RE: [PATCH v9] platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc

[Liming Sun]

Thanks Andy. Please see my response and questions on some of the comments below.

Regards,
Liming

> -----Original Message-----
> From: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> Sent: Wednesday, February 13, 2019 1:11 PM
> To: Liming Sun <lsun@xxxxxxxxxxxx>
> Cc: David Woods <dwoods@xxxxxxxxxxxx>; Andy Shevchenko <andy@xxxxxxxxxxxxx>; Darren Hart <dvhart@xxxxxxxxxxxxx>; Vadim
> Pasternak <vadimp@xxxxxxxxxxxx>; Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>; Platform Driver <platform-driver-
> x86@xxxxxxxxxxxxxxx>
> Subject: Re: [PATCH v9] platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc
> 
> On Wed, Feb 13, 2019 at 3:27 PM Liming Sun <lsun@xxxxxxxxxxxx> wrote:
> >
> ...
> 
> > +/* Use a union struction for 64-bit little/big endian. */
> 
> What does this mean?
> 
> > +union mlxbf_tmfifo_data_64bit {
> > +       u64 data;
> > +       __le64 data_le;
> > +};

The purpose is to send 8 bytes into the FIFO without data casting in writeq().

Below is the example with the cast.

u64 data = 0x1234;
__le64 data_le;
data_le = cpu_to_le64(data)
writeq(*(u64 *)&data_le, fifo->tx_base + MLXBF_TMFIFO_TX_DATA);

Below is the alternative trying to use union to avoid the cast.

mlxbf_tmfifo_data_64bit u;
u.data = 0x1234;
u. data_le = cpu_to_le64(u.data);
writeq(u.data_le, fifo->tx_base + MLXBF_TMFIFO_TX_DATA);

Which way might be better or any other suggestions?

> > +
> > +/* Message header used to demux data in the TmFifo. */
> > +union mlxbf_tmfifo_msg_hdr {
> > +       struct {
> > +               u8 type;                /* message type */
> > +               __be16 len;             /* payload length */
> > +               u8 unused[5];           /* reserved, set to 0 */
> > +       } __packed;
> 
> It's already packed. No?

The '__packed' is needed here. Without it the compiler will make the
structure size exceeding 8 bytes which is not desired.

>...
> > +       if (vdev_id == VIRTIO_ID_CONSOLE)
> 
> > +               tm_vdev->tx_buf = devm_kmalloc(dev,
> > +                                              MLXBF_TMFIFO_CONS_TX_BUF_SIZE,
> > +                                              GFP_KERNEL);
> 
> Are you sure devm_ suits here?

The 'tx_buf' are normal buffer to hold the console output. 
It seems ok to use devm_kmalloc so it could automatically freed
on driver detach. Please correct me if I am wrong.

>>...
> > +
> > +       fifo->tx_base = devm_ioremap(&pdev->dev, tx_res->start,
> > +                                    resource_size(tx_res));
> > +       if (!fifo->tx_base)
> > +               return -ENOMEM;
> 
> Switch to devm_ioremap_resource().
> However, I think you probably need memremap().

This is device registers accessed by arm64 core.
In arm64/include/asm/io.h, several apis are defined the same.

#define ioremap(addr, size)		__ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRE))
#define ioremap_nocache(addr, size)	__ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRE))
#define ioremap_wt(addr, size)		__ioremap((addr), (size), __pgprot(PROT_DEVICE_nGnRE))

How about using devm_ioremap_nocache()?
It could take advantage of the devm_xx() api.

>...
> Is it correct?
> 
> --
> With Best Regards,
> Andy Shevchenko