Commit 9bc4430d authored by Thorsten Leemhuis's avatar Thorsten Leemhuis Committed by Jonathan Corbet

docs: reporting-issues.rst: duplicate sections for reviewing purposes

This duplicates two section to make the diff in the next patch a bit
easier to gasp for humans.

Straight copy, no content changes.
Signed-off-by: default avatarThorsten Leemhuis <linux@leemhuis.info>
Link: https://lore.kernel.org/r/ef85edc8466f035eb243dd6629429ad4fd0565d8.1616181657.git.linux@leemhuis.infoSigned-off-by: default avatarJonathan Corbet <corbet@lwn.net>
parent 4f08d7ab
...@@ -328,6 +328,66 @@ those often get rejected or ignored, so consider yourself warned. But it's still ...@@ -328,6 +328,66 @@ those often get rejected or ignored, so consider yourself warned. But it's still
better than not reporting the issue at all: sometimes such reports directly or better than not reporting the issue at all: sometimes such reports directly or
indirectly will help to get the issue fixed over time. indirectly will help to get the issue fixed over time.
Search for existing reports
---------------------------
*Search the archives of the bug tracker or mailing list in question
thoroughly for reports that might match your issue. Also check if you find
something with your favorite internet search engine or in the Linux Kernel
Mailing List (LKML) archives. If you find anything, join the discussion
instead of sending a new report.*
Reporting an issue that someone else already brought forward is often a waste
of time for everyone involved, especially you as the reporter. So it's in your
own interest to thoroughly check if somebody reported the issue already. Thus
do not hurry with this step of the reporting process. Spending 30 to 60 minutes
or even more time can save you and others quite a lot of time and trouble.
The best place to search is the bug tracker or the mailing list where your
report needs to be filed. You'll find quite a few of those lists on
`lore.kernel.org <https://lore.kernel.org/>`_, but some are hosted in
different places. That for example is the case for the ath10k WiFi driver used
as example in the previous step. But you'll often find the archives for these
lists easily on the net. Searching for 'archive ath10k@lists.infradead.org' for
example will quickly lead you to the `Info page for the ath10k mailing list
<https://lists.infradead.org/mailman/listinfo/ath10k>`_, which at the top links
to its `list archives <https://lists.infradead.org/pipermail/ath10k/>`_.
Sadly this and quite a few other lists miss a way to search the archives. In
those cases use a regular internet search engine and add something like
'site:lists.infradead.org/pipermail/ath10k/' to your search terms, which limits
the results to the archives at that URL.
Additionally, search the internet and the `Linux Kernel Mailing List (LKML)
archives <https://lore.kernel.org/lkml/>`_, as maybe the real culprit might be
in some other subsystem. Searching in `bugzilla.kernel.org
<https://bugzilla.kernel.org/>`_ might also be a good idea, but if you find
anything there keep in mind: most subsystems expect reports in different
places, hence those you find there might have not even reached the people
responsible for the subsystem in question. Nevertheless, the data there might
provide valuable insights.
If you get flooded with results consider telling your search engine to limit
search timeframe to the past month or year. And wherever you search, make sure
to use good search terms; vary them a few times, too. While doing so try to
look at the issue from the perspective of someone else: that will help you to
come up with other words to use as search terms. Also make sure not to use too
many search terms at once. Remember to search with and without information like
the name of the kernel driver or the name of the affected hardware component.
But its exact brand name (say 'ASUS Red Devil Radeon RX 5700 XT Gaming OC')
often is not much helpful, as it is too specific. Instead try search terms like
the model line (Radeon 5700 or Radeon 5000) and the code name of the main chip
('Navi' or 'Navi10') with and without its manufacturer ('AMD').
In case you find an existing report about your issue, join the discussion, as
you might be able to provide valuable additional information. That can be
important even when a fix is prepared or in its final stages already, as
developers might look for people that can provide additional information or
test a proposed fix. Jump to the section 'Duties after the report went out' for
details on how to get properly involved.
Issue of high priority? Issue of high priority?
----------------------- -----------------------
...@@ -1523,6 +1583,130 @@ was already fixed in the latest release of version line you're interested in. ...@@ -1523,6 +1583,130 @@ was already fixed in the latest release of version line you're interested in.
This kernel needs to be vanilla and shouldn't be tainted before the issue This kernel needs to be vanilla and shouldn't be tainted before the issue
happens, as detailed outlined already above in the process of testing mainline. happens, as detailed outlined already above in the process of testing mainline.
Check code history and search for existing discussions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*Search the Linux kernel version control system for the change that fixed
the issue in mainline, as its commit message might tell you if the fix is
scheduled for backporting already. If you don't find anything that way,
search the appropriate mailing lists for posts that discuss such an issue
or peer-review possible fixes; then check the discussions if the fix was
deemed unsuitable for backporting. If backporting was not considered at
all, join the newest discussion, asking if it's in the cards.*
In a lot of cases the issue you deal with will have happened with mainline, but
got fixed there. The commit that fixed it would need to get backported as well
to get the issue solved. That's why you want to search for it or any
discussions abound it.
* First try to find the fix in the Git repository that holds the Linux kernel
sources. You can do this with the web interfaces `on kernel.org
<https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/>`_
or its mirror `on GitHub <https://github.com/torvalds/linux>`_; if you have
a local clone you alternatively can search on the command line with ``git
log --grep=<pattern>``.
If you find the fix, look if the commit message near the end contains a
'stable tag' that looks like this:
Cc: <stable@vger.kernel.org> # 5.4+
If that's case the developer marked the fix safe for backporting to version
line 5.4 and later. Most of the time it's getting applied there within two
weeks, but sometimes it takes a bit longer.
* If the commit doesn't tell you anything or if you can't find the fix, look
again for discussions about the issue. Search the net with your favorite
internet search engine as well as the archives for the `Linux kernel
developers mailing list <https://lore.kernel.org/lkml/>`_. Also read the
section `Locate kernel area that causes the issue` above and follow the
instructions to find the subsystem in question: its bug tracker or mailing
list archive might have the answer you are looking for.
* If you see a proposed fix, search for it in the version control system as
outlined above, as the commit might tell you if a backport can be expected.
* Check the discussions for any indicators the fix might be too risky to get
backported to the version line you care about. If that's the case you have
to live with the issue or switch to the kernel version line where the fix
got applied.
* If the fix doesn't contain a stable tag and backporting was not discussed,
join the discussion: mention the version where you face the issue and that
you would like to see it fixed, if suitable.
Details about reporting issues only occurring in older kernel version lines
---------------------------------------------------------------------------
This subsection provides details for steps you need to take if you could not
reproduce your issue with a mainline kernel, but want to see it fixed in older
version lines (aka stable and longterm kernels).
Some fixes are too complex
~~~~~~~~~~~~~~~~~~~~~~~~~~
*Prepare yourself for the possibility that going through the next few steps
might not get the issue solved in older releases: the fix might be too big
or risky to get backported there.*
Even small and seemingly obvious code-changes sometimes introduce new and
totally unexpected problems. The maintainers of the stable and longterm kernels
are very aware of that and thus only apply changes to these kernels that are
within rules outlined in 'Documentation/process/stable-kernel-rules.rst'.
Complex or risky changes for example do not qualify and thus only get applied
to mainline. Other fixes are easy to get backported to the newest stable and
longterm kernels, but too risky to integrate into older ones. So be aware the
fix you are hoping for might be one of those that won't be backported to the
version line your care about. In that case you'll have no other choice then to
live with the issue or switch to a newer Linux version, unless you want to
patch the fix into your kernels yourself.
Make sure the particular version line still gets support
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*Check if the kernel developers still maintain the Linux kernel version
line you care about: go to the front page of kernel.org and make sure it
mentions the latest release of the particular version line without an
'[EOL]' tag.*
Most kernel version lines only get supported for about three months, as
maintaining them longer is quite a lot of work. Hence, only one per year is
chosen and gets supported for at least two years (often six). That's why you
need to check if the kernel developers still support the version line you care
for.
Note, if kernel.org lists two 'stable' version lines on the front page, you
should consider switching to the newer one and forget about the older one:
support for it is likely to be abandoned soon. Then it will get a "end-of-life"
(EOL) stamp. Version lines that reached that point still get mentioned on the
kernel.org front page for a week or two, but are unsuitable for testing and
reporting.
Search stable mailing list
~~~~~~~~~~~~~~~~~~~~~~~~~~
*Check the archives of the Linux stable mailing list for existing reports.*
Maybe the issue you face is already known and was fixed or is about to. Hence,
`search the archives of the Linux stable mailing list
<https://lore.kernel.org/stable/>`_ for reports about an issue like yours. If
you find any matches, consider joining the discussion, unless the fix is
already finished and scheduled to get applied soon.
Reproduce issue with the newest release
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
*Install the latest release from the particular version line as a vanilla
kernel. Ensure this kernel is not tainted and still shows the problem, as
the issue might have already been fixed there.*
Before investing any more time in this process you want to check if the issue
was already fixed in the latest release of version line you're interested in.
This kernel needs to be vanilla and shouldn't be tainted before the issue
happens, as detailed outlined already above in the process of testing mainline.
Check code history and search for existing discussions Check code history and search for existing discussions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment