On 06/08/2016 08:34 AM, Brian Starkey wrote:
Hi,
I'm finding "usage_id" a bit confusing - there's not a very clear distinction between usage_id and heap ID.
For instance, ION_IOC_USAGE_CNT claims to return the number of usage IDs, but seems to return the number of heaps (i.e. number heap IDs, some of which might be usage_ids).
Similarly, alloc2 claims to allocate by usage_id, but will just as happily allocate by heap ID.
Maybe this should just be described as a single heap ID namespace, where some IDs represent groups of heap IDs.
For now I'm just going to focus on comments not about the heap ID mapping because I'm leaning towards dropping the heap ID mapping.
I've a few inline comments below.
-Brian
On Mon, Jun 06, 2016 at 11:23:33AM -0700, Laura Abbott wrote:
From: Laura Abbott labbott@fedoraproject.org
The Ion ABI for heaps is limiting to work with for more complex systems. Heaps have to be registered at boot time with known ids available to userspace. This becomes a tight ABI which is prone to breakage.
Introduce a new mechanism for registering heap ids dynamically. A base set of heap ids are registered at boot time but there is no knowledge of fallbacks. Fallback information can be remapped and changed dynamically. Information about available heaps can also be queried with an ioctl to avoid the need to have heap ids be an ABI with userspace.
Signed-off-by: Laura Abbott labbott@redhat.com
drivers/staging/android/ion/ion-ioctl.c | 109 +++++++++++++++++-- drivers/staging/android/ion/ion.c | 184 ++++++++++++++++++++++++++++++-- drivers/staging/android/ion/ion_priv.h | 15 +++ drivers/staging/android/uapi/ion.h | 135 +++++++++++++++++++++++ 4 files changed, 426 insertions(+), 17 deletions(-)
diff --git a/drivers/staging/android/ion/ion-ioctl.c b/drivers/staging/android/ion/ion-ioctl.c index 45b89e8..169cad8 100644 --- a/drivers/staging/android/ion/ion-ioctl.c +++ b/drivers/staging/android/ion/ion-ioctl.c @@ -22,6 +22,49 @@ #include "ion_priv.h" #include "compat_ion.h"
+union ion_ioctl_arg {
- struct ion_fd_data fd;
- struct ion_allocation_data allocation;
- struct ion_handle_data handle;
- struct ion_custom_data custom;
- struct ion_abi_version abi_version;
- struct ion_new_alloc_data allocation2;
- struct ion_usage_id_map id_map;
- struct ion_usage_cnt usage_cnt;
- struct ion_heap_query query;
+};
+static int validate_ioctl_arg(unsigned int cmd, union ion_ioctl_arg *arg) +{
- int ret = 0;
- switch (cmd) {
- case ION_IOC_ABI_VERSION:
ret = arg->abi_version.reserved != 0;
break;
- case ION_IOC_ALLOC2:
ret = arg->allocation2.reserved0 != 0;
ret |= arg->allocation2.reserved1 != 0;
ret |= arg->allocation2.reserved2 != 0;
break;
- case ION_IOC_ID_MAP:
ret = arg->id_map.reserved0 != 0;
ret |= arg->id_map.reserved1 != 0;
break;
- case ION_IOC_USAGE_CNT:
ret = arg->usage_cnt.reserved != 0;
break;
- case ION_IOC_HEAP_QUERY:
ret = arg->query.reserved0 != 0;
ret |= arg->query.reserved1 != 0;
ret |= arg->query.reserved2 != 0;
break;
- default:
break;
- }
- return ret ? -EINVAL : 0;
+}
/* fix up the cases where the ioctl direction bits are incorrect */ static unsigned int ion_ioctl_dir(unsigned int cmd) { @@ -42,14 +85,7 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) struct ion_handle *cleanup_handle = NULL; int ret = 0; unsigned int dir;
- union {
struct ion_fd_data fd;
struct ion_allocation_data allocation;
struct ion_handle_data handle;
struct ion_custom_data custom;
struct ion_abi_version abi_version;
- } data;
union ion_ioctl_arg data;
dir = ion_ioctl_dir(cmd);
@@ -60,7 +96,12 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) if (copy_from_user(&data, (void __user *)arg, _IOC_SIZE(cmd))) return -EFAULT;
- ret = validate_ioctl_arg(cmd, &data);
- if (ret)
return ret;
- switch (cmd) {
- /* Old ioctl */ case ION_IOC_ALLOC: { struct ion_handle *handle;
@@ -77,6 +118,7 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) cleanup_handle = handle; break; }
- /* Old ioctl */ case ION_IOC_FREE: { struct ion_handle *handle;
@@ -92,6 +134,7 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) mutex_unlock(&client->lock); break; }
- /* Old ioctl */ case ION_IOC_SHARE: case ION_IOC_MAP: {
@@ -106,6 +149,7 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) ret = data.fd.fd; break; }
- /* Old ioctl */ case ION_IOC_IMPORT: { struct ion_handle *handle;
@@ -117,11 +161,13 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) data.handle.handle = handle->id; break; }
- /* Old ioctl */ case ION_IOC_SYNC: { ret = ion_sync_for_device(client, data.fd.fd); break; }
- /* Old ioctl */ case ION_IOC_CUSTOM: { if (!dev->custom_ioctl)
@@ -135,6 +181,53 @@ long ion_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) data.abi_version.abi_version = ION_ABI_VERSION; break; }
- case ION_IOC_ALLOC2:
- {
struct ion_handle *handle;
handle = ion_alloc2(client, data.allocation2.len,
data.allocation2.align,
data.allocation2.usage_id,
data.allocation2.flags);
if (IS_ERR(handle))
return PTR_ERR(handle);
if (data.allocation2.flags & ION_FLAG_NO_HANDLE) {
data.allocation2.fd = ion_share_dma_buf_fd(client,
handle);
ion_handle_put(handle);
if (data.allocation2.fd < 0)
ret = data.allocation2.fd;
} else {
data.allocation2.handle = handle->id;
cleanup_handle = handle;
}
break;
- }
- case ION_IOC_ID_MAP:
- {
ret = ion_map_usage_ids(client,
(unsigned int __user *)data.id_map.usage_ids,
data.id_map.cnt);
if (ret > 0)
data.id_map.new_id = ret;
break;
- }
- case ION_IOC_USAGE_CNT:
- {
down_read(&client->dev->lock);
data.usage_cnt.cnt = client->dev->heap_cnt;
up_read(&client->dev->lock);
break;
- }
- case ION_IOC_HEAP_QUERY:
- {
ret = ion_query_heaps(client,
(struct ion_heap_data __user *)data.query.heaps,
data.query.cnt);
break;
- } default: return -ENOTTY; }
diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c index 129707f..c096cb9 100644 --- a/drivers/staging/android/ion/ion.c +++ b/drivers/staging/android/ion/ion.c @@ -459,6 +459,48 @@ static struct ion_handle *__ion_alloc(struct ion_client *client, size_t len, return handle; }
+struct ion_handle *ion_alloc2(struct ion_client *client, size_t len,
size_t align, unsigned int usage_id,
unsigned int flags)
+{
- struct ion_device *dev = client->dev;
- struct ion_heap *heap;
- struct ion_handle *handle = ERR_PTR(-ENODEV);
- down_read(&dev->lock);
- heap = idr_find(&dev->idr, usage_id);
- if (!heap) {
handle = ERR_PTR(-EINVAL);
goto out;
- }
- if (heap->type == ION_USAGE_ID_MAP) {
int i;
for (i = 0; i < heap->fallback_cnt; i++){
heap = idr_find(&dev->idr, heap->fallbacks[i]);
if (!heap)
continue;
/* Don't recurse for now? */
if (heap->type == ION_USAGE_ID_MAP)
continue;
handle = __ion_alloc(client, len, align, heap, flags);
if (IS_ERR(handle))
continue;
else
break;
}
- } else {
handle = __ion_alloc(client, len, align, heap, flags);
- }
+out:
- up_read(&dev->lock);
- return handle;
+} +EXPORT_SYMBOL(ion_alloc2);
struct ion_handle *ion_alloc(struct ion_client *client, size_t len, size_t align, unsigned int heap_mask, unsigned int flags) @@ -1243,6 +1285,55 @@ int ion_sync_for_device(struct ion_client *client, int fd) return 0; }
+struct ion_query_data {
- struct ion_heap_data __user *buffer;
- int cnt;
- int max_cnt;
+};
+int __ion_query(int id, void *p, void *data) +{
- struct ion_heap *heap = p;
- struct ion_query_data *query = data;
- struct ion_heap_data hdata = {0};
- if (query->cnt >= query->max_cnt)
return -ENOSPC;
- strncpy(hdata.name, heap->name, 20);
Why 20?
Arbitrary limit for a fixed size buffer. As noted below I need to make this a consistent #define.
- hdata.name[sizeof(hdata.name) - 1] = '\0';
- hdata.type = heap->type;
- hdata.usage_id = heap->id;
- return copy_to_user(&query->buffer[query->cnt++], &hdata, sizeof(hdata));
+}
+int ion_query_heaps(struct ion_client *client,
struct ion_heap_data __user *buffer,
int cnt)
+{
- struct ion_device *dev = client->dev;
- struct ion_query_data data;
- int ret;
- data.buffer = buffer;
- data.cnt = 0;
- data.max_cnt = cnt;
- down_read(&dev->lock);
- if (data.max_cnt < 0 || data.max_cnt > dev->heap_cnt) {
Is the check against heap_cnt necessary? You could update the cnt field in ion_heap_query to say how many were actually copied.
This was mostly a paranoid check to reduce risk of accidentally giving userspace access to something it shouldn't. Updating the count is still a good idea in case userspace wants fewer than are available.
I think you could probably drop ION_IOC_USAGE_COUNT by giving ION_IOC_HEAP_QUERY similar semantics to the DRM ioctls:
- First call with buffer pointer = NULL, kernel sets cnt to space required
- Userspace allocates buffer and calls the ioctl again, kernel fills the buffer
I'm not that familiar with the DRM ioctls but this sounds like a good suggestion. One less ioctl to worry about.
ret = -EINVAL;
goto out;
- }
- ret = idr_for_each(&dev->idr, __ion_query, &data);
+out:
- up_read(&dev->lock);
- return ret;
+}
static int ion_release(struct inode *inode, struct file *file) { struct ion_client *client = file->private_data; @@ -1407,13 +1498,69 @@ static int debug_shrink_get(void *data, u64 *val) DEFINE_SIMPLE_ATTRIBUTE(debug_shrink_fops, debug_shrink_get, debug_shrink_set, "%llu\n");
+int ion_map_usage_ids(struct ion_client *client,
unsigned int __user *usage_ids,
int cnt)
+{
- struct ion_heap *heap = kzalloc(sizeof(*heap), GFP_KERNEL);
- unsigned int *fallbacks;
- int i;
- int ret;
- if (!heap)
return -ENOMEM;
- fallbacks = kzalloc(sizeof(unsigned int)*cnt, GFP_KERNEL);
- if (!fallbacks) {
ret = -ENOMEM;
goto out1;
- }
- ret = copy_from_user(fallbacks, usage_ids, sizeof(unsigned int)*cnt);
- if (ret)
goto out2;
- down_read(&client->dev->lock);
- for (i = 0; i < cnt; i++) {
if (idr_find(&client->dev->idr, fallbacks[i]) == NULL) {
ret = -EINVAL;
goto out3;
}
- }
- up_read(&client->dev->lock);
- /*
* This is a racy check since the lock is dropped before the heap
* is actually added. It's okay though because ids are never actually
* deleted. Worst case some user gets an error back and an indication
* to fix races in their code.
*/
- heap->fallbacks = fallbacks;
- heap->fallback_cnt = cnt;
- heap->type = ION_USAGE_ID_MAP;
- heap->id = ION_DYNAMIC_HEAP_ASSIGN;
It would be nice for userspace to be able to give these usage heaps meaningful names - That could be their intended usage ("camera", "display") or their properties ("contiguous", "sram").
- ion_device_add_heap(client->dev, heap);
- return heap->id;
+out3:
- up_read(&client->dev->lock);
+out2:
- kfree(fallbacks);
+out1:
- kfree(heap);
- return ret;
+}
int ion_device_add_heap(struct ion_device *dev, struct ion_heap *heap) { struct dentry *debug_file; int ret;
- if (!heap->ops->allocate || !heap->ops->free || !heap->ops->map_dma ||
!heap->ops->unmap_dma) {
- if (heap->type != ION_USAGE_ID_MAP &&
(!heap->ops->allocate ||
!heap->ops->free ||
!heap->ops->map_dma ||
!heap->ops->unmap_dma)) { pr_err("%s: can not add heap with invalid ops struct.\n", __func__); return -EINVAL;
@@ -1425,18 +1572,36 @@ int ion_device_add_heap(struct ion_device *dev, struct ion_heap *heap) if (heap->flags & ION_HEAP_FLAG_DEFER_FREE) ion_heap_init_deferred_free(heap);
- if ((heap->flags & ION_HEAP_FLAG_DEFER_FREE) || heap->ops->shrink)
if ((heap->flags & ION_HEAP_FLAG_DEFER_FREE) || (heap->ops && heap->ops->shrink)) ion_heap_init_shrinker(heap);
heap->dev = dev; down_write(&dev->lock);
- ret = idr_alloc(&dev->idr, heap, heap->id, heap->id + 1, GFP_KERNEL);
- if (ret < 0 || ret != heap->id) {
pr_info("%s: Failed to add heap id, expected %d got %d\n",
__func__, heap->id, ret);
up_write(&dev->lock);
return ret < 0 ? ret : -EINVAL;
if (heap->id == ION_DYNAMIC_HEAP_ASSIGN) {
ret = idr_alloc(&dev->idr, heap,
ION_DYNAMIC_HEAP_ASSIGN + 1, 0, GFP_KERNEL);
if (ret < 0)
goto out_unlock;
heap->id = ret;
} else {
ret = idr_alloc(&dev->idr, heap, heap->id, heap->id + 1,
GFP_KERNEL);
if (ret < 0 || ret != heap->id) {
pr_info("%s: Failed to add heap id, expected %d got %d\n",
__func__, heap->id, ret);
ret = ret < 0 ? ret : -EINVAL;
goto out_unlock;
}
}
if (!heap->name) {
heap->name = kasprintf(GFP_KERNEL, "heap%d", heap->id);
if (!heap->name) {
ret = -ENOMEM;
goto out_unlock;
}
}
debug_file = debugfs_create_file(heap->name, 0664,
@@ -1467,6 +1632,7 @@ int ion_device_add_heap(struct ion_device *dev, struct ion_heap *heap) } } dev->heap_cnt++; +out_unlock: up_write(&dev->lock); return 0; } diff --git a/drivers/staging/android/ion/ion_priv.h b/drivers/staging/android/ion/ion_priv.h index b09d751..e03e940 100644 --- a/drivers/staging/android/ion/ion_priv.h +++ b/drivers/staging/android/ion/ion_priv.h @@ -255,6 +255,9 @@ struct ion_heap { wait_queue_head_t waitqueue; struct task_struct *task;
- unsigned int *fallbacks;
- int fallback_cnt;
- int (*debug_show)(struct ion_heap *heap, struct seq_file *, void *);
};
@@ -381,6 +384,10 @@ size_t ion_heap_freelist_shrink(struct ion_heap *heap, size_t ion_heap_freelist_size(struct ion_heap *heap);
+int ion_map_usage_ids(struct ion_client *client,
unsigned int __user *usage_ids,
int cnt);
/**
- functions for creating and destroying the built in ion heaps.
- architectures can add their own custom architecture specific
@@ -495,4 +502,12 @@ struct ion_handle *ion_handle_get_by_id(struct ion_client *client,
int ion_handle_put(struct ion_handle *handle);
+struct ion_handle *ion_alloc2(struct ion_client *client, size_t len,
size_t align, unsigned int usage_id,
unsigned int flags);
+int ion_query_heaps(struct ion_client *client,
struct ion_heap_data __user *buffer,
int cnt);
#endif /* _ION_PRIV_H */ diff --git a/drivers/staging/android/uapi/ion.h b/drivers/staging/android/uapi/ion.h index 145005f..d36b4e4 100644 --- a/drivers/staging/android/uapi/ion.h +++ b/drivers/staging/android/uapi/ion.h @@ -45,10 +45,17 @@ enum ion_heap_type { * must be last so device specific heaps always * are at the end of this enum */
- ION_USAGE_ID_MAP,
A short description for this would be nice
};
#define ION_NUM_HEAP_IDS (sizeof(unsigned int) * 8)
+/*
- This isn't completely random. The old Ion heap ID namespace goes up to
- 32 bits so take the first one after that to use as a dynamic setting
- */
+#define ION_DYNAMIC_HEAP_ASSIGN 33
The old comment says 32, but the implementation looks like 16.
Either way, shouldn't that be 32? If the previous max was 31, 32 is next available - but in ion_device_add_heap() you use ION_DYNAMIC_HEAP_ASSIGN + 1, which makes the first dynamically-assigned ID 34.
/**
- allocation flags - the lower 16 bits are used by core ion, the upper 16
- bits are reserved for use by the heaps themselves.
@@ -65,6 +72,11 @@ enum ion_heap_type { * caches must be managed * manually */ +#define ION_FLAG_NO_HANDLE 3 /*
* Return only an fd associated with
* this buffer and not a handle
*/
/**
- DOC: Ion Userspace API
@@ -142,6 +154,96 @@ struct ion_abi_version { __u32 reserved; };
+/**
- struct ion_new_alloc_data - metadata to/from usersapce allocation v2
- @len: size of the allocation
- @align: required alignment of the allocation
- @usage_id: mapping to heap id to allocate. Will try fallbacks
if specified in the heap mapping
- @flags: flags passed to heap
- @handle: pointer that will be populated with a cookie to use to
refer to this allocation
- @fd: optionally, may return just an fd with no handle
reference
- */
+struct ion_new_alloc_data {
- __u64 len;
- __u64 align;
- __u32 usage_id;
- __u32 flags;
- /*
* I'd like to add a flag to just return the fd instead of just
* a handle for those who want to skip the next step.
*/
- union {
__u32 fd;
__u32 handle;
- };
- /*
* Allocation has a definite problem of 'creep' so give more than
* enough extra bits for expansion
*/
- __u32 reserved0;
- __u32 reserved1;
- __u32 reserved2;
+};
+/**
- struct ion_usage_id_map - metadata to create a new usage map
- @usage_id - userspace allocated array of existing usage ids
- @cnt - number of ids to be mapped
- @new_id - will be populated with the new usage id
- */
+struct ion_usage_id_map {
- /* Array of usage ids provided by user */
- __u64 usage_ids;
- __u32 cnt;
- /* Returned on success */
- __u32 new_id;
- /* Fields for future growth */
- __u32 reserved0;
- __u32 reserved1;
+};
+/**
- struct ion_usage_cnt - meta data for the count of heaps
- @cnt - returned count of number of heaps present
- */
+struct ion_usage_cnt {
- __u32 cnt; /* cnt returned */
- __u32 reserved;
+};
+/**
- struct ion_heap_data - data about a heap
- @name - first 32 characters of the heap name
- @type - heap type
- @usage_id - usage id for the heap
- */
+struct ion_heap_data {
- char name[32];
I realized I upped this to 32 and didn't update the code. I'm going to make this consistent.
- __u32 type;
- __u32 usage_id;
For heaps with type ION_USAGE_ID_MAP I think userspace would want a way to turn a usage_id back into a list of fallbacks to figure out that usage_id == 34 means "only contiguous buffers" for instance. Otherwise only the process that made the mapping can ever know what its for.
- __u32 reserved0;
- __u32 reserved1;
- __u32 reserved2;
+};
+/**
- struct ion_heap_query - collection of data about all heaps
- @cnt - total number of heaps to be copied
- @heaps - buffer to copy heap data
- */
+struct ion_heap_query {
- __u32 cnt; /* Total number of heaps to be copied */
- __u64 heaps; /* buffer to be populated */
I guess this field needs explicit alignment to 64 bits
Yes, I was going by the suggestion in botching-up-ioctls.txt
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/Documen...
- __u32 reserved0;
- __u32 reserved1;
- __u32 reserved2;
+};
#define ION_IOC_MAGIC 'I'
/** @@ -216,5 +318,38 @@ struct ion_abi_version { #define ION_IOC_ABI_VERSION _IOR(ION_IOC_MAGIC, 8, \ struct ion_abi_version)
+/**
- DOC: ION_IOC_ALLOC2 - allocate memory via new API
- Same concept as ION_IOC_ALLOC except with updated ABI. Can return an fd
- directly in addition to a handle
- */
+#define ION_IOC_ALLOC2 _IOWR(ION_IOC_MAGIC, 9, \
struct ion_new_alloc_data)
+/**
- DOC: ION_IOC_ID_MAP - remap an array of heap IDs
- Takes an ion_usage_id_map structure populated with information about
- fallback information. Returns a new usage id for use in allocation.
Can you mention the implied priority of the fallbacks here?
- */
+#define ION_IOC_ID_MAP _IOWR(ION_IOC_MAGIC, 10, \
struct ion_usage_id_map)
+/**
- DOC: ION_IOC_USAGE_CNT - returns a count of the number of usage ids
- Takes an ion_usage_cnt structure and returns the total number of usage
- ids available.
Number of usage IDs or number of heaps?
- */
+#define ION_IOC_USAGE_CNT _IOR(ION_IOC_MAGIC, 11, \
struct ion_usage_cnt)
+/**
- DOC: ION_IOC_HEAP_QUERY - information about available heaps
- Takes an ion_heap_query structure and populates information about
- availble Ion heaps.
- */
+#define ION_IOC_HEAP_QUERY _IOWR(ION_IOC_MAGIC, 12, \
struct ion_heap_query)
#endif /* _UAPI_LINUX_ION_H */
2.5.5
Hi Laura,
On Wed, Jun 08, 2016 at 12:14:12PM -0700, Laura Abbott wrote:
For now I'm just going to focus on comments not about the heap ID mapping because I'm leaning towards dropping the heap ID mapping.
Fair enough. Like you said, giving userspace enough information to just figure out the right heap IDs is already a big improvement.
I'm not that familiar with the DRM ioctls but this sounds like a good suggestion. One less ioctl to worry about.
Sorry, I probably should have pointed to an example: http://lxr.free-electrons.com/source/drivers/gpu/drm/drm_crtc.c#L2325
- */
+struct ion_heap_query {
- __u32 cnt; /* Total number of heaps to be copied */
- __u64 heaps; /* buffer to be populated */
I guess this field needs explicit alignment to 64 bits
Yes, I was going by the suggestion in botching-up-ioctls.txt
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/Documen...
The type is fine, but you should add a 32-bit padding field before heaps: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/Documen...
Cheers,
-Brian
linaro-mm-sig@lists.linaro.org