pf *pf, u16 num_vfs)
        set_bit(ICE_OICR_INTR_DIS, pf->state);
        ice_flush(hw);
 
-       ret = pci_enable_sriov(pf->pdev, num_vfs);
-       if (ret)
-               goto err_unroll_intr;
-
        mutex_lock(&pf->vfs.table_lock);
 
        ret = ice_set_per_vf_res(pf, num_vfs);
        if (ret) {
                dev_err(dev, "Not enough resources for %d VFs, err %d. Try with 
fewer number of VFs\n",
                        num_vfs, ret);
-               goto err_unroll_sriov;
+               goto err_unroll_intr;
        }
 
        ret = ice_create_vf_entries(pf, num_vfs);
        if (ret) {
                dev_err(dev, "Failed to allocate VF entries for %d VFs\n",
                        num_vfs);
-               goto err_unroll_sriov;
+               goto err_unroll_intr;
        }
 
        ice_eswitch_reserve_cp_queues(pf, num_vfs);
@@ -906,6 +902,10 @@ static int ice_ena_vfs(struct ice_pf *pf, u16 num_vfs)
                goto err_unroll_vf_entries;
        }
 
+       ret = pci_enable_sriov(pf->pdev, num_vfs);
+       if (ret)
+               goto err_unroll_vf_entries;
+
        clear_bit(ICE_VF_DIS, pf->state);
 
        /* rearm global interrupts */
@@ -918,10 +918,8 @@ static int ice_ena_vfs(struct ice_pf *pf, u16 num_vfs)
 
 err_unroll_vf_entries:
        ice_free_vf_entries(pf);
-err_unroll_sriov:
-       mutex_unlock(&pf->vfs.table_lock);
-       pci_disable_sriov(pf->pdev);
 err_unroll_intr:
+       mutex_unlock(&pf->vfs.table_lock);
        /* rearm interrupts here */
        ice_irq_dynamic_ena(hw, NULL, NULL);
        clear_bit(ICE_OICR_INTR_DIS, pf->state);
-- 
2.43.1

Reply via email to