From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Inform userspace how many fences are in the sync_fence_info field.
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk Reviewed-by: Maarten Lankhorst maarten.lankhorst@linux.intel.com --- drivers/staging/android/sync.c | 2 ++ drivers/staging/android/uapi/sync.h | 2 ++ 2 files changed, 4 insertions(+)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index 3a8f210..31aa462 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -525,6 +525,8 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file, if (info->status >= 0) info->status = !info->status;
+ info->num_fences = sync_file->num_fences; + len = sizeof(struct sync_file_info);
for (i = 0; i < sync_file->num_fences; ++i) { diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index a0cf357..4ffb7cc 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -47,12 +47,14 @@ struct sync_fence_info { * userspace including pt_info. * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error + * @num_fences number of fences in the sync_file * @sync_fence_info: array of sync_fence_info for every fence in the sync_file */ struct sync_file_info { __u32 len; char name[32]; __s32 status; + __u32 num_fences;
__u8 sync_fence_info[0]; };
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
We don't use the 'fence' name to refer to sync_file anymore. So rename it to SYNC_IOC_FILE_INFO.
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk Reviewed-by: Maarten Lankhorst maarten.lankhorst@linux.intel.com --- drivers/staging/android/sync.c | 2 +- drivers/staging/android/uapi/sync.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index 31aa462..dc5f382 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -562,7 +562,7 @@ static long sync_file_ioctl(struct file *file, unsigned int cmd, case SYNC_IOC_MERGE: return sync_file_ioctl_merge(sync_file, arg);
- case SYNC_IOC_FENCE_INFO: + case SYNC_IOC_FILE_INFO: return sync_file_ioctl_fence_info(sync_file, arg);
default: diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index 4ffb7cc..dd0dd84 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -81,6 +81,6 @@ struct sync_file_info { * pt_info is a buffer containing sync_pt_infos for every sync_pt in the fence. * To iterate over the sync_pt_infos, use the sync_pt_info.len field. */ -#define SYNC_IOC_FENCE_INFO _IOWR(SYNC_IOC_MAGIC, 2, struct sync_file_info) +#define SYNC_IOC_FILE_INFO _IOWR(SYNC_IOC_MAGIC, 2, struct sync_file_info)
#endif /* _UAPI_LINUX_SYNC_H */
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
struct sync_merge_data already have documentation on top of the struct definition. No need to duplicate it.
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk Reviewed-by: Maarten Lankhorst maarten.lankhorst@linux.intel.com --- drivers/staging/android/uapi/sync.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index dd0dd84..f0b41ce 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -21,9 +21,9 @@ * @fence: returns the fd of the new fence to userspace */ struct sync_merge_data { - __s32 fd2; /* fd of second fence */ - char name[32]; /* name of new fence */ - __s32 fence; /* fd on newly created fence */ + __s32 fd2; + char name[32]; + __s32 fence; };
/**
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change order of the field to avoid alignment issues with 64 bits platforms.
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/uapi/sync.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index f0b41ce..a6c648c 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -16,13 +16,13 @@
/** * struct sync_merge_data - data passed to merge ioctl - * @fd2: file descriptor of second fence * @name: name of new fence + * @fd2: file descriptor of second fence * @fence: returns the fd of the new fence to userspace */ struct sync_merge_data { - __s32 fd2; char name[32]; + __s32 fd2; __s32 fence; };
2016-03-02 Gustavo Padovan gustavo@padovan.org:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change order of the field to avoid alignment issues with 64 bits platforms.
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk
drivers/staging/android/uapi/sync.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Please ignore this patch, it is not really needed.
Gustavo
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change SYNC_IOC_FILE_INFO behaviour to avoid future API breaks and optimize buffer
Now num_fences can be filled by the caller to inform how many fences it wants to retrieve from the kernel. If the num_fences passed is greater than zero info->sync_fence_info should point to a buffer with enough space to fit all fences.
However if num_fences passed to the kernel is 0, the kernel will reply with number of fences of the sync_file.
Sending first an ioctl with num_fences = 0 can optimize buffer allocation, in a first call with num_fences = 0 userspace will receive the actual number of fences in the num_fences filed.
Then it can allocate a buffer with the correct size on sync_fence_info and call SYNC_IOC_FILE_INFO again, but now with the actual value of num_fences in the sync_file.
Also, info->sync_fence_info was converted to __u64 pointer to prevent 32bit compatibility issues.
An example userspace code for the later would be:
struct sync_file_info *info; int err, size, num_fences;
info = malloc(sizeof(*info));
info.flags = 0; err = ioctl(fd, SYNC_IOC_FILE_INFO, info); num_fences = info->num_fences;
if (num_fences) { info.flags = 0; size = sizeof(struct sync_fence_info) * num_fences; info->num_fences = num_fences; info->sync_fence_info = (uint64_t) calloc(num_fences, sizeof(struct sync_fence_info));
err = ioctl(fd, SYNC_IOC_FILE_INFO, info); }
v2: fix fence_info memory leak
v3: Comments from Emil Velikov - improve commit message - remove __u64 cast - remove check for output fields in file_info - clean up sync_fill_fence_info()
Comments from Maarten Lankhorst - remove in.num_fences && !in.sync_fence_info check - remove info->len and use only num_fences to calculate size
Comments from Dan Carpenter - fix info->sync_fence_info documentation
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/sync.c | 64 ++++++++++++++++++++----------------- drivers/staging/android/uapi/sync.h | 9 ++---- 2 files changed, 38 insertions(+), 35 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index dc5f382..3604e453 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -479,13 +479,9 @@ err_put_fd: return err; }
-static int sync_fill_fence_info(struct fence *fence, void *data, int size) +static void sync_fill_fence_info(struct fence *fence, + struct sync_fence_info *info) { - struct sync_fence_info *info = data; - - if (size < sizeof(*info)) - return -ENOMEM; - strlcpy(info->obj_name, fence->ops->get_timeline_name(fence), sizeof(info->obj_name)); strlcpy(info->driver_name, fence->ops->get_driver_name(fence), @@ -495,28 +491,20 @@ static int sync_fill_fence_info(struct fence *fence, void *data, int size) else info->status = 0; info->timestamp_ns = ktime_to_ns(fence->timestamp); - - return sizeof(*info); }
static long sync_file_ioctl_fence_info(struct sync_file *sync_file, unsigned long arg) { - struct sync_file_info *info; + struct sync_file_info in, *info; + struct sync_fence_info *fence_info = NULL; __u32 size; - __u32 len = 0; int ret, i;
- if (copy_from_user(&size, (void __user *)arg, sizeof(size))) + if (copy_from_user(&in, (void __user *)arg, sizeof(in))) return -EFAULT;
- if (size < sizeof(struct sync_file_info)) - return -EINVAL; - - if (size > 4096) - size = 4096; - - info = kzalloc(size, GFP_KERNEL); + info = kzalloc(sizeof(*info), GFP_KERNEL); if (!info) return -ENOMEM;
@@ -525,29 +513,47 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file, if (info->status >= 0) info->status = !info->status;
- info->num_fences = sync_file->num_fences; - - len = sizeof(struct sync_file_info); + /* + * Passing num_fences = 0 means that userspace doesn't want to + * retrieve any sync_fence_info. If num_fences = 0 we skip filling + * sync_fence_info and return the actual number of fences on + * info->num_fences. + */ + if (!in.num_fences) + goto no_fences;
- for (i = 0; i < sync_file->num_fences; ++i) { - struct fence *fence = sync_file->cbs[i].fence; + if (in.num_fences < sync_file->num_fences) { + ret = -EINVAL; + goto out; + }
- ret = sync_fill_fence_info(fence, (u8 *)info + len, size - len); + size = sync_file->num_fences * sizeof(*fence_info); + fence_info = kzalloc(size, GFP_KERNEL); + if (!fence_info) { + ret = -ENOMEM; + goto out; + }
- if (ret < 0) - goto out; + for (i = 0; i < sync_file->num_fences; ++i) + sync_fill_fence_info(sync_file->cbs[i].fence, &fence_info[i]);
- len += ret; + if (copy_to_user((void __user *)in.sync_fence_info, fence_info, size)) { + ret = -EFAULT; + goto out; }
- info->len = len; + info->sync_fence_info = in.sync_fence_info; + +no_fences: + info->num_fences = sync_file->num_fences;
- if (copy_to_user((void __user *)arg, info, len)) + if (copy_to_user((void __user *)arg, info, sizeof(*info))) ret = -EFAULT; else ret = 0;
out: + kfree(fence_info); kfree(info);
return ret; diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index a6c648c..f064923 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -42,21 +42,18 @@ struct sync_fence_info {
/** * struct sync_file_info - data returned from fence info ioctl - * @len: ioctl caller writes the size of the buffer its passing in. - * ioctl returns length of sync_file_info returned to - * userspace including pt_info. * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error * @num_fences number of fences in the sync_file - * @sync_fence_info: array of sync_fence_info for every fence in the sync_file + * @sync_fence_info: pointer to array of structs sync_fence_info with all + * fences in the sync_file */ struct sync_file_info { - __u32 len; char name[32]; __s32 status; __u32 num_fences;
- __u8 sync_fence_info[0]; + __u64 sync_fence_info; };
#define SYNC_IOC_MAGIC '>'
Op 02-03-16 om 20:52 schreef Gustavo Padovan:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change SYNC_IOC_FILE_INFO behaviour to avoid future API breaks and optimize buffer
Now num_fences can be filled by the caller to inform how many fences it wants to retrieve from the kernel. If the num_fences passed is greater than zero info->sync_fence_info should point to a buffer with enough space to fit all fences.
However if num_fences passed to the kernel is 0, the kernel will reply with number of fences of the sync_file.
Sending first an ioctl with num_fences = 0 can optimize buffer allocation, in a first call with num_fences = 0 userspace will receive the actual number of fences in the num_fences filed.
Then it can allocate a buffer with the correct size on sync_fence_info and call SYNC_IOC_FILE_INFO again, but now with the actual value of num_fences in the sync_file.
Also, info->sync_fence_info was converted to __u64 pointer to prevent 32bit compatibility issues.
An example userspace code for the later would be:
struct sync_file_info *info; int err, size, num_fences;
info = malloc(sizeof(*info));
info.flags = 0; err = ioctl(fd, SYNC_IOC_FILE_INFO, info); num_fences = info->num_fences;
if (num_fences) { info.flags = 0; size = sizeof(struct sync_fence_info) * num_fences; info->num_fences = num_fences; info->sync_fence_info = (uint64_t) calloc(num_fences, sizeof(struct sync_fence_info));
err = ioctl(fd, SYNC_IOC_FILE_INFO, info);}
v2: fix fence_info memory leak
v3: Comments from Emil Velikov
- improve commit message
- remove __u64 cast
- remove check for output fields in file_info
- clean up sync_fill_fence_info()
Comments from Maarten Lankhorst
- remove in.num_fences && !in.sync_fence_info check
- remove info->len and use only num_fences to calculate size
Comments from Dan Carpenter
- fix info->sync_fence_info documentation
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk
drivers/staging/android/sync.c | 64 ++++++++++++++++++++----------------- drivers/staging/android/uapi/sync.h | 9 ++---- 2 files changed, 38 insertions(+), 35 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index dc5f382..3604e453 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -479,13 +479,9 @@ err_put_fd: return err; }
-static int sync_fill_fence_info(struct fence *fence, void *data, int size) +static void sync_fill_fence_info(struct fence *fence,
struct sync_fence_info *info){
- struct sync_fence_info *info = data;
- if (size < sizeof(*info))
return -ENOMEM;- strlcpy(info->obj_name, fence->ops->get_timeline_name(fence), sizeof(info->obj_name)); strlcpy(info->driver_name, fence->ops->get_driver_name(fence),
@@ -495,28 +491,20 @@ static int sync_fill_fence_info(struct fence *fence, void *data, int size) else info->status = 0; info->timestamp_ns = ktime_to_ns(fence->timestamp);
- return sizeof(*info);
}
static long sync_file_ioctl_fence_info(struct sync_file *sync_file, unsigned long arg) {
- struct sync_file_info *info;
- struct sync_file_info in, *info;
Why put one copy on the stack, and allocate a second? With sync_file_info now being a fixed size just put 1 copy on the stack from userspace, run some sanity checks first, put in the new values and copy it back to userspace without additional memory allocation.
The rest looks good now. :)
~Maarten
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change SYNC_IOC_FILE_INFO behaviour to avoid future API breaks and optimize buffer
Now num_fences can be filled by the caller to inform how many fences it wants to retrieve from the kernel. If the num_fences passed is greater than zero info->sync_fence_info should point to a buffer with enough space to fit all fences.
However if num_fences passed to the kernel is 0, the kernel will reply with number of fences of the sync_file.
Sending first an ioctl with num_fences = 0 can optimize buffer allocation, in a first call with num_fences = 0 userspace will receive the actual number of fences in the num_fences filed.
Then it can allocate a buffer with the correct size on sync_fence_info and call SYNC_IOC_FILE_INFO again, but now with the actual value of num_fences in the sync_file.
Also, info->sync_fence_info was converted to __u64 pointer to prevent 32bit compatibility issues.
An example userspace code for the later would be:
struct sync_file_info *info; int err, size, num_fences;
info = malloc(sizeof(*info));
info.flags = 0; err = ioctl(fd, SYNC_IOC_FILE_INFO, info); num_fences = info->num_fences;
if (num_fences) { info.flags = 0; size = sizeof(struct sync_fence_info) * num_fences; info->num_fences = num_fences; info->sync_fence_info = (uint64_t) calloc(num_fences, sizeof(struct sync_fence_info));
err = ioctl(fd, SYNC_IOC_FILE_INFO, info); }
v2: fix fence_info memory leak
v3: Comments from Emil Velikov - improve commit message - remove __u64 cast - remove check for output fields in file_info - clean up sync_fill_fence_info()
Comments from Maarten Lankhorst - remove in.num_fences && !in.sync_fence_info check - remove info->len and use only num_fences to calculate size
Comments from Dan Carpenter - fix info->sync_fence_info documentation
v4: remove allocated struct sync_file_info (comment from Maarten)
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/sync.c | 70 +++++++++++++++++-------------------- drivers/staging/android/uapi/sync.h | 9 ++--- 2 files changed, 36 insertions(+), 43 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index dc5f382..48ee175 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -479,13 +479,9 @@ err_put_fd: return err; }
-static int sync_fill_fence_info(struct fence *fence, void *data, int size) +static void sync_fill_fence_info(struct fence *fence, + struct sync_fence_info *info) { - struct sync_fence_info *info = data; - - if (size < sizeof(*info)) - return -ENOMEM; - strlcpy(info->obj_name, fence->ops->get_timeline_name(fence), sizeof(info->obj_name)); strlcpy(info->driver_name, fence->ops->get_driver_name(fence), @@ -495,60 +491,60 @@ static int sync_fill_fence_info(struct fence *fence, void *data, int size) else info->status = 0; info->timestamp_ns = ktime_to_ns(fence->timestamp); - - return sizeof(*info); }
static long sync_file_ioctl_fence_info(struct sync_file *sync_file, unsigned long arg) { - struct sync_file_info *info; + struct sync_file_info info; + struct sync_fence_info *fence_info = NULL; __u32 size; - __u32 len = 0; int ret, i;
- if (copy_from_user(&size, (void __user *)arg, sizeof(size))) + if (copy_from_user(&info, (void __user *)arg, sizeof(info))) return -EFAULT;
- if (size < sizeof(struct sync_file_info)) - return -EINVAL; + /* + * Passing num_fences = 0 means that userspace doesn't want to + * retrieve any sync_fence_info. If num_fences = 0 we skip filling + * sync_fence_info and return the actual number of fences on + * info->num_fences. + */ + if (!info.num_fences) + goto no_fences;
- if (size > 4096) - size = 4096; + if (info.num_fences < sync_file->num_fences) + return -EINVAL;
- info = kzalloc(size, GFP_KERNEL); - if (!info) + size = sync_file->num_fences * sizeof(*fence_info); + fence_info = kzalloc(size, GFP_KERNEL); + if (!fence_info) return -ENOMEM;
- strlcpy(info->name, sync_file->name, sizeof(info->name)); - info->status = atomic_read(&sync_file->status); - if (info->status >= 0) - info->status = !info->status; + for (i = 0; i < sync_file->num_fences; ++i) + sync_fill_fence_info(sync_file->cbs[i].fence, &fence_info[i]);
- info->num_fences = sync_file->num_fences; - - len = sizeof(struct sync_file_info); - - for (i = 0; i < sync_file->num_fences; ++i) { - struct fence *fence = sync_file->cbs[i].fence; - - ret = sync_fill_fence_info(fence, (u8 *)info + len, size - len); - - if (ret < 0) - goto out; - - len += ret; + if (copy_to_user((void __user *)info.sync_fence_info, fence_info, + size)) { + ret = -EFAULT; + goto out; }
- info->len = len; +no_fences: + strlcpy(info.name, sync_file->name, sizeof(info.name)); + info.status = atomic_read(&sync_file->status); + if (info.status >= 0) + info.status = !info.status; + + info.num_fences = sync_file->num_fences;
- if (copy_to_user((void __user *)arg, info, len)) + if (copy_to_user((void __user *)arg, &info, sizeof(info))) ret = -EFAULT; else ret = 0;
out: - kfree(info); + kfree(fence_info);
return ret; } diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index f0b41ce..a122bb5 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -42,21 +42,18 @@ struct sync_fence_info {
/** * struct sync_file_info - data returned from fence info ioctl - * @len: ioctl caller writes the size of the buffer its passing in. - * ioctl returns length of sync_file_info returned to - * userspace including pt_info. * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error * @num_fences number of fences in the sync_file - * @sync_fence_info: array of sync_fence_info for every fence in the sync_file + * @sync_fence_info: pointer to array of structs sync_fence_info with all + * fences in the sync_file */ struct sync_file_info { - __u32 len; char name[32]; __s32 status; __u32 num_fences;
- __u8 sync_fence_info[0]; + __u64 sync_fence_info; };
#define SYNC_IOC_MAGIC '>'
Op 03-03-16 om 15:34 schreef Gustavo Padovan:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Change SYNC_IOC_FILE_INFO behaviour to avoid future API breaks and optimize buffer
Now num_fences can be filled by the caller to inform how many fences it wants to retrieve from the kernel. If the num_fences passed is greater than zero info->sync_fence_info should point to a buffer with enough space to fit all fences.
However if num_fences passed to the kernel is 0, the kernel will reply with number of fences of the sync_file.
Sending first an ioctl with num_fences = 0 can optimize buffer allocation, in a first call with num_fences = 0 userspace will receive the actual number of fences in the num_fences filed.
Then it can allocate a buffer with the correct size on sync_fence_info and call SYNC_IOC_FILE_INFO again, but now with the actual value of num_fences in the sync_file.
Also, info->sync_fence_info was converted to __u64 pointer to prevent 32bit compatibility issues.
For this patch and 6/6:
Reviewed-by: Maarten Lankhorst maarten.lankhorst@linux.intel.com
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/sync.c | 8 ++++++++ drivers/staging/android/uapi/sync.h | 8 ++++++++ 2 files changed, 16 insertions(+)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index 3604e453..e635e6e 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -445,6 +445,11 @@ static long sync_file_ioctl_merge(struct sync_file *sync_file, goto err_put_fd; }
+ if (data.flags) { + err = -EINVAL; + goto err_put_fd; + } + fence2 = sync_file_fdget(data.fd2); if (!fence2) { err = -ENOENT; @@ -504,6 +509,9 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file, if (copy_from_user(&in, (void __user *)arg, sizeof(in))) return -EFAULT;
+ if (in.flags || in.pad) + return -EINVAL; + info = kzalloc(sizeof(*info), GFP_KERNEL); if (!info) return -ENOMEM; diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index f064923..4b4d079 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -19,11 +19,13 @@ * @name: name of new fence * @fd2: file descriptor of second fence * @fence: returns the fd of the new fence to userspace + * @flags: merge_data flags */ struct sync_merge_data { char name[32]; __s32 fd2; __s32 fence; + __u32 flags; };
/** @@ -31,12 +33,14 @@ struct sync_merge_data { * @obj_name: name of parent sync_timeline * @driver_name: name of driver implementing the parent * @status: status of the fence 0:active 1:signaled <0:error + * @flags: fence_info flags * @timestamp_ns: timestamp of status change in nanoseconds */ struct sync_fence_info { char obj_name[32]; char driver_name[32]; __s32 status; + __u32 flags; __u64 timestamp_ns; };
@@ -44,14 +48,18 @@ struct sync_fence_info { * struct sync_file_info - data returned from fence info ioctl * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error + * @flags: sync_file_info flags * @num_fences number of fences in the sync_file + * @pad padding for 64-bit alignment, should always be zero * @sync_fence_info: pointer to array of structs sync_fence_info with all * fences in the sync_file */ struct sync_file_info { char name[32]; __s32 status; + __u32 flags; __u32 num_fences; + __u32 pad;
__u64 sync_fence_info; };
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/sync.c | 8 ++++++++ drivers/staging/android/uapi/sync.h | 14 ++++++++++++-- 2 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index 3604e453..0196d3d 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -445,6 +445,11 @@ static long sync_file_ioctl_merge(struct sync_file *sync_file, goto err_put_fd; }
+ if (data.flags || data.pad) { + err = -EINVAL; + goto err_put_fd; + } + fence2 = sync_file_fdget(data.fd2); if (!fence2) { err = -ENOENT; @@ -504,6 +509,9 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file, if (copy_from_user(&in, (void __user *)arg, sizeof(in))) return -EFAULT;
+ if (in.flags || in.pad) + return -EINVAL; + info = kzalloc(sizeof(*info), GFP_KERNEL); if (!info) return -ENOMEM; diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index a122bb5..859977c 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -16,14 +16,18 @@
/** * struct sync_merge_data - data passed to merge ioctl - * @fd2: file descriptor of second fence * @name: name of new fence + * @fd2: file descriptor of second fence * @fence: returns the fd of the new fence to userspace + * @flags: merge_data flags + * @pad: padding for 64-bit alignment, should always be zero */ struct sync_merge_data { - __s32 fd2; char name[32]; + __s32 fd2; __s32 fence; + __u32 flags; + __u32 pad; };
/** @@ -31,12 +35,14 @@ struct sync_merge_data { * @obj_name: name of parent sync_timeline * @driver_name: name of driver implementing the parent * @status: status of the fence 0:active 1:signaled <0:error + * @flags: fence_info flags * @timestamp_ns: timestamp of status change in nanoseconds */ struct sync_fence_info { char obj_name[32]; char driver_name[32]; __s32 status; + __u32 flags; __u64 timestamp_ns; };
@@ -44,14 +50,18 @@ struct sync_fence_info { * struct sync_file_info - data returned from fence info ioctl * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error + * @flags: sync_file_info flags * @num_fences number of fences in the sync_file + * @pad: padding for 64-bit alignment, should always be zero * @sync_fence_info: pointer to array of structs sync_fence_info with all * fences in the sync_file */ struct sync_file_info { char name[32]; __s32 status; + __u32 flags; __u32 num_fences; + __u32 pad;
__u64 sync_fence_info; };
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Signed-off-by: Gustavo Padovan gustavo.padovan@collabora.co.uk --- drivers/staging/android/sync.c | 8 ++++++++ drivers/staging/android/uapi/sync.h | 14 ++++++++++++-- 2 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/android/sync.c b/drivers/staging/android/sync.c index 48ee175..ae81c95 100644 --- a/drivers/staging/android/sync.c +++ b/drivers/staging/android/sync.c @@ -445,6 +445,11 @@ static long sync_file_ioctl_merge(struct sync_file *sync_file, goto err_put_fd; }
+ if (data.flags || data.pad) { + err = -EINVAL; + goto err_put_fd; + } + fence2 = sync_file_fdget(data.fd2); if (!fence2) { err = -ENOENT; @@ -504,6 +509,9 @@ static long sync_file_ioctl_fence_info(struct sync_file *sync_file, if (copy_from_user(&info, (void __user *)arg, sizeof(info))) return -EFAULT;
+ if (info.flags || info.pad) + return -EINVAL; + /* * Passing num_fences = 0 means that userspace doesn't want to * retrieve any sync_fence_info. If num_fences = 0 we skip filling diff --git a/drivers/staging/android/uapi/sync.h b/drivers/staging/android/uapi/sync.h index a122bb5..859977c 100644 --- a/drivers/staging/android/uapi/sync.h +++ b/drivers/staging/android/uapi/sync.h @@ -16,14 +16,18 @@
/** * struct sync_merge_data - data passed to merge ioctl - * @fd2: file descriptor of second fence * @name: name of new fence + * @fd2: file descriptor of second fence * @fence: returns the fd of the new fence to userspace + * @flags: merge_data flags + * @pad: padding for 64-bit alignment, should always be zero */ struct sync_merge_data { - __s32 fd2; char name[32]; + __s32 fd2; __s32 fence; + __u32 flags; + __u32 pad; };
/** @@ -31,12 +35,14 @@ struct sync_merge_data { * @obj_name: name of parent sync_timeline * @driver_name: name of driver implementing the parent * @status: status of the fence 0:active 1:signaled <0:error + * @flags: fence_info flags * @timestamp_ns: timestamp of status change in nanoseconds */ struct sync_fence_info { char obj_name[32]; char driver_name[32]; __s32 status; + __u32 flags; __u64 timestamp_ns; };
@@ -44,14 +50,18 @@ struct sync_fence_info { * struct sync_file_info - data returned from fence info ioctl * @name: name of fence * @status: status of fence. 1: signaled 0:active <0:error + * @flags: sync_file_info flags * @num_fences number of fences in the sync_file + * @pad: padding for 64-bit alignment, should always be zero * @sync_fence_info: pointer to array of structs sync_fence_info with all * fences in the sync_file */ struct sync_file_info { char name[32]; __s32 status; + __u32 flags; __u32 num_fences; + __u32 pad;
__u64 sync_fence_info; };
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
And you just broke all existing userspace users of this code, why are you allowed to do that?
not ok...
Hi Greg,
2016-03-03 Greg Kroah-Hartman gregkh@linuxfoundation.org:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
There are few people who does this in drm, so I just followed that.
And you just broke all existing userspace users of this code, why are you allowed to do that?
Because we've discussed this extensively in the last versions of this patches. Most of the people from CC agreed with it.
We are cleaning up the Android sync and coming up with a better ABI for it before we de-stage it. Android people agreed with this and we will patch it.
Gustavo
2016-03-03 Gustavo Padovan gustavo.padovan@collabora.co.uk:
Hi Greg,
2016-03-03 Greg Kroah-Hartman gregkh@linuxfoundation.org:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
There are few people who does this in drm, so I just followed that.
Anyway, I've sent v7 without the vX things in the changelog.
Gustavo
On Thu, Mar 03, 2016 at 08:17:14AM -0800, Greg Kroah-Hartman wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
That's how we usually roll in gpu land.
On Thu, Mar 3, 2016 at 11:17 AM, Greg Kroah-Hartman gregkh@linuxfoundation.org wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
And you just broke all existing userspace users of this code, why are you allowed to do that?
not ok...
There are not really any users of this on an upstream kernel yet, so it makes sense to fix the ABI to something we can live with now, before that changes. If we are stuck not breaking ABI with android stuff pulled into staging as we destage it, then maybe we should be a *lot* slower at pulling android stuff into staging. (Ie. if that is the case, please kick it all out now and we'll re-add things properly.)
BR, -R
On Thu, Mar 3, 2016 at 3:54 PM, Rob Clark robdclark@gmail.com wrote:
On Thu, Mar 3, 2016 at 11:17 AM, Greg Kroah-Hartman gregkh@linuxfoundation.org wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
And you just broke all existing userspace users of this code, why are you allowed to do that?
not ok...
There are not really any users of this on an upstream kernel yet, so it makes sense to fix the ABI to something we can live with now, before that changes. If we are stuck not breaking ABI with android stuff pulled into staging as we destage it, then maybe we should be a *lot* slower at pulling android stuff into staging. (Ie. if that is the case, please kick it all out now and we'll re-add things properly.)
That all said, I suppose one sensible concession to practicality would be to burn the old ioctl numbers and pick new ioctl numbers. At least this way if someone did manage to take an old android userspace (with it's various dependencies on other non-upstream SoC specific platform drivers, etc) and run it on upstream kernel, at least things would fail in an obvious way. I could live with this as the mode-of-operations for fixing up and destaging staging/android stuff.
BR, -R
On Thu, Mar 03, 2016 at 08:17:14AM -0800, Greg Kroah-Hartman wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
Because this is drm and we're special ;-)
And you just broke all existing userspace users of this code, why are you allowed to do that?
not ok...
We could do fence2.h if you absolutely insist and just forget about the current one, but that seemed silly. Like Gustavo said, everyone who actually cares about this stuff is perfectly fine with this. And there's not a single user of this in upstream anyway, so the only trees we could break are vendor trees with massive amounts of additional stuff.
Is that reasonable ok for you, or do you insist we do a fences2.h without going through staging ? ;-)
Thanks, Daniel
On Fri, Mar 04, 2016 at 05:40:29PM +0100, Daniel Vetter wrote:
On Thu, Mar 03, 2016 at 08:17:14AM -0800, Greg Kroah-Hartman wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
Because this is drm and we're special ;-)
And you just broke all existing userspace users of this code, why are you allowed to do that?
not ok...
We could do fence2.h if you absolutely insist and just forget about the current one, but that seemed silly. Like Gustavo said, everyone who actually cares about this stuff is perfectly fine with this. And there's not a single user of this in upstream anyway, so the only trees we could break are vendor trees with massive amounts of additional stuff.
Is that reasonable ok for you, or do you insist we do a fences2.h without going through staging ? ;-)
Ok, if everyone is ok with this api changing, and will not get mad if it breaks things, I'm all for fixing this up.
I just want all of your signed-off-by lines on the series please. Please respond to the v7 of this series and I'll be glad to queue them up.
thanks,
greg k-h
Hi Greg,
Allow me to chip in as well.
On 3 March 2016 at 16:17, Greg Kroah-Hartman gregkh@linuxfoundation.org wrote:
On Thu, Mar 03, 2016 at 11:37:17AM -0300, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Play safe and add flags member to all structs. So we don't need to break API or create new IOCTL in the future if new features that requires flags arises.
v2: check if flags are valid (zero, in this case)
v3: return -EINVAL if flags are not zero'ed
v4: add padding for 64-bit alignment
v5: rebase to use only stacked sync_file_info
Why are these vX things here in the changelog?
And you just broke all existing userspace users of this code, why are you allowed to do that?
In all honesty, isn't 'fluid ABI' one of the reasons behind staging ? That is how it was used by a few drivers in the past, at least. If the rules have changed and/or Android is special in that regard, we ought to make it perfectly clear so that people are aware from day 1.
That aside, Android developers were clear that only internal, downstream components are using this code and they are OK with breaking the ABI [1]. Gustavo is in the process of rewriting their tests for upstream inclusion and he'll also update the Android side of things [2].
With those in mind, I think everything should be safe here. If you prefer to avoid the ABI break, which approach are you keen on - reassign new ioctl numbers (Rob suggestion) or use new header fence2.h (Daniel).
Thank you Emil
[1] https://lists.freedesktop.org/archives/dri-devel/2016-January/099592.html [2] https://lists.freedesktop.org/archives/dri-devel/2016-January/099726.html
dri-devel@lists.freedesktop.org
-
Daniel Vetter -
Emil Velikov -
Greg Kroah-Hartman -
Gustavo Padovan -
Gustavo Padovan -
Maarten Lankhorst -
Rob Clark -
Ville Syrjälä