2009-09-14 21:59:09

by Pavel Machek

[permalink] [raw]
Subject: [patch 3/6] add pmem driver



This adds android pmem driver, one of dependencies of camera driver.

Signed-off-by: Pavel Machek <[email protected]>

diff --git a/drivers/staging/dream/Kconfig b/drivers/staging/dream/Kconfig
index 5770ea6..67783f8 100644
--- a/drivers/staging/dream/Kconfig
+++ b/drivers/staging/dream/Kconfig
@@ -39,5 +39,9 @@ config INPUT_GPIO
Say Y here if you want to support gpio based keys, wheels etc...

config GENERIC_GPIO
- bool
+ bool
+ default y
+
+config ANDROID_PMEM
+ bool "Android pmem allocator"
default y
diff --git a/drivers/staging/dream/Makefile b/drivers/staging/dream/Makefile
index 2b79151..6afbadf 100644
--- a/drivers/staging/dream/Makefile
+++ b/drivers/staging/dream/Makefile
@@ -1,4 +1,5 @@
obj-$(CONFIG_MSM_ADSP) += qdsp5/ smd/
obj-$(CONFIG_MSM_CAMERA) += camera/
obj-$(CONFIG_INPUT_GPIO) += gpio_axis.o gpio_event.o gpio_input.o gpio_matrix.o gpio_output.o
+obj-$(CONFIG_ANDROID_PMEM) += pmem.o

diff --git a/drivers/staging/dream/pmem.c b/drivers/staging/dream/pmem.c
new file mode 100644
index 0000000..def6468
--- /dev/null
+++ b/drivers/staging/dream/pmem.c
@@ -0,0 +1,1335 @@
+/* drivers/android/pmem.c
+ *
+ * Copyright (C) 2007 Google, Inc.
+ *
+ * This software is licensed under the terms of the GNU General Public
+ * License version 2, as published by the Free Software Foundation, and
+ * may be copied, distributed, and modified under those terms.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ */
+
+#include <linux/miscdevice.h>
+#include <linux/platform_device.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/mm.h>
+#include <linux/list.h>
+#include <linux/debugfs.h>
+#include <linux/android_pmem.h>
+#include <linux/mempolicy.h>
+#include <linux/sched.h>
+#include <asm/io.h>
+#include <asm/uaccess.h>
+#include <asm/cacheflush.h>
+
+#define PMEM_MAX_DEVICES 10
+#define PMEM_MAX_ORDER 128
+#define PMEM_MIN_ALLOC PAGE_SIZE
+
+#define PMEM_DEBUG 1
+
+/* indicates that a refernce to this file has been taken via get_pmem_file,
+ * the file should not be released until put_pmem_file is called */
+#define PMEM_FLAGS_BUSY 0x1
+/* indicates that this is a suballocation of a larger master range */
+#define PMEM_FLAGS_CONNECTED 0x1 << 1
+/* indicates this is a master and not a sub allocation and that it is mmaped */
+#define PMEM_FLAGS_MASTERMAP 0x1 << 2
+/* submap and unsubmap flags indicate:
+ * 00: subregion has never been mmaped
+ * 10: subregion has been mmaped, reference to the mm was taken
+ * 11: subretion has ben released, refernece to the mm still held
+ * 01: subretion has been released, reference to the mm has been released
+ */
+#define PMEM_FLAGS_SUBMAP 0x1 << 3
+#define PMEM_FLAGS_UNSUBMAP 0x1 << 4
+
+
+struct pmem_data {
+ /* in alloc mode: an index into the bitmap
+ * in no_alloc mode: the size of the allocation */
+ int index;
+ /* see flags above for descriptions */
+ unsigned int flags;
+ /* protects this data field, if the mm_mmap sem will be held at the
+ * same time as this sem, the mm sem must be taken first (as this is
+ * the order for vma_open and vma_close ops */
+ struct rw_semaphore sem;
+ /* info about the mmaping process */
+ struct vm_area_struct *vma;
+ /* task struct of the mapping process */
+ struct task_struct *task;
+ /* process id of teh mapping process */
+ pid_t pid;
+ /* file descriptor of the master */
+ int master_fd;
+ /* file struct of the master */
+ struct file *master_file;
+ /* a list of currently available regions if this is a suballocation */
+ struct list_head region_list;
+ /* a linked list of data so we can access them for debugging */
+ struct list_head list;
+#if PMEM_DEBUG
+ int ref;
+#endif
+};
+
+struct pmem_bits {
+ unsigned allocated:1; /* 1 if allocated, 0 if free */
+ unsigned order:7; /* size of the region in pmem space */
+};
+
+struct pmem_region_node {
+ struct pmem_region region;
+ struct list_head list;
+};
+
+#define PMEM_DEBUG_MSGS 0
+#if PMEM_DEBUG_MSGS
+#define DLOG(fmt,args...) \
+ do { printk(KERN_INFO "[%s:%s:%d] "fmt, __FILE__, __func__, __LINE__, \
+ ##args); } \
+ while (0)
+#else
+#define DLOG(x...) do {} while (0)
+#endif
+
+struct pmem_info {
+ struct miscdevice dev;
+ /* physical start address of the remaped pmem space */
+ unsigned long base;
+ /* vitual start address of the remaped pmem space */
+ unsigned char __iomem *vbase;
+ /* total size of the pmem space */
+ unsigned long size;
+ /* number of entries in the pmem space */
+ unsigned long num_entries;
+ /* pfn of the garbage page in memory */
+ unsigned long garbage_pfn;
+ /* index of the garbage page in the pmem space */
+ int garbage_index;
+ /* the bitmap for the region indicating which entries are allocated
+ * and which are free */
+ struct pmem_bits *bitmap;
+ /* indicates the region should not be managed with an allocator */
+ unsigned no_allocator;
+ /* indicates maps of this region should be cached, if a mix of
+ * cached and uncached is desired, set this and open the device with
+ * O_SYNC to get an uncached region */
+ unsigned cached;
+ unsigned buffered;
+ /* in no_allocator mode the first mapper gets the whole space and sets
+ * this flag */
+ unsigned allocated;
+ /* for debugging, creates a list of pmem file structs, the
+ * data_list_sem should be taken before pmem_data->sem if both are
+ * needed */
+ struct semaphore data_list_sem;
+ struct list_head data_list;
+ /* pmem_sem protects the bitmap array
+ * a write lock should be held when modifying entries in bitmap
+ * a read lock should be held when reading data from bits or
+ * dereferencing a pointer into bitmap
+ *
+ * pmem_data->sem protects the pmem data of a particular file
+ * Many of the function that require the pmem_data->sem have a non-
+ * locking version for when the caller is already holding that sem.
+ *
+ * IF YOU TAKE BOTH LOCKS TAKE THEM IN THIS ORDER:
+ * down(pmem_data->sem) => down(bitmap_sem)
+ */
+ struct rw_semaphore bitmap_sem;
+
+ long (*ioctl)(struct file *, unsigned int, unsigned long);
+ int (*release)(struct inode *, struct file *);
+};
+
+static struct pmem_info pmem[PMEM_MAX_DEVICES];
+static int id_count;
+
+#define PMEM_IS_FREE(id, index) !(pmem[id].bitmap[index].allocated)
+#define PMEM_ORDER(id, index) pmem[id].bitmap[index].order
+#define PMEM_BUDDY_INDEX(id, index) (index ^ (1 << PMEM_ORDER(id, index)))
+#define PMEM_NEXT_INDEX(id, index) (index + (1 << PMEM_ORDER(id, index)))
+#define PMEM_OFFSET(index) (index * PMEM_MIN_ALLOC)
+#define PMEM_START_ADDR(id, index) (PMEM_OFFSET(index) + pmem[id].base)
+#define PMEM_LEN(id, index) ((1 << PMEM_ORDER(id, index)) * PMEM_MIN_ALLOC)
+#define PMEM_END_ADDR(id, index) (PMEM_START_ADDR(id, index) + \
+ PMEM_LEN(id, index))
+#define PMEM_START_VADDR(id, index) (PMEM_OFFSET(id, index) + pmem[id].vbase)
+#define PMEM_END_VADDR(id, index) (PMEM_START_VADDR(id, index) + \
+ PMEM_LEN(id, index))
+#define PMEM_REVOKED(data) (data->flags & PMEM_FLAGS_REVOKED)
+#define PMEM_IS_PAGE_ALIGNED(addr) (!((addr) & (~PAGE_MASK)))
+#define PMEM_IS_SUBMAP(data) ((data->flags & PMEM_FLAGS_SUBMAP) && \
+ (!(data->flags & PMEM_FLAGS_UNSUBMAP)))
+
+static int pmem_release(struct inode *, struct file *);
+static int pmem_mmap(struct file *, struct vm_area_struct *);
+static int pmem_open(struct inode *, struct file *);
+static long pmem_ioctl(struct file *, unsigned int, unsigned long);
+
+struct file_operations pmem_fops = {
+ .release = pmem_release,
+ .mmap = pmem_mmap,
+ .open = pmem_open,
+ .unlocked_ioctl = pmem_ioctl,
+};
+
+static int get_id(struct file *file)
+{
+ return MINOR(file->f_dentry->d_inode->i_rdev);
+}
+
+static int is_pmem_file(struct file *file)
+{
+ int id;
+
+ if (unlikely(!file || !file->f_dentry || !file->f_dentry->d_inode))
+ return 0;
+ id = get_id(file);
+ if (unlikely(id >= PMEM_MAX_DEVICES))
+ return 0;
+ if (unlikely(file->f_dentry->d_inode->i_rdev !=
+ MKDEV(MISC_MAJOR, pmem[id].dev.minor)))
+ return 0;
+ return 1;
+}
+
+static int has_allocation(struct file *file)
+{
+ struct pmem_data *data;
+ /* check is_pmem_file first if not accessed via pmem_file_ops */
+
+ if (unlikely(!file->private_data))
+ return 0;
+ data = (struct pmem_data *)file->private_data;
+ if (unlikely(data->index < 0))
+ return 0;
+ return 1;
+}
+
+static int is_master_owner(struct file *file)
+{
+ struct file *master_file;
+ struct pmem_data *data;
+ int put_needed, ret = 0;
+
+ if (!is_pmem_file(file) || !has_allocation(file))
+ return 0;
+ data = (struct pmem_data *)file->private_data;
+ if (PMEM_FLAGS_MASTERMAP & data->flags)
+ return 1;
+ master_file = fget_light(data->master_fd, &put_needed);
+ if (master_file && data->master_file == master_file)
+ ret = 1;
+ fput_light(master_file, put_needed);
+ return ret;
+}
+
+static int pmem_free(int id, int index)
+{
+ /* caller should hold the write lock on pmem_sem! */
+ int buddy, curr = index;
+ DLOG("index %d\n", index);
+
+ if (pmem[id].no_allocator) {
+ pmem[id].allocated = 0;
+ return 0;
+ }
+ /* clean up the bitmap, merging any buddies */
+ pmem[id].bitmap[curr].allocated = 0;
+ /* find a slots buddy Buddy# = Slot# ^ (1 << order)
+ * if the buddy is also free merge them
+ * repeat until the buddy is not free or end of the bitmap is reached
+ */
+ do {
+ buddy = PMEM_BUDDY_INDEX(id, curr);
+ if (PMEM_IS_FREE(id, buddy) &&
+ PMEM_ORDER(id, buddy) == PMEM_ORDER(id, curr)) {
+ PMEM_ORDER(id, buddy)++;
+ PMEM_ORDER(id, curr)++;
+ curr = min(buddy, curr);
+ } else {
+ break;
+ }
+ } while (curr < pmem[id].num_entries);
+
+ return 0;
+}
+
+static void pmem_revoke(struct file *file, struct pmem_data *data);
+
+static int pmem_release(struct inode *inode, struct file *file)
+{
+ struct pmem_data *data = (struct pmem_data *)file->private_data;
+ struct pmem_region_node *region_node;
+ struct list_head *elt, *elt2;
+ int id = get_id(file), ret = 0;
+
+
+ down(&pmem[id].data_list_sem);
+ /* if this file is a master, revoke all the memory in the connected
+ * files */
+ if (PMEM_FLAGS_MASTERMAP & data->flags) {
+ struct pmem_data *sub_data;
+ list_for_each(elt, &pmem[id].data_list) {
+ sub_data = list_entry(elt, struct pmem_data, list);
+ down_read(&sub_data->sem);
+ if (PMEM_IS_SUBMAP(sub_data) &&
+ file == sub_data->master_file) {
+ up_read(&sub_data->sem);
+ pmem_revoke(file, sub_data);
+ } else
+ up_read(&sub_data->sem);
+ }
+ }
+ list_del(&data->list);
+ up(&pmem[id].data_list_sem);
+
+
+ down_write(&data->sem);
+
+ /* if its not a conencted file and it has an allocation, free it */
+ if (!(PMEM_FLAGS_CONNECTED & data->flags) && has_allocation(file)) {
+ down_write(&pmem[id].bitmap_sem);
+ ret = pmem_free(id, data->index);
+ up_write(&pmem[id].bitmap_sem);
+ }
+
+ /* if this file is a submap (mapped, connected file), downref the
+ * task struct */
+ if (PMEM_FLAGS_SUBMAP & data->flags)
+ if (data->task) {
+ put_task_struct(data->task);
+ data->task = NULL;
+ }
+
+ file->private_data = NULL;
+
+ list_for_each_safe(elt, elt2, &data->region_list) {
+ region_node = list_entry(elt, struct pmem_region_node, list);
+ list_del(elt);
+ kfree(region_node);
+ }
+ BUG_ON(!list_empty(&data->region_list));
+
+ up_write(&data->sem);
+ kfree(data);
+ if (pmem[id].release)
+ ret = pmem[id].release(inode, file);
+
+ return ret;
+}
+
+static int pmem_open(struct inode *inode, struct file *file)
+{
+ struct pmem_data *data;
+ int id = get_id(file);
+ int ret = 0;
+
+ DLOG("current %u file %p(%d)\n", current->pid, file, file_count(file));
+ /* setup file->private_data to indicate its unmapped */
+ /* you can only open a pmem device one time */
+ if (file->private_data != NULL)
+ return -1;
+ data = kmalloc(sizeof(struct pmem_data), GFP_KERNEL);
+ if (!data) {
+ printk("pmem: unable to allocate memory for pmem metadata.");
+ return -1;
+ }
+ data->flags = 0;
+ data->index = -1;
+ data->task = NULL;
+ data->vma = NULL;
+ data->pid = 0;
+ data->master_file = NULL;
+#if PMEM_DEBUG
+ data->ref = 0;
+#endif
+ INIT_LIST_HEAD(&data->region_list);
+ init_rwsem(&data->sem);
+
+ file->private_data = data;
+ INIT_LIST_HEAD(&data->list);
+
+ down(&pmem[id].data_list_sem);
+ list_add(&data->list, &pmem[id].data_list);
+ up(&pmem[id].data_list_sem);
+ return ret;
+}
+
+static unsigned long pmem_order(unsigned long len)
+{
+ int i;
+
+ len = (len + PMEM_MIN_ALLOC - 1)/PMEM_MIN_ALLOC;
+ len--;
+ for (i = 0; i < sizeof(len)*8; i++)
+ if (len >> i == 0)
+ break;
+ return i;
+}
+
+static int pmem_allocate(int id, unsigned long len)
+{
+ /* caller should hold the write lock on pmem_sem! */
+ /* return the corresponding pdata[] entry */
+ int curr = 0;
+ int end = pmem[id].num_entries;
+ int best_fit = -1;
+ unsigned long order = pmem_order(len);
+
+ if (pmem[id].no_allocator) {
+ DLOG("no allocator");
+ if ((len > pmem[id].size) || pmem[id].allocated)
+ return -1;
+ pmem[id].allocated = 1;
+ return len;
+ }
+
+ if (order > PMEM_MAX_ORDER)
+ return -1;
+ DLOG("order %lx\n", order);
+
+ /* look through the bitmap:
+ * if you find a free slot of the correct order use it
+ * otherwise, use the best fit (smallest with size > order) slot
+ */
+ while (curr < end) {
+ if (PMEM_IS_FREE(id, curr)) {
+ if (PMEM_ORDER(id, curr) == (unsigned char)order) {
+ /* set the not free bit and clear others */
+ best_fit = curr;
+ break;
+ }
+ if (PMEM_ORDER(id, curr) > (unsigned char)order &&
+ (best_fit < 0 ||
+ PMEM_ORDER(id, curr) < PMEM_ORDER(id, best_fit)))
+ best_fit = curr;
+ }
+ curr = PMEM_NEXT_INDEX(id, curr);
+ }
+
+ /* if best_fit < 0, there are no suitable slots,
+ * return an error
+ */
+ if (best_fit < 0) {
+ printk("pmem: no space left to allocate!\n");
+ return -1;
+ }
+
+ /* now partition the best fit:
+ * split the slot into 2 buddies of order - 1
+ * repeat until the slot is of the correct order
+ */
+ while (PMEM_ORDER(id, best_fit) > (unsigned char)order) {
+ int buddy;
+ PMEM_ORDER(id, best_fit) -= 1;
+ buddy = PMEM_BUDDY_INDEX(id, best_fit);
+ PMEM_ORDER(id, buddy) = PMEM_ORDER(id, best_fit);
+ }
+ pmem[id].bitmap[best_fit].allocated = 1;
+ return best_fit;
+}
+
+static pgprot_t phys_mem_access_prot(struct file *file, pgprot_t vma_prot)
+{
+ int id = get_id(file);
+#ifdef pgprot_noncached
+ if (pmem[id].cached == 0 || file->f_flags & O_SYNC)
+ return pgprot_noncached(vma_prot);
+#endif
+#ifdef pgprot_ext_buffered
+ else if (pmem[id].buffered)
+ return pgprot_ext_buffered(vma_prot);
+#endif
+ return vma_prot;
+}
+
+static unsigned long pmem_start_addr(int id, struct pmem_data *data)
+{
+ if (pmem[id].no_allocator)
+ return PMEM_START_ADDR(id, 0);
+ else
+ return PMEM_START_ADDR(id, data->index);
+
+}
+
+static void *pmem_start_vaddr(int id, struct pmem_data *data)
+{
+ return pmem_start_addr(id, data) - pmem[id].base + pmem[id].vbase;
+}
+
+static unsigned long pmem_len(int id, struct pmem_data *data)
+{
+ if (pmem[id].no_allocator)
+ return data->index;
+ else
+ return PMEM_LEN(id, data->index);
+}
+
+static int pmem_map_garbage(int id, struct vm_area_struct *vma,
+ struct pmem_data *data, unsigned long offset,
+ unsigned long len)
+{
+ int i, garbage_pages = len >> PAGE_SHIFT;
+
+ vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP | VM_SHARED | VM_WRITE;
+ for (i = 0; i < garbage_pages; i++) {
+ if (vm_insert_pfn(vma, vma->vm_start + offset + (i * PAGE_SIZE),
+ pmem[id].garbage_pfn))
+ return -EAGAIN;
+ }
+ return 0;
+}
+
+static int pmem_unmap_pfn_range(int id, struct vm_area_struct *vma,
+ struct pmem_data *data, unsigned long offset,
+ unsigned long len)
+{
+ int garbage_pages;
+ DLOG("unmap offset %lx len %lx\n", offset, len);
+
+ BUG_ON(!PMEM_IS_PAGE_ALIGNED(len));
+
+ garbage_pages = len >> PAGE_SHIFT;
+ zap_page_range(vma, vma->vm_start + offset, len, NULL);
+ pmem_map_garbage(id, vma, data, offset, len);
+ return 0;
+}
+
+static int pmem_map_pfn_range(int id, struct vm_area_struct *vma,
+ struct pmem_data *data, unsigned long offset,
+ unsigned long len)
+{
+ DLOG("map offset %lx len %lx\n", offset, len);
+ BUG_ON(!PMEM_IS_PAGE_ALIGNED(vma->vm_start));
+ BUG_ON(!PMEM_IS_PAGE_ALIGNED(vma->vm_end));
+ BUG_ON(!PMEM_IS_PAGE_ALIGNED(len));
+ BUG_ON(!PMEM_IS_PAGE_ALIGNED(offset));
+
+ if (io_remap_pfn_range(vma, vma->vm_start + offset,
+ (pmem_start_addr(id, data) + offset) >> PAGE_SHIFT,
+ len, vma->vm_page_prot)) {
+ return -EAGAIN;
+ }
+ return 0;
+}
+
+static int pmem_remap_pfn_range(int id, struct vm_area_struct *vma,
+ struct pmem_data *data, unsigned long offset,
+ unsigned long len)
+{
+ /* hold the mm semp for the vma you are modifying when you call this */
+ BUG_ON(!vma);
+ zap_page_range(vma, vma->vm_start + offset, len, NULL);
+ return pmem_map_pfn_range(id, vma, data, offset, len);
+}
+
+static void pmem_vma_open(struct vm_area_struct *vma)
+{
+ struct file *file = vma->vm_file;
+ struct pmem_data *data = file->private_data;
+ int id = get_id(file);
+ /* this should never be called as we don't support copying pmem
+ * ranges via fork */
+ BUG_ON(!has_allocation(file));
+ down_write(&data->sem);
+ /* remap the garbage pages, forkers don't get access to the data */
+ pmem_unmap_pfn_range(id, vma, data, 0, vma->vm_start - vma->vm_end);
+ up_write(&data->sem);
+}
+
+static void pmem_vma_close(struct vm_area_struct *vma)
+{
+ struct file *file = vma->vm_file;
+ struct pmem_data *data = file->private_data;
+
+ DLOG("current %u ppid %u file %p count %d\n", current->pid,
+ current->parent->pid, file, file_count(file));
+ if (unlikely(!is_pmem_file(file) || !has_allocation(file))) {
+ printk(KERN_WARNING "pmem: something is very wrong, you are "
+ "closing a vm backing an allocation that doesn't "
+ "exist!\n");
+ return;
+ }
+ down_write(&data->sem);
+ if (data->vma == vma) {
+ data->vma = NULL;
+ if ((data->flags & PMEM_FLAGS_CONNECTED) &&
+ (data->flags & PMEM_FLAGS_SUBMAP))
+ data->flags |= PMEM_FLAGS_UNSUBMAP;
+ }
+ /* the kernel is going to free this vma now anyway */
+ up_write(&data->sem);
+}
+
+static struct vm_operations_struct vm_ops = {
+ .open = pmem_vma_open,
+ .close = pmem_vma_close,
+};
+
+static int pmem_mmap(struct file *file, struct vm_area_struct *vma)
+{
+ struct pmem_data *data;
+ int index;
+ unsigned long vma_size = vma->vm_end - vma->vm_start;
+ int ret = 0, id = get_id(file);
+
+ if (vma->vm_pgoff || !PMEM_IS_PAGE_ALIGNED(vma_size)) {
+#if PMEM_DEBUG
+ printk(KERN_ERR "pmem: mmaps must be at offset zero, aligned"
+ " and a multiple of pages_size.\n");
+#endif
+ return -EINVAL;
+ }
+
+ data = (struct pmem_data *)file->private_data;
+ down_write(&data->sem);
+ /* check this file isn't already mmaped, for submaps check this file
+ * has never been mmaped */
+ if ((data->flags & PMEM_FLAGS_MASTERMAP) ||
+ (data->flags & PMEM_FLAGS_SUBMAP) ||
+ (data->flags & PMEM_FLAGS_UNSUBMAP)) {
+#if PMEM_DEBUG
+ printk(KERN_ERR "pmem: you can only mmap a pmem file once, "
+ "this file is already mmaped. %x\n", data->flags);
+#endif
+ ret = -EINVAL;
+ goto error;
+ }
+ /* if file->private_data == unalloced, alloc*/
+ if (data && data->index == -1) {
+ down_write(&pmem[id].bitmap_sem);
+ index = pmem_allocate(id, vma->vm_end - vma->vm_start);
+ up_write(&pmem[id].bitmap_sem);
+ data->index = index;
+ }
+ /* either no space was available or an error occured */
+ if (!has_allocation(file)) {
+ ret = -EINVAL;
+ printk("pmem: could not find allocation for map.\n");
+ goto error;
+ }
+
+ if (pmem_len(id, data) < vma_size) {
+#if PMEM_DEBUG
+ printk(KERN_WARNING "pmem: mmap size [%lu] does not match"
+ "size of backing region [%lu].\n", vma_size,
+ pmem_len(id, data));
+#endif
+ ret = -EINVAL;
+ goto error;
+ }
+
+ vma->vm_pgoff = pmem_start_addr(id, data) >> PAGE_SHIFT;
+ vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_page_prot);
+
+ if (data->flags & PMEM_FLAGS_CONNECTED) {
+ struct pmem_region_node *region_node;
+ struct list_head *elt;
+ if (pmem_map_garbage(id, vma, data, 0, vma_size)) {
+ printk("pmem: mmap failed in kernel!\n");
+ ret = -EAGAIN;
+ goto error;
+ }
+ list_for_each(elt, &data->region_list) {
+ region_node = list_entry(elt, struct pmem_region_node,
+ list);
+ DLOG("remapping file: %p %lx %lx\n", file,
+ region_node->region.offset,
+ region_node->region.len);
+ if (pmem_remap_pfn_range(id, vma, data,
+ region_node->region.offset,
+ region_node->region.len)) {
+ ret = -EAGAIN;
+ goto error;
+ }
+ }
+ data->flags |= PMEM_FLAGS_SUBMAP;
+ get_task_struct(current->group_leader);
+ data->task = current->group_leader;
+ data->vma = vma;
+#if PMEM_DEBUG
+ data->pid = current->pid;
+#endif
+ DLOG("submmapped file %p vma %p pid %u\n", file, vma,
+ current->pid);
+ } else {
+ if (pmem_map_pfn_range(id, vma, data, 0, vma_size)) {
+ printk(KERN_INFO "pmem: mmap failed in kernel!\n");
+ ret = -EAGAIN;
+ goto error;
+ }
+ data->flags |= PMEM_FLAGS_MASTERMAP;
+ data->pid = current->pid;
+ }
+ vma->vm_ops = &vm_ops;
+error:
+ up_write(&data->sem);
+ return ret;
+}
+
+/* the following are the api for accessing pmem regions by other drivers
+ * from inside the kernel */
+int get_pmem_user_addr(struct file *file, unsigned long *start,
+ unsigned long *len)
+{
+ struct pmem_data *data;
+ if (!is_pmem_file(file) || !has_allocation(file)) {
+#if PMEM_DEBUG
+ printk(KERN_INFO "pmem: requested pmem data from invalid"
+ "file.\n");
+#endif
+ return -1;
+ }
+ data = (struct pmem_data *)file->private_data;
+ down_read(&data->sem);
+ if (data->vma) {
+ *start = data->vma->vm_start;
+ *len = data->vma->vm_end - data->vma->vm_start;
+ } else {
+ *start = 0;
+ *len = 0;
+ }
+ up_read(&data->sem);
+ return 0;
+}
+
+int get_pmem_addr(struct file *file, unsigned long *start,
+ unsigned long *vstart, unsigned long *len)
+{
+ struct pmem_data *data;
+ int id;
+
+ if (!is_pmem_file(file) || !has_allocation(file)) {
+ return -1;
+ }
+
+ data = (struct pmem_data *)file->private_data;
+ if (data->index == -1) {
+#if PMEM_DEBUG
+ printk(KERN_INFO "pmem: requested pmem data from file with no "
+ "allocation.\n");
+ return -1;
+#endif
+ }
+ id = get_id(file);
+
+ down_read(&data->sem);
+ *start = pmem_start_addr(id, data);
+ *len = pmem_len(id, data);
+ *vstart = (unsigned long)pmem_start_vaddr(id, data);
+ up_read(&data->sem);
+#if PMEM_DEBUG
+ down_write(&data->sem);
+ data->ref++;
+ up_write(&data->sem);
+#endif
+ return 0;
+}
+
+int get_pmem_file(int fd, unsigned long *start, unsigned long *vstart,
+ unsigned long *len, struct file **filp)
+{
+ struct file *file;
+
+ file = fget(fd);
+ if (unlikely(file == NULL)) {
+ printk(KERN_INFO "pmem: requested data from file descriptor "
+ "that doesn't exist.");
+ return -1;
+ }
+
+ if (get_pmem_addr(file, start, vstart, len))
+ goto end;
+
+ if (filp)
+ *filp = file;
+ return 0;
+end:
+ fput(file);
+ return -1;
+}
+
+void put_pmem_file(struct file *file)
+{
+ struct pmem_data *data;
+ int id;
+
+ if (!is_pmem_file(file))
+ return;
+ id = get_id(file);
+ data = (struct pmem_data *)file->private_data;
+#if PMEM_DEBUG
+ down_write(&data->sem);
+ if (data->ref == 0) {
+ printk("pmem: pmem_put > pmem_get %s (pid %d)\n",
+ pmem[id].dev.name, data->pid);
+ BUG();
+ }
+ data->ref--;
+ up_write(&data->sem);
+#endif
+ fput(file);
+}
+
+void flush_pmem_file(struct file *file, unsigned long offset, unsigned long len)
+{
+ struct pmem_data *data;
+ int id;
+ void *vaddr;
+ struct pmem_region_node *region_node;
+ struct list_head *elt;
+ void *flush_start, *flush_end;
+
+ if (!is_pmem_file(file) || !has_allocation(file)) {
+ return;
+ }
+
+ id = get_id(file);
+ data = (struct pmem_data *)file->private_data;
+ if (!pmem[id].cached)
+ return;
+
+ down_read(&data->sem);
+ vaddr = pmem_start_vaddr(id, data);
+ /* if this isn't a submmapped file, flush the whole thing */
+ if (unlikely(!(data->flags & PMEM_FLAGS_CONNECTED))) {
+ dmac_flush_range(vaddr, vaddr + pmem_len(id, data));
+ goto end;
+ }
+ /* otherwise, flush the region of the file we are drawing */
+ list_for_each(elt, &data->region_list) {
+ region_node = list_entry(elt, struct pmem_region_node, list);
+ if ((offset >= region_node->region.offset) &&
+ ((offset + len) <= (region_node->region.offset +
+ region_node->region.len))) {
+ flush_start = vaddr + region_node->region.offset;
+ flush_end = flush_start + region_node->region.len;
+ dmac_flush_range(flush_start, flush_end);
+ break;
+ }
+ }
+end:
+ up_read(&data->sem);
+}
+
+static int pmem_connect(unsigned long connect, struct file *file)
+{
+ struct pmem_data *data = (struct pmem_data *)file->private_data;
+ struct pmem_data *src_data;
+ struct file *src_file;
+ int ret = 0, put_needed;
+
+ down_write(&data->sem);
+ /* retrieve the src file and check it is a pmem file with an alloc */
+ src_file = fget_light(connect, &put_needed);
+ DLOG("connect %p to %p\n", file, src_file);
+ if (!src_file) {
+ printk("pmem: src file not found!\n");
+ ret = -EINVAL;
+ goto err_no_file;
+ }
+ if (unlikely(!is_pmem_file(src_file) || !has_allocation(src_file))) {
+ printk(KERN_INFO "pmem: src file is not a pmem file or has no "
+ "alloc!\n");
+ ret = -EINVAL;
+ goto err_bad_file;
+ }
+ src_data = (struct pmem_data *)src_file->private_data;
+
+ if (has_allocation(file) && (data->index != src_data->index)) {
+ printk("pmem: file is already mapped but doesn't match this"
+ " src_file!\n");
+ ret = -EINVAL;
+ goto err_bad_file;
+ }
+ data->index = src_data->index;
+ data->flags |= PMEM_FLAGS_CONNECTED;
+ data->master_fd = connect;
+ data->master_file = src_file;
+
+err_bad_file:
+ fput_light(src_file, put_needed);
+err_no_file:
+ up_write(&data->sem);
+ return ret;
+}
+
+static void pmem_unlock_data_and_mm(struct pmem_data *data,
+ struct mm_struct *mm)
+{
+ up_write(&data->sem);
+ if (mm != NULL) {
+ up_write(&mm->mmap_sem);
+ mmput(mm);
+ }
+}
+
+static int pmem_lock_data_and_mm(struct file *file, struct pmem_data *data,
+ struct mm_struct **locked_mm)
+{
+ int ret = 0;
+ struct mm_struct *mm = NULL;
+ *locked_mm = NULL;
+lock_mm:
+ down_read(&data->sem);
+ if (PMEM_IS_SUBMAP(data)) {
+ mm = get_task_mm(data->task);
+ if (!mm) {
+#if PMEM_DEBUG
+ printk("pmem: can't remap task is gone!\n");
+#endif
+ up_read(&data->sem);
+ return -1;
+ }
+ }
+ up_read(&data->sem);
+
+ if (mm)
+ down_write(&mm->mmap_sem);
+
+ down_write(&data->sem);
+ /* check that the file didn't get mmaped before we could take the
+ * data sem, this should be safe b/c you can only submap each file
+ * once */
+ if (PMEM_IS_SUBMAP(data) && !mm) {
+ pmem_unlock_data_and_mm(data, mm);
+ up_write(&data->sem);
+ goto lock_mm;
+ }
+ /* now check that vma.mm is still there, it could have been
+ * deleted by vma_close before we could get the data->sem */
+ if ((data->flags & PMEM_FLAGS_UNSUBMAP) && (mm != NULL)) {
+ /* might as well release this */
+ if (data->flags & PMEM_FLAGS_SUBMAP) {
+ put_task_struct(data->task);
+ data->task = NULL;
+ /* lower the submap flag to show the mm is gone */
+ data->flags &= ~(PMEM_FLAGS_SUBMAP);
+ }
+ pmem_unlock_data_and_mm(data, mm);
+ return -1;
+ }
+ *locked_mm = mm;
+ return ret;
+}
+
+int pmem_remap(struct pmem_region *region, struct file *file,
+ unsigned operation)
+{
+ int ret;
+ struct pmem_region_node *region_node;
+ struct mm_struct *mm = NULL;
+ struct list_head *elt, *elt2;
+ int id = get_id(file);
+ struct pmem_data *data = (struct pmem_data *)file->private_data;
+
+ /* pmem region must be aligned on a page boundry */
+ if (unlikely(!PMEM_IS_PAGE_ALIGNED(region->offset) ||
+ !PMEM_IS_PAGE_ALIGNED(region->len))) {
+#if PMEM_DEBUG
+ printk("pmem: request for unaligned pmem suballocation "
+ "%lx %lx\n", region->offset, region->len);
+#endif
+ return -EINVAL;
+ }
+
+ /* if userspace requests a region of len 0, there's nothing to do */
+ if (region->len == 0)
+ return 0;
+
+ /* lock the mm and data */
+ ret = pmem_lock_data_and_mm(file, data, &mm);
+ if (ret)
+ return 0;
+
+ /* only the owner of the master file can remap the client fds
+ * that back in it */
+ if (!is_master_owner(file)) {
+#if PMEM_DEBUG
+ printk("pmem: remap requested from non-master process\n");
+#endif
+ ret = -EINVAL;
+ goto err;
+ }
+
+ /* check that the requested range is within the src allocation */
+ if (unlikely((region->offset > pmem_len(id, data)) ||
+ (region->len > pmem_len(id, data)) ||
+ (region->offset + region->len > pmem_len(id, data)))) {
+#if PMEM_DEBUG
+ printk(KERN_INFO "pmem: suballoc doesn't fit in src_file!\n");
+#endif
+ ret = -EINVAL;
+ goto err;
+ }
+
+ if (operation == PMEM_MAP) {
+ region_node = kmalloc(sizeof(struct pmem_region_node),
+ GFP_KERNEL);
+ if (!region_node) {
+ ret = -ENOMEM;
+#if PMEM_DEBUG
+ printk(KERN_INFO "No space to allocate metadata!");
+#endif
+ goto err;
+ }
+ region_node->region = *region;
+ list_add(&region_node->list, &data->region_list);
+ } else if (operation == PMEM_UNMAP) {
+ int found = 0;
+ list_for_each_safe(elt, elt2, &data->region_list) {
+ region_node = list_entry(elt, struct pmem_region_node,
+ list);
+ if (region->len == 0 ||
+ (region_node->region.offset == region->offset &&
+ region_node->region.len == region->len)) {
+ list_del(elt);
+ kfree(region_node);
+ found = 1;
+ }
+ }
+ if (!found) {
+#if PMEM_DEBUG
+ printk("pmem: Unmap region does not map any mapped "
+ "region!");
+#endif
+ ret = -EINVAL;
+ goto err;
+ }
+ }
+
+ if (data->vma && PMEM_IS_SUBMAP(data)) {
+ if (operation == PMEM_MAP)
+ ret = pmem_remap_pfn_range(id, data->vma, data,
+ region->offset, region->len);
+ else if (operation == PMEM_UNMAP)
+ ret = pmem_unmap_pfn_range(id, data->vma, data,
+ region->offset, region->len);
+ }
+
+err:
+ pmem_unlock_data_and_mm(data, mm);
+ return ret;
+}
+
+static void pmem_revoke(struct file *file, struct pmem_data *data)
+{
+ struct pmem_region_node *region_node;
+ struct list_head *elt, *elt2;
+ struct mm_struct *mm = NULL;
+ int id = get_id(file);
+ int ret = 0;
+
+ data->master_file = NULL;
+ ret = pmem_lock_data_and_mm(file, data, &mm);
+ /* if lock_data_and_mm fails either the task that mapped the fd, or
+ * the vma that mapped it have already gone away, nothing more
+ * needs to be done */
+ if (ret)
+ return;
+ /* unmap everything */
+ /* delete the regions and region list nothing is mapped any more */
+ if (data->vma)
+ list_for_each_safe(elt, elt2, &data->region_list) {
+ region_node = list_entry(elt, struct pmem_region_node,
+ list);
+ pmem_unmap_pfn_range(id, data->vma, data,
+ region_node->region.offset,
+ region_node->region.len);
+ list_del(elt);
+ kfree(region_node);
+ }
+ /* delete the master file */
+ pmem_unlock_data_and_mm(data, mm);
+}
+
+static void pmem_get_size(struct pmem_region *region, struct file *file)
+{
+ struct pmem_data *data = (struct pmem_data *)file->private_data;
+ int id = get_id(file);
+
+ if (!has_allocation(file)) {
+ region->offset = 0;
+ region->len = 0;
+ return;
+ } else {
+ region->offset = pmem_start_addr(id, data);
+ region->len = pmem_len(id, data);
+ }
+ DLOG("offset %lx len %lx\n", region->offset, region->len);
+}
+
+
+static long pmem_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+ struct pmem_data *data;
+ int id = get_id(file);
+
+ switch (cmd) {
+ case PMEM_GET_PHYS:
+ {
+ struct pmem_region region;
+ DLOG("get_phys\n");
+ if (!has_allocation(file)) {
+ region.offset = 0;
+ region.len = 0;
+ } else {
+ data = (struct pmem_data *)file->private_data;
+ region.offset = pmem_start_addr(id, data);
+ region.len = pmem_len(id, data);
+ }
+ printk(KERN_INFO "pmem: request for physical address of pmem region "
+ "from process %d.\n", current->pid);
+ if (copy_to_user((void __user *)arg, &region,
+ sizeof(struct pmem_region)))
+ return -EFAULT;
+ break;
+ }
+ case PMEM_MAP:
+ {
+ struct pmem_region region;
+ if (copy_from_user(&region, (void __user *)arg,
+ sizeof(struct pmem_region)))
+ return -EFAULT;
+ data = (struct pmem_data *)file->private_data;
+ return pmem_remap(&region, file, PMEM_MAP);
+ }
+ break;
+ case PMEM_UNMAP:
+ {
+ struct pmem_region region;
+ if (copy_from_user(&region, (void __user *)arg,
+ sizeof(struct pmem_region)))
+ return -EFAULT;
+ data = (struct pmem_data *)file->private_data;
+ return pmem_remap(&region, file, PMEM_UNMAP);
+ break;
+ }
+ case PMEM_GET_SIZE:
+ {
+ struct pmem_region region;
+ DLOG("get_size\n");
+ pmem_get_size(&region, file);
+ if (copy_to_user((void __user *)arg, &region,
+ sizeof(struct pmem_region)))
+ return -EFAULT;
+ break;
+ }
+ case PMEM_GET_TOTAL_SIZE:
+ {
+ struct pmem_region region;
+ DLOG("get total size\n");
+ region.offset = 0;
+ get_id(file);
+ region.len = pmem[id].size;
+ if (copy_to_user((void __user *)arg, &region,
+ sizeof(struct pmem_region)))
+ return -EFAULT;
+ break;
+ }
+ case PMEM_ALLOCATE:
+ {
+ if (has_allocation(file))
+ return -EINVAL;
+ data = (struct pmem_data *)file->private_data;
+ data->index = pmem_allocate(id, arg);
+ break;
+ }
+ case PMEM_CONNECT:
+ DLOG("connect\n");
+ return pmem_connect(arg, file);
+ break;
+ default:
+ if (pmem[id].ioctl)
+ return pmem[id].ioctl(file, cmd, arg);
+ return -EINVAL;
+ }
+ return 0;
+}
+
+#if PMEM_DEBUG
+static ssize_t debug_open(struct inode *inode, struct file *file)
+{
+ file->private_data = inode->i_private;
+ return 0;
+}
+
+static ssize_t debug_read(struct file *file, char __user *buf, size_t count,
+ loff_t *ppos)
+{
+ struct list_head *elt, *elt2;
+ struct pmem_data *data;
+ struct pmem_region_node *region_node;
+ int id = (int)file->private_data;
+ const int debug_bufmax = 4096;
+ static char buffer[4096];
+ int n = 0;
+
+ DLOG("debug open\n");
+ n = scnprintf(buffer, debug_bufmax,
+ "pid #: mapped regions (offset, len) (offset,len)...\n");
+
+ down(&pmem[id].data_list_sem);
+ list_for_each(elt, &pmem[id].data_list) {
+ data = list_entry(elt, struct pmem_data, list);
+ down_read(&data->sem);
+ n += scnprintf(buffer + n, debug_bufmax - n, "pid %u:",
+ data->pid);
+ list_for_each(elt2, &data->region_list) {
+ region_node = list_entry(elt2, struct pmem_region_node,
+ list);
+ n += scnprintf(buffer + n, debug_bufmax - n,
+ "(%lx,%lx) ",
+ region_node->region.offset,
+ region_node->region.len);
+ }
+ n += scnprintf(buffer + n, debug_bufmax - n, "\n");
+ up_read(&data->sem);
+ }
+ up(&pmem[id].data_list_sem);
+
+ n++;
+ buffer[n] = 0;
+ return simple_read_from_buffer(buf, count, ppos, buffer, n);
+}
+
+static struct file_operations debug_fops = {
+ .read = debug_read,
+ .open = debug_open,
+};
+#endif
+
+#if 0
+static struct miscdevice pmem_dev = {
+ .name = "pmem",
+ .fops = &pmem_fops,
+};
+#endif
+
+int pmem_setup(struct android_pmem_platform_data *pdata,
+ long (*ioctl)(struct file *, unsigned int, unsigned long),
+ int (*release)(struct inode *, struct file *))
+{
+ int err = 0;
+ int i, index = 0;
+ int id = id_count;
+ id_count++;
+
+ pmem[id].no_allocator = pdata->no_allocator;
+ pmem[id].cached = pdata->cached;
+ pmem[id].buffered = pdata->buffered;
+ pmem[id].base = pdata->start;
+ pmem[id].size = pdata->size;
+ pmem[id].ioctl = ioctl;
+ pmem[id].release = release;
+ init_rwsem(&pmem[id].bitmap_sem);
+ init_MUTEX(&pmem[id].data_list_sem);
+ INIT_LIST_HEAD(&pmem[id].data_list);
+ pmem[id].dev.name = pdata->name;
+ pmem[id].dev.minor = id;
+ pmem[id].dev.fops = &pmem_fops;
+ printk(KERN_INFO "%s: %d init\n", pdata->name, pdata->cached);
+
+ err = misc_register(&pmem[id].dev);
+ if (err) {
+ printk(KERN_ALERT "Unable to register pmem driver!\n");
+ goto err_cant_register_device;
+ }
+ pmem[id].num_entries = pmem[id].size / PMEM_MIN_ALLOC;
+
+ pmem[id].bitmap = kmalloc(pmem[id].num_entries *
+ sizeof(struct pmem_bits), GFP_KERNEL);
+ if (!pmem[id].bitmap)
+ goto err_no_mem_for_metadata;
+
+ memset(pmem[id].bitmap, 0, sizeof(struct pmem_bits) *
+ pmem[id].num_entries);
+
+ for (i = sizeof(pmem[id].num_entries) * 8 - 1; i >= 0; i--) {
+ if ((pmem[id].num_entries) & 1<<i) {
+ PMEM_ORDER(id, index) = i;
+ index = PMEM_NEXT_INDEX(id, index);
+ }
+ }
+
+ if (pmem[id].cached)
+ pmem[id].vbase = ioremap_cached(pmem[id].base,
+ pmem[id].size);
+#ifdef ioremap_ext_buffered
+ else if (pmem[id].buffered)
+ pmem[id].vbase = ioremap_ext_buffered(pmem[id].base,
+ pmem[id].size);
+#endif
+ else
+ pmem[id].vbase = ioremap(pmem[id].base, pmem[id].size);
+
+ if (pmem[id].vbase == 0)
+ goto error_cant_remap;
+
+ pmem[id].garbage_pfn = page_to_pfn(alloc_page(GFP_KERNEL));
+ if (pmem[id].no_allocator)
+ pmem[id].allocated = 0;
+
+#if PMEM_DEBUG
+ debugfs_create_file(pdata->name, S_IFREG | S_IRUGO, NULL, (void *)id,
+ &debug_fops);
+#endif
+ return 0;
+error_cant_remap:
+ kfree(pmem[id].bitmap);
+err_no_mem_for_metadata:
+ misc_deregister(&pmem[id].dev);
+err_cant_register_device:
+ return -1;
+}
+
+static int pmem_probe(struct platform_device *pdev)
+{
+ struct android_pmem_platform_data *pdata;
+
+ if (!pdev || !pdev->dev.platform_data) {
+ printk(KERN_ALERT "Unable to probe pmem!\n");
+ return -1;
+ }
+ pdata = pdev->dev.platform_data;
+ return pmem_setup(pdata, NULL, NULL);
+}
+
+
+static int pmem_remove(struct platform_device *pdev)
+{
+ int id = pdev->id;
+ __free_page(pfn_to_page(pmem[id].garbage_pfn));
+ misc_deregister(&pmem[id].dev);
+ return 0;
+}
+
+static struct platform_driver pmem_driver = {
+ .probe = pmem_probe,
+ .remove = pmem_remove,
+ .driver = { .name = "android_pmem" }
+};
+
+
+static int __init pmem_init(void)
+{
+ return platform_driver_register(&pmem_driver);
+}
+
+static void __exit pmem_exit(void)
+{
+ platform_driver_unregister(&pmem_driver);
+}
+
+module_init(pmem_init);
+module_exit(pmem_exit);
+

--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

----- End forwarded message -----

--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html


2009-09-14 22:05:34

by Daniel Walker

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
>
> This adds android pmem driver, one of dependencies of camera driver.
>

I wouldn't even both with this one .. There's no way it could ever go
into mainline .. According to the android developers the camera , and
the frame buffer driver should work without it anyway..

Daniel

2009-09-14 22:08:34

by Pavel Machek

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Mon 2009-09-14 15:06:20, Daniel Walker wrote:
> On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
> >
> > This adds android pmem driver, one of dependencies of camera driver.
>
> I wouldn't even both with this one .. There's no way it could ever go
> into mainline .. According to the android developers the camera , and
> the frame buffer driver should work without it anyway..

Well, without it, it does not link. If you can provide better patch,
please go ahead. (But please let me fix the compilation in the meantime).

Plus... why do you think this is completely unacceptable? We may need
to strip the userland interface, but the rest...?
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

2009-09-14 22:19:13

by Brian Swetland

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Mon, Sep 14, 2009 at 3:06 PM, Daniel Walker <[email protected]> wrote:
> On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
>>
>> This adds android pmem driver, one of dependencies of camera driver.
>
> I wouldn't even both with this one .. There's no way it could ever go
> into mainline .. According to the android developers the camera , and
> the frame buffer driver should work without it anyway..

The latest camera driver should work without pmem (or be fixable to do
so -- I forget if that's done yet).

The framebuffer itself does not need pmem, but pmem is used heavily by
the android userspace on dream/magic for hardware compositable
surfaces (since the 2d compositor requires physically contiguous
memory).

Is there a general linux kernel solution for managing large (1-8MB+)
chunks of physically contiguous memory that needs to be shared between
kernel and userspace for media operations? This is a very common
problem with these SoCs -- the a/v subsystem and/or GPU don't have
scatter/gather or MMU support and as a result we need to carve out
memory for their use and provide some way for userspace to manipulate
it.

Brian

2009-09-14 22:22:18

by Daniel Walker

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Tue, 2009-09-15 at 00:08 +0200, Pavel Machek wrote:
> On Mon 2009-09-14 15:06:20, Daniel Walker wrote:
> > On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
> > >
> > > This adds android pmem driver, one of dependencies of camera driver.
> >
> > I wouldn't even both with this one .. There's no way it could ever go
> > into mainline .. According to the android developers the camera , and
> > the frame buffer driver should work without it anyway..
>
> Well, without it, it does not link. If you can provide better patch,
> please go ahead. (But please let me fix the compilation in the meantime).

Now that I think about it, maybe just the frame buffer driver can work
without it.. If it's really needed for the camera then the agenda should
be removing pmem from the camera ..

> Plus... why do you think this is completely unacceptable? We may need
> to strip the userland interface, but the rest..

The userland interface is the core of the driver tho .. According to the
android developers (my understanding anyway), this driver just allows
large blocks of memory to be allocated then shared between userspace and
kernel space .. Brian can explain it better, but from the sound of it
they are likely replicating interfaces that we already have ..

Daniel


2009-09-14 22:26:17

by Pavel Machek

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Mon 2009-09-14 15:23:05, Daniel Walker wrote:
> On Tue, 2009-09-15 at 00:08 +0200, Pavel Machek wrote:
> > On Mon 2009-09-14 15:06:20, Daniel Walker wrote:
> > > On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
> > > >
> > > > This adds android pmem driver, one of dependencies of camera driver.
> > >
> > > I wouldn't even both with this one .. There's no way it could ever go
> > > into mainline .. According to the android developers the camera , and
> > > the frame buffer driver should work without it anyway..
> >
> > Well, without it, it does not link. If you can provide better patch,
> > please go ahead. (But please let me fix the compilation in the meantime).
>
> Now that I think about it, maybe just the frame buffer driver can work
> without it.. If it's really needed for the camera then the agenda should
> be removing pmem from the camera ..
>
> > Plus... why do you think this is completely unacceptable? We may need
> > to strip the userland interface, but the rest..
>
> The userland interface is the core of the driver tho .. According to the
> android developers (my understanding anyway), this driver just allows
> large blocks of memory to be allocated then shared between userspace and
> kernel space .. Brian can explain it better, but from the sound of it
> they are likely replicating interfaces that we already have ..

AFAICT, pmem exports functionality that is required by the other
part. I agree that new userland interface is probably no-go, but I
also believe that it should be easy enough to remove (or maybe change)
in followup patch.

It will be worked on, eventually. For now, we'd like dream support to
compile and link, and then maybe to boot.
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

2009-09-14 22:54:43

by Daniel Walker

[permalink] [raw]
Subject: Re: [patch 3/6] add pmem driver

On Mon, 2009-09-14 at 15:19 -0700, Brian Swetland wrote:
> On Mon, Sep 14, 2009 at 3:06 PM, Daniel Walker <[email protected]> wrote:
> > On Mon, 2009-09-14 at 23:58 +0200, Pavel Machek wrote:
> >>
> >> This adds android pmem driver, one of dependencies of camera driver.
> >
> > I wouldn't even both with this one .. There's no way it could ever go
> > into mainline .. According to the android developers the camera , and
> > the frame buffer driver should work without it anyway..
>
> The latest camera driver should work without pmem (or be fixable to do
> so -- I forget if that's done yet).
>
> The framebuffer itself does not need pmem, but pmem is used heavily by
> the android userspace on dream/magic for hardware compositable
> surfaces (since the 2d compositor requires physically contiguous
> memory).
>
> Is there a general linux kernel solution for managing large (1-8MB+)
> chunks of physically contiguous memory that needs to be shared between
> kernel and userspace for media operations? This is a very common
> problem with these SoCs -- the a/v subsystem and/or GPU don't have
> scatter/gather or MMU support and as a result we need to carve out
> memory for their use and provide some way for userspace to manipulate
> it.

After out last conversation I found the the system calls
splice/vm-splice .. I don't know a great deal about those system call,
but from what I read that sounded like it would be the appropriate way
to transfer memory between two kernel drivers (i.e. camera to
framebuffer) ..

As for sharing it with userspace, I haven't really researched that
enough to give a good answer. I would think you could build something
into the camera driver that would expose the memory to userspace
temporarily then the camera could reclaim the memory later.

Daniel