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

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

On Tue, 11 Feb 2025, Keith Busch wrote:

> On Tue, Feb 11, 2025 at 03:02:35PM -0600, Bjorn Helgaas wrote:
> > 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.
> 
> The original code is actually correct despite using a flexible array of
> a struct that contains a flexible array. That arrangement just means you
> can't index into it, but the code is only doing pointer arithmetic, so
> should be fine.
> 
> With this struct:
> 
> struct pci_saved_state {
>  	u32 config_space[16];
> 	struct pci_cap_saved_data cap[];
> };
> 
> Accessing "cap" field returns the address right after the config_space[]
> member. When it's changed to a pointer type, though, it needs to be
> initialized to *something* but the code doesn't do that. The code just
> expects the cap to follow right after the config.
> 
> Anyway, to silence the warning we can just make it an anonymous member
> and add 1 to the state to get to the same place in memory as before.
> 
> ---
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index 869d204a70a37..e562037644fd0 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -1929,7 +1929,6 @@ EXPORT_SYMBOL(pci_restore_state);
>  
>  struct pci_saved_state {
>  	u32 config_space[16];
> -	struct pci_cap_saved_data cap[];

Can't just [] be dropped from it (and removed from the size calculation)?

It's not a real flex array because the second pci_cap_saved_data is not at 
->cap[1] but calculated by the loop by adding in the data in between. But 
there's one entry at ->cap[0] which is same as ->cap, no need to make it 
a flex array at all, IMO.

>  };
>  
>  /**
> @@ -1961,7 +1960,7 @@ struct pci_saved_state *pci_store_saved_state(struct pci_dev *dev)
>  	memcpy(state->config_space, dev->saved_config_space,
>  	       sizeof(state->config_space));
>  
> -	cap = state->cap;
> +	cap = (void *)(state + 1);
>  	hlist_for_each_entry(tmp, &dev->saved_cap_space, next) {
>  		size_t len = sizeof(struct pci_cap_saved_data) + tmp->cap.size;
>  		memcpy(cap, &tmp->cap, len);
> @@ -1991,7 +1990,7 @@ int pci_load_saved_state(struct pci_dev *dev,
>  	memcpy(dev->saved_config_space, state->config_space,
>  	       sizeof(state->config_space));
>  
> -	cap = state->cap;
> +	cap = (void *)(state + 1);
>  	while (cap->size) {
>  		struct pci_cap_saved_state *tmp;
>  
> --
> 

-- 
 i.