From: Daniel Scheller <d.scheller@xxxxxxx>
Still, target: 4.14.
Changes from v3 to v4:
* IRQ handler splitoff removed
* Prefixed irq_handler{,0,1} with ddb_ to avoid potential conflicts in
the global namespace wrt things like allyesconfig due to very generic
function naming
* Removed MSI_IRQHANDLERS #ifdef'ery
Changes from v2 to v3:
* (private) IOCTLs (temporarily) removed. Proper implementation via
UAPI and some documentation will be (re)added with a separate patch.
Even with this functionality removed (which most prominently is used
to update the card FPGA) everything is still working as intended.
With this, also temporarily removed the existing FLASHIO IOCTL.
* Code move/split already merged thus no longer part of the series
Changes from v1 to v2:
* I2C access functions (ie. i2c_read() et al) refactored from
ddbridge-i2c.c into ddbridge-i2c.h and declared static, and needed
include added to all .c files making use of them. This fixes symbol
conflicts in the global namespace with other drivers (kbuild test
robot reported a conflict with an infiniband driver) when compiling
into the kernel blob. While at it, code style has been made proper
(in ddbridge-i2c.h) and the 0.9.29 code bump patch was updated to
re-use the "plural" functions from their "singular" equivalents.
* The IRQ_HANDLE_BYTE() macro was removed. It is used nowhere, even
in the unmodified/unstripped upstream driver.
* Shortened the buffer overflow fix in ddb_ports_init() from checking
"p > 0" to only check "p".
Changes from original series to the resend:
* rebased on latest mediatree-master wrt
commit 618e8aac3d7c ("media: ddbridge: constify i2c_algorithm structure")
* build error in ddbridge-core.c fixed wrt
commit dcda9b04713c ("mm, tree wide: replace __GFP_REPEAT by __GFP_RETRY_MAYFAIL with more useful semantic")
* useless return removed from void calc_con()
* UTF8 in ddbridge-regs.h removed
* Tested-by's added to commit messages
Previous Tested-by testimonials still apply. v4 only has structural
changes which were build and runtime tested (built from the full tree and
via media_build).
Unless any important changes are neccessary, I consider this v4 the final
version. IOCTL refactor/rework won't be part of the code bump but will
be done as separate patch instead.
Note: DD have already picked up most of the changes that were introduced
during cleanup of the ddbridge driver so both versions are quite on par
in this regard. DD already released a 0.9.31, patches to bump this 0.9.29
to 0.9.31 are in preparation and even partly finished to get things fully
synced after this series hits media_tree.
>From previous cover letters:
Mauro/Media maintainers, this updates drivers/media/pci/ddbridge to the
very latest code that DD carry in their vendor driver package as of
version 0.9.29, in the "once, the big-bang-way is ok" way as discussed at
[2] (compared to the incremental, awkward to do variant since that
involves dissecting all available release archives and having to - try
to - build proper commits out of this, which will always be inaccurate;
a start was done at [3], however - and please understand - I definitely
don't want to continue doing that...)
In the last patch, I add myself to MAINTAINERS. This means I will care
about getting driver updates as they're released by DD into mainline,
starting from this (0.9.29) version, which is definitely doable in an
incremental way. So, I'll make sure the in-kernel driver won't bit-rot
again, and it will receive new hardware support as it becomes available
in a timely manner.
While the driver code bump looks massive, judging from the diff, there's
mostly a whole lot of refactoring and restructuring of variables, port/
link management and all such stuff in it. Feature-wise, this is most
notable:
- Support for all (PCIe) CI (single/duo) cards and Flex addons
- Support for MSI (Message Signaled Interrupts), though disabled by
default since there were still reports of problems with this
- TS Loopback support (set up ports to behave as if a CI is connected,
without decryption of course)
- As mentioned: Heavy code reordering, and split up into multiple files
Stripped functionality compared to dddvb:
- DVB-C modulator card support removed (requires DVB core API)
- OctoNET SAT>IP server/box support removed (requires API aswell)
- with this, GT link support was removed (only on OctoNET hardware)
- MaxS8 4/8 DVB-S/S2 card support (temporarily) removed (requires an
additional Demod driver; subject for another, later, series)
- (temp) driver IOCTLs/API removed and subject for readdition.
A note on the patches:
The bump starts by aligning the code "order-wise" to the updated driver,
to keep the diff a bit cleaner. Next, the code split is applied, without
actually changing any functionality. Compared to upstream, this isn't done
by moving functions into different C files and then do an include on them,
but we're handling them with the Makefile, building separate objects, and
having proper prototypes in ddbridge.h. After the code bump, further split
up is applied to increase readability and maintainability (also, for the
MaxS8 support, there will be another object with another ~400 LoC, which
originally lives in ddbridge-core aswell). Then, all issues found by W=1
and smatch are resolved, one by one. This is kept separate since those
fixes will be proposed for upstream inclusion. The last thing is the
addition of the MSI default Kconfig options which will mainly inform users
that there's something that might(!) cause issues but is still being
worked on - the default is "off" to provide a proper OotB experience.
To distinguish from the original unchanged vendor driver, "-integrated" is
suffixed to the version code.
Note on checkpatch:
First two patches are solely code-moving, so checkpatch will complain on
them. With the ddbridge code bump, all non-strict style issues are
resolved. "--strict" checking will receive another round of patches
afterwards.
Yes, you will hate me for this large code drop, but at least we sort-of
discussed this beforehand, and we have to start *somewhere*.
Thanks in advance for reviewing and (optimally) getting this merged and
getting the DD driver dilemma solved hopefully once and for all.
[1] http://www.spinics.net/lists/linux-media/msg117946.html
[2] http://www.spinics.net/lists/linux-media/msg117358.html
[3] https://github.com/herrnst/dddvb-linux-kernel/compare/4226861...mediatree/master-ddbupdate
Daniel Scheller (11):
[media] ddbridge: bump ddbridge code to version 0.9.29
[media] ddbridge: split I/O related functions off from ddbridge.h
[media] ddbridge: split off hardware definitions and mappings
[media] ddbridge: check pointers before dereferencing
[media] ddbridge: only register frontends in fe2 if fe is not NULL
[media] ddbridge: fix possible buffer overflow in ddb_ports_init()
[media] ddbridge: remove unreachable code
[media] ddbridge: fix impossible condition warning
[media] ddbridge: fix dereference before check
[media] ddbridge: Kconfig option to control the MSI modparam default
[media] MAINTAINERS: add entry for ddbridge
MAINTAINERS | 8 +
drivers/media/pci/ddbridge/Kconfig | 15 +
drivers/media/pci/ddbridge/Makefile | 3 +-
drivers/media/pci/ddbridge/ddbridge-core.c | 3235 +++++++++++++++++++++-------
drivers/media/pci/ddbridge/ddbridge-hw.c | 299 +++
drivers/media/pci/ddbridge/ddbridge-hw.h | 52 +
drivers/media/pci/ddbridge/ddbridge-i2c.c | 218 +-
drivers/media/pci/ddbridge/ddbridge-i2c.h | 41 +-
drivers/media/pci/ddbridge/ddbridge-io.h | 71 +
drivers/media/pci/ddbridge/ddbridge-main.c | 525 ++---
drivers/media/pci/ddbridge/ddbridge-regs.h | 138 +-
drivers/media/pci/ddbridge/ddbridge.h | 332 ++-
12 files changed, 3636 insertions(+), 1301 deletions(-)
create mode 100644 drivers/media/pci/ddbridge/ddbridge-hw.c
create mode 100644 drivers/media/pci/ddbridge/ddbridge-hw.h
create mode 100644 drivers/media/pci/ddbridge/ddbridge-io.h
--
2.13.0
