Re: [PATCH] powerpc/fsl_pci: Fix pci stack build bug with FRAME_WARN

[Kim Phillips]

On Wed, 21 Jan 2015 21:02:27 -0600
Scott Wood <scottwood@xxxxxxxxxxxxx> wrote:

> On Wed, 2015-01-21 at 20:48 -0600, Kim Phillips wrote:
> > On Tue, 20 Jan 2015 18:31:32 -0600
> > Scott Wood <scottwood@xxxxxxxxxxxxx> wrote:
> > 
> > > On Tue, 2015-01-20 at 14:03 -0600, Kim Phillips wrote:
> > > > Fix this:
> > > > 
> > > >   CC      arch/powerpc/sysdev/fsl_pci.o
> > > > arch/powerpc/sysdev/fsl_pci.c: In function 'fsl_pcie_check_link':
> > > > arch/powerpc/sysdev/fsl_pci.c:91:1: error: the frame size of 1360 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]
> > > > 
> > > > when configuring FRAME_WARN, by converting the allocation from the
> > > > stack to the heap.  We use GFP_ATOMIC since this function can be
> > > > called with interrupts disabled.
> > > > 
> > > > Signed-off-by: Kim Phillips <kim.phillips@xxxxxxxxxxxxx>
> > > > ---
> > > >  arch/powerpc/sysdev/fsl_pci.c | 12 +++++++-----
> > > >  1 file changed, 7 insertions(+), 5 deletions(-)
> > > > 
> > > > diff --git a/arch/powerpc/sysdev/fsl_pci.c b/arch/powerpc/sysdev/fsl_pci.c
> > > > index 6455c1e..635d743 100644
> > > > --- a/arch/powerpc/sysdev/fsl_pci.c
> > > > +++ b/arch/powerpc/sysdev/fsl_pci.c
> > > > @@ -69,11 +69,13 @@ static int fsl_pcie_check_link(struct pci_controller *hose)
> > > >  
> > > >  	if (hose->indirect_type & PPC_INDIRECT_TYPE_FSL_CFG_REG_LINK) {
> > > >  		if (hose->ops->read == fsl_indirect_read_config) {
> > > > -			struct pci_bus bus;
> > > > -			bus.number = hose->first_busno;
> > > > -			bus.sysdata = hose;
> > > > -			bus.ops = hose->ops;
> > > > -			indirect_read_config(&bus, 0, PCIE_LTSSM, 4, &val);
> > > > +			struct pci_bus *bus;
> > > > +			bus = kmalloc(sizeof(*bus), GFP_ATOMIC);
> > > > +			bus->number = hose->first_busno;
> > > 
> > > Missing check for allocation failure.
> > 
> > thanks.
> > 
> > > Do we not have a real struct pci_bus we can use here?  Or refactor
> > > indirect_read_config() to take hose and bus number instead?
> > 
> > indirect_read_config() can't be refactored because it is also used
> > in the generic struct pci_ops.  Unless you mean making an
> > __indirect_read_config that the original would call, 
> 
> Yes, that's what I mean.
> 
> > but that doesn't look that trivial given it calls pci_exclude_device with a
> > struct pci_controller hose.
> 
> Check for excluded devices in indirect_read_config(), not
> __indirect_read_config().

turns out it wasn't an issue; see below for a v2.

> > > If putting a pci_bus struct on the stack is no longer OK, then
> > > fake_pci_bus() should be fixed as well.  I wonder if GCC is allocating
> > > separate pci_bus structs on the stack for this one and the one that
> > > early_read_config_dword() uses...
> > 
> > fake_pci_bus()' version is static, so it's not on the stack.
> > 
> > given that, maybe fsl_pcie_check_link()'s should be static too?
> 
> Oh.  How would you ensure that it's only called once at a time?  It
> doesn't look like this is only called during early boot.
> fsl_pcie_check_link() is called every time we do any config read through
> the normal interface.  This is also a concern for the call to
> early_read_config_dword().

I really don't know how that works: that code has been there since
before linux was maintained in git.

Below is the v2.

Thanks,

Kim.