Re: [PATCH v3 net-next 01/11] net: stmmac: prepare dma op mode config for multiple queues

[Joao Pinto]

Ãs 11:12 AM de 5/8/2017, Andy Shevchenko escreveu:
> On Mon, May 8, 2017 at 12:54 PM, Joao Pinto <Joao.Pinto@xxxxxxxxxxxx> wrote:
>> Hi Andy and Jan,
>>
>> Ãs 10:36 AM de 5/8/2017, Andy Shevchenko escreveu:
>>> On Mon, May 8, 2017 at 9:56 AM, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
>>>> On 2017-03-15 12:04, Joao Pinto wrote:
>>>>> This patch prepares DMA Operation Mode configuration for multiple queues.
>>>>> The work consisted on breaking the DMA operation Mode configuration function
>>>>> into RX and TX scope and adapting its mechanism in stmmac_main.
>>>
>>>> Starting with this patch, the stmmac-based network adapters of the Intel
>>>> Quark SoC stop working. I'm getting an IP via DHCP, I can ping, but TCP
>>>> connections can no longer be established.
> 
>>> JFYI: With today's linux-next when _kexec:ed_ kernel boots I tried and
>>> got the following:
>>>

snip (...)

>>>
>>>
>>
>> Are you using the same version of Ethernet IP, 10/100?
> 
> I'm running on Intel Galileo Gen2 board (v4.11 by the way works fine
> with direct boot from SD card)
> 
>> Could you please verify if the crash you are experiencing is this place?
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__git.kernel.org_pub_scm_linux_kernel_git_torvalds_linux.git_tree_drivers_net_ethernet_stmicro_stmmac_stmmac-5Fmain.c-23n2956&d=DwIFaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=s2fO0hii0OGNOv9qQy_HRXy-xAJUD1NNoEcc3io_kx0&m=UF269QZ9ExFRw1XXpgdvO2QeTCLEp-GquRe8OqZwRf0&s=yZu3uME5PK-3nJlxz-H-HfHh3Shjzg0je5If_jSXVb4&e= 
>>
>> I would say that for rather old IPs, the napi is not capable of giving a valid
>> queue number. Could you please print the queue index returned by this line?
>>
>> https://urldefense.proofpoint.com/v2/url?u=https-3A__git.kernel.org_pub_scm_linux_kernel_git_torvalds_linux.git_tree_drivers_net_ethernet_stmicro_stmmac_stmmac-5Fmain.c-23n2948&d=DwIFaQ&c=DPL6_X_6JkXFx7AXWqB0tg&r=s2fO0hii0OGNOv9qQy_HRXy-xAJUD1NNoEcc3io_kx0&m=UF269QZ9ExFRw1XXpgdvO2QeTCLEp-GquRe8OqZwRf0&s=p_TgHODJum23I2N4AldR4oIaOPffSDpk9agmbRMQgoM&e= 
> 
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -2953,7 +2953,9 @@ static netdev_tx_t stmmac_xmit(struct sk_buff
> *skb, struct net_device *dev)
>        unsigned int enh_desc;
>        unsigned int des;
> 
> +       pr_info("%s <<< 1: priv %p, queue: %u\n", __func__, priv, queue);
>        tx_q = &priv->tx_queue[queue];
> +       pr_info("%s <<< 2: priv %p, queue: %u tx_q: %p\n", __func__,
> priv, queue, tx_q);
> 
> 
> [  101.591040] stmmac_xmit <<< 1: priv cdd1c4c0, queue: 7
> [  101.596377] stmmac_xmit <<< 2: priv cdd1c4c0, queue: 7 tx_q: cdd1caac
> 

I assume that the queue index is always 7 right? By return 7, the napi interface
'thinks' that your setup is using 8 TX queues which I assume it is not and thats
the problem causing your board to malfuntion.

Could you please check the values of the 'real' tx and rx queues count in this line?
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c#n4107

For default they are =1, so napi should be assuming 1RX and 1TX, and so you
should be getting queue index =0 in reception and transmission.

In terms of reception, could you print the queue index that stmmac_poll is using
here:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c#n3468

> 
> Also noticed warning that have to be addressed:
> 
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2504:49: warning:
> incorrect type in argument 1 (different address spaces)
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2504:49:    expected
> void [noderef] <asn:2>*ioaddr
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:2504:49:    got
> struct mac_device_info *hw

This one was well caught! Although it has no influence in your setup, since you
don't have this callback implemented, eQOS (>= 4.00) and 1000 cores will have
issues if using PCS. I can make a patch for this one.

> 
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c: In function
> âinit_dma_rx_desc_ringsâ:
> drivers/net/ethernet/stmicro/stmmac/stmmac_main.c:1274:15: warning:
> comparison of
> unsigned expression >= 0 is always true [-Wtype-limits]
>  while (queue >= 0) {
>               ^~

This one I have in my agenda to improve it, I also talked about it with Dan
Carpenter about it.