Re: [PATCH v13] platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc

[Liming Sun <lsun@mellanox.com>]

Thanks Andy!  I just posted v14, which addresses all the comments you mentioned below for v13.

Regards,
Liming

> -----Original Message-----
> From: Andy Shevchenko <andy.shevchenko@gmail.com>
> Sent: Friday, April 5, 2019 11:44 AM
> To: Liming Sun <lsun@mellanox.com>
> Cc: David Woods <dwoods@mellanox.com>; Andy Shevchenko <andy@infradead.org>; Darren Hart <dvhart@infradead.org>; Vadim
> Pasternak <vadimp@mellanox.com>; Linux Kernel Mailing List <linux-kernel@vger.kernel.org>; Platform Driver <platform-driver-
> x86@vger.kernel.org>
> Subject: Re: [PATCH v13] platform/mellanox: Add TmFifo driver for Mellanox BlueField Soc
>
> On Thu, Apr 4, 2019 at 10:36 PM Liming Sun <lsun@mellanox.com> wrote:
> > This commit adds the TmFifo platform driver for Mellanox BlueField
> > Soc. TmFifo is a shared FIFO which enables external host machine
> > to exchange data with the SoC via USB or PCIe. The driver is based
> > on virtio framework and has console and network access enabled.
>
> Thanks for an update. Almost good.
> My comments below.
>
> Meanwhile I pushed this to my review and testing queue, thanks!
>
> > +#include <linux/acpi.h>
> > +#include <linux/bitfield.h>
> > +#include <linux/circ_buf.h>
> > +#include <linux/efi.h>
> > +#include <linux/irq.h>
> > +#include <linux/module.h>
> > +#include <linux/mutex.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/types.h>
>
> Perhaps blank line here. Would be more clear that this is utilizing
> virtio framework.

Updated in v14.

>
> > +#include <linux/virtio_config.h>
> > +#include <linux/virtio_console.h>
> > +#include <linux/virtio_ids.h>
> > +#include <linux/virtio_net.h>
> > +#include <linux/virtio_ring.h>
>
> > +/**
> > + * mlxbf_tmfifo_msg_hdr - Structure of the TmFifo message header
> > + * @type: message type
> > + * @len: payload length
> > + * @u: 64-bit union data
> > + */
> > +union mlxbf_tmfifo_msg_hdr {
> > +       struct {
> > +               u8 type;
> > +               __be16 len;
> > +               u8 unused[5];
> > +       } __packed;
> > +       u64 data;
>
> I'm not sure I understand how you can distinguish which field of union to use?
> Isn't here some type missed?

Updated the comment in v14.

This message header is a union of struct and u64 data.
The 'struct' has
type and length field which are used to encode & decode the message. 
The 'data' field is used to read/write the message header from/to the FIFO.

>
> > +};
>
> > +static u8 mlxbf_tmfifo_net_default_mac[ETH_ALEN] = {
>
> > +       0x00, 0x1A, 0xCA, 0xFF, 0xFF, 0x01};
>
> This should be two lines.

Updated in v14.

>
> > +/* Supported virtio-net features. */
> > +#define MLXBF_TMFIFO_NET_FEATURES      (BIT_ULL(VIRTIO_NET_F_MTU) | \
> > +                                        BIT_ULL(VIRTIO_NET_F_STATUS) | \
> > +                                        BIT_ULL(VIRTIO_NET_F_MAC))
>
> Better to write as
>
> #define FOO \
> (BIT(x) | BIT(y) ...)
>
> I think I told this earlier?

Updated in v14.

>
> > +/* Allocate vrings for the fifo. */
>
> fifo -> FIFO (and check all occurrences)

Updated in v14.

>
> > +static int mlxbf_tmfifo_alloc_vrings(struct mlxbf_tmfifo *fifo,
> > +                                    struct mlxbf_tmfifo_vdev *tm_vdev)
> > +{
> > +       struct mlxbf_tmfifo_vring *vring;
> > +       struct device *dev;
> > +       dma_addr_t dma;
> > +       int i, size;
> > +       void *va;
> > +
> > +       for (i = 0; i < ARRAY_SIZE(tm_vdev->vrings); i++) {
> > +               vring = &tm_vdev->vrings[i];
> > +               vring->fifo = fifo;
> > +               vring->num = MLXBF_TMFIFO_VRING_SIZE;
> > +               vring->align = SMP_CACHE_BYTES;
> > +               vring->index = i;
> > +               vring->vdev_id = tm_vdev->vdev.id.device;
> > +               dev = &tm_vdev->vdev.dev;
> > +
> > +               size = vring_size(vring->num, vring->align);
> > +               va = dma_alloc_coherent(dev->parent, size, &dma, GFP_KERNEL);
> > +               if (!va) {
>
> > +                       dev_err(dev->parent, "dma_alloc_coherent failed\n");
>
> I don't see how this will free the allocated entries.
> I think I told about this either.

Updated in v14.
It's not a memory leak since the caller will release them
in case of failures. I added one line in this function to
call the mlxbf_tmfifo_free_vrings() to be more clear.

>
> > +                       return -ENOMEM;
> > +               }
> > +
> > +               vring->va = va;
> > +               vring->dma = dma;
> > +       }
> > +
> > +       return 0;
> > +}
>
> > +/* House-keeping timer. */
> > +static void mlxbf_tmfifo_timer(struct timer_list *arg)
> > +{
>
> > +       struct mlxbf_tmfifo *fifo = container_of(arg, struct mlxbf_tmfifo,
> > +                                                timer);
>
> One line would be still good enough.

Updated in v14.

>
> > +       int more;
> > +
> > +       more = !test_and_set_bit(MLXBF_TM_RX_HWM_IRQ, &fifo->pend_events) ||
> > +                   !test_and_set_bit(MLXBF_TM_TX_LWM_IRQ, &fifo->pend_events);
> > +
> > +       if (more)
> > +               schedule_work(&fifo->work);
> > +
> > +       mod_timer(&fifo->timer, jiffies + MLXBF_TMFIFO_TIMER_INTERVAL);
> > +}
>
> > +       status = efi.get_variable(mlxbf_tmfifo_efi_name, &guid, NULL, &size,
> > +                                 buf);
> > +       if (status == EFI_SUCCESS && size == ETH_ALEN)
> > +               ether_addr_copy(mac, buf);
> > +       else
>
> > +               memcpy(mac, mlxbf_tmfifo_net_default_mac, ETH_ALEN);
>
> ether_addr_copy() as well.

Updated in v14.

>
> > +}
>
> > +       fifo->pdev = pdev;
>
> Do you really need to keep pdev there? Isn't struct device pointer enough?

Not needed. Updated in v14. Thanks!

>
>
> > +       /* Create the console vdev. */
> > +       ret = mlxbf_tmfifo_create_vdev(&pdev->dev, fifo, VIRTIO_ID_CONSOLE, 0,
> > +                                      NULL, 0);
>
> If you define temporary variable
>   struct device *dev = &pdev->dev;
> these lines can be merged into one.

Yes, updated in v14.

>
> > +       if (ret)
> > +               goto fail;
>
> --
> With Best Regards,
> Andy Shevchenko