Hi, So a recent commit that disable 64 bit dma addressing for ASM1142: https://github.com/torvalds/linux/commit/ec37198acca7b4c17b96247697406e47aafe0605 I notice that it also causes (hw) max_sectors of uas drives connected to the controller to drop from 1024 blocks (SCSI_DEFAULT_MAX_SECTORS) to 512 blocks, because we clamp it against dma_max_mapping_size($some_device). For uas drives, $some_device is `sdev->host->dma_dev`: https://github.com/torvalds/linux/blob/v5.8/drivers/scsi/scsi_lib.c#L1778 (`shost` is `sdev->host`: https://github.com/torvalds/linux/blob/v5.8/drivers/scsi/scsi_lib.c#L1873) But for bot/msc drives: $some_device is some other device: https://github.com/torvalds/linux/blob/v5.8/drivers/usb/storage/scsiglue.c#L92 In that case, max_sectors remains to be 2048 blocks for SS BOT drives connected to the ASM1142 controller. I tried to change it to `sdev->host->dma_dev`. It lowered their max_sectors to 512 blocks. So my questions are: 1. Do we really need to do any clamping to max_sectors in either (uas/bot) case? 2. If so, should we change the $some_device in either case? For the record, in the original commit (for BOT drives): https://github.com/torvalds/linux/commit/d74ffae8b8dd17eaa8b82fc163e6aa2076dc8fb1 The commit message states that the clamping is needed because of some other change, which is supposed to cause max_segment_size to raise from 0x10000 to 0xffffffff, but when I check the sysfs files for both kinds of drives, all of the values are 0x10000. Also see https://github.com/virtio-win/kvm-guest-drivers-windows/issues/498
