Re: [PATCH v6 1/6] PCI/IOV: Restore VF resizable BAR state after reset

[Ilpo Järvinen]

On Thu, 20 Mar 2025, Michał Winiarski wrote:

> Similar to regular resizable BAR, VF BAR can also be resized, e.g. by
> the system firmware or the PCI subsystem itself.
> 
> Add the capability ID and restore it as a part of IOV state.
>
> See PCIe r4.0, sec 9.3.7.4.

Usually it's best o refer to latest gen doc, the section number seems to 
be the same also in r6.2.

This didn't refer to spec section that specified VF Rebar ext capability
(7.8.7) though. I think it should and it would also be good to mention the 
capability layout is the same as with the rebar cap.

> Signed-off-by: Michał Winiarski <michal.winiarski@xxxxxxxxx>
> Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>
> Reviewed-by: Christian König <christian.koenig@xxxxxxx>
> ---
>  drivers/pci/iov.c             | 30 +++++++++++++++++++++++++++++-
>  drivers/pci/pci.h             |  1 +
>  include/uapi/linux/pci_regs.h |  1 +
>  3 files changed, 31 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c
> index 121540f57d4bf..bf95387993cd5 100644
> --- a/drivers/pci/iov.c
> +++ b/drivers/pci/iov.c
> @@ -7,6 +7,7 @@
>   * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@xxxxxxxxx>
>   */
>  
> +#include <linux/bitfield.h>
>  #include <linux/pci.h>
>  #include <linux/slab.h>
>  #include <linux/export.h>
> @@ -830,6 +831,7 @@ static int sriov_init(struct pci_dev *dev, int pos)
>  	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
>  	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
>  		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
> +	iov->vf_rebar_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_VF_REBAR);
>  
>  	if (pdev)
>  		iov->dev = pci_dev_get(pdev);
> @@ -868,6 +870,30 @@ static void sriov_release(struct pci_dev *dev)
>  	dev->sriov = NULL;
>  }
>  
> +static void sriov_restore_vf_rebar_state(struct pci_dev *dev)
> +{
> +	unsigned int pos, nbars, i;
> +	u32 ctrl;
> +
> +	pos = dev->sriov->vf_rebar_cap;
> +	if (!pos)
> +		return;
> +
> +	pci_read_config_dword(dev, pos + PCI_REBAR_CTRL, &ctrl);
> +	nbars = FIELD_GET(PCI_REBAR_CTRL_NBAR_MASK, ctrl);
> +
> +	for (i = 0; i < nbars; i++, pos += 8) {
> +		int bar_idx, size;
> +
> +		pci_read_config_dword(dev, pos + PCI_REBAR_CTRL, &ctrl);
> +		bar_idx = FIELD_GET(PCI_REBAR_CTRL_BAR_IDX, ctrl);
> +		size = pci_rebar_bytes_to_size(dev->sriov->barsz[bar_idx]);
> +		ctrl &= ~PCI_REBAR_CTRL_BAR_SIZE;
> +		ctrl |= FIELD_PREP(PCI_REBAR_CTRL_BAR_SIZE, size);
> +		pci_write_config_dword(dev, pos + PCI_REBAR_CTRL, ctrl);

I started to wonder if we'd still want to have the VF Rebar ones in 
uapi/linux/pci_regs.h (despite the same capability layout):

/*
 * PCI Resizable BAR and PCI VF Resizable BAR extended capabilities have 
 * the same layout of fields.
 */
#define PCI_VF_REBAR_CTRL		PCI_REBAR_CTRL
#define PCI_VF_REBAR_CTRL_BAR_IDX	PCI_REBAR_CTRL_BAR_IDX
etc.

as then it would be possible grep to pick up only the relevant lines.

I'd not duplicate _SHIFT defines though. FIELD_PREP/GET() in general does 
not need _SHIFT defines at all and they are just duplicated information.

> +	}
> +}
> +
>  static void sriov_restore_state(struct pci_dev *dev)
>  {
>  	int i;
> @@ -1027,8 +1053,10 @@ resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
>   */
>  void pci_restore_iov_state(struct pci_dev *dev)
>  {
> -	if (dev->is_physfn)
> +	if (dev->is_physfn) {
> +		sriov_restore_vf_rebar_state(dev);
>  		sriov_restore_state(dev);
> +	}
>  }
>  
>  /**
> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
> index b81e99cd4b62a..adc54bb2c8b34 100644
> --- a/drivers/pci/pci.h
> +++ b/drivers/pci/pci.h
> @@ -482,6 +482,7 @@ struct pci_sriov {
>  	u16		subsystem_vendor; /* VF subsystem vendor */
>  	u16		subsystem_device; /* VF subsystem device */
>  	resource_size_t	barsz[PCI_SRIOV_NUM_BARS];	/* VF BAR size */
> +	u16		vf_rebar_cap;	/* VF Resizable BAR capability offset */
>  	bool		drivers_autoprobe; /* Auto probing of VFs by driver */
>  };
>  
> diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
> index ba326710f9c8b..bb2a334e50386 100644
> --- a/include/uapi/linux/pci_regs.h
> +++ b/include/uapi/linux/pci_regs.h
> @@ -745,6 +745,7 @@
>  #define PCI_EXT_CAP_ID_L1SS	0x1E	/* L1 PM Substates */
>  #define PCI_EXT_CAP_ID_PTM	0x1F	/* Precision Time Measurement */
>  #define PCI_EXT_CAP_ID_DVSEC	0x23	/* Designated Vendor-Specific */
> +#define PCI_EXT_CAP_ID_VF_REBAR 0x24	/* VF Resizable BAR */
>  #define PCI_EXT_CAP_ID_DLF	0x25	/* Data Link Feature */
>  #define PCI_EXT_CAP_ID_PL_16GT	0x26	/* Physical Layer 16.0 GT/s */
>  #define PCI_EXT_CAP_ID_NPEM	0x29	/* Native PCIe Enclosure Management */

Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx>

-- 
 i.