On Thu, 2025-04-17 at 12:56 +0000, Vamsi Krishna Attunuru wrote:
>
>
> > -----Original Message-----
> > From: Philipp Stanner <pha...@mailbox.org>
> > Sent: Thursday, April 17, 2025 4:09 PM
> > To: Vamsi Krishna Attunuru <vattun...@marvell.com>; Philipp Stanner
> > <pha...@kernel.org>; Srujana Challa <scha...@marvell.com>; Michael
> > S.
> > Tsirkin <m...@redhat.com>; Jason Wang <jasow...@redhat.com>; Xuan
> > Zhuo <xuanz...@linux.alibaba.com>; Eugenio Pérez
> > <epere...@redhat.com>; Shijith Thotton <sthot...@marvell.com>; Dan
> > Carpenter <dan.carpen...@linaro.org>; Satha Koteswara Rao Kottidi
> > <skotesh...@marvell.com>
> > Cc: virtualizat...@lists.linux.dev; linux-kernel@vger.kernel.org
> > Subject: Re: [EXTERNAL] [PATCH] vdpa/octeon_ep: Use non-hybrid PCI
> > devres API
> >
> > On Thu, 2025-04-17 at 09: 02 +0000, Vamsi Krishna Attunuru wrote: >
> > > > > -----
> > Original Message----- > > From: Philipp Stanner
> > <phasta@ kernel. org> > >
> > Sent: Thursday, April 17, 2025 1: 32 PM > > To: Srujana
> >
> > On Thu, 2025-04-17 at 09:02 +0000, Vamsi Krishna Attunuru wrote:
> > >
> > >
> > > > -----Original Message-----
> > > > From: Philipp Stanner <pha...@kernel.org>
> > > > Sent: Thursday, April 17, 2025 1:32 PM
> > > > To: Srujana Challa <scha...@marvell.com>; Vamsi Krishna
> > > > Attunuru
> > > > <vattun...@marvell.com>; Michael S. Tsirkin <m...@redhat.com>;
> > > > Jason
> > > > Wang <jasow...@redhat.com>; Xuan Zhuo
> > <xuanz...@linux.alibaba.com>;
> > > > Eugenio Pérez <epere...@redhat.com>; Shijith Thotton
> > > > <sthot...@marvell.com>; Dan Carpenter
> > > > <dan.carpen...@linaro.org>;
> > > > Philipp Stanner <pha...@kernel.org>; Satha Koteswara Rao
> > > > Kottidi
> > > > <skotesh...@marvell.com>
> > > > Cc: virtualizat...@lists.linux.dev;
> > > > linux-kernel@vger.kernel.org
> > > > Subject: [EXTERNAL] [PATCH] vdpa/octeon_ep: Use non-hybrid PCI
> > > > devres API
> > > >
> > > > octeon enables its PCI device with pcim_enable_device(). This,
> > > > implicitly, switches the function pci_request_region() into
> > > > managed
> > > > mode, where it becomes a devres function. The PCI subsystem
> > > > wants to
> > > > remove this hybrid nature from its interfaces.
> > > > octeon enables its PCI device with pcim_enable_device(). This,
> > > > implicitly, switches the function pci_request_region() into
> > > > managed
> > > > mode, where it becomes a devres function.
> > > >
> > > > The PCI subsystem wants to remove this hybrid nature from its
> > > > interfaces. To do so, users of the aforementioned combination
> > > > of
> > > > functions must be ported to non-hybrid functions.
> > > >
> > > > Moreover, since both functions are already managed in this
> > > > driver,
> > > > the calls to
> > > > pci_release_region() are unnecessary.
> > > >
> > > > Remove the calls to pci_release_region().
> > > >
> > > > Replace the call to sometimes-managed pci_request_region() with
> > > > one
> > > > to the always-managed pcim_request_region().
> > >
> >
> > Hi,
> >
> > > Thanks you, Philipps, for the patch. The Octeon EP driver does
> > > not use
> > > managed calls for handling resource regions.
> > > This is because the PCIe PF function reduces its resources to
> > > share
> > > them with its VFs and later restores them to original size once
> > > the
> > > VFs are released. Due to this resource sharing requirement, the
> > > driver
> > > cannot use the always-managed request regions calls.
> >
> > so this would mean that the driver is already broken.
> > pci_request_region() in your driver is always-managed since you
> > call
> > pcim_enable_device(). Or am I missing something?
>
> Driver is not actually broken. Yes, pci_request_region is always
> managed in the driver due to pcim_enable_device().
> But inside pcim_request_region(), res->type is considered as
> PCIM_ADDR_DEVRES_TYPE_REGION and
> pcim_addr_resource_release() releases entire bar. Whereas my driver
> needs type=PCIM_ADDR_DEVRES_TYPE_MAPPING
> so that pci_iounmap() get called. If I use this patch, driver reload
> was failing for VF devices with below errors
>
> [90662.789921] octep_vdpa 0000:17:02.0: BAR 4: can't reserve [mem
> 0x207ff0000000-0x207ff07fffff 64bit pref]
> [90662.789922] octep_vdpa 0000:17:02.0: Failed to request BAR:4
> region
>
> As you suggested below, I prefer to have non-managed version (use
> pci_request_region()) and explicit remove() at required
> places for now.
Would you then mind replacing pcim_enable_device() with
pci_enable_device()? Should I send you a patch for that or do you want
to do that yourself?
That should do the trick.
P.
>
> >
> > The only way for you to, currently, have non-managed versions is by
> > using
> > pci_enable_device() instead and then doing pci_disable_device()
> > manually in
> > remove() and the other appropriate places.
> >
> > This patch should not change behavior in any way.
> >
> > If you're sure that having no management is not a problem, then we
> > could
> > simply drop this patch and I later remove the hybrid feature from
> > PCI. Then
> > your call to pci_request_region() will become unmanaged, even if
> > you keep
> > the pcim_enable_device().
> >
> > Tell me if you have a preference.
> >
> > P.
> >
> > >
> > > >
> > > > Signed-off-by: Philipp Stanner <pha...@kernel.org>
> > > > ---
> > > > drivers/vdpa/octeon_ep/octep_vdpa_main.c | 4 +---
> > > > 1 file changed, 1 insertion(+), 3 deletions(-)
> > > >
> > > > diff --git a/drivers/vdpa/octeon_ep/octep_vdpa_main.c
> > > > b/drivers/vdpa/octeon_ep/octep_vdpa_main.c
> > > > index f3d4dda4e04c..e0da6367661e 100644
> > > > --- a/drivers/vdpa/octeon_ep/octep_vdpa_main.c
> > > > +++ b/drivers/vdpa/octeon_ep/octep_vdpa_main.c
> > > > @@ -391,7 +391,7 @@ static int octep_iomap_region(struct
> > > > pci_dev
> > > > *pdev,
> > > > u8 __iomem **tbl, u8 bar) {
> > > > int ret;
> > > >
> > > > - ret = pci_request_region(pdev, bar,
> > > > OCTEP_VDPA_DRIVER_NAME);
> > > > + ret = pcim_request_region(pdev, bar,
> > > > OCTEP_VDPA_DRIVER_NAME);
> > > > if (ret) {
> > > > dev_err(&pdev->dev, "Failed to request BAR:%u
> > > > region\n",
> > bar);
> > > > return ret;
> > > > @@ -400,7 +400,6 @@ static int octep_iomap_region(struct
> > > > pci_dev
> > > > *pdev,
> > > > u8 __iomem **tbl, u8 bar)
> > > > tbl[bar] = pci_iomap(pdev, bar, pci_resource_len(pdev,
> > > > bar));
> > > > if (!tbl[bar]) {
> > > > dev_err(&pdev->dev, "Failed to iomap
> > > > BAR:%u\n", bar);
> > > > - pci_release_region(pdev, bar);
> > > > ret = -ENOMEM;
> > > > }
> > > >
> > > > @@ -410,7 +409,6 @@ static int octep_iomap_region(struct
> > > > pci_dev
> > > > *pdev,
> > > > u8 __iomem **tbl, u8 bar) static void
> > > > octep_iounmap_region(struct
> > > > pci_dev *pdev, u8 __iomem **tbl, u8 bar) {
> > > > pci_iounmap(pdev, tbl[bar]);
> > > > - pci_release_region(pdev, bar);
> > > > }
> > > >
> > > > static void octep_vdpa_pf_bar_shrink(struct octep_pf *octpf)
> > > > --
> > > > 2.48.1
> > >
>
