checkpatch.pl: Add warning for new __packed additions
PATCH 1/1 checkpatch.pl: Add warning for new __packed additions
While there are valid reasons to use __packed, often the answer is that you should be doing something else here instead.
--- scripts/checkpatch.pl | 5 +++++ 1 file changed, 5 insertions(+)
scripts/checkpatch.pl +5 -0
@@ -4010,6 +4010,11 @@ sub process { WARN("PREFER_PACKED", "__packed is preferred over __attribute__((packed))\n" . $herecurr); } +# Check for new packed usage, warn to use care + if ($line =~ /\b(__attribute__\s*\(\s*\(.*\bpacked|__packed)\b/) { + WARN("NEW_PACKED", + "Adding new packed members is to be done with care\n" . $herecurr); + } # Check for __attribute__ aligned, prefer __aligned if ($realfile !~ m@\binclude/uapi/@ &&
Signature · 1 line
1.7.9.5
Joe Perches
maintainerQuoted 2 lines · Tom Rini
While there are valid reasons to use
__packed, often the answer is that you should be doing something else here instead.
[]
scripts/checkpatch.pl
Quoted 1 line
scripts/checkpatch.pl +0
-0
[]
scripts/checkpatch.pl sub process { · line 4017
Quoted 9 lines
scripts/checkpatch.pl +5
-0@@ -4010,6 +4010,11 @@ sub process { WARN("PREFER_PACKED", "__packed is preferred over __attribute__((packed))\n" . $herecurr); } +# Check for new packed usage, warn to use care + if ($line =~ /\b(__attribute__\s*\(\s*\(.*\bpacked|__packed)\b/) { + WARN("NEW_PACKED", + "Adding new packed members is to be done with care\n" . $herecurr); + }
Hi Tom.
This might be too noisy as it will warn on any use of __packed, including existing ones if checking using -f/--file.
(there are some odd, probably broken ones in drivers/char/tpm/tpm_acpi.c)
How often is this actually a problem?
It'll warn on any use of packed not just packed members.
This may be better as
"Using 'packed' can impact performance\n"
and only tested when not in --file mode.
Quoted 18 lines · Joe Perches
Quoted 2 lines · Tom Rini
While there are valid reasons to use
__packed, often the answer is that you should be doing something else here instead.[]
scripts/checkpatch.pl
Quoted 1 line
scripts/checkpatch.pl +0
-0[]
scripts/checkpatch.pl sub process { · line 4017
Quoted 9 lines
scripts/checkpatch.pl +5
-0@@ -4010,6 +4010,11 @@ sub process { WARN("PREFER_PACKED", "__packed is preferred over __attribute__((packed))\n" . $herecurr); } +# Check for new packed usage, warn to use care + if ($line =~ /\b(__attribute__\s*\(\s*\(.*\bpacked|__packed)\b/) { + WARN("NEW_PACKED", + "Adding new packed members is to be done with care\n" . $herecurr); + }Hi Tom.
Quoted 2 lines
This might be too noisy as it will warn on any use of
__packed, including existing ones if checking using -f/--file.
Well, in my mind this is like typedef and some other things that can be warned on in existing files. I did a quick and lazy git grep and got ~1100 files in v3.14-rc1 out of ~36000 files.
Quoted 3 lines
(there are some odd, probably broken ones in
drivers/char/tpm/tpm_acpi.c)How often is this actually a problem?
I think the first line answers the second one, honestly. If one wants to get pedantic about things and really investigate there's probably some unneeded usages scattered about, and that's generally the type of thing one wants to address when checking whole files, right?
Quoted 5 lines
It'll warn on any use of packed not just packed members.
This may be better as
"Using 'packed' can impact performance\n"
and only tested when not in --file mode.
I can also make this change, sure, just point me off-list for an example to crib from and test?
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 4 lines · Tom Rini
Quoted 3 lines · Joe Perches
Quoted 2 lines · Tom Rini
While there are valid reasons to use
__packed, often the answer is that you should be doing something else here instead.
[]
Quoted 6 lines
Quoted 1 line
How often is this actually a problem?
I think the first line answers the second one, honestly. If one wants to get pedantic about things and really investigate there's probably some unneeded usages scattered about, and that's generally the type of thing one wants to address when checking whole files, right?
Maybe not.
That entirely depends on the correct and necessary uses of packed vs the incorrect usage rates.
I think almost all packed uses are correct and there might be a lot of patches submitted to remove them by over-zealous advocates of checkpatch -f.
Quoted 6 lines
Quoted 3 lines
This may be better as
"Using 'packed' can impact performance\n"
and only tested when not in --file mode.I can also make this change, sure, just point me off-list for an example to crib from and test?
Look at the FSF mailing address test as an example:
my $msg_type = \&ERROR;
$msg_type = \&CHK if ($file);
&{$msg_type}("FSF_MAILING_ADDRESS",Quoted 13 lines · Joe Perches
Quoted 4 lines · Tom Rini
Quoted 3 lines · Joe Perches
Quoted 2 lines · Tom Rini
While there are valid reasons to use
__packed, often the answer is that you should be doing something else here instead.[]
Quoted 6 lines
Quoted 1 line
How often is this actually a problem?
I think the first line answers the second one, honestly. If one wants to get pedantic about things and really investigate there's probably some unneeded usages scattered about, and that's generally the type of thing one wants to address when checking whole files, right?
Quoted 8 lines
Maybe not.
That entirely depends on the correct and necessary uses of packed vs the incorrect usage rates.
I think almost all packed uses are correct and there might be a lot of patches submitted to remove them by over-zealous advocates of checkpatch -f.
To try and also answer Josh's feedback as well, I've been lead to believe that most cases now people should be using regmap instead, which just leaves the case of having to match on-disk formats or similar cases I believe as the things that must stay __packed.
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
This may be better as
"Using 'packed' can impact performance\n"
and only tested when not in --file mode.
Quoted 8 lines
Quoted 2 lines
I can also make this change, sure, just point me off-list for an example to crib from and test?
Look at the FSF mailing address test as an example:
my $msg_type = \&ERROR; $msg_type = \&CHK if ($file); &{$msg_type}("FSF_MAILING_ADDRESS",
OK, thanks, I'll make something happen, and drop it to a CHK too.
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 4 lines · Tom Rini
I've been lead to
believe that most cases now people should be using regmap instead, which just leaves the case of having to match on-disk formats or similar cases I believe as the things that must stay__packed.
__packed is also necessary for on-wire networking protocols.
Quoted 7 lines · Joe Perches
Quoted 4 lines · Tom Rini
I've been lead to
believe that most cases now people should be using regmap instead, which just leaves the case of having to match on-disk formats or similar cases I believe as the things that must stay__packed.
__packedis also necessary for on-wire networking protocols.
Indeed, and there's probably a few other valid cases I'm also forgetting. But the common case is hardware, where regmap is now preferred.
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 2 lines · Tom Rini
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
Probably not, but I'm still not convinced it's useful.
Have you found a case where it's currently specified but not useful?
Quoted 8 lines · Joe Perches
Quoted 2 lines · Tom Rini
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
Probably not, but I'm still not convinced it's useful.
Have you found a case where it's currently specified but not useful?
Well, U-Boot and the kernel both share the dubious to incorrect __packed horror of cros_ec things (see include/linux/mfd/cros_ec_commands.h).
If this is really not seen as useful for the kernel, that's fine, I'll drop it. I mainly did this for U-Boot where we do want a bit more loud screaming going on when people add __packed to make sure it's for a good reason. Wanted to be a good neighbor so to speak and see if upstream wants it too.
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 5 lines · Tom Rini
Quoted 4 lines · Joe Perches
Quoted 2 lines · Tom Rini
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
Quoted 7 lines
Quoted 4 lines
Probably not, but I'm still not convinced it's useful.
Have you found a case where it's currently specified but not useful?
Well, U-Boot and the kernel both share the dubious to incorrect
__packedhorror ofcros_ecthings (seeinclude/linux/mfd/cros_ec_commands.h).
Are the __packed entries in cros_ec dubious?
Maybe the ones that don't seem to need them
because the are naturally 32 bit aligned, but
the others that are u16 aligned probably _do_
need __packed.
Quoted 5 lines
If this is really not seen as useful for the kernel, that's fine, I'll drop it. I mainly did this for U-Boot where we do want a bit more loud screaming going on when people add
__packedto make sure it's for a good reason. Wanted to be a good neighbor so to speak and see if upstream wants it too.
I don't mind keeping checkpatch compatible with U-Boot requirements, but probably not on by default.
Maybe there would be some "UBOOT-<foo>" type specific entries that could be default off but enabled with some wildcard or a .checkpatch.conf specific file for U-Boot.
Quoted 14 lines · Joe Perches
Quoted 5 lines · Tom Rini
Quoted 4 lines · Joe Perches
Quoted 2 lines · Tom Rini
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
Quoted 7 lines
Quoted 4 lines
Probably not, but I'm still not convinced it's useful.
Have you found a case where it's currently specified but not useful?
Well, U-Boot and the kernel both share the dubious to incorrect
__packedhorror ofcros_ecthings (seeinclude/linux/mfd/cros_ec_commands.h).
Quoted 6 lines
Are the
__packedentries incros_ecdubious?Maybe the ones that don't seem to need them
because the are naturally 32 bit aligned, but
the others that are u16 aligned probably _do_
need__packed.
There's so many unused entries in there that I cannot honestly say if things need to be packed or not.
Quoted 6 lines
Quoted 5 lines
If this is really not seen as useful for the kernel, that's fine, I'll drop it. I mainly did this for U-Boot where we do want a bit more loud screaming going on when people add
__packedto make sure it's for a good reason. Wanted to be a good neighbor so to speak and see if upstream wants it too.
Quoted 6 lines
I don't mind keeping checkpatch compatible with U-Boot requirements, but probably not on by default.
Maybe there would be some "UBOOT-<foo>" type specific entries that could be default off but enabled with some wildcard or a .checkpatch.conf specific file for U-Boot.
My perl is quite limited, so however much effort you're interested in putting in here is greatly appreciated (even if it's pointing out something else already in the script to copy and modify). We already ship a .checkpatch.conf so having to enable something by default is fine. Thanks!
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 5 lines · Tom Rini
My perl is quite limited, so however much effort you're interested in putting in here is greatly appreciated (even if it's pointing out something else already in the script to copy and modify). We already ship a .checkpatch.conf so having to enable something by default is fine. Thanks!
The U-Boot checkpatch is not the same as the current linux version and would need updating.
Tell me a little more please about how often or possible it would be to update the current linux version of checkpatch to the version used in U-Boot.
Quoted 7 lines · Joe Perches
Quoted 5 lines · Tom Rini
My perl is quite limited, so however much effort you're interested in putting in here is greatly appreciated (even if it's pointing out something else already in the script to copy and modify). We already ship a .checkpatch.conf so having to enable something by default is fine. Thanks!
Quoted 6 lines
The U-Boot checkpatch is not the same as the current linux version and would need updating.
Tell me a little more please about how often or possible it would be to update the current linux version of checkpatch to the version used in U-Boot.
The only intentional differences we have today are adding debug/printf to the list of log functions and pointing people at boards.cfg not CHECKPATCH in MAINTAINERS, so I can re-sync us no problem as, yup, we're about a year out of date there. Oh and I see now someone snuck in a check for whitespace before semi-colon to catch 'return 0 ;' which I'll check real quick isn't something already caught and if not post that as well.
Signature · 1 line
Tom
Joe Perches
maintainerQuoted 2 lines · Tom Rini
The only intentional differences we have today are adding debug/printf to the list of log functions
That seems fine and trivial to keep current.
Quoted 2 lines
and pointing people at boards.cfg not
CHECKPATCH inMAINTAINERS,
That seems wrong.
The idea of that message is to fix defects in checkpatch, not any other code.
Maybe that could be reworded to
"See the CHECKPATCH entry in the MAINTAINERS file of linux git" or some such.
Quoted 5 lines
so I can re-sync us no problem as, yup, we're
about a year out of date there. Oh and I see now someone snuck in a check for whitespace before semi-colon to catch 'return 0 ;' which I'll check real quick isn't something already caught and if not post that as well.
Does anyone know of other projects that use checkpatch?
Quoted 5 lines · Joe Perches
Quoted 2 lines · Tom Rini
The only intentional differences we have today are adding debug/printf to the list of log functions
That seems fine and trivial to keep current.
Agreed,
Quoted 11 lines
Quoted 2 lines
and pointing people at boards.cfg not
CHECKPATCH inMAINTAINERS,That seems wrong.
The idea of that message is to fix defects in checkpatch, not any other code.
Maybe that could be reworded to
"See the CHECKPATCH entry in theMAINTAINERSfile of linux git" or some such.
I've dropped that change now in a re-sync to current checkpatch.pl. Happy to pull in a re-wording of this part from a near-future resync with the kernel :)
Signature · 1 line
Tom
Quoted 11 lines · Tom Rini
Quoted 7 lines · Joe Perches
Quoted 4 lines · Tom Rini
I've been lead to
believe that most cases now people should be using regmap instead, which just leaves the case of having to match on-disk formats or similar cases I believe as the things that must stay__packed.
__packedis also necessary for on-wire networking protocols.Indeed, and there's probably a few other valid cases I'm also forgetting. But the common case is hardware, where regmap is now preferred.
Except on s390 which doesn't use regmap. So this will give us a lot of false positives with a warning message that isn't helpful at all.
It doesn't even tell us why it could be problematic and what would be the better choice.
Quoted 2 lines
I've got this modified to a CHK and only for non-file usage. Anything else we want to talk about before I repost?
/me thinks this patch should be dropped. It would cause too many false positives, since there are lot of valid use cases for __packed.
Quoted 16 lines · Tom Rini
While there are valid reasons to use
__packed, often the answer is that you should be doing something else here instead.Cc Andrew Morton <[email protected]>Cc Joe Perches <[email protected]>Cc Josh Triplett <[email protected]>Signed-off-by Tom Rini <[email protected]>--- scripts/checkpatch.pl | 5 +++++ 1 file changed, 5 insertions(+)scripts/checkpatch.pl +0
-0
scripts/checkpatch.pl sub process { · line 4017
Quoted 8 lines
scripts/checkpatch.pl +5
-0@@ -4010,6 +4010,11 @@ sub process { WARN("PREFER_PACKED", "__packed is preferred over __attribute__((packed))\n" . $herecurr); } +# Check for new packed usage, warn to use care + if ($line =~ /\b(__attribute__\s*\(\s*\(.*\bpacked|__packed)\b/) { + WARN("NEW_PACKED", + "Adding new packed members is to be done with care\n" . $herecurr); + }
This seems wrong; "is to be done with care" is the very definition of a false positive. At *best* this should always be CHK, and even then it seems excessive.
- Josh Triplett