ide: merge do_rw_taskfile() and flagged_taskfile() into do_taskfile()
PATCH 1/1 ide: merge do_rw_taskfile() and flagged_taskfile() into do_taskfile()
Hello, Bartlomiej.
This is a new version of ide_do_taskfile.patch. Compared to the
original do_rw_task(), only one more 'if' is used in the hot path, so I think the performance issue can be ignored now. Also, there's no userland visible change with this patch. Everything should work just as it did with do_rw_taskfile()/flagged_taskfile().
do_taskfile() is different from do_rw_taskfile() in that
- It uses task->data_phase to determine whether it's a DMA command or not.
do_taskfile() is different from flagged_taskfile() in that
- No (TASKFILE_MULTI_IN && !mult_count) check. ide_taskfile_ioctl() checks the same thing, so it doesn't change anything.
- No task->tf_out_flags handling. ide_end_drive_cmd() ignores it anyway, so, again, it doesn't change anything.
So, what do you think?
drivers/ide/ide-disk.c +6 -6
@@ -535,7 +535,7 @@ static ide_startstop_t idedisk_special ( args.tfRegister[IDE_COMMAND_OFFSET ] =WIN_SPECIFY ; args.command_type =IDE_DRIVE_TASK_NO_DATA ; args.handler = &set_geometry_intr; -do_rw_taskfile (drive, &args); +do_taskfile (drive, &args); } }else if (s->b.recalibrate) { s->b.recalibrate =0 ;
@@ -546,7 +546,7 @@ static ide_startstop_t idedisk_special ( args.tfRegister[IDE_COMMAND_OFFSET ] =WIN_RESTORE ; args.command_type =IDE_DRIVE_TASK_NO_DATA ; args.handler = &recal_intr; -do_rw_taskfile (drive, &args); +do_taskfile (drive, &args); } }else if (s->b.set_multmode) { s->b.set_multmode =0 ;
@@ -559,7 +559,7 @@ static ide_startstop_t idedisk_special ( args.tfRegister[IDE_COMMAND_OFFSET ] =WIN_SETMULT ; args.command_type =IDE_DRIVE_TASK_NO_DATA ; args.handler = &set_multmode_intr; -do_rw_taskfile (drive, &args); +do_taskfile (drive, &args); } }else if (s->all) {int special = s->all;
@@ -898,19 +898,19 @@ static ide_startstop_t idedisk_start_pow args->tfRegister[IDE_COMMAND_OFFSET ] =WIN_FLUSH_CACHE ; args->command_type =IDE_DRIVE_TASK_NO_DATA ; args->handler = &task_no_data_intr; -return do_rw_taskfile (drive, args); +return do_taskfile (drive, args);case idedisk_pm_standby:/* Suspend step 2 (standby) */ args->tfRegister[IDE_COMMAND_OFFSET ] =WIN_STANDBYNOW1 ; args->command_type =IDE_DRIVE_TASK_NO_DATA ; args->handler = &task_no_data_intr; -return do_rw_taskfile (drive, args); +return do_taskfile (drive, args);case idedisk_pm_idle:/* Resume step 1 (idle) */ args->tfRegister[IDE_COMMAND_OFFSET ] =WIN_IDLEIMMEDIATE ; args->command_type =IDE_DRIVE_TASK_NO_DATA ; args->handler = task_no_data_intr; -return do_rw_taskfile (drive, args); +return do_taskfile (drive, args);case idedisk_pm_restore_dma:/* Resume step 2 (restore DMA) */ /*
drivers/ide/ide-io.c +1 -3
@@ -682,9 +682,7 @@ static ide_startstop_t execute_drive_tasbreak ; } -if (args->tf_out_flags.all !=0 ) -return flagged_taskfile (drive, args); -return do_rw_taskfile (drive, args); +return do_taskfile (drive, args); }/**
drivers/ide/ide-taskfile.c +83 -162
274 lines not on this page · raw message
include/linux/ide.h +2 -7
@@ -1090,7 +1090,7 @@ static inline void destroy_proc_ide_inte* - ide_started : In this case, the channel is left busy until an * async event (interrupt) occurs. * Typically, start_power_step() will issue a taskfile request with -* do_rw_taskfile(). +* do_taskfile(). * * Upon reception of the interrupt, the core will call complete_power_step() * with the error code if any. This routine should update the step value
@@ -1298,12 +1298,7 @@ extern int wait_for_ready(ide_drive_t *,/* * taskfile io for disks for now...and builds request from ide_ioctl */ -extern ide_startstop_t do_rw_taskfile (ide_drive_t *,ide_task_t *); - -/* -* Special Flagged Register Validation Caller -*/ -extern ide_startstop_t flagged_taskfile (ide_drive_t *,ide_task_t *); +extern ide_startstop_t do_taskfile (ide_drive_t *,ide_task_t *);extern ide_startstop_t set_multmode_intr (ide_drive_t *);extern ide_startstop_t set_geometry_intr (ide_drive_t *);
Quoted 1 line · Tejun Heo
Hello, Bartlomiej.
Hi,
Quoted 7 lines
This is a new version of ide_do_taskfile.patch. Compared to the
originaldo_rw_task(), only one more 'if' is used in the hot path, so I think the performance issue can be ignored now. Also, there's no userland visible change with this patch. Everything should work just as it did with do_rw_taskfile()/flagged_taskfile().
do_taskfile()is different fromdo_rw_taskfile()in that
Is there any gain in changing name to do_taskfile()?
Quoted 2 lines
- It uses
task->data_phaseto determine whether it's a DMA command or not.
this is user-space visible change
(it is right thing to do, I just wanted to point the fact)
Quoted 4 lines
do_taskfile()is different fromflagged_taskfile()in that- No (
TASKFILE_MULTI_IN&& !mult_count) check.ide_taskfile_ioctl()checks the same thing, so it doesn't change anything.
The check may be needed. AFAIR drive->mult_count may change before our taskfile request is started.
Quoted 2 lines
- No
task->tf_out_flagshandling.ide_end_drive_cmd()ignores it anyway, so, again, it doesn't change anything.
I guess you mean ->tf_in_flags?
Quoted 1 line
So, what do you think?
This patch looks much better but could you move writing taskfile registers to separate helpers (one for non-flagged and one for flagged)?
Probably splitting non-flagged taskfile load helper off do_rw_taskfile() should be done in separate patch. We can then use this helper in ide-disk.c for __ide_do_rw_taskfile() (we can't do direct conversion to do_rw_taskfile() yet for various reasons).
Thanks,
Bartlomiej
Hello,
Quoted 14 lines · Bartlomiej Zolnierkiewicz
Quoted 1 line · Tejun Heo
Hello, Bartlomiej.
Hi,
Quoted 6 lines
This is a new version of ide_do_taskfile.patch. Compared to the original
do_rw_task(), only one more 'if' is used in the hot path, so I think the performance issue can be ignored now. Also, there's no userland visible change with this patch. Everything should work just as it did with do_rw_taskfile()/flagged_taskfile().
Quoted 5 lines
Quoted 1 line
do_taskfile()is different fromdo_rw_taskfile()in thatIs there any gain in changing name to
do_taskfile()?
Well, I was just thinking
do_rw_taskfile + (do_)flagged_taskfile -> do_taskfile.
If you like do_rw_taskfile better, I guess that's okay too. :-)
Quoted 8 lines
Quoted 3 lines
- It uses
task->data_phaseto determine whether it's a DMA command or not.this is user-space visible change
(it is right thing to do, I just wanted to point the fact)
Oops, I forgot that. Still, it was kind of weird before. If any of
tf_{in|out}_flags is set, flagged_taskfile() is called and ->data_phase was used, but, if none of the flags was set, do_rw_taskfile() was called and the command is used to determine the same thing. But, yeah, it's user-visible.
Quoted 10 lines
Quoted 5 lines
do_taskfile()is different fromflagged_taskfile()in that- No (
TASKFILE_MULTI_IN&& !mult_count) check.ide_taskfile_ioctl()checks the same thing, so it doesn't change anything.The check may be needed. AFAIR
drive->mult_countmay change before our taskfile request is started.
Okay, I'll resurrect that test.
Quoted 7 lines
Quoted 3 lines
- No
task->tf_out_flagshandling.ide_end_drive_cmd()ignores it anyway, so, again, it doesn't change anything.I guess you mean ->tf_in_flags?
Yes.
Quoted 11 lines
Quoted 2 lines
So, what do you think?
This patch looks much better but could you move writing taskfile registers to separate helpers (one for non-flagged and one for flagged)?
Probably splitting non-flagged taskfile load helper off
do_rw_taskfile()should be done in separate patch. We can then use this helper inide-disk.cfor__ide_do_rw_taskfile()(we can't do direct conversion todo_rw_taskfile()yet for various reasons).
Sure, I'll do it now.
Thanks.
Signature · 1 line
tejun