| Message ID | 20221114102500.88892-1-a.kovaleva@yadro.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Make target send correct io limits | expand |
On 11/14/22 4:24 AM, Anastasia Kovaleva wrote: > Currently iblock always reports MAXIMUM TRANSFER LENGTH in 512b units > regardless of the LU block size. > > Target block size: > target:~ # fdisk -l /dev/nullb0 > Disk /dev/nullb0: 250 GiB, 268435456000 bytes, 65536000 sectors > Units: sectors of 1 * 4096 = 4096 bytes > Sector size (logical/physical): 4096 bytes / 4096 bytes > I/O size (minimum/optimal): 4096 bytes / 4096 bytes > > Initiator block size: > initiator:~ # fdisk -l /dev/sdc > Disk /dev/sdc: 250 GiB, 268435456000 bytes, 65536000 sectors > Disk model: nullb0 > Units: sectors of 1 * 4096 = 4096 bytes > Sector size (logical/physical): 4096 bytes / 4096 bytes > I/O size (minimum/optimal): 4096 bytes / 131072 bytes > > target max transfer length limit: > target:~ # cat /sys/block/nullb0/queue/max_hw_sectors_kb > 128 > > So the maximum transfer length should be 128 * 1024 / 4096 = 32 blocks > But the target sends MTL in 512b units, so the initiators see 256 blocks > instead. > > initiator:~ # sg_inq -p 0xb0 /dev/sdc > VPD INQUIRY: Block limits page (SBC) > Maximum compare and write length: 1 blocks > Optimal transfer length granularity: 1 blocks > Maximum transfer length: 256 blocks > Optimal transfer length: 32 blocks > Maximum prefetch transfer length: 0 blocks [ignored] > > This happens because MAXIMUM TRANSFER LENGTH field in VPD Block Limits > page is derived from dev->dev_attrib.hw_max_sectors which happens to be > in 512b units for iblock. This patch series fixes this issue and removes > some special-casing for other backstores. > > Changes since v1: > Sinse the v1 patch series, some variables was renamed, the checkpatch > script was run and issues with debug logs was fixed, some refactoring > was done. > You forgot the tabbing/spacing issue I mentioned (checkpatch --strict reports them for example). It's probably not a big enough deal to hold this up though. Reviewed-by: Mike Christie <michael.christie@oracle.com>
On Mon, 14 Nov 2022 13:24:57 +0300, Anastasia Kovaleva wrote: > Currently iblock always reports MAXIMUM TRANSFER LENGTH in 512b units > regardless of the LU block size. > > Target block size: > target:~ # fdisk -l /dev/nullb0 > Disk /dev/nullb0: 250 GiB, 268435456000 bytes, 65536000 sectors > Units: sectors of 1 * 4096 = 4096 bytes > Sector size (logical/physical): 4096 bytes / 4096 bytes > I/O size (minimum/optimal): 4096 bytes / 4096 bytes > > [...] Applied to 6.2/scsi-queue, thanks! [1/3] target: core: Send mtl in blocks https://git.kernel.org/mkp/scsi/c/7870d2481789 [2/3] target: core: make hw_max_sectors store the sectors amount in blocks https://git.kernel.org/mkp/scsi/c/9375031ee40b [3/3] target: core: Change the way target_xcopy_do_work sets restiction on max io https://git.kernel.org/mkp/scsi/c/689d94ec208c
Currently iblock always reports MAXIMUM TRANSFER LENGTH in 512b units regardless of the LU block size. Target block size: target:~ # fdisk -l /dev/nullb0 Disk /dev/nullb0: 250 GiB, 268435456000 bytes, 65536000 sectors Units: sectors of 1 * 4096 = 4096 bytes Sector size (logical/physical): 4096 bytes / 4096 bytes I/O size (minimum/optimal): 4096 bytes / 4096 bytes Initiator block size: initiator:~ # fdisk -l /dev/sdc Disk /dev/sdc: 250 GiB, 268435456000 bytes, 65536000 sectors Disk model: nullb0 Units: sectors of 1 * 4096 = 4096 bytes Sector size (logical/physical): 4096 bytes / 4096 bytes I/O size (minimum/optimal): 4096 bytes / 131072 bytes target max transfer length limit: target:~ # cat /sys/block/nullb0/queue/max_hw_sectors_kb 128 So the maximum transfer length should be 128 * 1024 / 4096 = 32 blocks But the target sends MTL in 512b units, so the initiators see 256 blocks instead. initiator:~ # sg_inq -p 0xb0 /dev/sdc VPD INQUIRY: Block limits page (SBC) Maximum compare and write length: 1 blocks Optimal transfer length granularity: 1 blocks Maximum transfer length: 256 blocks Optimal transfer length: 32 blocks Maximum prefetch transfer length: 0 blocks [ignored] This happens because MAXIMUM TRANSFER LENGTH field in VPD Block Limits page is derived from dev->dev_attrib.hw_max_sectors which happens to be in 512b units for iblock. This patch series fixes this issue and removes some special-casing for other backstores. Changes since v1: Sinse the v1 patch series, some variables was renamed, the checkpatch script was run and issues with debug logs was fixed, some refactoring was done. Changes since v2: Since the v2 patch series, a kernel bot found some issues with building a kernel for i386. The problem was that division 64 bit number make GCC generate __udivdi3 which is missing on i386. Anastasia Kovaleva (3): target: core: Send mtl in blocks target: core: make hw_max_sectors store the sectors amount in blocks target: core: Change the way target_xcopy_do_work sets restiction on max io drivers/target/target_core_configfs.c | 2 - drivers/target/target_core_file.c | 1 - drivers/target/target_core_iblock.c | 4 +- drivers/target/target_core_spc.c | 6 +- drivers/target/target_core_xcopy.c | 97 +++++++++++++++------------ drivers/target/target_core_xcopy.h | 2 +- include/target/target_core_base.h | 1 - 7 files changed, 62 insertions(+), 51 deletions(-)