| Message ID | 20211004120650.153218-1-adrian.hunter@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | scsi: ufs: Start to make driver synchronization easier to understand | expand |
> Driver synchronization would be easier to understand if we used the > clk_scaling_lock as the only lock to provide either shared (down/up_read) > or exclusive (down/up_write) access to the host. > > These patches make changes with that in mind, finally resulting in being > able to hold the down_write() lock for the entire error handler duration. > > If this approach is acceptable, it could be extended to simplify the > the synchronization of PM vs error handler and Shutdown vs sysfs. Given that UFSHCD_CAP_CLK_SCALING is only set for ufs-qcom: If extending its use, wouldn't that become a source of contention for them? Thanks, Avri
On 04/10/2021 15:42, Avri Altman wrote: > >> Driver synchronization would be easier to understand if we used the >> clk_scaling_lock as the only lock to provide either shared (down/up_read) >> or exclusive (down/up_write) access to the host. >> >> These patches make changes with that in mind, finally resulting in being >> able to hold the down_write() lock for the entire error handler duration. >> >> If this approach is acceptable, it could be extended to simplify the >> the synchronization of PM vs error handler and Shutdown vs sysfs. > Given that UFSHCD_CAP_CLK_SCALING is only set for ufs-qcom: > If extending its use, wouldn't that become a source of contention for them? It is a good question. The error handler already seems to stop clock scaling. PM / Shutdown seem to suspend / resume clock scaling. So it doesn't *look* like it should make much difference.
Hi Driver synchronization would be easier to understand if we used the clk_scaling_lock as the only lock to provide either shared (down/up_read) or exclusive (down/up_write) access to the host. These patches make changes with that in mind, finally resulting in being able to hold the down_write() lock for the entire error handler duration. If this approach is acceptable, it could be extended to simplify the the synchronization of PM vs error handler and Shutdown vs sysfs. Adrian Hunter (6): scsi: ufs: Encapsulate clk_scaling_lock by inline functions scsi: ufs: Rename clk_scaling_lock to host_rw_sem scsi: ufs: Let ufshcd_[down/up]_read be nested within ufshcd_[down/up]_write scsi: ufs: Fix a possible dead lock in clock scaling scsi: ufs: Reorder dev_cmd locking scsi: ufs: Hold ufshcd_down_write() lock for entire error handler duration drivers/scsi/ufs/ufshcd.c | 104 +++++++++++++++++++++------------------------- drivers/scsi/ufs/ufshcd.h | 41 +++++++++++++++++- 2 files changed, 87 insertions(+), 58 deletions(-) Regards Adrian