On Sat, 10 Apr 2010, Ramya Desai wrote: > On Fri, Apr 9, 2010 at 9:14 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > > On Fri, 9 Apr 2010, Sarah Sharp wrote: > > > >> > However there's nothing wrong with capping things at some reasonable > >> > size. If someone tries to submit a request that violates the driver's > >> > specified limit, it's their own fault. Was Ramya's very large transfer > >> > a single bulk URB? Did it use scatter-gather? > >> > >> Yes, Ramya's transfer was one URB with a scatter-gather list of ~120 > >> entries. He was working on a UASP driver that sets the max_sectors > >> field in scsi_host_template to 960 or 1024, which makes some of the > >> sglists very large. > > > > Then his driver is buggy. It should limit the length of scatter-gather > > lists allowed by the host according to what the HCD can support, the > > way usb-storage does: > > > > host->sg_tablesize = usb_stor_sg_tablesize(intf); > > Dear Alan, > > My device is a UASP device, however I am testing the default usb > storage driver for maximum transfer buffer sizes as my device supports > upto 1 MB transfer buffer sizes. There is a thread about the "Maximum > data size in a single transfer for MS driver", which we have discussed > in the month February, 2010. The link for the thread is as follows. > http://marc.info/?t=126641036700002&r=3&w=2 > > NOTE: As we discussed in the above thread, the function, named > store_max_sectors, has changed in 2.6.34-rc2 kernel version. That should not have made any difference. We are talking about sg_tablesize, not max_sectors. In xhci-pci.c, sg_tablesize is set to TRBS_PER_SEGMENT - 1, or 63. Therefore it should not be possible for you to receive a transfer request containing more than 63 scatterlist elements. But Sarah said (see above) that you submitted a request with ~120 elements. This means something is wrong somewhere. > I checked out the 2.6.34-rc2 kernel version from Sarah Sharp git as it > supports the streams. Then, I tested my UASP device for default usb > storage driver for 960 sectors. The xhci-large-tx branch, in Sarah > Sharp git, has every thing set properly like max-sectors in the host > template is set to 960. So, I did NOT change any thing and I am NOT > using the UASP driver for testing the maximum sectors (960 or 1024 > sector transfer buffer sizes) issue. I am using only default usb > storage driver for testing the maximum sectors like 960 or 1024 issue. The max_sectors value has nothing to do with the problem we are discussing. Number of sectors != number of scatterlist elements. The two are related, but they are not the same thing. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
