summaryrefslogtreecommitdiff
path: root/components/drivers/sdio/block_dev.c
diff options
context:
space:
mode:
authorWayne Lin <[email protected]>2021-12-07 13:04:41 +0800
committerBernard Xiong <[email protected]>2021-12-24 05:36:12 +0800
commitbac75e24bd9d67d36cb42b3be8221817aa09eb2e (patch)
tree88f048a5f9407f9e38c56dad55a65759d88fd75d /components/drivers/sdio/block_dev.c
parent600e67f952ab5ffa41084668520fe09015ac37c0 (diff)
[SDIO] Create SD block device.
1. New naming rule for multiple cards: sdXpY. Ex: sd0, sd0p0, sd0p1, sd1, sd1p0, sd1p1, .... sdXpY. 2. Always create super-node for backward-compatible and fdisk-like utility. EX: sd0, sd1, ..... sdX. 3. Shrink code.
Diffstat (limited to 'components/drivers/sdio/block_dev.c')
-rw-r--r--components/drivers/sdio/block_dev.c192
1 files changed, 99 insertions, 93 deletions
diff --git a/components/drivers/sdio/block_dev.c b/components/drivers/sdio/block_dev.c
index 4a49075d27..c546f4a99a 100644
--- a/components/drivers/sdio/block_dev.c
+++ b/components/drivers/sdio/block_dev.c
@@ -351,14 +351,83 @@ const static struct rt_device_ops mmcsd_blk_ops =
};
#endif
+
+static struct mmcsd_blk_device * rt_mmcsd_create_blkdev(struct rt_mmcsd_card *card, const char* dname, struct dfs_partition* psPart, rt_bool_t bMount)
+{
+ struct mmcsd_blk_device *blk_dev;
+ char sname[12];
+
+ blk_dev = rt_calloc(1, sizeof(struct mmcsd_blk_device));
+ if (!blk_dev)
+ {
+ LOG_E("mmcsd:malloc memory failed!");
+ return RT_NULL;
+ }
+
+ if (psPart != RT_NULL)
+ {
+ rt_memcpy(&blk_dev->part, psPart, sizeof(struct dfs_partition));
+ blk_dev->geometry.sector_count = psPart->size;
+ }
+ else
+ {
+ blk_dev->part.offset = 0;
+ blk_dev->part.size = 0;
+ blk_dev->geometry.sector_count = card->card_capacity * (1024 / 512);
+ }
+
+ blk_dev->geometry.bytes_per_sector = 1<<9;
+ blk_dev->geometry.block_size = card->card_blksize;
+
+ blk_dev->part.lock = rt_sem_create(sname, 1, RT_IPC_FLAG_FIFO);
+
+ blk_dev->max_req_size = BLK_MIN((card->host->max_dma_segs *
+ card->host->max_seg_size) >> 9,
+ (card->host->max_blk_count *
+ card->host->max_blk_size) >> 9);
+
+ rt_snprintf(sname, sizeof(sname), "sem_%s", dname);
+
+ /* register mmcsd device */
+ blk_dev->dev.type = RT_Device_Class_Block;
+#ifdef RT_USING_DEVICE_OPS
+ blk_dev->dev.ops = &mmcsd_blk_ops;
+#else
+ blk_dev->dev.init = rt_mmcsd_init;
+ blk_dev->dev.open = rt_mmcsd_open;
+ blk_dev->dev.close = rt_mmcsd_close;
+ blk_dev->dev.read = rt_mmcsd_read;
+ blk_dev->dev.write = rt_mmcsd_write;
+ blk_dev->dev.control = rt_mmcsd_control;
+#endif
+ blk_dev->dev.user_data = blk_dev;
+
+ blk_dev->card = card;
+
+
+ rt_device_register(&blk_dev->dev, dname,
+ RT_DEVICE_FLAG_RDWR | RT_DEVICE_FLAG_REMOVABLE | RT_DEVICE_FLAG_STANDALONE);
+
+ /* Insert to list. */
+ rt_list_insert_after(&card->blk_devices, &blk_dev->list);
+
+#ifdef RT_USING_DFS_MNTTABLE
+ if ( bMount && blk_dev )
+ {
+ LOG_I("try to mount file system!");
+ /* try to mount file system on this block device */
+ dfs_mount_device(&(blk_dev->dev));
+ }
+#endif
+
+ return blk_dev;
+}
+
rt_int32_t rt_mmcsd_blk_probe(struct rt_mmcsd_card *card)
{
rt_int32_t err = 0;
- rt_uint8_t i, status;
+ rt_uint8_t status;
rt_uint8_t *sector;
- char dname[8];
- char sname[12];
- struct mmcsd_blk_device *blk_dev = RT_NULL;
err = mmcsd_set_blksize(card);
if(err)
@@ -380,108 +449,43 @@ rt_int32_t rt_mmcsd_blk_probe(struct rt_mmcsd_card *card)
status = rt_mmcsd_req_blk(card, 0, sector, 1, 0);
if (status == RT_EOK)
{
+ rt_uint8_t i;
+ char dname[8];
+ struct dfs_partition part;
+ struct mmcsd_blk_device *blk_dev = RT_NULL;
rt_int32_t host_id = card->host->id;
/* Initial blk_device link-list. */
rt_list_init(&card->blk_devices);
+
for (i = 0; i < RT_MMCSD_MAX_PARTITION; i++)
{
- blk_dev = rt_calloc(1, sizeof(struct mmcsd_blk_device));
- if (!blk_dev)
- {
- LOG_E("mmcsd:malloc memory failed!");
- break;
- }
-
- /* Given block device name is with allocated host id and its partition index. */
- rt_snprintf(dname, sizeof(dname), "sd%c%d", (host_id+'a'), i);
- rt_snprintf(sname, sizeof(sname), "sem_sd%c%d", (host_id+'a'), i);
-
- blk_dev->max_req_size = BLK_MIN((card->host->max_dma_segs *
- card->host->max_seg_size) >> 9,
- (card->host->max_blk_count *
- card->host->max_blk_size) >> 9);
-
- /* get the first partition */
- status = dfs_filesystem_get_partition(&blk_dev->part, sector, i);
+ /* Get the first partition */
+ status = dfs_filesystem_get_partition(&part, sector, i);
if (status == RT_EOK)
{
- blk_dev->part.lock = rt_sem_create(sname, 1, RT_IPC_FLAG_FIFO);
-
- /* register mmcsd device */
- blk_dev->dev.type = RT_Device_Class_Block;
-#ifdef RT_USING_DEVICE_OPS
- blk_dev->dev.ops = &mmcsd_blk_ops;
-#else
- blk_dev->dev.init = rt_mmcsd_init;
- blk_dev->dev.open = rt_mmcsd_open;
- blk_dev->dev.close = rt_mmcsd_close;
- blk_dev->dev.read = rt_mmcsd_read;
- blk_dev->dev.write = rt_mmcsd_write;
- blk_dev->dev.control = rt_mmcsd_control;
-#endif
- blk_dev->dev.user_data = blk_dev;
-
- blk_dev->card = card;
-
- blk_dev->geometry.bytes_per_sector = 1<<9;
- blk_dev->geometry.block_size = card->card_blksize;
- blk_dev->geometry.sector_count = blk_dev->part.size;
-
- rt_device_register(&blk_dev->dev, dname,
- RT_DEVICE_FLAG_RDWR | RT_DEVICE_FLAG_REMOVABLE | RT_DEVICE_FLAG_STANDALONE);
- rt_list_insert_after(&card->blk_devices, &blk_dev->list);
- }
- else
- {
- if (i == 0)
+ /* Given name is with allocated host id and its partition index. */
+ rt_snprintf(dname, sizeof(dname), "sd%dp%d", host_id, i);
+ blk_dev = rt_mmcsd_create_blkdev(card, (const char*)dname, &part, RT_TRUE);
+ if ( blk_dev == RT_NULL )
{
- /* there is no partition table */
- blk_dev->part.offset = 0;
- blk_dev->part.size = 0;
- blk_dev->part.lock = rt_sem_create(sname, 1, RT_IPC_FLAG_FIFO);
-
- /* register mmcsd device */
- blk_dev->dev.type = RT_Device_Class_Block;
-#ifdef RT_USING_DEVICE_OPS
- blk_dev->dev.ops = &mmcsd_blk_ops;
-#else
- blk_dev->dev.init = rt_mmcsd_init;
- blk_dev->dev.open = rt_mmcsd_open;
- blk_dev->dev.close = rt_mmcsd_close;
- blk_dev->dev.read = rt_mmcsd_read;
- blk_dev->dev.write = rt_mmcsd_write;
- blk_dev->dev.control = rt_mmcsd_control;
-#endif
- blk_dev->dev.user_data = blk_dev;
-
- blk_dev->card = card;
-
- blk_dev->geometry.bytes_per_sector = 1<<9;
- blk_dev->geometry.block_size = card->card_blksize;
- blk_dev->geometry.sector_count =
- card->card_capacity * (1024 / 512);
-
- rt_device_register(&blk_dev->dev, dname,
- RT_DEVICE_FLAG_RDWR | RT_DEVICE_FLAG_REMOVABLE | RT_DEVICE_FLAG_STANDALONE);
- rt_list_insert_after(&card->blk_devices, &blk_dev->list);
- }
- else
- {
- rt_free(blk_dev);
- blk_dev = RT_NULL;
- break;
+ err = -RT_ENOMEM;
+ goto exit_rt_mmcsd_blk_probe;
}
}
-
-#ifdef RT_USING_DFS_MNTTABLE
- if (blk_dev)
+ else
{
- LOG_I("try to mount file system!");
- /* try to mount file system on this block device */
- dfs_mount_device(&(blk_dev->dev));
+ break;
}
-#endif
+ } // for
+
+ /* Always create the super node, given name is with allocated host id. */
+ rt_snprintf(dname, sizeof(dname), "sd%d", host_id);
+ blk_dev = rt_mmcsd_create_blkdev(card, (const char*)dname, RT_NULL, RT_TRUE /*(i==0)?RT_TRUE:RT_FALSE*/);
+ if ( blk_dev == RT_NULL )
+ {
+ err = -RT_ENOMEM;
+ goto exit_rt_mmcsd_blk_probe;
}
}
else
@@ -490,6 +494,8 @@ rt_int32_t rt_mmcsd_blk_probe(struct rt_mmcsd_card *card)
err = -RT_ERROR;
}
+exit_rt_mmcsd_blk_probe:
+
/* release sector buffer */
rt_free(sector);