Re: [PATCH] drivers: pci: Fix flexible array usage

From: Bjorn Helgaas
Date: Tue Feb 11 2025 - 16:02:51 EST


On Mon, Feb 10, 2025 at 04:03:26PM -0700, Keith Busch wrote:
> On Mon, Feb 10, 2025 at 06:57:40PM +0530, Purva Yeshi wrote:
> > Fix warning detected by smatch tool:
> > Array of flexible structure occurs in 'pci_saved_state' struct
> >
> > The warning occurs because struct pci_saved_state contains struct
> > pci_cap_saved_data cap[], where cap[] has a flexible array member (data[]).
> > Arrays of structures with flexible members are not allowed, leading to this
> > warning.
> >
> > Replaced cap[] with a pointer (*cap), allowing dynamic memory allocation
> > instead of embedding an invalid array of flexible structures.
> >
> > Signed-off-by: Purva Yeshi <purvayeshi550@xxxxxxxxx>
> > ---
> > drivers/pci/pci.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > index 869d204a7..648a080ef 100644
> > --- a/drivers/pci/pci.c
> > +++ b/drivers/pci/pci.c
> > @@ -1929,7 +1929,7 @@ EXPORT_SYMBOL(pci_restore_state);
> >
> > struct pci_saved_state {
> > u32 config_space[16];
> > - struct pci_cap_saved_data cap[];
> > + struct pci_cap_saved_data *cap;
> > };
>
> I don't think this is right. Previously the space for "cap" was
> allocated at the end of the pci_saved_state, but now it's just an
> uninitialized pointer.

Thanks, I think you're right. Dropped pending fix or better
explanation.

This is kind of a complicated data structure. IIUC, a struct
pci_saved_state is allocated only in pci_store_saved_state(), where
the size is determined by the sum of the sizes of all the entries in
the dev->saved_cap_space list.

The pci_saved_state is filled by copying from entries in the
dev->saved_cap_space list. The entries need not be all the same size
because we copy each entry manually based on its size.

So cap[] is really just the base of this buffer of variable-sized
entries. Maybe "struct pci_cap_saved_data cap[]" is not the best
representation of this, but *cap (a pointer) doesn't seem better.

Bjorn