netdev - Re: [PATCH net-next] dpaa2-eth: add ethtool MAC counters

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 7 Nov 2019 07:34:41 +0000
From:   Ioana Ciornei <ioana.ciornei@....com>
To:     Andrew Lunn <andrew@...n.ch>
CC:     "davem@...emloft.net" <davem@...emloft.net>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [PATCH net-next] dpaa2-eth: add ethtool MAC counters

On 11/7/19 3:48 AM, Andrew Lunn wrote:
> On Thu, Nov 07, 2019 at 02:49:08AM +0200, Ioana Ciornei wrote:
>> +void dpaa2_mac_get_ethtool_stats(struct dpaa2_mac *mac, u64 *data)
>> +{
>> +	struct fsl_mc_device *dpmac_dev = mac->mc_dev;
>> +	int i, err;
>> +	u64 value;
>> +
>> +	for (i = 0; i < DPAA2_MAC_NUM_STATS; i++) {
>> +		err = dpmac_get_counter(mac->mc_io, 0, dpmac_dev->mc_handle,
>> +					i, &value);
>> +		if (err) {
>> +			netdev_err(mac->net_dev,
>> +				   "dpmac_get_counter error %d\n", err);
>> +			return;
> 
> Hi Ioana
> 
> I've seen quite a few drivers set *data to U64_MAX when there is an
> error. A value like that should stand out. The kernel message might
> not be seen.

Hi Andrew,

I was wondering also how to make the error stand out. I myself missed 
the error message, especially when the error is printed before a full 
screen of counters.

Thanks, I'll do this in v2.

> 
>> +/**
>> + * dpmac_get_counter() - Read a specific DPMAC counter
>> + * @mc_io:	Pointer to opaque I/O object
>> + * @cmd_flags:	Command flags; one or more of 'MC_CMD_FLAG_'
>> + * @token:	Token of DPMAC object
>> + * @id:		The requested counter ID
>> + * @value:	Returned counter value
>> + *
>> + * Return:	The requested counter; '0' otherwise.
>> + */
>> +int dpmac_get_counter(struct fsl_mc_io *mc_io, u32 cmd_flags, u16 token,
>> +		      enum dpmac_counter_id id, u64 *value)
>> +{
>> +	struct dpmac_cmd_get_counter *dpmac_cmd;
>> +	struct dpmac_rsp_get_counter *dpmac_rsp;
>> +	struct fsl_mc_command cmd = { 0 };
>> +	int err = 0;
>> +
>> +	cmd.header = mc_encode_cmd_header(DPMAC_CMDID_GET_COUNTER,
>> +					  cmd_flags,
>> +					  token);
>> +	dpmac_cmd = (struct dpmac_cmd_get_counter *)cmd.params;
>> +	dpmac_cmd->id = id;
>> +
>> +	err = mc_send_command(mc_io, &cmd);
>> +	if (err)
>> +		return err;
>> +
>> +	dpmac_rsp = (struct dpmac_rsp_get_counter *)cmd.params;
>> +	*value = le64_to_cpu(dpmac_rsp->counter);
>> +
>> +	return 0;
>> +}
> 
> How expensive is getting a single value? Is there a way to just get
> them all in a single command? The ethtool API always returns them all,
> so maybe you can optimise the firmware API to do the same?
> 
>     Andrew
> 

I don't know exact numbers for how expensive a single fw command is but 
you definitely do not wait for the output of 'ethtool -S'.
At the moment, we do not have an API to just get all the counters at 
once (as we have for the DPNI object). I would argue that even though I 
can optimise the fw API, I still want to have counters on older firmware 
images.
Anyhow, I'll add this to the list of things to improve in the fw :)

   Ioana