Message-ID: <20200326211253.GC11304@nanopsycho.orion>
Date:   Thu, 26 Mar 2020 22:12:53 +0100
From:   Jiri Pirko <jiri@...nulli.us>
To:     Jacob Keller <jacob.e.keller@...el.com>
Cc:     netdev@...r.kernel.org, Jakub Kicinski <kuba@...nel.org>
Subject: Re: [net-next v2 00/11] implement DEVLINK_CMD_REGION_NEW

Thu, Mar 26, 2020 at 05:15:05PM CET, jacob.e.keller@...el.com wrote:
>
>
>On 3/26/2020 12:51 AM, Jiri Pirko wrote:
>>> Changes to patches since v1:
>>>
>>> * devlink: prepare to support region operations
>>>
>>>   No changes
>>>
>>> * devlink: convert snapshot destructor callback to region op
>>>
>>>   No changes
>>>
>>> * devlink: trivial: fix tab in function documentation
>>>
>>>   No changes
>>>
>>> * devlink: add function to take snapshot while locked
>>>
>>>   Added Jakub's Reviewed-by tag.
>>>
>>> * <NEW> devlink: use -ENOSPC to indicate no more room for snapshots
>>>
>>>   New patch added to convert confusing -ENOMEM to -ENOSPC, as suggested by
>>>   Jiri.
>>>
>>> * devlink: extract snapshot id allocation to helper function
>>>
>>>   No changes
>>>
>>> * devlink: convert snapshot id getter to return an error
>>>
>>>   Changed title to "devlink: report error once U32_MAX snapshot ids have
>>>   been used".
>>>
>>>   Refactored this patch to make devlink_region_snapshot_id_get take a
>>>   pointer to u32, so that the error value and id value are separated. This
>>>   means that we can remove the INT_MAX limitation on id values.
>>>
>>> * devlink: track snapshot id usage count using an xarray
>>>
>>>   Fixed the xa_init to use xa_init_flags with XA_FLAGS_ALLOC, so that
>>>   xa_alloc can properly be used.
>>>
>>>   Changed devlink_region_snapshot_id_get to use an initial count of 1
>>>   instead of 0. Added a new devlink_region_snapshot_id_put function, used
>>>   to release this initial count. This closes the race condition and issues
>>>   caused if the driver either doesn't create a snapshot, or if userspace
>>>   deletes the first snapshot before others are created.
>>>
>>>   Used WARN_ON in a few more checks that should not occur, such as if the
>>>   xarray entry is not a value, or when the id isn't yet in the xarray.
>>>
>>>   Removed an unnecessary if (err) { return err; } construction.
>>>
>>>   Use xa_limit_32b instead of xa_limit_31b now that we don't return the
>>>   snapshot id directly.
>>>
>>>   Cleanup the label used in __devlink_region_snapshot_create to indicate the
>>>   failure cause, rather than the cleanup step.
>>>
>>>   Removed the unnecessary locking around xa_destroy
>>>
>>> * devlink: implement DEVLINK_CMD_REGION_NEW
>>>
>>>   Added a WARN_ON to the check in snapshot_id_insert in case the id already
>>>   exists.
>>>
>>>   Removed an unnecessary "if (err) { return err; }" construction
>>>
>>>   Use -ENOSPC instead of -ENOMEM when max_snapshots is reached.
>>>
>>>   Cleanup label names to match style of the other labels in the file,
>>>   naming after the failure cause rather than the cleanup step. Also fix a
>>>   bug in the label ordering.
>>>
>>>   Call the new devlink_region_snapshot_id_put function in the mlx4 and
>>>   netdevsim drivers.
>>>
>>> * netdevsim: support taking immediate snapshot via devlink
>>>
>>>   Create a local devlink pointer instead of calling priv_to_devlink
>>>   multiple times.
>>>
>>>   Removed previous selftest for devlink region new without a snapshot id,
>>>   as this is no longer supported. Adjusted and verified that the tests pass
>>>   now.
>>>
>>> * ice: add a devlink region for dumping NVM contents
>>>
>>>   Use "dev_err" instead of "dev_warn" for a message about failure to create
>>>   the devlink region.
>> 
>> Could you please have the changelog per-patch, as I suggested for v1?
>> Much easier to review then.
>
>What do you mean? I already broke this into a description of the change
>for each patch, I thought that's what you wanted..
>
>Do you want me to move this into the individual commit emails?

Yes, that is what I mean. Much more convenient in my opinion.

Powered by blists - more mailing lists