On Fri, 2019-01-18 at 10:48 -0700, Jens Axboe wrote: +AD4 It's UFS that totally buggy, if you look at its queuecommand, it does: +AD4 +AD4 if (+ACE-down+AF8-read+AF8-trylock(+ACY-hba-+AD4-clk+AF8-scaling+AF8-lock)) +AD4 return SCSI+AF8-MLQUEUE+AF8-HOST+AF8-BUSY+ADs +AD4 +AD4 UFS either needs to get fixed up, or we'll want a way to do something like +AD4 the below. +AD4 +AD4 Marc, can you test this? +AD4 +AD4 +AD4 diff --git a/drivers/scsi/hosts.c b/drivers/scsi/hosts.c +AD4 index eaf329db3973..e28c3420a9d9 100644 +AD4 --- a/drivers/scsi/hosts.c +AD4 +-+-+- b/drivers/scsi/hosts.c +AD4 +AEAAQA -412,6 +-412,7 +AEAAQA struct Scsi+AF8-Host +ACo-scsi+AF8-host+AF8-alloc(struct scsi+AF8-host+AF8-template +ACo-sht, int privsize) +AD4 shost-+AD4-hostt +AD0 sht+ADs +AD4 shost-+AD4-this+AF8-id +AD0 sht-+AD4-this+AF8-id+ADs +AD4 shost-+AD4-can+AF8-queue +AD0 sht-+AD4-can+AF8-queue+ADs +AD4 +- shost-+AD4-queue+AF8-may+AF8-block +AD0 sht-+AD4-queue+AF8-may+AF8-block+ADs +AD4 shost-+AD4-sg+AF8-tablesize +AD0 sht-+AD4-sg+AF8-tablesize+ADs +AD4 shost-+AD4-sg+AF8-prot+AF8-tablesize +AD0 sht-+AD4-sg+AF8-prot+AF8-tablesize+ADs +AD4 shost-+AD4-cmd+AF8-per+AF8-lun +AD0 sht-+AD4-cmd+AF8-per+AF8-lun+ADs +AD4 diff --git a/drivers/scsi/scsi+AF8-lib.c b/drivers/scsi/scsi+AF8-lib.c +AD4 index b13cc9288ba0..4e266af2871f 100644 +AD4 --- a/drivers/scsi/scsi+AF8-lib.c +AD4 +-+-+- b/drivers/scsi/scsi+AF8-lib.c +AD4 +AEAAQA -1902,6 +-1902,8 +AEAAQA int scsi+AF8-mq+AF8-setup+AF8-tags(struct Scsi+AF8-Host +ACo-shost) +AD4 shost-+AD4-tag+AF8-set.flags +AD0 BLK+AF8-MQ+AF8-F+AF8-SHOULD+AF8-MERGE +AHw BLK+AF8-MQ+AF8-F+AF8-SG+AF8-MERGE+ADs +AD4 shost-+AD4-tag+AF8-set.flags +AHwAPQ +AD4 BLK+AF8-ALLOC+AF8-POLICY+AF8-TO+AF8-MQ+AF8-FLAG(shost-+AD4-hostt-+AD4-tag+AF8-alloc+AF8-policy)+ADs +AD4 +- if (shost-+AD4-queue+AF8-may+AF8-blocK) +AD4 +- shost-+AD4-tag+AF8-set.flags +AHwAPQ BLK+AF8-MQ+AF8-F+AF8-BLOCKING+ADs +AD4 shost-+AD4-tag+AF8-set.driver+AF8-data +AD0 shost+ADs +AD4 +AD4 return blk+AF8-mq+AF8-alloc+AF8-tag+AF8-set(+ACY-shost-+AD4-tag+AF8-set)+ADs +AD4 diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c +AD4 index 9ba7671b84f8..9ab354e43630 100644 +AD4 --- a/drivers/scsi/ufs/ufshcd.c +AD4 +-+-+- b/drivers/scsi/ufs/ufshcd.c +AD4 +AEAAQA -6981,6 +-6981,7 +AEAAQA static struct scsi+AF8-host+AF8-template ufshcd+AF8-driver+AF8-template +AD0 +AHs +AD4 .sg+AF8-tablesize +AD0 SG+AF8-ALL, +AD4 .cmd+AF8-per+AF8-lun +AD0 UFSHCD+AF8-CMD+AF8-PER+AF8-LUN, +AD4 .can+AF8-queue +AD0 UFSHCD+AF8-CAN+AF8-QUEUE, +AD4 +- .queue+AF8-may+AF8-block +AD0 1, +AD4 .max+AF8-host+AF8-blocked +AD0 1, +AD4 .track+AF8-queue+AF8-depth +AD0 1, +AD4 .sdev+AF8-groups +AD0 ufshcd+AF8-driver+AF8-groups, +AD4 diff --git a/include/scsi/scsi+AF8-host.h b/include/scsi/scsi+AF8-host.h +AD4 index 6ca954e9f752..30aa7b6c4342 100644 +AD4 --- a/include/scsi/scsi+AF8-host.h +AD4 +-+-+- b/include/scsi/scsi+AF8-host.h +AD4 +AEAAQA -339,6 +-339,11 +AEAAQA struct scsi+AF8-host+AF8-template +AHs +AD4 +ACo-/ +AD4 int can+AF8-queue+ADs +AD4 +AD4 +- /+ACo +AD4 +- +ACo If the -+AD4-queuecommand() ever blocks, this should be set +AD4 +- +ACo-/ +AD4 +- int queue+AF8-may+AF8-block+ADs +AD4 +- +AD4 /+ACo +AD4 +ACo In many instances, especially where disconnect / reconnect are +AD4 +ACo supported, our host also has an ID on the SCSI bus. If this is +AD4 +AEAAQA -584,6 +-589,7 +AEAAQA struct Scsi+AF8-Host +AHs +AD4 +AD4 int this+AF8-id+ADs +AD4 int can+AF8-queue+ADs +AD4 +- int queue+AF8-may+AF8-block+ADs +AD4 short cmd+AF8-per+AF8-lun+ADs +AD4 short unsigned int sg+AF8-tablesize+ADs +AD4 short unsigned int sg+AF8-prot+AF8-tablesize+ADs Hi Jens, Did you perhaps want to include a change from down+AF8-read+AF8-trylock() into down+AF8-read() in the UFS queuecommand function in this patch? Thanks, Bart.
