Re: [PATCH] ftgmac100: Implement variable descriptor size

[Cédric Le Goater]

On 6/3/20 9:08 AM, Philippe Mathieu-Daudé wrote:
> On 6/2/20 6:47 PM, Erik Smit wrote:
>> The hardware supports variable descriptor sizes, configured with the DBLAC
>> register.
>>
>> Most drivers use the default 2*8, which is currently hardcoded in qemu, but
>> the implementation of the driver in Supermicro BMC SMT_X11_158 uses 4*8.
>>
>> --
>> The implementation of the driver in Supermicro BMC SMT_X11_158 adds 4 extra
>> 4-bytes entries:
>> https://github.com/ya-mouse/openwrt-linux-aspeed/blob/master/drivers/net/ftgmac100_26.h#L387-L391
>>
>> And sets DBLAC to 0x44f97:
>> https://github.com/ya-mouse/openwrt-linux-aspeed/blob/master/drivers/net/ftgmac100_26.c#L449
>>
>> There's not a lot of public documentation on this hardware, but the
>> current linux driver shows the meaning of these registers:
>>
>> https://github.com/torvalds/linux/blob/master/drivers/net/ethernet/faraday/ftgmac100.c#L280-L281
>>
>>         iowrite32(FTGMAC100_DBLAC_RXDES_SIZE(2) |   /* 2*8 bytes RX descs */
>>                   FTGMAC100_DBLAC_TXDES_SIZE(2) |   /* 2*8 bytes TX descs */
>>
>> Without this patch, networking in SMT_X11_158 does not pass data.
> 
> Does it really 'pass' *all* the data?
> 
> This patch seems incomplete...
> 
> IMO you should 1/ declare FTGMAC100Desc as:
> 
> typedef struct {
>     uint32_t        des0;
>     uint32_t        des1;
> } FTGMAC100Desc;

The TX and RX descriptors have 4 words in the architecture which are
used by HW but software can use bigger size for its own purpose. This 
is what is doing the Aspeed SDK.

> 2/ Replace the code using static '2' by dynamic use of
> FTGMAC100_DBLAC_xXDES_SIZE(dblac):
> 
>   static int ftgmac100_read_bd(FTGMAC100Desc **bd, dma_addr_t addr)
>   {
>       unsigned bd_idx;
> 
>       if (dma_memory_read(&address_space_memory, addr, bd, sizeof(*bd))) {
>           qemu_log_mask(LOG_GUEST_ERROR, "%s: failed to read descriptor
> @ 0x%"
>                         HWADDR_PRIx "\n", __func__, addr);
>           return -1;
>       }
>       for (bd_idx = 0; bd_idx< FTGMAC100_DBLAC_RXDES_SIZE(s->dblac);
> bd_idx++) {
>           bd[bd_idx]->des0 = le32_to_cpu(bd[bd_idx]->des0);
>           bd[bd_idx]->des1 = le32_to_cpu(bd[bd_idx]->des1);
>       }
> 
>       return 0;
>   }
> 
> Etc...
> 

The current ftgmac100_read_bd() and ftgmac100_write_bd() routines model 
the HW perspective and they are fine as they are I think.

C.

> Maybe worth introduce the bd_to_cpu()/cpu_to_bd() helpers too
> (respectively calling le32_to_cpu & cpu_to_le32).
>>
>> Signed-off-by: Erik Smit <erik.lucas.smit@gmail.com
>> <mailto:erik.lucas.smit@gmail.com>>
>> ---
>>  hw/net/ftgmac100.c | 17 +++++++++++++++--
>>  1 file changed, 15 insertions(+), 2 deletions(-)
>>
>> diff --git a/hw/net/ftgmac100.c b/hw/net/ftgmac100.c
>> index 25ebee7ec2..1640b24b23 100644
>> --- a/hw/net/ftgmac100.c
>> +++ b/hw/net/ftgmac100.c
>> @@ -79,6 +79,19 @@
>>  #define FTGMAC100_APTC_TXPOLL_CNT(x)        (((x) >> 8) & 0xf)
>>  #define FTGMAC100_APTC_TXPOLL_TIME_SEL      (1 << 12)
>>
>> +/*
>> + * DMA burst length and arbitration control register
>> + */
>> +#define FTGMAC100_DBLAC_RXFIFO_LTHR(x)      (((x) >> 0) & 0x7)
>> +#define FTGMAC100_DBLAC_RXFIFO_HTHR(x)      (((x) >> 3) & 0x7)
>> +#define FTGMAC100_DBLAC_RX_THR_EN           (1 << 6)
>> +#define FTGMAC100_DBLAC_RXBURST_SIZE(x)     (((x) >> 8) & 0x3)
>> +#define FTGMAC100_DBLAC_TXBURST_SIZE(x)     (((x) >> 10) & 0x3)
>> +#define FTGMAC100_DBLAC_RXDES_SIZE(x)       (((x) >> 12) & 0xf)
>> +#define FTGMAC100_DBLAC_TXDES_SIZE(x)       (((x) >> 16) & 0xf)
>> +#define FTGMAC100_DBLAC_IFG_CNT(x)          (((x) >> 20) & 0x7)
>> +#define FTGMAC100_DBLAC_IFG_INC             (1 << 23)
>> +
>>  /*
>>   * PHY control register
>>   */
>> @@ -553,7 +566,7 @@ static void ftgmac100_do_tx(FTGMAC100State *s,
>> uint32_t tx_ring,
>>          if (bd.des0 & s->txdes0_edotr) {
>>              addr = tx_ring;
>>          } else {
>> -            addr += sizeof(FTGMAC100Desc);
>> +            addr += (FTGMAC100_DBLAC_TXDES_SIZE(s->dblac)) * 8;
> 
> Extra parenthesis not needed.
> 
> After doing 1/ you can now replace '8' by sizeof(FTGMAC100Desc).
> 
>>          }
>>      }
>>
>> @@ -982,7 +995,7 @@ static ssize_t ftgmac100_receive(NetClientState *nc,
>> const uint8_t *buf,
>>          if (bd.des0 & s->rxdes0_edorr) {
>>              addr = s->rx_ring;
>>          } else {
>> -            addr += sizeof(FTGMAC100Desc);
>> +            addr += (FTGMAC100_DBLAC_RXDES_SIZE(s->dblac)) * 8;
>>          }
>>      }
>>      s->rx_descriptor = addr;
>> --
>> 2.25.1
>