Re: [PATCH 04/15] block: add an API to atomically update queue limits

[Christian Lamparter <chunkeey@gmail.com>]

Hi,

got a WARNING splatch (=> boot harddrive is inaccessible - device fails to boot) 

------------[ cut here ]------------
WARNING: CPU: 0 PID: 29 at block/blk-settings.c:185 blk_validate_limits+0x154/0x294
Modules linked in:
CPU: 0 PID: 29 Comm: kworker/u4:2 Tainted: G        W          6.10.0+ #1
Hardware name: MyBook Live APM821XX 0x12c41c83 PowerPC 44x Platform
Workqueue: async async_run_entry_fn
NIP:  c02f1f00 LR: c02eef3c CTR: 00000000
REGS: c114bbc0 TRAP: 0700   Tainted: G        W           (6.10.0+)
MSR:  0002b000 <CE,EE,FP,ME>  CR: 84000008  XER: 00000000

GPR00: c02eef28 c114bcb0 c116cf40 c114bda8 00000082 ffffffff ffffffff 00000200
GPR08: 00000200 0000ffff 00001fff c114bc80 44000008 00000000 c00433f0 c119b440
GPR16: 00000000 00000000 00000000 00000000 c105d505 c1101880 c11b6250 00000001
GPR24: 00000000 c0ab0000 c114bda8 ffffffff c0a1eb68 00000000 00000014 c1b683d0
NIP [c02f1f00] blk_validate_limits+0x154/0x294
LR [c02eef3c] blk_alloc_queue+0x80/0x1f0
Call Trace:
[c114bcb0] [c02ffa54] blk_alloc_queue_stats+0x20/0x48 (unreliable)
[c114bcc0] [c02eef28] blk_alloc_queue+0x6c/0x1f0
[c114bcf0] [c02fde24] blk_mq_alloc_queue+0x50/0xa8
[c114bd90] [c04393b4] scsi_alloc_sdev+0x190/0x2b8
[c114be40] [c04395a8] scsi_probe_and_add_lun+0xcc/0x2a0
[c114bea0] [c043a008] __scsi_add_device+0xe4/0x134
[c114bee0] [c045296c] ata_scsi_scan_host+0x84/0x27c
[c114bf30] [c0048158] async_run_entry_fn+0x34/0xcc
[c114bf50] [c003c800] process_scheduled_works+0x170/0x244
[c114bf90] [c003cc48] worker_thread+0x184/0x1d4
[c114bfc0] [c00434bc] kthread+0xcc/0xd0
[c114bff0] [c000c210] start_kernel_thread+0x10/0x14
Code: 81430050 7c085040 40810008 91030050 81430004 2c0a0000 40820010 3940ffff 91430004 48000014 280a3ffe 41a1000c <0fe00000> 48000130 80c30008 81430020
---[ end trace 0000000000000000 ]---
scsi_alloc_sdev: Allocation failure during SCSI scanning, some SCSI devices might not be configured

---

This is due to this patch adds
| /*
|   * By default there is no limit on the segment boundary alignment,
|   * but if there is one it can't be smaller than the page size as
|   * that would break all the normal I/O patterns.
|   */
|   if (!lim->seg_boundary_mask)
|           lim->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
|   if (WARN_ON_ONCE(lim->seg_boundary_mask < PAGE_SIZE - 1)) <----- this warning gets triggered
|          return -EINVAL;

My guess is that this is caused by the kernel has a 16K page size.

CONFIG_HAVE_PAGE_SIZE_16KB=y
CONFIG_PAGE_SIZE_16KB=y
CONFIG_PAGE_SIZE_LESS_THAN_64KB=y
CONFIG_PAGE_SIZE_LESS_THAN_256KB=y
CONFIG_PAGE_SHIFT=14

This worked fine (sata driver is sata_dwc_460ex.c) in the past (and using 16K pages was
slightly faster than 4k pages)... and yes: using a 4K page size works (as in: device boots again).

Regards,
Christian