[PATCH libbsd v2] mmcsd: Don't handle multiple hardware partitions.

Christian Mauderer christian.mauderer at embedded-brains.de
Wed Apr 1 09:45:23 UTC 2020


The rtems_bsd_mmcsd_attach_worker acquired the bus without releasing it.
If a MMC device has multiple hardware partitions (like eMMC which
typically has at least one boot partition) the
rtems_bsd_mmcsd_attach_worker would try to acquire the bus multiple
times. This doesn't work.

Doing it right would mean to acquire and release the bus for each
access which would have an performance impact. Beneath that it would
mean that partition switching has to be supported by the RTEMS code too.

There is currently no known use case where the access would be
necessary. Therefore this patch prefers the performance and just avoids
all further hardware partitions.
---
 freebsd/sys/dev/mmc/mmcsd.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/freebsd/sys/dev/mmc/mmcsd.c b/freebsd/sys/dev/mmc/mmcsd.c
index 077f8d74..21f54046 100644
--- a/freebsd/sys/dev/mmc/mmcsd.c
+++ b/freebsd/sys/dev/mmc/mmcsd.c
@@ -436,6 +436,15 @@ rtems_bsd_mmcsd_attach_worker(rtems_media_state state, const char *src, char **d
 			goto error;
 		}
 
+		/*
+		 * FIXME: There is no release for this acquire. Implementing
+		 * this would be necessary for:
+		 * - multiple hardware partitions of eMMC chips
+		 * - multiple devices on one bus
+		 *
+		 * On the other hand it would mean that the bus has to be
+		 * acquired on every read which would decrease the performance.
+		 */
 		MMCBUS_ACQUIRE_BUS(device_get_parent(dev), dev);
 
 		status_code = rtems_bsd_mmcsd_set_block_size(dev, block_size);
@@ -764,6 +773,10 @@ mmcsd_add_part(struct mmcsd_softc *sc, u_int type, const char *name, u_int cnt,
 			free(part, M_DEVBUF);
 			return;
 		}
+#ifdef __rtems__
+	} else if (type != EXT_CSD_PART_CONFIG_ACC_DEFAULT) {
+		printf("%s: Additional partition. This is currently not supported in RTEMS.", part->name);
+#endif /* __rtems__ */
 	} else {
 		MMCSD_DISK_LOCK_INIT(part);
 
-- 
2.16.4



More information about the devel mailing list