On 2/24/21 12:53 PM, Bodo Stroesser wrote: > This small series is based on Martin's for-next. > > Future patches will need something like the - meanwhile removed - > radix_tree_for_each_contig macro. > Since general direction is to use xarray as replacement for > radix_tree and IDR, instead of re-introducing or open coding the > removed macro, with this series we switch over to xarray API. > Based on xarray a future patch easily can implement an analog > to radix_tree_for_each_contig. > > Bodo Stroesser (2): > scsi: target: tcmu: Replace IDR by XArray > scsi: target: tcmu: Replace radix_tree with XArray > > drivers/target/target_core_user.c | 64 +++++++++++++++++++-------------------- > 1 file changed, 31 insertions(+), 33 deletions(-) > Looks ok to me. Reviewed-by: Mike Christie <michael.christie@xxxxxxxxxx> I think in a separate patch we need to change: + if (xa_store(&udev->data_blocks, dbi, page, GFP_KERNEL)) goto err_insert; to GFP_NOIO. There were some users doing tcm loop + tcmu and were hitting issues when GFP_KERNEL lead to write backs back on to the same tcmu device. We tried to change all the gfp flags but we missed that one, because it was hidden in the radix tree's xa_flags.
