mbox series

[v5,0/6] FWU: Handle meta-data in common code

Message ID 20230228005135.1635726-1-jassisinghbrar@gmail.com
Headers show
Series FWU: Handle meta-data in common code | expand

Message

Jassi Brar Feb. 28, 2023, 12:51 a.m. UTC
From: Jassi Brar <jaswinder.singh@linaro.org>

The patchset reduces ~400 lines of code, while keeping the functionality same and making
meta-data operations much faster (by using cached structures).

Issue:
 meta-data copies (primary and secondary) are being handled by the backend/storage layer
instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
could by make smaller, cleaner and optimised.

Basic idea:
 Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
as a result we can get rid of every other callback .get_mdata() .update_mdata()
.get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
corresponding wrapper functions thereby making the code 100s of LOC smaller.

Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
layer to manage and verify mdata copies.
Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
verifies and, if needed, fixes the meta-data copies.

Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
low-level expensive read and parse calls.
gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.

Changes since v4:
        * Change fwu-mdata-mtd bindings to not require external changes
        * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
        * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata

Changes since v3:
        * Fix error log wording
        * call fwu_write_mdata() with part & PRIMARY_PART ? true: false

Changes since v2:
        * Drop whitespace changes
        * Fix missing mdata copy before return

Changes since v1:
        * Fix typos and misc cosmetic changes
        * Catch error returns

Jassi Brar (6):
  dt/bindings: fwu-mdata-mtd: drop changes outside FWU
  fwu: gpt: use cached meta-data partition numbers
  fwu: move meta-data management in core
  fwu: gpt: implement read_mdata and write_mdata callbacks
  fwu: meta-data: switch to management by common code
  fwu: rename fwu_get_verified_mdata to fwu_get_mdata

 cmd/fwu_mdata.c                               |  17 +-
 .../firmware/fwu-mdata-mtd.yaml               | 105 ++++++-
 drivers/fwu-mdata/fwu-mdata-uclass.c          | 151 +--------
 drivers/fwu-mdata/gpt_blk.c                   | 175 +++--------
 include/fwu.h                                 | 198 ++----------
 lib/fwu_updates/fwu.c                         | 295 +++++++-----------
 6 files changed, 295 insertions(+), 646 deletions(-)

Comments

Tom Rini Feb. 28, 2023, 12:58 a.m. UTC | #1
On Mon, Feb 27, 2023 at 06:51:35PM -0600, jassisinghbrar@gmail.com wrote:

> From: Jassi Brar <jaswinder.singh@linaro.org>
> 
> The patchset reduces ~400 lines of code, while keeping the functionality same and making
> meta-data operations much faster (by using cached structures).
> 
> Issue:
>  meta-data copies (primary and secondary) are being handled by the backend/storage layer
> instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
> meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
> could by make smaller, cleaner and optimised.
> 
> Basic idea:
>  Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
> meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
> as a result we can get rid of every other callback .get_mdata() .update_mdata()
> .get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
> corresponding wrapper functions thereby making the code 100s of LOC smaller.
> 
> Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
> layer to manage and verify mdata copies.
> Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
> verifies and, if needed, fixes the meta-data copies.
> 
> Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
> low-level expensive read and parse calls.
> gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.
> 
> Changes since v4:
>         * Change fwu-mdata-mtd bindings to not require external changes
>         * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
>         * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata

Did you run this through CI / build sandbox? This doesn't read like you
fixed the problem I reported in CI, when I was trying to merge v4.
Jassi Brar Feb. 28, 2023, 1 a.m. UTC | #2
On Mon, 27 Feb 2023 at 18:58, Tom Rini <trini@konsulko.com> wrote:
>
> On Mon, Feb 27, 2023 at 06:51:35PM -0600, jassisinghbrar@gmail.com wrote:
>
> > From: Jassi Brar <jaswinder.singh@linaro.org>
> >
> > The patchset reduces ~400 lines of code, while keeping the functionality same and making
> > meta-data operations much faster (by using cached structures).
> >
> > Issue:
> >  meta-data copies (primary and secondary) are being handled by the backend/storage layer
> > instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
> > meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
> > could by make smaller, cleaner and optimised.
> >
> > Basic idea:
> >  Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
> > meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
> > as a result we can get rid of every other callback .get_mdata() .update_mdata()
> > .get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
> > corresponding wrapper functions thereby making the code 100s of LOC smaller.
> >
> > Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
> > layer to manage and verify mdata copies.
> > Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
> > verifies and, if needed, fixes the meta-data copies.
> >
> > Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
> > low-level expensive read and parse calls.
> > gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.
> >
> > Changes since v4:
> >         * Change fwu-mdata-mtd bindings to not require external changes
> >         * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
> >         * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata
>
> Did you run this through CI / build sandbox? This doesn't read like you
> fixed the problem I reported in CI, when I was trying to merge v4.
>
I know that remains to be done.
The dt-bindings for fwu-mdata is changed in this patchset and I
thought any testcase may be impacted by it.

thanks.
Tom Rini Feb. 28, 2023, 1:28 a.m. UTC | #3
On Mon, Feb 27, 2023 at 07:00:10PM -0600, Jassi Brar wrote:
> On Mon, 27 Feb 2023 at 18:58, Tom Rini <trini@konsulko.com> wrote:
> >
> > On Mon, Feb 27, 2023 at 06:51:35PM -0600, jassisinghbrar@gmail.com wrote:
> >
> > > From: Jassi Brar <jaswinder.singh@linaro.org>
> > >
> > > The patchset reduces ~400 lines of code, while keeping the functionality same and making
> > > meta-data operations much faster (by using cached structures).
> > >
> > > Issue:
> > >  meta-data copies (primary and secondary) are being handled by the backend/storage layer
> > > instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
> > > meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
> > > could by make smaller, cleaner and optimised.
> > >
> > > Basic idea:
> > >  Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
> > > meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
> > > as a result we can get rid of every other callback .get_mdata() .update_mdata()
> > > .get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
> > > corresponding wrapper functions thereby making the code 100s of LOC smaller.
> > >
> > > Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
> > > layer to manage and verify mdata copies.
> > > Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
> > > verifies and, if needed, fixes the meta-data copies.
> > >
> > > Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
> > > low-level expensive read and parse calls.
> > > gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.
> > >
> > > Changes since v4:
> > >         * Change fwu-mdata-mtd bindings to not require external changes
> > >         * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
> > >         * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata
> >
> > Did you run this through CI / build sandbox? This doesn't read like you
> > fixed the problem I reported in CI, when I was trying to merge v4.
> >
> I know that remains to be done.
> The dt-bindings for fwu-mdata is changed in this patchset and I
> thought any testcase may be impacted by it.

So you're not expecting this iteration to be merged, as CI doesn't pass,
and that's known? OK.
Jassi Brar Feb. 28, 2023, 1:51 a.m. UTC | #4
On Mon, Feb 27, 2023 at 7:28 PM Tom Rini <trini@konsulko.com> wrote:
>
> On Mon, Feb 27, 2023 at 07:00:10PM -0600, Jassi Brar wrote:
> > On Mon, 27 Feb 2023 at 18:58, Tom Rini <trini@konsulko.com> wrote:
> > >
> > > On Mon, Feb 27, 2023 at 06:51:35PM -0600, jassisinghbrar@gmail.com wrote:
> > >
> > > > From: Jassi Brar <jaswinder.singh@linaro.org>
> > > >
> > > > The patchset reduces ~400 lines of code, while keeping the functionality same and making
> > > > meta-data operations much faster (by using cached structures).
> > > >
> > > > Issue:
> > > >  meta-data copies (primary and secondary) are being handled by the backend/storage layer
> > > > instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
> > > > meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
> > > > could by make smaller, cleaner and optimised.
> > > >
> > > > Basic idea:
> > > >  Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
> > > > meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
> > > > as a result we can get rid of every other callback .get_mdata() .update_mdata()
> > > > .get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
> > > > corresponding wrapper functions thereby making the code 100s of LOC smaller.
> > > >
> > > > Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
> > > > layer to manage and verify mdata copies.
> > > > Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
> > > > verifies and, if needed, fixes the meta-data copies.
> > > >
> > > > Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
> > > > low-level expensive read and parse calls.
> > > > gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.
> > > >
> > > > Changes since v4:
> > > >         * Change fwu-mdata-mtd bindings to not require external changes
> > > >         * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
> > > >         * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata
> > >
> > > Did you run this through CI / build sandbox? This doesn't read like you
> > > fixed the problem I reported in CI, when I was trying to merge v4.
> > >
> > I know that remains to be done.
> > The dt-bindings for fwu-mdata is changed in this patchset and I
> > thought any testcase may be impacted by it.
>
> So you're not expecting this iteration to be merged, as CI doesn't pass,
> and that's known? OK.
>
I am more concerned if the bindings displease someone and I have to
drop any code.
If I get acks on all patches, the next revision will only add the test cases.

thanks
Jassi Brar March 6, 2023, 11:21 p.m. UTC | #5
On Mon, Feb 27, 2023 at 7:28 PM Tom Rini <trini@konsulko.com> wrote:
>
> On Mon, Feb 27, 2023 at 07:00:10PM -0600, Jassi Brar wrote:
> > On Mon, 27 Feb 2023 at 18:58, Tom Rini <trini@konsulko.com> wrote:
> > >
> > > On Mon, Feb 27, 2023 at 06:51:35PM -0600, jassisinghbrar@gmail.com wrote:
> > >
> > > > From: Jassi Brar <jaswinder.singh@linaro.org>
> > > >
> > > > The patchset reduces ~400 lines of code, while keeping the functionality same and making
> > > > meta-data operations much faster (by using cached structures).
> > > >
> > > > Issue:
> > > >  meta-data copies (primary and secondary) are being handled by the backend/storage layer
> > > > instead of the common core in fwu.c (as also noted by Ilias)  that is, gpt_blk.c manages
> > > > meta-data and similarly raw_mtd.c will have to do the same when it arrives. The code
> > > > could by make smaller, cleaner and optimised.
> > > >
> > > > Basic idea:
> > > >  Introduce  .read_mdata() and .write_mdata() in fwu_mdata_ops  that simply read/write
> > > > meta-data copy. The core code takes care of integrity and redundancy of the meta-data,
> > > > as a result we can get rid of every other callback .get_mdata() .update_mdata()
> > > > .get_mdata_part_num()  .read_mdata_partition()  .write_mdata_partition() and the
> > > > corresponding wrapper functions thereby making the code 100s of LOC smaller.
> > > >
> > > > Get rid of fwu_check_mdata_validity() and fwu_mdata_check() which expected underlying
> > > > layer to manage and verify mdata copies.
> > > > Implement  fwu_get_verified_mdata(struct fwu_mdata *mdata) public function that reads,
> > > > verifies and, if needed, fixes the meta-data copies.
> > > >
> > > > Verified copy of meta-data is now cached as 'g_mdata' in fwu.c, which avoids multiple
> > > > low-level expensive read and parse calls.
> > > > gpt meta-data partition numbers are now cached in gpt_blk.c, so that we don't have to do expensive part_get_info() and uid ops.
> > > >
> > > > Changes since v4:
> > > >         * Change fwu-mdata-mtd bindings to not require external changes
> > > >         * Handle 'part == BOTH_PARTS' in fwu_sync_mdata
> > > >         * use parts_ok[] and parts_mdata[] instead of pri/sec_ok and p/s_mdata
> > >
> > > Did you run this through CI / build sandbox? This doesn't read like you
> > > fixed the problem I reported in CI, when I was trying to merge v4.
> > >
> > I know that remains to be done.
> > The dt-bindings for fwu-mdata is changed in this patchset and I
> > thought any testcase may be impacted by it.
>
> So you're not expecting this iteration to be merged, as CI doesn't pass,
> and that's known? OK.
>
Sorry I got confused. I thought new test cases are expected to be
written, whereas you only wanted the sandbox compilation breakage
fixed. My replies must have sounded so stupid :(
I have a patch to fix sandbox compilation in the v6 revision.

thanks.