On Fri, Sep 21 2012, Leo Song wrote: > If connect an UMS device (for example, an Android phone) to Linux PC and > Windows PC, then copy a big file (~500MB) to the device, you will find that > Linux PC's write speed is much faster than Windows PC. This issue can be fixed > by hiding the "caching mode page" on UMS device. > > Previously in f_mass_storage.c: do_mode_sense(), caching mode page was the > only mode page, and it was configured as: > Write cache enabled (WCE=1) > Read cache not disabled > No cache retention priorities > > The "Write cache enable"(WCE) bit has different influence for Linux PC and > Windows PC. > > For Linux PC, whether WCE is 0 or 1, the write speed from Linux PC to UMS > device has no difference. Linux PC just ignores this flag. > > For Windows PC, if > 1. the UMS devices has a valid "caching mode page" > 2. the WCE bit in "caching mode page" is set to 1 > 3. the "write-cache buffer flushing" on Windows PC is not turned off > > With these three conditions together, Windows PC will send SCSI WRITE COMMAND > with FUA=1 (FUA, force unit access), and FUA=1 will cause UMS device to wait > until the blocks have been written onto the medium, which will drag down write > speed a lot. Unluckily, this is Windows PC's default behavior. > > On Windows, the FUA=1 flag can be changed by turning off "write-cache buffer > flushing" under UMS device's properties, but if we do so, Windows will pop up > error message as following: > > Windows could not change the write-caching setting for the device. > Your device might not support this feature or changing the setting. > [ OK ] > > Windows need to change UMS device's caching mode page, but this page is hard > coded and can't be changed, so there will be error pop up. > > To fix this issue, we can just hide the "caching mode page". With this patch, > there will be no influence for Linux PC, because Linux PC does not handle UMS > device's caching mode page; for Windows PC, if it does see the "caching mode > page", it will not send SCSI WRITE COMMAND with FUA=1, and write speed will > become as fast as Linux PC.
How does it interact with the nofua attribute? With f_mass_storage, you
can write 1 to nofua file in sysfs's directory for given lun to make it
ignore the FUA bit.
> Signed-off-by: Leo Song <lia...@marvell.com>
> ---
> drivers/usb/gadget/f_mass_storage.c | 27 +++------------------------
> 1 files changed, 3 insertions(+), 24 deletions(-)
>
> diff --git a/drivers/usb/gadget/f_mass_storage.c
> b/drivers/usb/gadget/f_mass_storage.c
> index 4f1142e..70f3047 100644
> --- a/drivers/usb/gadget/f_mass_storage.c
> +++ b/drivers/usb/gadget/f_mass_storage.c
> @@ -1296,30 +1296,9 @@ static int do_mode_sense(struct fsg_common *common,
> struct fsg_buffhd *bh)
>
> /* No block descriptors */
>
> - /*
> - * The mode pages, in numerical order. The only page we support
> - * is the Caching page.
> - */
> - if (page_code == 0x08 || all_pages) {
> - valid_page = 1;
> - buf[0] = 0x08; /* Page code */
> - buf[1] = 10; /* Page length */
> - memset(buf+2, 0, 10); /* None of the fields are changeable */
> -
> - if (!changeable_values) {
> - buf[2] = 0x04; /* Write cache enable, */
> - /* Read cache not disabled */
> - /* No cache retention priorities */
> - put_unaligned_be16(0xffff, &buf[4]);
> - /* Don't disable prefetch */
> - /* Minimum prefetch = 0 */
> - put_unaligned_be16(0xffff, &buf[8]);
> - /* Maximum prefetch */
> - put_unaligned_be16(0xffff, &buf[10]);
> - /* Maximum prefetch ceiling */
> - }
> - buf += 12;
> - }
> + /* No mode pages */
> +
> + valid_page = 1;
>
> /*
> * Check that a valid page was requested and the mode data length
--
Best regards, _ _
.o. | Liege of Serenely Enlightened Majesty of o' \,=./ `o
..o | Computer Science, Michał “mina86” Nazarewicz (o o)
ooo +----<email/xmpp: m...@google.com>--------------ooO--(_)--Ooo--
pgp7KZUX6yfOW.pgp
Description: PGP signature
