linux-kernel - Re: [PATCH 1/2] fpga: dfl: map feature mmio resources in their own feature drivers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200721074025.GF17091@yilunxu-OptiPlex-7050>
Date:   Tue, 21 Jul 2020 15:40:25 +0800
From:   Xu Yilun <yilun.xu@...el.com>
To:     "Wu, Hao" <hao.wu@...el.com>
Cc:     "mdf@...nel.org" <mdf@...nel.org>,
        "linux-fpga@...r.kernel.org" <linux-fpga@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "trix@...hat.com" <trix@...hat.com>,
        "lgoncalv@...hat.com" <lgoncalv@...hat.com>,
        Matthew Gerlach <matthew.gerlach@...ux.intel.com>,
        "Weight, Russell H" <russell.h.weight@...el.com>
Subject: Re: [PATCH 1/2] fpga: dfl: map feature mmio resources in their own
  feature drivers

On Mon, Jul 20, 2020 at 05:09:35PM +0800, Wu, Hao wrote:
> > Subject: [PATCH 1/2] fpga: dfl: map feature mmio resources in their own
> > feature drivers
> >
> > This patch makes preparation for modularization of DFL sub feature
> > drivers.
> >
> > Currently, if we need to support a new DFL sub feature, an entry should
> > be added to fme/port_feature_drvs[] in dfl-fme/port-main.c. And we need
> > to re-compile the whole DFL modules. That make the DFL drivers hard to be
> > extended.
> >
> > Another consideration is that DFL may contain some IP blocks which are
> > already supported by kernel, most of them are supported by platform
> > device drivers. We could create platform devices for these IP blocks and
> > get them supported by these drivers.
> >
> > An important issue is that platform device drivers usually requests mmio
> > resources on probe. But now dfl mmio is mapped in dfl bus driver (e.g.
> > dfl-pci) as a whole region. Then platform device drivers for sub features
> > can't request their own mmio resources again. This is what the patch
> > trying to resolve.
> >
> > This patch changes the DFL enumeration. DFL bus driver will unmap mmio
> > resources after first step enumeration and pass enumeration info to DFL
> > framework. Then DFL framework will map the mmio resources again, do 2nd
> > step enumeration, and also unmap the mmio resources. In this way, sub
> > feature drivers could then request their own mmio resources as needed.
> >
> > An exception is that mmio resource of FIU headers are still mapped in dfl
> > bus driver. The FIU headers have some fundamental functions (sriov set,
> > port enable/disable) needed for dfl bus devices and other sub features.
> > They should not be unmapped as long as dfl bus device is alive.
> >
> > Signed-off-by: Xu Yilun <yilun.xu@...el.com>
> > Signed-off-by: Wu Hao <hao.wu@...el.com>
> > Signed-off-by: Matthew Gerlach <matthew.gerlach@...ux.intel.com>
> > Signed-off-by: Russ Weight <russell.h.weight@...el.com>
> > ---
> >  drivers/fpga/dfl-pci.c |  21 ++++--
> >  drivers/fpga/dfl.c     | 187 +++++++++++++++++++++++++++++++++++-----------
> > ---
> >  drivers/fpga/dfl.h     |   6 +-
> >  3 files changed, 152 insertions(+), 62 deletions(-)
> >
> > diff --git a/drivers/fpga/dfl-pci.c b/drivers/fpga/dfl-pci.c
> > index e220bec..22dc025 100644
> > --- a/drivers/fpga/dfl-pci.c
> > +++ b/drivers/fpga/dfl-pci.c
> > @@ -39,6 +39,11 @@ static void __iomem *cci_pci_ioremap_bar(struct
> > pci_dev *pcidev, int bar)
> >  return pcim_iomap_table(pcidev)[bar];
> >  }
> >
> > +static void cci_pci_iounmap_bars(struct pci_dev *pcidev, int mapped_bars)
> > +{
> > +pcim_iounmap_regions(pcidev, mapped_bars);
> > +}
> > +
> >  static int cci_pci_alloc_irq(struct pci_dev *pcidev)
> >  {
> >  int ret, nvec = pci_msix_vec_count(pcidev);
> > @@ -123,7 +128,7 @@ static int *cci_pci_create_irq_table(struct pci_dev
> > *pcidev, unsigned int nvec)
> >  static int cci_enumerate_feature_devs(struct pci_dev *pcidev)
> >  {
> >  struct cci_drvdata *drvdata = pci_get_drvdata(pcidev);
> > -int port_num, bar, i, nvec, ret = 0;
> > +int port_num, bar, i, nvec, mapped_bars, ret = 0;
> >  struct dfl_fpga_enum_info *info;
> >  struct dfl_fpga_cdev *cdev;
> >  resource_size_t start, len;
> > @@ -163,6 +168,8 @@ static int cci_enumerate_feature_devs(struct pci_dev
> > *pcidev)
> >  goto irq_free_exit;
> >  }
> >
> > +mapped_bars = BIT(0);
> > +
> >  /*
> >   * PF device has FME and Ports/AFUs, and VF device only has one
> >   * Port/AFU. Check them and add related "Device Feature List" info
> > @@ -172,7 +179,7 @@ static int cci_enumerate_feature_devs(struct pci_dev
> > *pcidev)
> >  start = pci_resource_start(pcidev, 0);
> >  len = pci_resource_len(pcidev, 0);
> >
> > -dfl_fpga_enum_info_add_dfl(info, start, len, base);
> > +dfl_fpga_enum_info_add_dfl(info, start, len);
> >
> >  /*
> >   * find more Device Feature Lists (e.g. Ports) per information
> > @@ -200,22 +207,26 @@ static int cci_enumerate_feature_devs(struct
> > pci_dev *pcidev)
> >  if (!base)
> >  continue;
> >
> > +mapped_bars |= BIT(bar);
> > +
> 
> One more place,
> 
> As you removed base addr usage below, so ideally we don't need to map
> more bars for port here? is my understanding correct?

Exactly, thanks for the catching. This makes the code much simpler.

Thanks,
Yilun

> 
> >  start = pci_resource_start(pcidev, bar) + offset;
> >  len = pci_resource_len(pcidev, bar) - offset;
> >
> > -dfl_fpga_enum_info_add_dfl(info, start, len,
> > -   base + offset);
> > +dfl_fpga_enum_info_add_dfl(info, start, len);
> 
> Thanks
> Hao