Skip to contentSkip to conversation

selftests: gpio: rework and port to GPIO uAPI v2

PATCH Superseded Kent Gibson 3 participants 18 messages Jan 2, 2021 – Jan 5, 2021 1 message not in the archive

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Kent Gibson

author reviewer
#0 lore raw

cover selftests: gpio: rework and port to GPIO uAPI v2

Initially I just wanted to port the selftests to the latest GPIO uAPI, but on finding that, due to dependency issues, the selftests are not built for the buildroot environments that I do most of my GPIO testing in, I decided to take a closer look.

The first patch is essentially a rewrite of the exising test suite. It uses a simplified abstraction of the uAPI interfaces to allow a common test suite to test the gpio-mockup using either of the uAPI interfaces. The simplified cdev interface is implemented in gpio-mockup.sh, with the actual driving of the uAPI implemented in gpio-mockup-cdev.c. The simplified sysfs interface replaces gpio-mockup-sysfs.sh and is loaded over the cdev implementation when selected.

The new tests should also be simpler to extend to cover new mockup interfaces, such as the one Bart has been working on.

I have dropped support for testing modules other than gpio-mockup from the command line options, as the tests are very gpio-mockup specific so I didn't see any calling for it.

I have also tried to emphasise in the test output that the tests are covering the gpio-mockup itself. They do perform some implicit testing of gpiolib and the uAPI interfaces, and so can be useful as smoke tests for those, but their primary focus is the gpio-mockup.

Patches 2 through 5 do some cleaning up that is now possible with the new implementation, including enabling building in buildroot environments. Patch 4 doesn't strictly clean up all the old gpio references that it could - the gpio was the only Level 1 test, so the Level 1 tests could potentially be removed, but I was unsure if there may be other implications to removing a whole test level, or that it may be useful as a placeholder in case other static LDLIBS tests are added in the future??

Patch 6 finally gets around to porting the tests to the latest GPIO uAPI.

And Patch 7 updates the config to set the CONFIG_GPIO_CDEV option that was added in v5.10.

Cheers,
Kent.

Kent Gibson (7):
selftests: gpio: rework and simplify test implementation
selftests: gpio: remove obsolete gpio-mockup-chardev.c
selftests: remove obsolete build restriction for gpio
selftests: remove obsolete gpio references from kselftest_deps.sh
tools: gpio: remove uAPI v1 code no longer used by selftests
selftests: gpio: port to GPIO uAPI v2
selftests: gpio: add CONFIG_GPIO_CDEV to config

 tools/gpio/gpio-utils.c                       |  89 ----
 tools/gpio/gpio-utils.h                       |   6 -
 tools/testing/selftests/Makefile              |   9 -
 tools/testing/selftests/gpio/Makefile         |  26 +-
 tools/testing/selftests/gpio/config           |   1 +
 .../testing/selftests/gpio/gpio-mockup-cdev.c | 198 ++++++++
 .../selftests/gpio/gpio-mockup-chardev.c      | 323 ------------
 .../selftests/gpio/gpio-mockup-sysfs.sh       | 168 ++-----
 tools/testing/selftests/gpio/gpio-mockup.sh   | 469 ++++++++++++------
 tools/testing/selftests/kselftest_deps.sh     |   4 +-
 10 files changed, 573 insertions(+), 720 deletions(-)
 create mode 100644 tools/testing/selftests/gpio/gpio-mockup-cdev.c
 delete mode 100644 tools/testing/selftests/gpio/gpio-mockup-chardev.c
Signature · 1 line
2.30.0
author reviewer
#1 lore raw

PATCH 1/7 selftests: gpio: rework and simplify test implementation

The GPIO mockup selftests are overly complicated with separate implementations of the tests for sysfs and cdev uAPI, and with the cdev implementation being dependent on tools/gpio and libmount.

Rework the test implementation to provide a common test suite with a simplified pluggable uAPI interface. The cdev implementation utilises the GPIO uAPI directly to remove the dependence on tools/gpio. The simplified uAPI interface removes the need for any file system mount checks in C, and so removes the dependence on libmount.

The rework also fixes the sysfs test implementation which has been broken since the device created in the multiple gpiochip case was split into separate devices.

Fixes commit 8a39f597bcfd ("gpio: mockup: rework device probing")
---
 tools/testing/selftests/gpio/Makefile         |  26 +-
 .../testing/selftests/gpio/gpio-mockup-cdev.c | 139 ++++++
 .../selftests/gpio/gpio-mockup-sysfs.sh       | 168 ++-----
 tools/testing/selftests/gpio/gpio-mockup.sh   | 462 ++++++++++++------
 4 files changed, 505 insertions(+), 290 deletions(-)
 create mode 100644 tools/testing/selftests/gpio/gpio-mockup-cdev.c
tools/testing/selftests/gpio/Makefile +1 -25
index 32bdc978a711..e4363c64d40d 100644
@@ -1,32 +1,8 @@
 # SPDX-License-Identifier: GPL-2.0
 
-VAR_CFLAGS := $(shell pkg-config --cflags mount 2>/dev/null)
-VAR_LDLIBS := $(shell pkg-config --libs mount 2>/dev/null)
-ifeq ($(VAR_LDLIBS),)
-VAR_LDLIBS := -lmount -I/usr/include/libmount
-endif
-
-CFLAGS += -O2 -g -std=gnu99 -Wall -I../../../../usr/include/ $(VAR_CFLAGS)
-LDLIBS += $(VAR_LDLIBS)
-
 TEST_PROGS := gpio-mockup.sh
 TEST_FILES := gpio-mockup-sysfs.sh
-TEST_PROGS_EXTENDED := gpio-mockup-chardev
-
-GPIODIR := $(realpath ../../../gpio)
-GPIOOBJ := gpio-utils.o
+TEST_GEN_PROGS_EXTENDED := gpio-mockup-cdev
 
-all: $(TEST_PROGS_EXTENDED)
-
-override define CLEAN
-	$(RM) $(TEST_PROGS_EXTENDED)
-	$(MAKE) -C $(GPIODIR) OUTPUT=$(GPIODIR)/ clean
-endef
-
-KSFT_KHDR_INSTALL := 1
 include ../lib.mk
 
-$(TEST_PROGS_EXTENDED): $(GPIODIR)/$(GPIOOBJ)
-
-$(GPIODIR)/$(GPIOOBJ):
-	$(MAKE) OUTPUT=$(GPIODIR)/ -C $(GPIODIR)
tools/testing/selftests/gpio/gpio-mockup-cdev.c new file +139 -0

139 lines not on this page · raw message

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +52 -116

187 lines not on this page · raw message

tools/testing/selftests/gpio/gpio-mockup.sh +313 -149

517 lines not on this page · raw message

Signature · 1 line
2.30.0

A message that is not in the archive (sent off-list or not yet received): CAHp75VdPdRRm+YQ-FzcFV5=XcNL6dXHDROutkgUbPLbj4xa8SQ@mail.gmail.com

author reviewer
#3 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 12 lines · Andy Shevchenko
Quoted 10 lines · Kent Gibson

The GPIO mockup selftests are overly complicated with separate implementations of the tests for sysfs and cdev uAPI, and with the cdev implementation being dependent on tools/gpio and libmount.

Rework the test implementation to provide a common test suite with a simplified pluggable uAPI interface. The cdev implementation utilises the GPIO uAPI directly to remove the dependence on tools/gpio. The simplified uAPI interface removes the need for any file system mount checks in C, and so removes the dependence on libmount.

Quoted 9 lines
Quoted 5 lines

The rework also fixes the sysfs test implementation which has been broken since the device created in the multiple gpiochip case was split into separate devices.

I briefly looked at code in shell below... there are places to improve (useless use of: cat, test, negation, etc).

My shell is clearly pretty poor, so I would really appreciate a pointer to an example of each, and I'll then hunt down the rest.

Thanks,
Kent.

#4 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

In reply to Kent Gibson
Quoted 11 lines · Kent Gibson

The GPIO mockup selftests are overly complicated with separate implementations of the tests for sysfs and cdev uAPI, and with the cdev implementation being dependent on tools/gpio and libmount.

Rework the test implementation to provide a common test suite with a simplified pluggable uAPI interface. The cdev implementation utilises the GPIO uAPI directly to remove the dependence on tools/gpio. The simplified uAPI interface removes the need for any file system mount checks in C, and so removes the dependence on libmount.

Quoted 3 lines

The rework also fixes the sysfs test implementation which has been broken since the device created in the multiple gpiochip case was split into separate devices.

Okay, I commented something, not sure if everything is correct, needs double checking.
Shell is quite a hard programming language. Everyday I found something new about it.

...

Quoted 1 line

+#include <linux/gpio.h>

Perhaps include it after system headers?

Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-cdev.c new file +7 -0
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <unistd.h>

...

Quoted 1 line

+SYSFS=mount -t sysfs | head -1 | awk '{ print $3 }'

Oh, would below be better?
grep -w sysfs /proc/mounts | cut -f2 -d' '

...

Quoted 1 line

+[ ! -d "$SYSFS" ] && skip "sysfs is not mounted"

[ -d ... ] || skip "..."

...

Quoted 1 line

+[ ! -d "$GPIO_SYSFS" ] && skip "CONFIG_GPIO_SYSFS is not selected"

Ditto.

...

Quoted 1 line

+ local platform=cat $SYSFS/kernel/debug/gpio | grep "$chip:" | tr -d ',' | awk '{print $5}'

Besides useless use of cat (and tr + awk can be simplified) why are you simply not using
/sys/bus/gpio/devices/$chip ?

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 21

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       # e.g. /sys/class/gpio/gpiochip508/device/gpiochip0/dev
+       local syschip=`ls -d $GPIO_SYSFS/gpiochip*/device/$chip/dev`

ls -d is fragile, better to use find ...

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 23

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       syschip=${syschip#$GPIO_SYSFS}
+       syschip=${syschip%/device/$chip/dev}

How does this handle more than one gpiochip listed?
Also, can you consider optimizing these to get whatever you want easily?

Quoted 1 line

+ sysfs_nr=cat $SYSFS/devices/$platform/gpio/$syschip/base

(It's probably fine here, but this doesn't work against PCI bus, for example, see above for the fix)

Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       sysfs_nr=$(($sysfs_nr + $offset))
+       sysfs_ldir=$GPIO_SYSFS/gpio$sysfs_nr
 }

...

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 41

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +4 -0
+set_line()
 {
+       if [ -z "$sysfs_nr" ]; then
+               find_sysfs_nr
+               echo $sysfs_nr > $GPIO_SYSFS/export
        fi

It sounds like a separate function (you have release_line(), perhaps acquire_line() is good to have).

Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +5 -0
+release_line()
 {
+       [ -z "$sysfs_nr" ] && return
+       echo $sysfs_nr > $GPIO_SYSFS/unexport
+       sysfs_nr=
+       sysfs_ldir=
 }

...

Quoted 1 line

+BASE=dirname $0

Can be used via shell substitutions.

...

tools/testing/selftests/gpio/gpio-mockup.sh

Quoted 4 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+skip()
 {
 
+       echo $* >&2

In all cases better to use "$*" (note surrounding double quotes).

Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       echo GPIO $module test SKIP
+       exit $ksft_skip
 }

...

Quoted 1 line

+ [ ! which modprobe > /dev/null 2>&1 ] && skip "need modprobe installed"

AFAIR which can be optional on some systems.

...

tools/testing/selftests/gpio/gpio-mockup.sh line 51

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       DEBUGFS=`mount -t debugfs | head -1 | awk '{ print $3 }'`
+       [ ! -d "$DEBUGFS" ] && skip "debugfs is not mounted"

Same as per sysfs in another script.

...

tools/testing/selftests/gpio/gpio-mockup.sh remove_module() · line 78

Quoted 5 lines
tools/testing/selftests/gpio/gpio-mockup.sh +5 -0
+try_insert_module()
+{
+       modprobe -q $module $1
+       err=$?
+       [ $err -ne 0 ] && fail "insert $module failed with error $err"

I guess it's as simple as `modprobe ... || fail "... $?"

Quoted 1 line

+}

...

Quoted 1 line

+ [ ! -e "$mock_line" ] && fail "missing line $chip:$offset"

[ -e ... ] || ...

...

tools/testing/selftests/gpio/gpio-mockup.sh remove_module() · line 265

Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +3 -0
+       local ranges=$1
+       local gc=
+       shift

I found that combination
local ranges=$1; shift
is better to read.

...

Quoted 1 line

+ gpiochip=ls -d $DEBUGFS/$module/gpiochip* 2>/dev/null

find ... is a better choice.

tools/testing/selftests/gpio/gpio-mockup.sh remove_module()

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup.sh +5 -0
+       for chip in $gpiochip; do
+               gc=`basename $chip`
+               [ -z "$1" ] && fail "unexpected chip - $gc"
+               test_line $gc 0
 
+               if [ "$random" ] && [ $1 -gt 2 ]; then

You call the test twice, while you may do it in one go.

Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup.sh +6 -0
+                       test_line $gc $((( RANDOM % ($1 - 2) + 1)))
+               fi
+               test_line $gc $(($1 - 1))
+               test_no_line $gc $1
                shift
+       done
+       [ "$1" ] && fail "missing expected chip of width $1"

...

Quoted 2 lines

+# manual gpio allocation tests fail if a physical chip already exists
+[ "$full_test" ] && [ -e "/dev/gpiochip0" ] && skip "full tests conflict with gpiochip0"

I guess it should be rather something like

[ "$full_test" = "true" -a -e "/dev/gpiochip0" ]

P.S. Also you may use #!/bin/sh -efu as shebang and fix other problems.

--
With Best Regards,
Andy Shevchenko

author reviewer
#5 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 12 lines · Andy Shevchenko
Quoted 11 lines · Kent Gibson

The GPIO mockup selftests are overly complicated with separate implementations of the tests for sysfs and cdev uAPI, and with the cdev implementation being dependent on tools/gpio and libmount.

Rework the test implementation to provide a common test suite with a simplified pluggable uAPI interface. The cdev implementation utilises the GPIO uAPI directly to remove the dependence on tools/gpio. The simplified uAPI interface removes the need for any file system mount checks in C, and so removes the dependence on libmount.

Quoted 9 lines
Quoted 3 lines

The rework also fixes the sysfs test implementation which has been broken since the device created in the multiple gpiochip case was split into separate devices.

Okay, I commented something, not sure if everything is correct, needs double checking.
Shell is quite a hard programming language. Everyday I found something new about it.

You are telling me - there are about six million ways to do even the most trivial tasks. Makes you appreciate more constrained languages.

Quoted 6 lines

...

Quoted 1 line

+#include <linux/gpio.h>

Perhaps include it after system headers?

hehe, I blindly sorted them.
Should it matter?

Quoted 10 lines
Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-cdev.c new file +7 -0
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <unistd.h>

...

Quoted 5 lines
Quoted 1 line

+SYSFS=mount -t sysfs | head -1 | awk '{ print $3 }'

Oh, would below be better?
grep -w sysfs /proc/mounts | cut -f2 -d' '

That looks good - the other is a carry over from the old gpio-mockup.sh.

Quoted 6 lines

...

Quoted 1 line

+[ ! -d "$SYSFS" ] && skip "sysfs is not mounted"

[ -d ... ] || skip "..."

Yeah, those were if [ .. ]; then fi originally. I did the first step of simplification and missed the second :-(.

Quoted 11 lines

...

Quoted 1 line

+[ ! -d "$GPIO_SYSFS" ] && skip "CONFIG_GPIO_SYSFS is not selected"

Ditto.

...

Quoted 1 line

+ local platform=cat $SYSFS/kernel/debug/gpio | grep "$chip:" | tr -d ',' | awk '{print $5}'

Besides useless use of cat (and tr + awk can be simplified) why are

What do you suggest for the tr/awk simplification?

Quoted 3 lines

you simply not using
/sys/bus/gpio/devices/$chip ?

Cos that shows all the gpiochips, not just the ones created by gpio-mockup. And I certainly don't want to go messing with real hardware. The default tests should still run on real hardware - but only accessing the mockup devices.

Got a better way to filter out real hardware?

Quoted 5 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 21

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       # e.g. /sys/class/gpio/gpiochip508/device/gpiochip0/dev
+       local syschip=`ls -d $GPIO_SYSFS/gpiochip*/device/$chip/dev`

ls -d is fragile, better to use find ...

OK

Quoted 4 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 23

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       syschip=${syschip#$GPIO_SYSFS}
+       syschip=${syschip%/device/$chip/dev}

How does this handle more than one gpiochip listed?

It is filtered by $chip so there can only be one.
Or is that a false assumption?

Quoted 2 lines

Also, can you consider optimizing these to get whatever you want easily?

Sadly that IS my optimized way - I don't know of an easier way to find the sysfs GPIO number given the gpiochip and offset :-(. Happy to learn of any alternative.

Quoted 5 lines
Quoted 1 line

+ sysfs_nr=cat $SYSFS/devices/$platform/gpio/$syschip/base

(It's probably fine here, but this doesn't work against PCI bus, for example, see above for the fix)

Not sure what you mean here.

Quoted 13 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       sysfs_nr=$(($sysfs_nr + $offset))
+       sysfs_ldir=$GPIO_SYSFS/gpio$sysfs_nr
 }

...

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 41

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +4 -0
+set_line()
 {
+       if [ -z "$sysfs_nr" ]; then
+               find_sysfs_nr
+               echo $sysfs_nr > $GPIO_SYSFS/export
        fi
Quoted 3 lines

It sounds like a separate function (you have release_line(), perhaps acquire_line() is good to have).

The cdev implementation has to release and re-acquire in the background as there is no simple way to perform a set_config on a requested line from shell - just holding the requested line for a set is painful enough, and the goal here was to keep the tests simple.

I didn't want to make line acquisition/release explicit in the gpio-mockup tests, as that would make them needlessly complicated, so the acquire is bundled into the set_line - and anywhere else the uAPI implementation needs it. There is an implicit assumption that a set_line will always be called before a get_line, but that is always true - there is no "as-is" being tested here.

Of course you still need the release_line at the end of the test, so that is still there.

Quoted 8 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 71

Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +5 -0
+release_line()
 {
+       [ -z "$sysfs_nr" ] && return
+       echo $sysfs_nr > $GPIO_SYSFS/unexport
+       sysfs_nr=
+       sysfs_ldir=
 }
Quoted 6 lines

...

Quoted 1 line

+BASE=dirname $0

Can be used via shell substitutions.

Yup

Quoted 9 lines

...

tools/testing/selftests/gpio/gpio-mockup.sh

Quoted 4 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+skip()
 {
 
+       echo $* >&2

In all cases better to use "$*" (note surrounding double quotes).

Agreed - except where

for option in $*; do

is used to parse parameters.

Quoted 10 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       echo GPIO $module test SKIP
+       exit $ksft_skip
 }

...

Quoted 1 line

+ [ ! which modprobe > /dev/null 2>&1 ] && skip "need modprobe installed"

AFAIR which can be optional on some systems.

That is how other selftests check for availability of modprobe. e.g. selftests/kmod/kmod.sh and selftests/vm/test_hmm.sh, so I assumed it was acceptable.

Is there an alternative?

Quoted 9 lines

...

tools/testing/selftests/gpio/gpio-mockup.sh line 51

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       DEBUGFS=`mount -t debugfs | head -1 | awk '{ print $3 }'`
+       [ ! -d "$DEBUGFS" ] && skip "debugfs is not mounted"

Same as per sysfs in another script.

...

Quoted 8 lines

tools/testing/selftests/gpio/gpio-mockup.sh remove_module() · line 78

Quoted 5 lines
tools/testing/selftests/gpio/gpio-mockup.sh +5 -0
+try_insert_module()
+{
+       modprobe -q $module $1
+       err=$?
+       [ $err -ne 0 ] && fail "insert $module failed with error $err"

I guess it's as simple as `modprobe ... || fail "... $?"

Yup

Quoted 10 lines
Quoted 1 line

+}

...

Quoted 1 line

+ [ ! -e "$mock_line" ] && fail "missing line $chip:$offset"

[ -e ... ] || ...

...

Quoted 8 lines

tools/testing/selftests/gpio/gpio-mockup.sh remove_module() · line 265

Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +3 -0
+       local ranges=$1
+       local gc=
+       shift

I found that combination
local ranges=$1; shift
is better to read.

Agreed - the gc certainly shouldn't be splitting the two.

Quoted 6 lines

...

Quoted 1 line

+ gpiochip=ls -d $DEBUGFS/$module/gpiochip* 2>/dev/null

find ... is a better choice.

Quoted 9 lines

tools/testing/selftests/gpio/gpio-mockup.sh remove_module()

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup.sh +5 -0
+       for chip in $gpiochip; do
+               gc=`basename $chip`
+               [ -z "$1" ] && fail "unexpected chip - $gc"
+               test_line $gc 0
 
+               if [ "$random" ] && [ $1 -gt 2 ]; then

You call the test twice, while you may do it in one go.

Ahh, replacing the && with -a. Good to know.

Quoted 8 lines

tools/testing/selftests/gpio/gpio-mockup.sh remove_module() · line 282

Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup.sh +6 -0
+                       test_line $gc $((( RANDOM % ($1 - 2) + 1)))
+               fi
+               test_line $gc $(($1 - 1))
+               test_no_line $gc $1
                shift
+       done
+       [ "$1" ] && fail "missing expected chip of width $1"
Quoted 9 lines

...

Quoted 2 lines

+# manual gpio allocation tests fail if a physical chip already exists
+[ "$full_test" ] && [ -e "/dev/gpiochip0" ] && skip "full tests conflict with gpiochip0"

I guess it should be rather something like

[ "$full_test" = "true" -a -e "/dev/gpiochip0" ]

I'm going with empty for false, so you can drop the = "true" here.

Quoted 2 lines

P.S. Also you may use #!/bin/sh -efu as shebang and fix other problems.

A shebang or a set -efu?
I don't see shebang options used anywhere in the selftest scripts, but I agree with a set.

Either way I am unsure what the shebang should be.
The majority of the selftest scripts use bash as the shebang, with the remainder using plain sh.
These scripts do use some bash extensions, and it was originally bash, so I left it as that.
My test setups mainly use busybox, and don't have bash, so they complain about the bash shebang - though the ash(??) busybox is using still runs the script fine.

Thanks again for the review - always a learning experience.

Cheers,
Kent.

#6 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 2 lines · Kent Gibson
Quoted 1 line · Andy Shevchenko

On Sat, Jan 2, 2021 at 4:32 AM Kent Gibson <[email protected]> wrote:

...

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

+#include <linux/gpio.h>

Perhaps include it after system headers?

hehe, I blindly sorted them.
Should it matter?

I would include more particular headers later.
Btw system headers can not always be in order because of dependencies.

...

Quoted 5 lines
Quoted 3 lines
Quoted 1 line

+ local platform=cat $SYSFS/kernel/debug/gpio | grep "$chip:" | tr -d ',' | awk '{print $5}'

Besides useless use of cat (and tr + awk can be simplified) why are

What do you suggest for the tr/awk simplification?

You have awk, you can easily switch the entire pipeline to a little awk scriptlet.

Quoted 4 lines
Quoted 2 lines

you simply not using
/sys/bus/gpio/devices/$chip ?

Cos that shows all the gpiochips, not just the ones created by gpio-mockup.

I didn't get this. What is the content of $chip in your case?

Quoted 5 lines

And I certainly don't want to go messing with real hardware. The default tests should still run on real hardware - but only accessing the mockup devices.

Got a better way to filter out real hardware?

I probably have to understand what is the input and what is the expected output. It's possible I missed something here.

Quoted 7 lines
Quoted 4 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 21

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       # e.g. /sys/class/gpio/gpiochip508/device/gpiochip0/dev
+       local syschip=`ls -d $GPIO_SYSFS/gpiochip*/device/$chip/dev`

ls -d is fragile, better to use find ...

OK

Quoted 7 lines
Quoted 4 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 23

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       syschip=${syschip#$GPIO_SYSFS}
+       syschip=${syschip%/device/$chip/dev}

How does this handle more than one gpiochip listed?

It is filtered by $chip so there can only be one.
Or is that a false assumption?

When you have glob() in use it may return any number of results (starting from 0) and your script should be prepared for that.

Quoted 5 lines
Quoted 1 line

Also, can you consider optimizing these to get whatever you want easily?

Sadly that IS my optimized way - I don't know of an easier way to find the sysfs GPIO number given the gpiochip and offset :-(. Happy to learn of any alternative.

I'm talking about getting $syschip. I think there is a way to get it without all those shell substitutions from somewhere else.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

+ sysfs_nr=cat $SYSFS/devices/$platform/gpio/$syschip/base

(It's probably fine here, but this doesn't work against PCI bus, for example, see above for the fix)

Not sure what you mean here.

When GPIO is a PCI device the above won't give a proper path. If we wish to give an example to somebody, it would be better to have it good enough.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       sysfs_nr=$(($sysfs_nr + $offset))
+       sysfs_ldir=$GPIO_SYSFS/gpio$sysfs_nr
 }

...

Quoted 23 lines
Quoted 10 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 41

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +4 -0
+set_line()
 {
+       if [ -z "$sysfs_nr" ]; then
+               find_sysfs_nr
+               echo $sysfs_nr > $GPIO_SYSFS/export
        fi

It sounds like a separate function (you have release_line(), perhaps acquire_line() is good to have).

The cdev implementation has to release and re-acquire in the background as there is no simple way to perform a set_config on a requested line from shell - just holding the requested line for a set is painful enough, and the goal here was to keep the tests simple.

I didn't want to make line acquisition/release explicit in the gpio-mockup tests, as that would make them needlessly complicated, so the acquire is bundled into the set_line - and anywhere else the uAPI implementation needs it. There is an implicit assumption that a set_line will always be called before a get_line, but that is always true - there is no "as-is" being tested here.

Quoted 2 lines

Of course you still need the release_line at the end of the test, so that is still there.

Yes and to me logically correct to distinguish acquire_line() with set_line(). Then wherever you need to set_line(), you may call acquire_line() which should be idempotent (the same way as release_line() call).

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +5 -0
+release_line()
 {
+       [ -z "$sysfs_nr" ] && return
+       echo $sysfs_nr > $GPIO_SYSFS/unexport
+       sysfs_nr=
+       sysfs_ldir=
 }

...

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+skip()
 {
 
+       echo $* >&2
Quoted 8 lines
Quoted 2 lines

In all cases better to use "$*" (note surrounding double quotes).

Agreed - except where

for option in $*; do

is used to parse parameters.

Exactly! And "" helps with that.

If I put parameters as a b c "d e", your case will take them wrongly.

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       echo GPIO $module test SKIP
+       exit $ksft_skip
 }

...

Quoted 1 line

+ [ ! which modprobe > /dev/null 2>&1 ] && skip "need modprobe installed"

Quoted 8 lines
Quoted 2 lines

AFAIR which can be optional on some systems.

That is how other selftests check for availability of modprobe. e.g. selftests/kmod/kmod.sh and selftests/vm/test_hmm.sh, so I assumed it was acceptable.

Is there an alternative?

OK. Just replace it with a dropped useless test call. which ... || skip ...

...

Quoted 3 lines
Quoted 1 line

P.S. Also you may use #!/bin/sh -efu as shebang and fix other problems.

A shebang or a set -efu?

Shebang. The difference is that with shebang you don't need to edit the script each time you want to change that.
sh -x /path/to/the/script will give different results.

Quoted 2 lines

I don't see shebang options used anywhere in the selftest scripts, but I agree with a set.

Because shell scripts in the kernel are really badly written (so does Python ones).
Again, even senior developers can't get shell right (including me).

Quoted 8 lines

Either way I am unsure what the shebang should be.
The majority of the selftest scripts use bash as the shebang, with the remainder using plain sh.
These scripts do use some bash extensions, and it was originally bash, so I left it as that.
My test setups mainly use busybox, and don't have bash, so they complain about the bash shebang - though the ash(??) busybox is using still runs the script fine.

I'm using busybox on an everyday basis and mentioned shebang works there if I'm not mistaken.
Because all flags are listed in the standard.
https://pubs.opengroup.org/onlinepubs/007904875/utilities/sh.html

Quoted 1 line

Thanks again for the review - always a learning experience.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author reviewer
#7 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 18 lines · Andy Shevchenko
Quoted 2 lines · Kent Gibson
Quoted 1 line · Andy Shevchenko

On Sat, Jan 2, 2021 at 4:32 AM Kent Gibson <[email protected]> wrote:

...

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

+#include <linux/gpio.h>

Perhaps include it after system headers?

hehe, I blindly sorted them.
Should it matter?

I would include more particular headers later.
Btw system headers can not always be in order because of dependencies.

...

Quoted 9 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

+ local platform=cat $SYSFS/kernel/debug/gpio | grep "$chip:" | tr -d ',' | awk '{print $5}'

Besides useless use of cat (and tr + awk can be simplified) why are

What do you suggest for the tr/awk simplification?

You have awk, you can easily switch the entire pipeline to a little awk scriptlet.

Ah ok - I was actually going the other way to do away with the awk, so had replaced it with a pair of cuts, though I'm still looking for better alternatives for the whole gpiochipN:offset -> sysfs_nr mapping problem
- see below.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

you simply not using
/sys/bus/gpio/devices/$chip ?

Cos that shows all the gpiochips, not just the ones created by gpio-mockup.

I didn't get this. What is the content of $chip in your case?

$chip is the gpiochipN name, so gpiochip0, gpiochip1 etc.

What we are trying to find here is the base of the GPIO numbering for the chip so we can export/unexport them to sysfs (after adding the offset for the particular line).

Quoted 19 lines
Quoted 5 lines

And I certainly don't want to go messing with real hardware. The default tests should still run on real hardware - but only accessing the mockup devices.

Got a better way to filter out real hardware?

I probably have to understand what is the input and what is the expected output. It's possible I missed something here.

Quoted 10 lines
Quoted 4 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 21

Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       # e.g. /sys/class/gpio/gpiochip508/device/gpiochip0/dev
+       local syschip=`ls -d $GPIO_SYSFS/gpiochip*/device/$chip/dev`

ls -d is fragile, better to use find ...

OK

Quoted 3 lines
Quoted 2 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       syschip=${syschip#$GPIO_SYSFS}
+       syschip=${syschip%/device/$chip/dev}
Quoted 8 lines
Quoted 4 lines
Quoted 1 line

How does this handle more than one gpiochip listed?

It is filtered by $chip so there can only be one.
Or is that a false assumption?

When you have glob() in use it may return any number of results (starting from 0) and your script should be prepared for that.

Yeah, we really don't want to be using globs at all.

Quoted 9 lines
Quoted 5 lines
Quoted 1 line

Also, can you consider optimizing these to get whatever you want easily?

Sadly that IS my optimized way - I don't know of an easier way to find the sysfs GPIO number given the gpiochip and offset :-(. Happy to learn of any alternative.

I'm talking about getting $syschip. I think there is a way to get it without all those shell substitutions from somewhere else.

$syschip is just an intermediate that I'm not really interested in - it just helps find the base, and so the nr.

I've been playing with alternatives and my current one is:

	# e.g. /sys/devices/platform/gpio-mockup.1/gpiochip1
	local platform=$(find $SYSFS/devices/platform/ -name $chip -type d -maxdepth 2)
	[ "$platform" ] || fail "can't find platform of $chip"
	# e.g. /sys/devices/platform/gpio-mockup.1/gpio/gpiochip508/base
	local base=$(find $(dirname $platform)/gpio/ -name base -type f -maxdepth 2)
	[ "$base" ] || fail "can't find base of $chip"
	sysfs_nr=$(< $base)
	sysfs_nr=$(($sysfs_nr + $offset))

which works, though still doesn't handle the possibility of multiple matches returned by the finds.

Quoted 11 lines
Quoted 6 lines
Quoted 4 lines
Quoted 1 line

+ sysfs_nr=cat $SYSFS/devices/$platform/gpio/$syschip/base

(It's probably fine here, but this doesn't work against PCI bus, for example, see above for the fix)

Not sure what you mean here.

When GPIO is a PCI device the above won't give a proper path. If we wish to give an example to somebody, it would be better to have it good enough.

How would it appear for PCI bus?

Quoted 29 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +2 -0
+       sysfs_nr=$(($sysfs_nr + $offset))
+       sysfs_ldir=$GPIO_SYSFS/gpio$sysfs_nr
 }

...

Quoted 23 lines
Quoted 10 lines

tools/testing/selftests/gpio/gpio-mockup-sysfs.sh line 41

Quoted 6 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +4 -0
+set_line()
 {
+       if [ -z "$sysfs_nr" ]; then
+               find_sysfs_nr
+               echo $sysfs_nr > $GPIO_SYSFS/export
        fi

It sounds like a separate function (you have release_line(), perhaps acquire_line() is good to have).

The cdev implementation has to release and re-acquire in the background as there is no simple way to perform a set_config on a requested line from shell - just holding the requested line for a set is painful enough, and the goal here was to keep the tests simple.

I didn't want to make line acquisition/release explicit in the gpio-mockup tests, as that would make them needlessly complicated, so the acquire is bundled into the set_line - and anywhere else the uAPI implementation needs it. There is an implicit assumption that a set_line will always be called before a get_line, but that is always true - there is no "as-is" being tested here.

Quoted 7 lines
Quoted 2 lines

Of course you still need the release_line at the end of the test, so that is still there.

Yes and to me logically correct to distinguish acquire_line() with set_line(). Then wherever you need to set_line(), you may call acquire_line() which should be idempotent (the same way as release_line() call).

Oh, ok - it would only be called from set_line - I thought you meant expose it as part of the uAPI test interface (currently get_line/set_line/release_line).

Quoted 39 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/gpio/gpio-mockup-sysfs.sh +5 -0
+release_line()
 {
+       [ -z "$sysfs_nr" ] && return
+       echo $sysfs_nr > $GPIO_SYSFS/unexport
+       sysfs_nr=
+       sysfs_ldir=
 }

...

Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+skip()
 {
 
+       echo $* >&2
Quoted 8 lines
Quoted 2 lines

In all cases better to use "$*" (note surrounding double quotes).

Agreed - except where

for option in $*; do

is used to parse parameters.

Exactly! And "" helps with that.

If I put parameters as a b c "d e", your case will take them wrongly.

Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
tools/testing/selftests/gpio/gpio-mockup.sh +2 -0
+       echo GPIO $module test SKIP
+       exit $ksft_skip
 }

...

Quoted 1 line

+ [ ! which modprobe > /dev/null 2>&1 ] && skip "need modprobe installed"

AFAIR which can be optional on some systems.

Quoted 9 lines
Quoted 5 lines

That is how other selftests check for availability of modprobe. e.g. selftests/kmod/kmod.sh and selftests/vm/test_hmm.sh, so I assumed it was acceptable.

Is there an alternative?

OK. Just replace it with a dropped useless test call. which ... || skip ...

Yup - I've since replaced it with a test call to modprobe -h, so no which required.

Quoted 10 lines

...

Quoted 3 lines
Quoted 1 line

P.S. Also you may use #!/bin/sh -efu as shebang and fix other problems.

A shebang or a set -efu?

Shebang. The difference is that with shebang you don't need to edit the script each time you want to change that.
sh -x /path/to/the/script will give different results.

OK, didn't consider that.
Have got the scripts running with the -efu flags set - that was entertaining.

Quoted 16 lines
Quoted 2 lines

I don't see shebang options used anywhere in the selftest scripts, but I agree with a set.

Because shell scripts in the kernel are really badly written (so does Python ones).
Again, even senior developers can't get shell right (including me).

Quoted 8 lines

Either way I am unsure what the shebang should be.
The majority of the selftest scripts use bash as the shebang, with the remainder using plain sh.
These scripts do use some bash extensions, and it was originally bash, so I left it as that.
My test setups mainly use busybox, and don't have bash, so they complain about the bash shebang - though the ash(??) busybox is using still runs the script fine.

Quoted 5 lines

I'm using busybox on an everyday basis and mentioned shebang works there if I'm not mistaken.
Because all flags are listed in the standard.
https://pubs.opengroup.org/onlinepubs/007904875/utilities/sh.html

I meant the actual /bin/bash, not the flags.
Though I now build bash in my buildroots, so I don't get that warning anymore.

Cheers,
Kent.

author reviewer
#8 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

In reply to Andy Shevchenko
Quoted 4 lines · Andy Shevchenko
Quoted 2 lines · Kent Gibson
Quoted 1 line · Andy Shevchenko

On Sat, Jan 2, 2021 at 4:32 AM Kent Gibson <[email protected]> wrote:

[snip]

Quoted 12 lines

...

Quoted 5 lines
Quoted 3 lines
Quoted 1 line

+ local platform=cat $SYSFS/kernel/debug/gpio | grep "$chip:" | tr -d ',' | awk '{print $5}'

Besides useless use of cat (and tr + awk can be simplified) why are

What do you suggest for the tr/awk simplification?

You have awk, you can easily switch the entire pipeline to a little awk scriptlet.

Baah, the number that I'm after is in the $SYSFS/kernel/debug/gpio that I was pulling the platform from, so I can just pull it directly from there.

No need to go hunting through the file system for the base file - the range of GPIOs assigned to the chip is right there.

In this example it is the 508:

# e.g. gpiochip0: GPIOs 508-511, parent: platform/gpio-mockup.0, gpio-mockup-A:

So I'll use that - unless it is unreliable for some reason?

Cheers,
Kent.

#9 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 1 line · Kent Gibson

On Sun, Jan 03, 2021 at 05:10:10PM +0200, Andy Shevchenko wrote:

...

Quoted 5 lines

In this example it is the 508:

# e.g. gpiochip0: GPIOs 508-511, parent: platform/gpio-mockup.0, gpio-mockup-A:

So I'll use that - unless it is unreliable for some reason?

debugfs is not an ABI and tomorrow this can be changed without notice.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author reviewer
#10 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 5 lines · Andy Shevchenko
Quoted 1 line · Kent Gibson

On Sun, Jan 03, 2021 at 05:10:10PM +0200, Andy Shevchenko wrote:

...

Quoted 8 lines
Quoted 5 lines

In this example it is the 508:

# e.g. gpiochip0: GPIOs 508-511, parent: platform/gpio-mockup.0, gpio-mockup-A:

So I'll use that - unless it is unreliable for some reason?

debugfs is not an ABI and tomorrow this can be changed without notice.

I had a bad feeling that might be the case, and all my current solutions use debugfs one way or another, so back to the drawing board on that one.

Thanks,
Kent.

author reviewer
#11 lore raw

Re: [PATCH 1/7] selftests: gpio: rework and simplify test implementation

Quoted 10 lines · Kent Gibson
Quoted 9 lines · Andy Shevchenko
Quoted 1 line · Kent Gibson

On Sun, Jan 03, 2021 at 05:10:10PM +0200, Andy Shevchenko wrote:

...

Quoted 4 lines

In this example it is the 508:

# e.g. gpiochip0: GPIOs 508-511, parent: platform/gpio-mockup.0, gpio-mockup-A:
Quoted 8 lines
Quoted 4 lines
Quoted 1 line

So I'll use that - unless it is unreliable for some reason?

debugfs is not an ABI and tomorrow this can be changed without notice.

I had a bad feeling that might be the case, and all my current solutions use debugfs one way or another, so back to the drawing board on that one.

Hang on - the find approach that I was looking at previously only uses /sys/devices/platform, so I'll revert to that one - and add handling for multi-match.

Cheers,
Kent.

author reviewer
#12 lore raw

PATCH 2/7 selftests: gpio: remove obsolete gpio-mockup-chardev.c

In reply to Kent Gibson

GPIO selftests have changed to new gpio-mockup-cdev helper, so remove old gpio-mockup-chardev helper.

---
 .../selftests/gpio/gpio-mockup-chardev.c      | 323 ------------------
 1 file changed, 323 deletions(-)
 delete mode 100644 tools/testing/selftests/gpio/gpio-mockup-chardev.c
tools/testing/selftests/gpio/gpio-mockup-chardev.c deleted +0 -323
deleted file mode 100644
index 73ead8828d3a..000000000000
@@ -1,323 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-only
-/*
- * GPIO chardev test helper
- *
- * Copyright (C) 2016 Bamvor Jian Zhang
- */
-
-#define _GNU_SOURCE
-#include <unistd.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <errno.h>
-#include <string.h>
-#include <fcntl.h>
-#include <getopt.h>
-#include <sys/ioctl.h>
-#include <libmount.h>
-#include <err.h>
-#include <dirent.h>
-#include <linux/gpio.h>
-#include "../../../gpio/gpio-utils.h"
-
-#define CONSUMER	"gpio-selftest"
-#define	GC_NUM		10
-enum direction {
-	OUT,
-	IN
-};
-
-static int get_debugfs(char **path)
-{
-	struct libmnt_context *cxt;
-	struct libmnt_table *tb;
-	struct libmnt_iter *itr = NULL;
-	struct libmnt_fs *fs;
-	int found = 0, ret;
-
-	cxt = mnt_new_context();
-	if (!cxt)
-		err(EXIT_FAILURE, "libmount context allocation failed");
-
-	itr = mnt_new_iter(MNT_ITER_FORWARD);
-	if (!itr)
-		err(EXIT_FAILURE, "failed to initialize libmount iterator");
-
-	if (mnt_context_get_mtab(cxt, &tb))
-		err(EXIT_FAILURE, "failed to read mtab");
-
-	while (mnt_table_next_fs(tb, itr, &fs) == 0) {
-		const char *type = mnt_fs_get_fstype(fs);
-
-		if (!strcmp(type, "debugfs")) {
-			found = 1;
-			break;
-		}
-	}
-	if (found) {
-		ret = asprintf(path, "%s/gpio", mnt_fs_get_target(fs));
-		if (ret < 0)
-			err(EXIT_FAILURE, "failed to format string");
-	}
-
-	mnt_free_iter(itr);
-	mnt_free_context(cxt);
-
-	if (!found)
-		return -1;
-
-	return 0;
-}
-
-static int gpio_debugfs_get(const char *consumer, int *dir, int *value)
-{
-	char *debugfs;
-	FILE *f;
-	char *line = NULL;
-	size_t len = 0;
-	char *cur;
-	int found = 0;
-
-	if (get_debugfs(&debugfs) != 0)
-		err(EXIT_FAILURE, "debugfs is not mounted");
-
-	f = fopen(debugfs, "r");
-	if (!f)
-		err(EXIT_FAILURE, "read from gpio debugfs failed");
-
-	/*
-	 * gpio-2   (                    |gpio-selftest               ) in  lo
-	 */
-	while (getline(&line, &len, f) != -1) {
-		cur = strstr(line, consumer);
-		if (cur == NULL)
-			continue;
-
-		cur = strchr(line, ')');
-		if (!cur)
-			continue;
-
-		cur += 2;
-		if (!strncmp(cur, "out", 3)) {
-			*dir = OUT;
-			cur += 4;
-		} else if (!strncmp(cur, "in", 2)) {
-			*dir = IN;
-			cur += 4;
-		}
-
-		if (!strncmp(cur, "hi", 2))
-			*value = 1;
-		else if (!strncmp(cur, "lo", 2))
-			*value = 0;
-
-		found = 1;
-		break;
-	}
-	free(debugfs);
-	fclose(f);
-	free(line);
-
-	if (!found)
-		return -1;
-
-	return 0;
-}
-
-static struct gpiochip_info *list_gpiochip(const char *gpiochip_name, int *ret)
-{
-	struct gpiochip_info *cinfo;
-	struct gpiochip_info *current;
-	const struct dirent *ent;
-	DIR *dp;
-	char *chrdev_name;
-	int fd;
-	int i = 0;
-
-	cinfo = calloc(sizeof(struct gpiochip_info) * 4, GC_NUM + 1);
-	if (!cinfo)
-		err(EXIT_FAILURE, "gpiochip_info allocation failed");
-
-	current = cinfo;
-	dp = opendir("/dev");
-	if (!dp) {
-		*ret = -errno;
-		goto error_out;
-	} else {
-		*ret = 0;
-	}
-
-	while (ent = readdir(dp), ent) {
-		if (check_prefix(ent->d_name, "gpiochip")) {
-			*ret = asprintf(&chrdev_name, "/dev/%s", ent->d_name);
-			if (*ret < 0)
-				goto error_out;
-
-			fd = open(chrdev_name, 0);
-			if (fd == -1) {
-				*ret = -errno;
-				fprintf(stderr, "Failed to open %s\n",
-					chrdev_name);
-				goto error_close_dir;
-			}
-			*ret = ioctl(fd, GPIO_GET_CHIPINFO_IOCTL, current);
-			if (*ret == -1) {
-				perror("Failed to issue CHIPINFO IOCTL\n");
-				goto error_close_dir;
-			}
-			close(fd);
-			if (strcmp(current->label, gpiochip_name) == 0
-			    || check_prefix(current->label, gpiochip_name)) {
-				*ret = 0;
-				current++;
-				i++;
-			}
-		}
-	}
-
-	if ((!*ret && i == 0) || *ret < 0) {
-		free(cinfo);
-		cinfo = NULL;
-	}
-	if (!*ret && i > 0) {
-		cinfo = realloc(cinfo, sizeof(struct gpiochip_info) * 4 * i);
-		*ret = i;
-	}
-
-error_close_dir:
-	closedir(dp);
-error_out:
-	if (*ret < 0)
-		err(EXIT_FAILURE, "list gpiochip failed: %s", strerror(*ret));
-
-	return cinfo;
-}
-
-int gpio_pin_test(struct gpiochip_info *cinfo, int line, int flag, int value)
-{
-	struct gpiohandle_data data;
-	unsigned int lines[] = {line};
-	int fd;
-	int debugfs_dir = IN;
-	int debugfs_value = 0;
-	int ret;
-
-	data.values[0] = value;
-	ret = gpiotools_request_linehandle(cinfo->name, lines, 1, flag, &data,
-					   CONSUMER);
-	if (ret < 0)
-		goto fail_out;
-	else
-		fd = ret;
-
-	ret = gpio_debugfs_get(CONSUMER, &debugfs_dir, &debugfs_value);
-	if (ret) {
-		ret = -EINVAL;
-		goto fail_out;
-	}
-	if (flag & GPIOHANDLE_REQUEST_INPUT) {
-		if (debugfs_dir != IN) {
-			errno = -EINVAL;
-			ret = -errno;
-		}
-	} else if (flag & GPIOHANDLE_REQUEST_OUTPUT) {
-		if (flag & GPIOHANDLE_REQUEST_ACTIVE_LOW)
-			debugfs_value = !debugfs_value;
-
-		if (!(debugfs_dir == OUT && value == debugfs_value)) {
-			errno = -EINVAL;
-			ret = -errno;
-		}
-	}
-	gpiotools_release_linehandle(fd);
-
-fail_out:
-	if (ret)
-		err(EXIT_FAILURE, "gpio<%s> line<%d> test flag<0x%x> value<%d>",
-		    cinfo->name, line, flag, value);
-
-	return ret;
-}
-
-void gpio_pin_tests(struct gpiochip_info *cinfo, unsigned int line)
-{
-	printf("line<%d>", line);
-	gpio_pin_test(cinfo, line, GPIOHANDLE_REQUEST_OUTPUT, 0);
-	printf(".");
-	gpio_pin_test(cinfo, line, GPIOHANDLE_REQUEST_OUTPUT, 1);
-	printf(".");
-	gpio_pin_test(cinfo, line,
-		      GPIOHANDLE_REQUEST_OUTPUT | GPIOHANDLE_REQUEST_ACTIVE_LOW,
-		      0);
-	printf(".");
-	gpio_pin_test(cinfo, line,
-		      GPIOHANDLE_REQUEST_OUTPUT | GPIOHANDLE_REQUEST_ACTIVE_LOW,
-		      1);
-	printf(".");
-	gpio_pin_test(cinfo, line, GPIOHANDLE_REQUEST_INPUT, 0);
-	printf(".");
-}
-
-/*
- * ./gpio-mockup-chardev gpio_chip_name_prefix is_valid_gpio_chip
- * Return 0 if successful or exit with EXIT_FAILURE if test failed.
- * gpio_chip_name_prefix: The prefix of gpiochip you want to test. E.g.
- *			  gpio-mockup
- * is_valid_gpio_chip:	  Whether the gpio_chip is valid. 1 means valid,
- *			  0 means invalid which could not be found by
- *			  list_gpiochip.
- */
-int main(int argc, char *argv[])
-{
-	char *prefix;
-	int valid;
-	struct gpiochip_info *cinfo;
-	struct gpiochip_info *current;
-	int i;
-	int ret;
-
-	if (argc < 3) {
-		printf("Usage: %s prefix is_valid", argv[0]);
-		exit(EXIT_FAILURE);
-	}
-
-	prefix = argv[1];
-	valid = strcmp(argv[2], "true") == 0 ? 1 : 0;
-
-	printf("Test gpiochip %s: ", prefix);
-	cinfo = list_gpiochip(prefix, &ret);
-	if (!cinfo) {
-		if (!valid && ret == 0) {
-			printf("Invalid test successful\n");
-			ret = 0;
-			goto out;
-		} else {
-			ret = -EINVAL;
-			goto out;
-		}
-	} else if (cinfo && !valid) {
-		ret = -EINVAL;
-		goto out;
-	}
-	current = cinfo;
-	for (i = 0; i < ret; i++) {
-		gpio_pin_tests(current, 0);
-		gpio_pin_tests(current, current->lines - 1);
-		gpio_pin_tests(current, random() % current->lines);
-		current++;
-	}
-	ret = 0;
-	printf("successful\n");
-
-out:
-	if (ret)
-		fprintf(stderr, "gpio<%s> test failed\n", prefix);
-
-	if (cinfo)
-		free(cinfo);
-
-	if (ret)
-		exit(EXIT_FAILURE);
-
-	return ret;
-}
Signature · 1 line
2.30.0
author reviewer
#13 lore raw

PATCH 3/7 selftests: remove obsolete build restriction for gpio

In reply to Kent Gibson

Build restrictions related to the gpio-mockup-chardev helper are no longer relevant so remove them.

---
 tools/testing/selftests/Makefile | 9 ---------
 1 file changed, 9 deletions(-)
tools/testing/selftests/Makefile +0 -9
index d9c283503159..5411041e63a0 100644
@@ -121,15 +121,6 @@ ARCH           ?= $(SUBARCH)
 export KSFT_KHDR_INSTALL_DONE := 1
 export BUILD
 
-# build and run gpio when output directory is the src dir.
-# gpio has dependency on tools/gpio and builds tools/gpio
-# objects in the src directory in all cases making the src
-# repo dirty even when objects are relocated.
-ifneq (1,$(DEFAULT_INSTALL_HDR_PATH))
-	TMP := $(filter-out gpio, $(TARGETS))
-	TARGETS := $(TMP)
-endif
-
 # set default goal to all, so make without a target runs all, even when
 # all isn't the first target in the file.
 .DEFAULT_GOAL := all
Signature · 1 line
2.30.0
author reviewer
#14 lore raw

PATCH 4/7 selftests: remove obsolete gpio references from kselftest_deps.sh

In reply to Kent Gibson

GPIO Makefile has been greatly simplified so remove references to lines which no longer exist.

---
 tools/testing/selftests/kselftest_deps.sh | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tools/testing/selftests/kselftest_deps.sh +1 -3
index bbc04646346b..00e60d6eb16b 100755
@@ -129,13 +129,11 @@ l2_tests=$(grep -r --include=Makefile ": LDLIBS" | \
 		grep -v "VAR_LDLIBS" | awk -F: '{print $1}')
 
 # Level 3
-# gpio,  memfd and others use pkg-config to find mount and fuse libs
+# memfd and others use pkg-config to find mount and fuse libs
 # respectively and save it in VAR_LDLIBS. If pkg-config doesn't find
 # any, VAR_LDLIBS set to default.
 # Use the default value and filter out pkg-config for dependency check.
 # e.g:
-# gpio/Makefile
-#	VAR_LDLIBS := $(shell pkg-config --libs mount) 2>/dev/null)
 # memfd/Makefile
 #	VAR_LDLIBS := $(shell pkg-config fuse --libs 2>/dev/null)
 
Signature · 1 line
2.30.0
author reviewer
#15 lore raw

PATCH 5/7 tools: gpio: remove uAPI v1 code no longer used by selftests

In reply to Kent Gibson

gpio-mockup-chardev helper has been obsoleted and removed, so also remove the tools/gpio code that it, and nothing else, was using.

---
 tools/gpio/gpio-utils.c | 89 -----------------------------------------
 tools/gpio/gpio-utils.h |  6 ---
 2 files changed, 95 deletions(-)
tools/gpio/gpio-utils.c +0 -89
index 37187e056c8b..1639b4d832cd 100644
@@ -32,74 +32,6 @@
  * following api will request gpio lines, do the operation and then
  * release these lines.
  */
-/**
- * gpiotools_request_linehandle() - request gpio lines in a gpiochip
- * @device_name:	The name of gpiochip without prefix "/dev/",
- *			such as "gpiochip0"
- * @lines:		An array desired lines, specified by offset
- *			index for the associated GPIO device.
- * @num_lines:		The number of lines to request.
- * @flag:		The new flag for requsted gpio. Reference
- *			"linux/gpio.h" for the meaning of flag.
- * @data:		Default value will be set to gpio when flag is
- *			GPIOHANDLE_REQUEST_OUTPUT.
- * @consumer_label:	The name of consumer, such as "sysfs",
- *			"powerkey". This is useful for other users to
- *			know who is using.
- *
- * Request gpio lines through the ioctl provided by chardev. User
- * could call gpiotools_set_values() and gpiotools_get_values() to
- * read and write respectively through the returned fd. Call
- * gpiotools_release_linehandle() to release these lines after that.
- *
- * Return:		On success return the fd;
- *			On failure return the errno.
- */
-int gpiotools_request_linehandle(const char *device_name, unsigned int *lines,
-				 unsigned int num_lines, unsigned int flag,
-				 struct gpiohandle_data *data,
-				 const char *consumer_label)
-{
-	struct gpiohandle_request req;
-	char *chrdev_name;
-	int fd;
-	int i;
-	int ret;
-
-	ret = asprintf(&chrdev_name, "/dev/%s", device_name);
-	if (ret < 0)
-		return -ENOMEM;
-
-	fd = open(chrdev_name, 0);
-	if (fd == -1) {
-		ret = -errno;
-		fprintf(stderr, "Failed to open %s, %s\n",
-			chrdev_name, strerror(errno));
-		goto exit_free_name;
-	}
-
-	for (i = 0; i < num_lines; i++)
-		req.lineoffsets[i] = lines[i];
-
-	req.flags = flag;
-	strcpy(req.consumer_label, consumer_label);
-	req.lines = num_lines;
-	if (flag & GPIOHANDLE_REQUEST_OUTPUT)
-		memcpy(req.default_values, data, sizeof(req.default_values));
-
-	ret = ioctl(fd, GPIO_GET_LINEHANDLE_IOCTL, &req);
-	if (ret == -1) {
-		ret = -errno;
-		fprintf(stderr, "Failed to issue %s (%d), %s\n",
-			"GPIO_GET_LINEHANDLE_IOCTL", ret, strerror(errno));
-	}
-
-	if (close(fd) == -1)
-		perror("Failed to close GPIO character device file");
-exit_free_name:
-	free(chrdev_name);
-	return ret < 0 ? ret : req.fd;
-}
 
 /**
  * gpiotools_request_line() - request gpio lines in a gpiochip
@@ -215,27 +147,6 @@ int gpiotools_get_values(const int fd, struct gpio_v2_line_values *values)
 	return ret;
 }
 
-/**
- * gpiotools_release_linehandle(): Release the line(s) of gpiochip
- * @fd:			The fd returned by
- *			gpiotools_request_linehandle().
- *
- * Return:		On success return 0;
- *			On failure return the errno.
- */
-int gpiotools_release_linehandle(const int fd)
-{
-	int ret;
-
-	ret = close(fd);
-	if (ret == -1) {
-		perror("Failed to close GPIO LINEHANDLE device file");
-		ret = -errno;
-	}
-
-	return ret;
-}
-
 /**
  * gpiotools_release_line(): Release the line(s) of gpiochip
  * @fd:			The fd returned by
tools/gpio/gpio-utils.h +0 -6
index 6c69a9f1c253..8af7c8ee19ce 100644
@@ -24,12 +24,6 @@ static inline int check_prefix(const char *str, const char *prefix)
 		strncmp(str, prefix, strlen(prefix)) == 0;
 }
 
-int gpiotools_request_linehandle(const char *device_name, unsigned int *lines,
-				 unsigned int num_lines, unsigned int flag,
-				 struct gpiohandle_data *data,
-				 const char *consumer_label);
-int gpiotools_release_linehandle(const int fd);
-
 int gpiotools_request_line(const char *device_name,
 			   unsigned int *lines,
 			   unsigned int num_lines,
Signature · 1 line
2.30.0
author reviewer
#16 lore raw

PATCH 6/7 selftests: gpio: port to GPIO uAPI v2

In reply to Kent Gibson

Add a port to the GPIO uAPI v2 interface and make it the default.

---
 .../testing/selftests/gpio/gpio-mockup-cdev.c | 75 +++++++++++++++++--
 tools/testing/selftests/gpio/gpio-mockup.sh   | 11 ++-
 2 files changed, 76 insertions(+), 10 deletions(-)
tools/testing/selftests/gpio/gpio-mockup-cdev.c +67 -8
index 3bfd876a8b6a..e8e3d2ec662c 100644
@@ -18,6 +18,44 @@
 
 #define CONSUMER	"gpio-mockup-cdev"
 
+static int request_line_v2(int cfd, unsigned int offset,
+			   uint64_t flags, unsigned int val)
+{
+	struct gpio_v2_line_request req;
+	int ret;
+
+	memset(&req, 0, sizeof(req));
+	req.num_lines = 1;
+	req.offsets[0] = offset;
+	req.config.flags = flags;
+	strcpy(req.consumer, CONSUMER);
+	if (flags & GPIO_V2_LINE_FLAG_OUTPUT) {
+		req.config.num_attrs = 1;
+		req.config.attrs[0].mask = 1;
+		req.config.attrs[0].attr.id = GPIO_V2_LINE_ATTR_ID_OUTPUT_VALUES;
+		if (val)
+			req.config.attrs[0].attr.values = 1;
+	}
+	ret = ioctl(cfd, GPIO_V2_GET_LINE_IOCTL, &req);
+	if (ret == -1)
+		return -errno;
+	return req.fd;
+}
+
+
+static int get_value_v2(int lfd)
+{
+	struct gpio_v2_line_values vals;
+	int ret;
+
+	memset(&vals, 0, sizeof(vals));
+	vals.mask = 1;
+	ret = ioctl(lfd, GPIO_V2_LINE_GET_VALUES_IOCTL, &vals);
+	if (ret == -1)
+		return -errno;
+	return vals.bits & 0x1;
+}
+
 static int request_line_v1(int cfd, unsigned int offset,
 			   uint32_t flags, unsigned int val)
 {
@@ -57,6 +95,7 @@ static void usage(char *prog)
 	printf("               (default is to leave bias unchanged):\n");
 	printf("        -l: set line active low (default is active high)\n");
 	printf("        -s: set line value (default is to get line value)\n");
+	printf("        -u: uAPI version to use (default is 2)\n");
 	exit(-1);
 }
 
@@ -78,29 +117,42 @@ int main(int argc, char *argv[])
 {
 	char *chip;
 	int opt, ret, cfd, lfd;
-	unsigned int offset, val;
+	unsigned int offset, val, abiv;
 	uint32_t flags_v1;
+	uint64_t flags_v2;
 
+	abiv = 2;
 	ret = 0;
 	flags_v1 = GPIOHANDLE_REQUEST_INPUT;
+	flags_v2 = GPIO_V2_LINE_FLAG_INPUT;
 
 	while ((opt = getopt(argc, argv, "lb:s:u:")) != -1) {
 		switch (opt) {
 		case 'l':
 			flags_v1 |= GPIOHANDLE_REQUEST_ACTIVE_LOW;
+			flags_v2 |= GPIO_V2_LINE_FLAG_ACTIVE_LOW;
 			break;
 		case 'b':
-			if (strcmp("pull-up", optarg) == 0)
+			if (strcmp("pull-up", optarg) == 0) {
 				flags_v1 |= GPIOHANDLE_REQUEST_BIAS_PULL_UP;
-			else if (strcmp("pull-down", optarg) == 0)
+				flags_v2 |= GPIO_V2_LINE_FLAG_BIAS_PULL_UP;
+			} else if (strcmp("pull-down", optarg) == 0) {
 				flags_v1 |= GPIOHANDLE_REQUEST_BIAS_PULL_DOWN;
-			else if (strcmp("disabled", optarg) == 0)
+				flags_v2 |= GPIO_V2_LINE_FLAG_BIAS_PULL_DOWN;
+			} else if (strcmp("disabled", optarg) == 0) {
 				flags_v1 |= GPIOHANDLE_REQUEST_BIAS_DISABLE;
+				flags_v2 |= GPIO_V2_LINE_FLAG_BIAS_DISABLED;
+			}
 			break;
 		case 's':
 			val = atoi(optarg);
 			flags_v1 &= ~GPIOHANDLE_REQUEST_INPUT;
 			flags_v1 |= GPIOHANDLE_REQUEST_OUTPUT;
+			flags_v2 &= ~GPIO_V2_LINE_FLAG_INPUT;
+			flags_v2 |= GPIO_V2_LINE_FLAG_OUTPUT;
+			break;
+		case 'u':
+			abiv = atoi(optarg);
 			break;
 		default:
 			usage(argv[0]);
@@ -119,7 +171,10 @@ int main(int argc, char *argv[])
 		return -errno;
 	}
 
-	lfd = request_line_v1(cfd, offset, flags_v1, val);
+	if (abiv == 1)
+		lfd = request_line_v1(cfd, offset, flags_v1, val);
+	else
+		lfd = request_line_v2(cfd, offset, flags_v2, val);
 
 	close(cfd);
 
@@ -128,10 +183,14 @@ int main(int argc, char *argv[])
 		return lfd;
 	}
 
-	if (flags_v1 & GPIOHANDLE_REQUEST_OUTPUT)
+	if (flags_v2 & GPIO_V2_LINE_FLAG_OUTPUT) {
 		wait_signal();
-	else
-		ret = get_value_v1(lfd);
+	} else {
+		if (abiv == 1)
+			ret = get_value_v1(lfd);
+		else
+			ret = get_value_v2(lfd);
+	}
 
 	close(lfd);
 
tools/testing/selftests/gpio/gpio-mockup.sh +9 -2
index 66eed9b60963..843f829b3dd8 100755
@@ -13,6 +13,7 @@ dev_type="cdev"
 module="gpio-mockup"
 verbose=
 random=
+uapi_opt=
 active_opt=
 bias_opt=
 line_set_pid=
@@ -29,6 +30,7 @@ usage()
 	echo "-r:  test random lines as well as fence posts"
 	echo "-t:  interface type:"
 	echo "      cdev (character device ABI) - default"
+	echo "      cdev_v1 (deprecated character device ABI)"
 	echo "      sysfs (deprecated SYSFS ABI)"
 	echo "-v:  verbose progress reporting"
 	exit $ksft_fail
@@ -102,7 +104,8 @@ get_line()
 {
 	release_line
 
-	$BASE/gpio-mockup-cdev $active_opt /dev/$chip $offset > /dev/null 2>&1
+	local cdev_opts=${uapi_opt}${active_opt}
+	$BASE/gpio-mockup-cdev $cdev_opts /dev/$chip $offset > /dev/null 2>&1
 	echo $?
 }
 
@@ -144,7 +147,7 @@ set_line()
 		esac
 	done
 
-	local cdev_opts=${active_opt}
+	local cdev_opts=${uapi_opt}${active_opt}
 	if [ "$val" ]; then
 		$BASE/gpio-mockup-cdev $cdev_opts -s$val /dev/$chip $offset 2>&1 >/dev/null &
 		# failure to set is detected by reading mock and toggling values
@@ -314,6 +317,10 @@ sysfs)
 	source $BASE/gpio-mockup-sysfs.sh
 	echo "WARNING: gpio sysfs ABI is deprecated."
 	;;
+cdev_v1)
+	echo "WARNING: gpio cdev ABI v1 is deprecated."
+	uapi_opt="-u1 "
+	;;
 cdev)
 	;;
 *)
Signature · 1 line
2.30.0
author reviewer
#17 lore raw

PATCH 7/7 selftests: gpio: add CONFIG_GPIO_CDEV to config

In reply to Kent Gibson

GPIO CDEV is now optional and required for the selftests so add it to the config.

---
 tools/testing/selftests/gpio/config | 1 +
 1 file changed, 1 insertion(+)
tools/testing/selftests/gpio/config +1 -0
index abaa6902b7b6..ce100342c20b 100644
@@ -1,2 +1,3 @@
 CONFIG_GPIOLIB=y
+CONFIG_GPIO_CDEV=y
 CONFIG_GPIO_MOCKUP=m
Signature · 1 line
2.30.0
Acked-by on the v1 series
#18 lore raw
In reply to Kent Gibson
Quoted 4 lines · Kent Gibson

Initially I just wanted to port the selftests to the latest GPIO uAPI, but on finding that, due to dependency issues, the selftests are not built for the buildroot environments that I do most of my GPIO testing in, I decided to take a closer look.

All patches look good to me, I see Andy is helping you to hash out some shell script, anyway:

Yours,
Linus Walleij