Re: [PATCH net-next] net/mlx5: Add reset_prepare and reset_done callbacks

[Moshe Shemesh <moshe@nvidia.com>]

On 10/14/2026 9:38 AM, Narcisa Vasile wrote:
> On Sun, Oct 11, 2026 at 12:48:09PM +0300, Tariq Toukan wrote:
>> From: Akiva Goldberger <agoldberger@nvidia.com>
>>
>> Currently, the mlx5_core driver registers only the error_detected,
>> slot_reset and resume PCI error handlers. Add the reset_prepare and
>> reset_done callbacks, which are triggered by the PCI core before and
>> after a PCI reset, respectively. These callbacks let mlx5 quiesce and
>> restore device resources and commands around a PCI reset, so the driver
>> no longer relies on the reset happening while the device is otherwise
>> idle.
>>
>> Additionally, add a device flag to indicate when a PCI reset is in
>> progress to serialize PCI reset flows with other reset flows, namely FW
>> reset. This flag also indicates to the reset_prepare and reset_done
>> callbacks when to do a full unload/reload of the driver. When the
>> callbacks are triggered by the hot-reset FW reset method, they leave the
>> teardown and reload to the FW reset.
>>
>> Signed-off-by: Akiva Goldberger <agoldberger@nvidia.com>
>> Reviewed-by: Moshe Shemesh <moshe@nvidia.com>
>> Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
>> ---
>>
>> +static void mlx5_pci_reset_prepare(struct pci_dev *pdev)
>> +{
>> +     mlx5_drain_health_wq(dev);
>> +     mlx5_unload_one(dev, false);
>> +     mlx5_pci_disable_device(dev);
> This disables bus-mastering, after which pci_save_state snapshots it in
> the disabled state. reset_done() re-enables it but the snapshot remains
> stale.
> 

Good point, will send v2.
Thanks.

> If a FW error later triggers the sw recovery through mlx5_pci_slot_reset(),
> the stale snapshot is re-applied with pci_restore_state().
> Would this disable bus mastering again and block the driver recovery?
> Should pci_set_master(pdev); and pci_restore_state(pdev); be reordered
> in mlx5_pci_slot_reset()?
>> +}
>> +
>> +static void mlx5_pci_reset_done(struct pci_dev *pdev)
>> +{
>> +     err = mlx5_pci_enable_device(dev);
>> +     if (err) {
>> +             mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed, err = %d\n",
>> +                           __func__, err);
> nit: mlx5_core_err already prints __func__
>> +             goto clear_pci_reset_flag;
>> +     }