This patch introduces a configuration option USBMSC_WRMULTIPLE,
which is used to store multiple blocks into a larger chunk that
then gets written via the mmcsd_writemultiple() (in case mmcsd
is used).
The bottleneck with the current implementation is the poor
performance due to short block writes. USBMSC_DRVR_WRITE()
always writes only one sector (with eMMC that's usually 512 bytes).
eMMC devices usually erase much larger regions with near constant
time (see Jedec JESD84-B51, Extended CSD register byte [225],
SUPER_PAGE_SIZE): 'This register defines one or multiple of
programmable boundary unit that is programmed at the same time.'
If USBMSC_WRMULTIPLE is defined, then USBMSC_NWRREQS is used to
allocate the write buffer size. We don't want this to be the
default behavior yet as this may reveal unseen bugs in usb drivers
due to the faster overall performance.
Sample configurations with measured performance:
- Without USBMSC_WRMULTIPLE: 470 Kb/s
- With USBMSC_WRMULTIPLE, CONFIG_USBMSC_NWRREQS=4: 1.1 Mb/s
(dd with bs=2k)
- With USBMSC_WRMULTIPLE, CONFIG_USBMSC_NWRREQS=16: 5.2 Mb/s
(dd with bs=8k)
No doubt, this feature alone may make the mass storage work 10
times faster than before with eMMC cards.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
usbmsc.h uses the CONFIG_USBMSC_IFNOBASE configuration option,
but it cannot be set anywhere. Provide means to configure this
option which is useful when it's other than the default zero.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
iSerialNumber field in the device descriptor can be used to determining the
board when multiple boards connected to the same host. So add feature to change
serial string by board unique ID dynamically.
To use this feature, user must be implement the board_usbdev_serialstr() logic.
refs #13909
The new configuration CONFIG_USBMSC_NOT_STALL_BULKEP avoids to go into
stall state when :
- Receiving SCSI_CMD_INQUIRY with flags or pagecode != 0
- Receiving SCSI_CMD_MODESENSE6 and return the result data shorter than
the requested data size.
Instead of stall, it just ignores the error condition.
Originally the usb driver goes into the stall under the conditions
mentioned above to inform to the USB host that the requested SCSI feature
is not supported, or the result data is shorter than the requested data
size.
But, at this moment, RP2040 USB device driver cannot handle entering and
leaving stall state of the bulk endpoints well. Once stalls, the host -
device communication stops and cannot be resumed.
I have investigated to solve the issue and found that the existing
RP2040 USB device driver implementation, TinyUSB
(https://github.com/hathach/tinyusb) mass-storage class driver doesn't
enter the stall state like the treatment of this patch.
So, I introduce this new configuration as the RP2040 workaround.
In the future, when the RP2040 USB device driver is fixed and can handle
the bulk endpoint stall correctly, this patch should be reverted.
Gregory Nutt is the copyright holder for those files and he has submitted the
SGA as a result we can migrate the licenses to Apache.
Signed-off-by: Alin Jerpelea <alin.jerpelea@sony.com>
Several licenses were missed in the initial work
David Sidrane has submitted the ICLA and we can migrate the licenses
to Apache.
Gregory Nutt has submitted the SGA and we can migrate the licenses
to Apache.
Signed-off-by: Alin Jerpelea <alin.jerpelea@sony.com>
Michael Jung has submitted the ICLA and we can migrate the licenses
to Apache.
Gregory Nutt has submitted the SGA and we can migrate the licenses
to Apache.
Signed-off-by: Alin Jerpelea <alin.jerpelea@sony.com>
Previous fix to RNDIS response queueing caused unaligned access
to response buffer in some configurations. On ARM Cortex-M, this
would cause hardfault if optimizer used strd instruction.
Fix by allocating the buffer as uint32_t. RNDIS messages are always
a multiple of 4 bytes in size, so this ensures alignment.
Gracefully handle output queue full conditions. This shouldn't happen
in practice as the host is supposed to limit the number of commands
simultaneously in execution.
Reset the response queue on RNDIS_RESET_MSG. This way communication
can recover even if host and device get out of sync.
Sometimes Windows would send RNDIS_KEEPALIVE_MSG and RNDIS_QUERY_MSG close
to each other. This would cause the latter command to overwrite the reply for
the prior command. This in turn will cause Windows to drop the connection after
a 20 second timeout.
Easy way to reproduce the issue is to open the Windows "Adapter Status" dialog that
shows the realtime TX/RX byte counts. This causes multiple RNDIS_QUERY_MSGs per
second, and the connection will drop in less than an hour.
This commit fixes this issue, and other potential race conditions (such as USB
descriptor read in middle on RNDIS query) by using a separate queue for the reply
packets.
to save the preserved space(1KB) and also avoid the heap overhead
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Change-Id: I694073f68e1bd63960cedeea1ddec441437be025
Summary:
- Change default value for RNDIS_NWRREQS based on NET_TCP_WRITE_BUFFERS
Impact:
- This commit affects RNDIS use cases
Testing:
- spresense:rndis with CONFIG_NET_TCP_WRITE_BUFFERS=y
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
