[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240905151225.GA387183@bhelgaas>
Date: Thu, 5 Sep 2024 10:12:25 -0500
From: Bjorn Helgaas <helgaas@...nel.org>
To: Wei Huang <wei.huang2@....com>
Cc: linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
linux-doc@...r.kernel.org, netdev@...r.kernel.org,
Jonathan.Cameron@...wei.com, corbet@....net, davem@...emloft.net,
edumazet@...gle.com, kuba@...nel.org, pabeni@...hat.com,
alex.williamson@...hat.com, gospo@...adcom.com,
michael.chan@...adcom.com, ajit.khaparde@...adcom.com,
somnath.kotur@...adcom.com, andrew.gospodarek@...adcom.com,
manoj.panicker2@....com, Eric.VanTassell@....com,
vadim.fedorenko@...ux.dev, horms@...nel.org, bagasdotme@...il.com,
bhelgaas@...gle.com, lukas@...ner.de, paul.e.luse@...el.com,
jing2.liu@...el.com
Subject: Re: [PATCH V4 03/12] PCI/TPH: Add pcie_tph_modes() to query TPH modes
On Thu, Sep 05, 2024 at 09:46:44AM -0500, Wei Huang wrote:
> On 9/4/24 14:40, Bjorn Helgaas wrote:
> > On Thu, Aug 22, 2024 at 03:41:11PM -0500, Wei Huang wrote:
> >> Add pcie_tph_modes() to allow drivers to query the TPH modes supported
> >> by an endpoint device, as reported in the TPH Requester Capability
> >> register. The modes are reported as a bitmask and current supported
> >> modes include:
> >>
> >> - PCI_TPH_CAP_NO_ST: NO ST Mode Supported
> >> - PCI_TPH_CAP_INT_VEC: Interrupt Vector Mode Supported
> >> - PCI_TPH_CAP_DEV_SPEC: Device Specific Mode Supported
> >
> >> + * pcie_tph_modes - Get the ST modes supported by device
> >> + * @pdev: PCI device
> >> + *
> >> + * Returns a bitmask with all TPH modes supported by a device as shown in the
> >> + * TPH capability register. Current supported modes include:
> >> + * PCI_TPH_CAP_NO_ST - NO ST Mode Supported
> >> + * PCI_TPH_CAP_INT_VEC - Interrupt Vector Mode Supported
> >> + * PCI_TPH_CAP_DEV_SPEC - Device Specific Mode Supported
> >> + *
> >> + * Return: 0 when TPH is not supported, otherwise bitmask of supported modes
> >> + */
> >> +int pcie_tph_modes(struct pci_dev *pdev)
> >> +{
> >> + if (!pdev->tph_cap)
> >> + return 0;
> >> +
> >> + return get_st_modes(pdev);
> >> +}
> >> +EXPORT_SYMBOL(pcie_tph_modes);
> >
> > I'm not sure I see the need for pcie_tph_modes(). The new bnxt code
> > looks like this:
> >
> > bnxt_request_irq
> > if (pcie_tph_modes(bp->pdev) & PCI_TPH_CAP_INT_VEC)
> > rc = pcie_enable_tph(bp->pdev, PCI_TPH_CAP_INT_VEC);
> >
> > What is the advantage of this over just this?
> >
> > bnxt_request_irq
> > rc = pcie_enable_tph(bp->pdev, PCI_TPH_CAP_INT_VEC);
> >
> > It seems like drivers could just ask for what they want since
> > pcie_enable_tph() has to verify support for it anyway. If that fails,
> > the driver can fall back to another mode.
>
> I can get rid of pcie_tph_modes() if unnecessary.
>
> The design logic was that a driver can be used on various devices from
> the same company. Some of these devices might not be TPH capable. So
> instead of using trial-and-error (i.e. try INT_VEC ==> DEV_SPEC ==> give
> up), we provide a way for the driver to query the device TPH
> capabilities and pick a mode explicitly. IMO the code will be a bit cleaner.
>
> > Returning a bitmask of supported modes might be useful if the driver
> > could combine them, but IIUC the modes are all mutually exclusive, so
> > the driver can't request a combination of them.
>
> In the real world cases I saw, this is true. In the spec I didn't find
> that these bits are mutually exclusive though.
A device may advertise *support* for multiple modes in TPH Requester
Capability, of course.
What I meant by "driver can't request a combination" is that we can
only *select* one of them via the ST Mode select in TPH Requester
Control.
Powered by blists - more mailing lists