Messages in this thread Patch in this message | | | From | Coiby Xu <> | Date | Fri, 25 Sep 2020 22:07:32 +0800 | Subject | Re: [PATCH 3/3] [PATCH] staging: greybus: __u8 is sufficient for snd_ctl_elem_type_t and snd_ctl_elem_iface_t |
| |
On Thu, Sep 24, 2020 at 01:00:57PM +0200, Greg Kroah-Hartman wrote: >On Thu, Sep 24, 2020 at 06:20:39PM +0800, Coiby Xu wrote: >> Use __8 to replace int and remove the unnecessary __bitwise type attribute. >> >> Found by sparse, >> >> $ make C=2 drivers/staging/greybus/ >> drivers/staging/greybus/audio_topology.c:185:24: warning: cast to restricted snd_ctl_elem_type_t >> drivers/staging/greybus/audio_topology.c:679:14: warning: restricted snd_ctl_elem_iface_t degrades to integer >> drivers/staging/greybus/audio_topology.c:906:14: warning: restricted snd_ctl_elem_iface_t degrades to integer >> >> Signed-off-by: Coiby Xu <coiby.xu@gmail.com> >> --- >> drivers/staging/greybus/audio_topology.c | 2 +- >> include/uapi/sound/asound.h | 4 ++-- >> 2 files changed, 3 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/staging/greybus/audio_topology.c b/drivers/staging/greybus/audio_topology.c >> index 56bf1a4f95ad..f6023ff390c2 100644 >> --- a/drivers/staging/greybus/audio_topology.c >> +++ b/drivers/staging/greybus/audio_topology.c >> @@ -182,7 +182,7 @@ static int gbcodec_mixer_ctl_info(struct snd_kcontrol *kcontrol, >> /* update uinfo */ >> uinfo->access = data->access; >> uinfo->count = data->vcount; >> - uinfo->type = (snd_ctl_elem_type_t)info->type; >> + uinfo->type = info->type; >> >> switch (info->type) { >> case GB_AUDIO_CTL_ELEM_TYPE_BOOLEAN: >> diff --git a/include/uapi/sound/asound.h b/include/uapi/sound/asound.h >> index 535a7229e1d9..8e71a95644ab 100644 >> --- a/include/uapi/sound/asound.h >> +++ b/include/uapi/sound/asound.h >> @@ -950,7 +950,7 @@ struct snd_ctl_card_info { >> unsigned char components[128]; /* card components / fine identification, delimited with one space (AC97 etc..) */ >> }; >> >> -typedef int __bitwise snd_ctl_elem_type_t; >> +typedef __u8 snd_ctl_elem_type_t; >> #define SNDRV_CTL_ELEM_TYPE_NONE ((__force snd_ctl_elem_type_t) 0) /* invalid */ >> #define SNDRV_CTL_ELEM_TYPE_BOOLEAN ((__force snd_ctl_elem_type_t) 1) /* boolean type */ >> #define SNDRV_CTL_ELEM_TYPE_INTEGER ((__force snd_ctl_elem_type_t) 2) /* integer type */ >> @@ -960,7 +960,7 @@ typedef int __bitwise snd_ctl_elem_type_t; >> #define SNDRV_CTL_ELEM_TYPE_INTEGER64 ((__force snd_ctl_elem_type_t) 6) /* 64-bit integer type */ >> #define SNDRV_CTL_ELEM_TYPE_LAST SNDRV_CTL_ELEM_TYPE_INTEGER64 >> >> -typedef int __bitwise snd_ctl_elem_iface_t; >> +typedef __u8 snd_ctl_elem_iface_t; >> #define SNDRV_CTL_ELEM_IFACE_CARD ((__force snd_ctl_elem_iface_t) 0) /* global control */ >> #define SNDRV_CTL_ELEM_IFACE_HWDEP ((__force snd_ctl_elem_iface_t) 1) /* hardware dependent device */ >> #define SNDRV_CTL_ELEM_IFACE_MIXER ((__force snd_ctl_elem_iface_t) 2) /* virtual mixer device */ > >I can't take a uapi sound header file patch along with a driver change, >these need to be independant.
Thank you and Alex for reminding me this is a change of public header! > >And this is a userspace api, you just changed the size of an externally >facing variable, are you _SURE_ that is ok to do?
The reasons I make this change are, 1) using int to represent 7 enumeration values seems to be overkill; 2) using one type could avoid worries about byte order.
I examine one userspace example (libalsa-intf/alsa_mixer.c [1]). This change won't cause compiling breakage. But I don't the experience to imagine any other bad consequence.
Another way to avoid userspace API change is to let "struct gb_audio_ctl_elem_info" use snd_ctl_elem_iface_t type which may be more elegant than using "__force" as suggested by Alex,
$ git diff diff --git a/include/linux/greybus/greybus_protocols.h b/include/linux/greybus/greybus_protocols.h index aeb8f9243545..7f6753ec7ef7 100644 --- a/include/linux/greybus/greybus_protocols.h +++ b/include/linux/greybus/greybus_protocols.h @@ -8,6 +8,7 @@ #define __GREYBUS_PROTOCOLS_H
#include <linux/types.h> +#include <sound/asound.h>
/* Fixed IDs for control/svc protocols */
@@ -1997,7 +1998,7 @@ struct gb_audio_enumerated { } __packed;
struct gb_audio_ctl_elem_info { /* See snd_ctl_elem_info in Linux source */ - __u8 type; /* GB_AUDIO_CTL_ELEM_TYPE_* */ + snd_ctl_elem_type_t type; /* GB_AUDIO_CTL_ELEM_TYPE_* */ __le16 dimen[4]; union { struct gb_audio_integer integer My only concern is if greybus authors have any special reason to not include sound/asound.h in the first place and re-define things like SNDRV_CTL_ELEM_TYPE_*,
/* See SNDRV_CTL_ELEM_TYPE_* in Linux source */ #define GB_AUDIO_CTL_ELEM_TYPE_BOOLEAN 0x01 #define GB_AUDIO_CTL_ELEM_TYPE_INTEGER 0x02 /* See SNDRV_CTL_ELEM_IFACE_* in Linux source */ #define GB_AUDIO_CTL_ELEM_IFACE_CARD 0x00 ... /* SNDRV_CTL_ELEM_ACCESS_* in Linux source */ #define GB_AUDIO_ACCESS_READ BIT(0) ... [1] https://gitlab.com/Codeaurora/platform_hardware_qcom_audio/-/blob/aee6032826ec7c8b6edda404422fda0ef40ec2db/libalsa-intf/alsa_mixer.c alsa_mixer.c > >thanks, > >greg k-h
-- Best regards, Coiby
| |