[U-Boot] [Patch v2 08/16] net/fm: fix compile warnings for 64-bit platform

[Hou Zhiqiang]

> -----Original Message-----
> From: Wood Scott-B07421
> Sent: 2015年9月18日 2:14
> To: Gong Qianyu-B52263
> Cc: u-boot at lists.denx.de; Hu Mingkai-B21284; Sun York-R58495; Hou
> Zhiqiang-B48286; Song Wenbin-B53747; Xie Shaohui-B21989; Wood Scott-
> B07421
> Subject: Re: [Patch v2 08/16] net/fm: fix compile warnings for 64-bit
> platform
> 
> On Thu, 2015-09-17 at 15:06 +0800, Gong Qianyu wrote:
> > This patch fixes such compile warnings:
> >
> > drivers/net/fm/eth.c: In function 'fm_eth_recv':
> > drivers/net/fm/eth.c:549:11: warning: cast to pointer from integer of
> > different size [-Wint-to-pointer-cast]
> >       data = (u8 *)in_be32(&rxbd->buf_ptr_lo);
> > drivers/net/fm/fm.c: In function 'fm_muram_alloc':
> > drivers/net/fm/fm.c:52:9: warning: cast to pointer from integer of
> > different size [-Wint-to-pointer-cast]
> >       memset((void *)ret, 0, size);
> > drivers/net/fm/fm.c: In function 'fm_init_muram':
> > drivers/net/fm/fm.c:59:13: warning: cast from pointer to integer of
> > different size [-Wpointer-to-int-cast]
> >       u32 base = (u32)reg;
> >
> > Just make the cast explicit for them.
> >
> > Signed-off-by: Gong Qianyu <Qianyu.Gong at freescale.com>
> > ---
> >  drivers/net/fm/eth.c | 31 ++++++++++++++++---------------
> > drivers/net/fm/fm.c  |  4 ++--
> >  2 files changed, 18 insertions(+), 17 deletions(-)
> >
> > diff --git a/drivers/net/fm/eth.c b/drivers/net/fm/eth.c index
> > 12eb9b8..6ef0afb 100644
> > --- a/drivers/net/fm/eth.c
> > +++ b/drivers/net/fm/eth.c
> > @@ -120,12 +120,12 @@ static int tgec_is_fibre(struct eth_device *dev)
> >
> >  static u16 muram_readw(u16 *addr)
> >  {
> > -     u32 base = (u32)addr & ~0x3;
> > +     ulong base = (ulong)addr & ~0x3;
> 
> This will still truncate the address at 32 bits.  It needs to be ~0x3UL.
> 
> 
> >       u32 val32 = in_be32((u32 *)base);
> >       int byte_pos;
> >       u16 ret;
> >
> > -     byte_pos = (u32)addr & 0x3;
> > +     byte_pos = (ulong)addr & 0x3;
> >       if (byte_pos)
> >               ret = (u16)(val32 & 0x0000ffff);
> >       else
> > @@ -136,12 +136,12 @@ static u16 muram_readw(u16 *addr)
> >
> >  static void muram_writew(u16 *addr, u16 val)  {
> > -     u32 base = (u32)addr & ~0x3;
> > +     ulong base = (ulong)addr & ~0x3;
> >       u32 org32 = in_be32((u32 *)base);
> >       u32 val32;
> >       int byte_pos;
> >
> > -     byte_pos = (u32)addr & 0x3;
> > +     byte_pos = (ulong)addr & 0x3;
> >       if (byte_pos)
> >               val32 = (org32 & 0xffff0000) | val;
> >       else
> > @@ -217,12 +217,12 @@ static int fm_eth_rx_port_parameter_init(struct
> > fm_eth *fm_eth)
> >       int i;
> >
> >       /* alloc global parameter ram at MURAM */
> > -     pram = (struct fm_port_global_pram *)fm_muram_alloc(fm_eth-
> >fm_index,
> > -             FM_PRAM_SIZE, FM_PRAM_ALIGN);
> > +     pram = (struct fm_port_global_pram *)(ulong)fm_muram_alloc(
> > +                     fm_eth->fm_index, FM_PRAM_SIZE, FM_PRAM_ALIGN);
> 
> Make fm_muram_alloc() return a pointer instead.  If muram were >= 4 GiB
> the above would fail.
> 
 
The muram is a region included in CCSR.
So does we take muram >= 4GiB into account make sense?

> >       fm_eth->rx_pram = pram;
> >
> >       /* parameter page offset to MURAM */
> > -     pram_page_offset = (u32)pram - fm_muram_base(fm_eth->fm_index);
> > +     pram_page_offset = (u32)(ulong)pram -
> > + fm_muram_base(fm_eth->fm_index);
> 
> Get rid of the u32 cast -- again, if the muram base were above >= 4 GiB
> this would fail because you're dropping the high bits before the
> subtraction rather than after.
> 
> >
> >       /* enable global mode- snooping data buffers and BDs */
> >       out_be32(&pram->mode, PRAM_MODE_GLOBAL); @@ -258,7 +258,8 @@
> > static int fm_eth_rx_port_parameter_init(struct fm_eth
> > *fm_eth)
> >               muram_writew(&rxbd->status, RxBD_EMPTY);
> >               muram_writew(&rxbd->len, 0);
> >               muram_writew(&rxbd->buf_ptr_hi, 0);
> > -             out_be32(&rxbd->buf_ptr_lo, (u32)rx_buf_pool + i *
> MAX_RXBUF_LEN);
> > +             out_be32(&rxbd->buf_ptr_lo, (u32)(ulong)rx_buf_pool +
> > +                                                     i *
> > + MAX_RXBUF_LEN);
> >               rxbd++;
> 
> Use virt_to_phys() and lower_32_bits().  Is there a "hi" register to
> handle the upper 32 bits?
> 
> Likewise elsewhere.  Don't just apply the minimum bandage to get rid of
> the warning.  Make the code actually be 64-bit clean.
> 

Thanks,
Zhiqiang