RE: [PATCH net-next] net: mana: Add a driver for Microsoft Azure Network Adapter (MANA)

[Haiyang Zhang <haiyangz@microsoft.com>]

> -----Original Message-----
> From: Leon Romanovsky <leon@kernel.org>
> Sent: Wednesday, April 7, 2021 10:55 AM
> To: Haiyang Zhang <haiyangz@microsoft.com>
> Cc: Dexuan Cui <decui@microsoft.com>; davem@davemloft.net;
> kuba@kernel.org; KY Srinivasan <kys@microsoft.com>; Stephen Hemminger
> <sthemmin@microsoft.com>; wei.liu@kernel.org; Wei Liu
> <liuwe@microsoft.com>; netdev@vger.kernel.org; linux-
> kernel@vger.kernel.org; linux-hyperv@vger.kernel.org
> Subject: Re: [PATCH net-next] net: mana: Add a driver for Microsoft Azure
> Network Adapter (MANA)
> 
> On Wed, Apr 07, 2021 at 02:41:45PM +0000, Haiyang Zhang wrote:
> >
> >
> > > -----Original Message-----
> > > From: Leon Romanovsky <leon@kernel.org>
> > > Sent: Wednesday, April 7, 2021 8:51 AM
> > > To: Dexuan Cui <decui@microsoft.com>
> > > Cc: davem@davemloft.net; kuba@kernel.org; KY Srinivasan
> > > <kys@microsoft.com>; Haiyang Zhang <haiyangz@microsoft.com>;
> Stephen
> > > Hemminger <sthemmin@microsoft.com>; wei.liu@kernel.org; Wei Liu
> > > <liuwe@microsoft.com>; netdev@vger.kernel.org; linux-
> > > kernel@vger.kernel.org; linux-hyperv@vger.kernel.org
> > > Subject: Re: [PATCH net-next] net: mana: Add a driver for Microsoft
> > > Azure Network Adapter (MANA)
> > >
> > > On Wed, Apr 07, 2021 at 08:40:13AM +0000, Dexuan Cui wrote:
> > > > > From: Leon Romanovsky <leon@kernel.org>
> > > > > Sent: Wednesday, April 7, 2021 1:10 AM
> > > > >
> > > > > <...>
> > > > >
> > > > > > +int gdma_verify_vf_version(struct pci_dev *pdev) {
> > > > > > +	struct gdma_context *gc = pci_get_drvdata(pdev);
> > > > > > +	struct gdma_verify_ver_req req = { 0 };
> > > > > > +	struct gdma_verify_ver_resp resp = { 0 };
> > > > > > +	int err;
> > > > > > +
> > > > > > +	gdma_init_req_hdr(&req.hdr,
> GDMA_VERIFY_VF_DRIVER_VERSION,
> > > > > > +			  sizeof(req), sizeof(resp));
> > > > > > +
> > > > > > +	req.protocol_ver_min = GDMA_PROTOCOL_FIRST;
> > > > > > +	req.protocol_ver_max = GDMA_PROTOCOL_LAST;
> > > > > > +
> > > > > > +	err = gdma_send_request(gc, sizeof(req), &req, sizeof(resp),
> &resp);
> > > > > > +	if (err || resp.hdr.status) {
> > > > > > +		pr_err("VfVerifyVersionOutput: %d, status=0x%x\n",
> err,
> > > > > > +		       resp.hdr.status);
> > > > > > +		return -EPROTO;
> > > > > > +	}
> > > > > > +
> > > > > > +	return 0;
> > > > > > +}
> > > > >
> > > > > <...>
> > > > > > +	err = gdma_verify_vf_version(pdev);
> > > > > > +	if (err)
> > > > > > +		goto remove_irq;
> > > > >
> > > > > Will this VF driver be used in the guest VM? What will prevent
> > > > > from users
> > > to
> > > > > change it?
> > > > > I think that such version negotiation scheme is not allowed.
> > > >
> > > > Yes, the VF driver is expected to run in a Linux VM that runs on Azure.
> > > >
> > > > Currently gdma_verify_vf_version() just tells the PF driver that
> > > > the VF
> > > driver
> > > > is only able to support GDMA_PROTOCOL_V1, and want to use
> > > > GDMA_PROTOCOL_V1's message formats to talk to the PF driver later.
> > > >
> > > > enum {
> > > >         GDMA_PROTOCOL_UNDEFINED = 0,
> > > >         GDMA_PROTOCOL_V1 = 1,
> > > >         GDMA_PROTOCOL_FIRST = GDMA_PROTOCOL_V1,
> > > >         GDMA_PROTOCOL_LAST = GDMA_PROTOCOL_V1,
> > > >         GDMA_PROTOCOL_VALUE_MAX
> > > > };
> > > >
> > > > The PF driver is supposed to always support GDMA_PROTOCOL_V1, so I
> > > expect
> > > > here gdma_verify_vf_version() should succeed. If a user changes
> > > > the Linux
> > > VF
> > > > driver and try to use a protocol version not supported by the PF
> > > > driver,
> > > then
> > > > gdma_verify_vf_version() will fail; later, if the VF driver tries
> > > > to talk to the
> > > PF
> > > > driver using an unsupported message format, the PF driver will
> > > > return a
> > > failure.
> > >
> > > The worry is not for the current code, but for the future one when
> > > you will support v2, v3 e.t.c. First, your code will look like a
> > > spaghetti and second, users will try and mix vX with "unsupported"
> > > commands just for the fun.
> >
> > In the future, if the protocol version updated on the host side,
> > guests need to support different host versions because not all hosts
> > are updated (simultaneously). So this negotiation is necessary to know
> > the supported version, and decide the proper command version to use.
> 
> And how do other paravirtual drivers solve this negotiation scheme?

I saw some other drivers used version negotiation too, for example:

/**
 *  ixgbevf_negotiate_api_version_vf - Negotiate supported API version
 *  @hw: pointer to the HW structure
 *  @api: integer containing requested API version
 **/
static int ixgbevf_negotiate_api_version_vf(struct ixgbe_hw *hw, int api)
{

Thanks,
- Haiyang