RE: [RESEND PATCH v3 0/2] change UIC command handling

From: Kiwoong Kim
Date: Mon Sep 11 2023 - 02:32:13 EST


> >> ufs: poll HCS.UCRDY before issuing a UIC command
> >
> > [ 4671.226480] [3: kworker/u20:29:17140] BUG: scheduling while atomic:
> > kworker/u20:29/17140/0x00000002
> > ..
> > [ 4671.228723] [3: kworker/u20:29:17140] panic+0x16c/0x388 [
> > 4671.228745] [3: kworker/u20:29:17140] check_panic_on_warn+0x60/0x94
> > [ 4671.228764] [3: kworker/u20:29:17140] __schedule_bug+0x6c/0x94 [
> > 4671.228786] [3: kworker/u20:29:17140] __schedule+0x6f4/0xa64 [
> > 4671.228806] [3: kworker/u20:29:17140] schedule+0x7c/0xe8 [
> > 4671.228824] [3: kworker/u20:29:17140]
> > schedule_hrtimeout_range_clock+0x98/0x114
> > [ 4671.228841] [3: kworker/u20:29:17140]
> > schedule_hrtimeout_range+0x14/0x24
> > [ 4671.228856] [3: kworker/u20:29:17140] usleep_range_state+0x60/0x94
> > [ 4671.228871] [3: kworker/u20:29:17140]
> > __ufshcd_send_uic_cmd+0xa0/0x1c4 [ 4671.228893] [3:
> > kworker/u20:29:17140] ufshcd_uic_pwr_ctrl+0x15c/0x390 [ 4671.228908]
> > [3: kworker/u20:29:17140] ufshcd_uic_hibern8_enter+0x9c/0x25c
> > [ 4671.228922] [3: kworker/u20:29:17140]
> > ufshcd_link_state_transition+0x34/0xb0
> > [ 4671.228939] [3: kworker/u20:29:17140]
> > __ufshcd_wl_suspend+0x3f0/0x4b4
>
> Do you know what is in that path that makes it an atomic context?

Hi,
This made that.

static int ufshcd_uic_pwr_ctrl(struct ufs_hba *hba, struct uic_command *cmd)
..
bool reenable_intr = false;

mutex_lock(&hba->uic_cmd_mutex); <<<<


At first, I was willing to post together w/ the following patch but I've got a suggestion to split the patch set because of different topic and I split the patch set.
- This patch removes the mutex, so it can fix the issue.
https://lore.kernel.org/linux-scsi/1694051306-172962-1-git-send-email-kwmad.kim@xxxxxxxxxxx/


But now I'm thinking again that simply removing the mutex could hurt atomicity of UIC command process
that the original code intended for the first time.
So I think this polling UCRDY should be modified rather than applying removal of the mutex.