Re: [PATCH] libata: Use per port sync for detach

From: Kai-Heng Feng
Date: Fri May 15 2020 - 13:49:36 EST



> On May 15, 2020, at 20:38, John Garry <john.garry@xxxxxxxxxx> wrote:
>
> On 15/05/2020 12:09, Kai-Heng Feng wrote:
>> Commit 130f4caf145c ("libata: Ensure ata_port probe has completed before
>> detach") may cause system freeze during suspend.
>> Using async_synchronize_full() in PM callbacks is wrong, since async
>> callbacks that are already scheduled may wait for not-yet-scheduled
>> callbacks, causes a circular dependency.
>
> It would be nice to describe this circular dependency a bit more.

Hmm, I think it's quite self-explanatory. Which part do you want me to add?

>
>> Instead of using big hammer like async_synchronize_full(), use async
>> cookie to make sure port probe are synced, without affecting other
>> scheduled PM callbacks.
>
> oh, I thought that we could avoid the hassle of per-port cookie management. Sorry.
>
> Did you check if the original issue which I tried to remedy is still suppressed?
>
> I tried your patch, and got this:
>
> [ 28.190587] ------------[ cut here ]------------
> [ 28.195194] WARNING: CPU: 79 PID: 1 at drivers/ata/libata-core.c:5888 ata_hos
> t_detach+0x238/0x248
> [ 28.204025] Modules linked in:
> [ 28.207072] CPU: 79 PID: 1 Comm: swapper/0 Tainted: G W 5.7.0-
> rc5-g644cd6f-dirty #84
> [ 28.216076] Hardware name: Huawei TaiShan 2280 V2/BC82AMDC, BIOS 2280-V2 CS V
> 3.B220.02 03/27/2020
> [ 28.224906] pstate: 20c00009 (nzCv daif +PAN +UAO)
> [ 28.229677] pc : ata_host_detach+0x238/0x248
> [ 28.233929] lr : ata_host_detach+0x12c/0x248
> [ 28.238181] sp : ffff0026dc74f980
> [ 28.241481] x29: ffff0026dc74f980 x28: 0000000000000000
> [ 28.246769] x27: ffff2026c541c010 x26: 00000000000038e0
> [ 28.252059] x25: 0000000000003590 x24: ffff0026d3a0a018
> [ 28.257350] x23: 0000000000000001 x22: ffff0026d3a0a000
> [ 28.262638] x21: 0000000000013ec0 x20: ffff2026c541c000
> [ 28.267928] x19: ffff2026c541c020 x18: 0000000000000000
> [ 28.273215] x17: 0000000000000000 x16: 0000000000000000
> [ 28.278504] x15: 00000000000006c0 x14: 0000000000000000
> [ 28.283792] x13: 0000000000000000 x12: 1fffe004da744317
> [ 28.289081] x11: ffff8004da744317 x10: dfffa00000000000
> [ 28.294370] x9 : ffff8004da744318 x8 : ffff0026d3a218bc
> [ 28.299657] x7 : 0000000000000001 x6 : ffff8004da744318
> [ 28.304945] x5 : ffff8004da744318 x4 : dfffa00000000000
> [ 28.310233] x3 : ffffa00075ea18b4 x2 : 0000000000000003
> [ 28.315521] x1 : 0000000000000000 x0 : 0000000000400200
> [ 28.320808] Call trace:
> [ 28.323246] ata_host_detach+0x238/0x248
> [ 28.327153] ata_pci_remove_one+0x24/0x38
> [ 28.331147] ahci_remove_one+0x54/0x88
> [ 28.334881] pci_device_remove+0x70/0x148
> [ 28.338874] really_probe+0x17c/0x570
> [ 28.342522] driver_probe_device+0x80/0x150
> [ 28.346690] device_driver_attach+0x9c/0xa8
> [ 28.350856] __driver_attach+0xac/0x118
> [ 28.354677] bus_for_each_dev+0xf0/0x168
> [ 28.358584] driver_attach+0x34/0x48
> [ 28.362146] bus_add_driver+0x240/0x300
> [ 28.365966] driver_register+0xc0/0x1e0
> [ 28.369787] __pci_register_driver+0xb4/0xd0
> [ 28.374039] ahci_pci_driver_init+0x24/0x30
> [ 28.378205] do_one_initcall+0xb8/0x268
> [ 28.382027] kernel_init_freeable+0x294/0x30c
> [ 28.386366] kernel_init+0x14/0x120
> [ 28.389841] ret_from_fork+0x10/0x1c
> [ 28.393400] ---[ end trace 9972785c7052048f ]---
> [ 28.435826] ahci 0000:b4:03.0: SSS flag set, parallel bus scan disabled

Can you please test the following patch:

diff --git a/drivers/ata/libata-eh.c b/drivers/ata/libata-eh.c
index 474c6c34fe02..51ee0cc4d414 100644
--- a/drivers/ata/libata-eh.c
+++ b/drivers/ata/libata-eh.c
@@ -3583,8 +3583,10 @@ int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
rc = 0;

/* if UNLOADING, finish immediately */
- if (ap->pflags & ATA_PFLAG_UNLOADING)
+ if (ap->pflags & ATA_PFLAG_UNLOADING) {
+ ap->pflags |= ATA_PFLAG_UNLOADED;
goto out;
+ }

It's only compile-tested, many drivers panic with CONFIG_DEBUG_TEST_DRIVER_REMOVE enabled, so the system I have can't even boot properly :(

Probably worth some time to fix them one by one...

Kai-Heng

>
>
> Thanks,
> John
>
>> Fixes: 130f4caf145c ("libata: Ensure ata_port probe has completed before detach")
>> BugLink: https://bugs.launchpad.net/bugs/1867983
>> Signed-off-by: Kai-Heng Feng <kai.heng.feng@xxxxxxxxxxxxx>
>> ---
>> drivers/ata/libata-core.c | 6 +++---
>> include/linux/libata.h | 3 +++
>> 2 files changed, 6 insertions(+), 3 deletions(-)
>> diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
>> index beca5f91bb4c..4a698f6388cd 100644
>> --- a/drivers/ata/libata-core.c
>> +++ b/drivers/ata/libata-core.c
>> @@ -42,7 +42,6 @@
>> #include <linux/workqueue.h>
>> #include <linux/scatterlist.h>
>> #include <linux/io.h>
>> -#include <linux/async.h>
>> #include <linux/log2.h>
>> #include <linux/slab.h>
>> #include <linux/glob.h>
>> @@ -5778,7 +5777,7 @@ int ata_host_register(struct ata_host *host, struct scsi_host_template *sht)
>> /* perform each probe asynchronously */
>> for (i = 0; i < host->n_ports; i++) {
>> struct ata_port *ap = host->ports[i];
>> - async_schedule(async_port_probe, ap);
>> + ap->cookie = async_schedule(async_port_probe, ap);
>> }
>> return 0;
>> @@ -5921,7 +5920,8 @@ void ata_host_detach(struct ata_host *host)
>> int i;
>> /* Ensure ata_port probe has completed */
>> - async_synchronize_full();
>> + for (i = 0; i < host->n_ports; i++)
>> + async_synchronize_cookie(host->ports[i]->cookie);
>> for (i = 0; i < host->n_ports; i++)
>
> Is it possible to combine these "for" loops?
>
>> ata_port_detach(host->ports[i]);
>> diff --git a/include/linux/libata.h b/include/linux/libata.h
>> index cffa4714bfa8..ae6dfc107ea8 100644
>> --- a/include/linux/libata.h
>> +++ b/include/linux/libata.h
>> @@ -22,6 +22,7 @@
>> #include <linux/acpi.h>
>> #include <linux/cdrom.h>
>> #include <linux/sched.h>
>> +#include <linux/async.h>
>>
>
> alphabetic?
>
>> /*
>> * Define if arch has non-standard setup. This is a _PCI_ standard
>> @@ -872,6 +873,8 @@ struct ata_port {
>> struct timer_list fastdrain_timer;
>> unsigned long fastdrain_cnt;
>> + async_cookie_t cookie;
>> +
>> int em_message_type;
>> void *private_data;