Re: [PATCH 4/5] staging: rtl8192e: Remove unused local variable irq_line

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Fri, Apr 07, 2023 at 10:08:50AM +0200, Philipp Hortmann wrote:
> On 4/6/23 22:06, Greg Kroah-Hartman wrote:
> > On Wed, Apr 05, 2023 at 11:48:20PM +0200, Philipp Hortmann wrote:
> > > Remove unused local variable irq_line.
> > > 
> > > Signed-off-by: Philipp Hortmann <philipp.g.hortmann@gmail.com>
> > > ---
> > >   drivers/staging/rtl8192e/rtl8192e/rtl_pci.c | 2 --
> > >   1 file changed, 2 deletions(-)
> > > 
> > > diff --git a/drivers/staging/rtl8192e/rtl8192e/rtl_pci.c b/drivers/staging/rtl8192e/rtl8192e/rtl_pci.c
> > > index 0bc3e013001e..1c3ccd2aabc3 100644
> > > --- a/drivers/staging/rtl8192e/rtl8192e/rtl_pci.c
> > > +++ b/drivers/staging/rtl8192e/rtl8192e/rtl_pci.c
> > > @@ -33,11 +33,9 @@ bool rtl92e_check_adapter(struct pci_dev *pdev, struct net_device *dev)
> > >   	struct r8192_priv *priv = (struct r8192_priv *)rtllib_priv(dev);
> > >   	u16 device_id;
> > >   	u8  revision_id;
> > > -	u16 irq_line;
> > >   	device_id = pdev->device;
> > >   	revision_id = pdev->revision;
> > > -	pci_read_config_word(pdev, 0x3C, &irq_line);
> > 
> > As was pointed out, that might not be "unused".
> > 
> > When doing a PCI write, the only way to know it has completed is to
> > issue a read.  Are you sure this isn't a read that is flushing out a
> > pending write?  Look at who calls this function and verify that this
> > read really is safe to remove and then document that really really well
> > in the changelog text when doing so.
> > 
> > Be careful about removing seemingly-unused PCI read/writes, they often
> > times are actually needed.
> > 
> > thanks,
> > 
> > greg k-h
> 
> Hi,
> 
> The first function that is called of the driver is: _rtl92e_pci_probe().
> 66 Lines later the function rtl92e_check_adapter() is called. The line
> pci_read_config_word() is the 10th line of this function.
> 
> No other pci function before in function rtl92e_check_adapter().
> 
> The following functions are used in the probe function before the
> rtl92e_check_adapter():
> 
> pci_enable_device()		Does not require a pci read.
> pci_set_master()		Does not require a pci read.
> dma_set_mask()			Does not require a pci read.
> alloc_rtllib()			Does not require a pci read.
> pci_set_drvdata()		Does not require a pci read.
> SET_NETDEV_DEV()		Does not require a pci read.
> rtllib_priv()			Does not require a pci read.
> pci_resource_start()		Does not require a pci read.
> pci_resource_len()		Does not require a pci read.
> pci_resource_flags()		Does not require a pci read.
> request_mem_region()		Does not require a pci read.
> ioremap()			Does not require a pci read.
> pci_resource_len()		Does not require a pci read.
> pci_read_config_byte()		Which is deleted in this
> 				patch series. So Gregs hint is
>  				applicable for this patch as well.
> 
> I am going to issue a v2.

How do you know that this device does not require that register to be
read from in order to work properly?  Remember, reads can change
device's state as well as writes.

thanks,

greg k-h