Re: [net-next 09/15] net/mlx5: Create FW devlink health reporter

[Jiri Pirko <jiri@resnulli.us>]

Mon, May 06, 2019 at 11:46:43PM CEST, alexei.starovoitov@gmail.com wrote:
>On Mon, May 06, 2019 at 07:52:18PM +0000, Saeed Mahameed wrote:
>> On Mon, 2019-05-06 at 13:38 +0200, Jiri Pirko wrote:
>> > Mon, May 06, 2019 at 12:45:44PM CEST, moshe@mellanox.com wrote:
>> > > 
>> > > On 5/5/2019 6:42 PM, Jiri Pirko wrote:
>> > > > Sun, May 05, 2019 at 02:33:23AM CEST, saeedm@mellanox.com wrote:
>> > > > > From: Moshe Shemesh <moshe@mellanox.com>
>> > > > > 
>> > > > > Create mlx5_devlink_health_reporter for FW reporter. The FW
>> > > > > reporter
>> > > > > implements devlink_health_reporter diagnose callback.
>> > > > > 
>> > > > > The fw reporter diagnose command can be triggered any time by
>> > > > > the user
>> > > > > to check current fw status.
>> > > > > In healthy status, it will return clear syndrome. Otherwise it
>> > > > > will dump
>> > > > > the health info buffer.
>> > > > > 
>> > > > > Command example and output on healthy status:
>> > > > > $ devlink health diagnose pci/0000:82:00.0 reporter fw
>> > > > > Syndrome: 0
>> > > > > 
>> > > > > Command example and output on non healthy status:
>> > > > > $ devlink health diagnose pci/0000:82:00.0 reporter fw
>> > > > > diagnose data:
>> > > > > assert_var[0] 0xfc3fc043
>> > > > > assert_var[1] 0x0001b41c
>> > > > > assert_var[2] 0x00000000
>> > > > > assert_var[3] 0x00000000
>> > > > > assert_var[4] 0x00000000
>> > > > > assert_exit_ptr 0x008033b4
>> > > > > assert_callra 0x0080365c
>> > > > > fw_ver 16.24.1000
>> > > > > hw_id 0x0000020d
>> > > > > irisc_index 0
>> > > > > synd 0x8: unrecoverable hardware error
>> > > > > ext_synd 0x003d
>> > > > > raw fw_ver 0x101803e8
>> > > > > 
>> > > > > Signed-off-by: Moshe Shemesh <moshe@mellanox.com>
>> > > > > Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
>> > > > > Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
>> > > > 
>> > > > 	
>> > > > [...]	
>> > > > 	
>> > > > 	
>> > > > > +static int
>> > > > > +mlx5_fw_reporter_diagnose(struct devlink_health_reporter
>> > > > > *reporter,
>> > > > > +			  struct devlink_fmsg *fmsg)
>> > > > > +{
>> > > > > +	struct mlx5_core_dev *dev =
>> > > > > devlink_health_reporter_priv(reporter);
>> > > > > +	struct mlx5_core_health *health = &dev->priv.health;
>> > > > > +	u8 synd;
>> > > > > +	int err;
>> > > > > +
>> > > > > +	mutex_lock(&health->info_buf_lock);
>> > > > > +	mlx5_get_health_info(dev, &synd);
>> > > > > +
>> > > > > +	if (!synd) {
>> > > > > +		mutex_unlock(&health->info_buf_lock);
>> > > > > +		return devlink_fmsg_u8_pair_put(fmsg,
>> > > > > "Syndrome", synd);
>> > > > > +	}
>> > > > > +
>> > > > > +	err = devlink_fmsg_string_pair_put(fmsg, "diagnose
>> > > > > data",
>> > > > > +					   health->info_buf);
>> > > > 
>> > > > No! This is wrong! You are sneaking in text blob. Please put the
>> > > > info in
>> > > > structured form using proper fmsg helpers.
>> > > > 
>> > > This is the fw output format, it is already in use, I don't want
>> > > to 
>> > > change it, just have it here in the diagnose output too.
>> > 
>> > Already in use where? in dmesg? Sorry, but that is not an argument.
>> > Please format the message properly.
>> > 
>> 
>> What is wrong here ? 
>> 
>> Unlike binary dump data, I thought diagnose data is allowed to be
>> developer friendly free text format, if not then let's enforce it in
>> devlink API. Jiri,  you can't audit each and every use of
>> devlink_fmsg_string_pair_put, it must be done by design.
>
>I agree with the purpose that Jiri is trying to get out of this infra,
>but I disagree that strings should be disallowed.

Not strings in general, I have nothing against them. If they are plain.
But when you push structured data into strings, for example using
sprintf for ints etc, that is a problem I'm pointing at.
Instead of that, the structured data should be formatted using
proper fmsg helpers. We already have them in kernel, it is just about
using them and not taking shortcuts.


>
>The example messages from commit log are not useful at all:
>$ devlink health diagnose pci/0000:82:00.0 reporter fw
>diagnose data:
>assert_var[0] 0xfc3fc043
>assert_var[1] 0x0001b41c
>assert_var[2] 0x00000000
>assert_var[3] 0x00000000
>assert_var[4] 0x00000000
>assert_exit_ptr 0x008033b4
>assert_callra 0x0080365c
>fw_ver 16.24.1000
>hw_id 0x0000020d
>irisc_index 0
>synd 0x8: unrecoverable hardware error
>ext_synd 0x003d
>raw fw_ver 0x101803e8
>
>Firmware dumps some magic numbers. How this is any better than
>what we have today?
>Every bit has to have meaningful message.
>The point of devlink diag is not to replicate dmesg, but to give
>users a clean way to understand and debug the hw.
>
>It seems mlx is adding a new interface and immediately misusing it.
>This not an example to give to other vendors.

I agree.