From: Vasiliy T. <v.t...@se...> - 2017-08-07 06:58:27
|
2017-08-06 3:20 GMT+03:00 Bart Van Assche <Bar...@wd...>: > Several years ago I experimented with adding AEN support to both the ib_srpt > and ib_srp kernel drivers. However, while testing these changes I noticed > that if ib_srpt sends an AEN (called SRP_AER_REQ in the SRP protocol spec) > that this crashes any connected Windows SRP initiator systems. Although this > is cleary a bug in the Windows SRP initiator driver, that's why AEN support > is not present in the ib_srpt driver. > So as i understand AEN not fully working... > But even if ib_srpt sends no SRP_AER_REQ requests, any Linux system that is > connected to it should poll the SRP target system periodically. Both the Linux > SCSI disk (sd) and CD-ROM (sr) drivers poll the target system periodically by > sending a TUR (TEST UNIT READY) command (see also the sd_check_events() and > sr_check_events() functions in drivers/scsi/s[dr].c). How often SCSI devices > are polled can be controlled through /sys/block/*/events_poll_msecs. A udev > rule (/usr/lib/udev/rules.d/60-block.rules) sets the default polling interval. > > Are you sure that you really need asynchronous notifications and that the > polling by the Linux SCSI initiator is not sufficient? Thanks for the info about polling. Yes i'm fine with polling (i'm use 1000 msec). Can you says, what values i need for dev_loss_tmo/fast_io_fail_tmo if i use disk for qemu and need to notify quemu as fast as possible when device is lost? (in qemu i'm freeze vm while io failed) -- Vasiliy Tolstov, e-mail: v.t...@se... |