Re: [PATCH] zram: Using GFP_ATOMIC instead of GFP_KERNEL to allocate bitmap memory in backing_dev_store

[Dongyun Liu]

On 2023/12/3 9:23, Sergey Senozhatsky wrote:
> On (23/12/02 23:50), Dongyun Liu wrote:
> > On 2023/12/1 23:39, Sergey Senozhatsky wrote:
> > > On (23/11/30 23:20), Dongyun Liu wrote:
> > > > INFO: task init:331 blocked for more than 120 seconds.  "echo 0 >
> > > > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> > > > task:init   state:D stack:    0 pid:    1 ppid:     0 flags:0x04000000
> > > > Call trace:
> > > >     __switch_to+0x244/0x4e4
> > > >     __schedule+0x5bc/0xc48
> > > >     schedule+0x80/0x164
> > > >     rwsem_down_read_slowpath+0x4fc/0xf9c
> > > >     __down_read+0x140/0x188
> > > >     down_read+0x14/0x24
> > > >     try_wakeup_wbd_thread+0x78/0x1ec [zram]
> > > >     __zram_bvec_write+0x720/0x878 [zram]
> > > >     zram_bvec_rw+0xa8/0x234 [zram]
> > > >     zram_submit_bio+0x16c/0x268 [zram]
> > > >     submit_bio_noacct+0x128/0x3c8
> > > >     submit_bio+0x1cc/0x3d0
> > > >     __swap_writepage+0x5c4/0xd4c
> > > >     swap_writepage+0x130/0x158
> > > >     pageout+0x1f4/0x478
> > > >     shrink_page_list+0x9b4/0x1eb8
> > > >     shrink_inactive_list+0x2f4/0xaa8
> > > >     shrink_lruvec+0x184/0x340
> > > >     shrink_node_memcgs+0x84/0x3a0
> > > >     shrink_node+0x2c4/0x6c4
> > > >     shrink_zones+0x16c/0x29c
> > > >     do_try_to_free_pages+0xe4/0x2b4
> > > >     try_to_free_pages+0x388/0x7b4
> > > >     __alloc_pages_direct_reclaim+0x88/0x278
> > > >     __alloc_pages_slowpath+0x4ec/0xf6c
> > > >     __alloc_pages_nodemask+0x1f4/0x3dc
> > > >     kmalloc_order+0x54/0x338
> > > >     kmalloc_order_trace+0x34/0x1bc
> > > >     __kmalloc+0x5e8/0x9c0
> > > >     kvmalloc_node+0xa8/0x264
> > > >     backing_dev_store+0x1a4/0x818 [zram]
> > > >     dev_attr_store+0x38/0x8c
> > > >     sysfs_kf_write+0x64/0xc4
> > >
> > > Hmm, I'm not really following this backtrace. Backing device
> > > configuration is only possible on un-initialized zram device.
> > > If it's uninitialized, then why is it being used for swapout
> > > later in the call stack?
> >
> > Uh, at this moment, zram has finished initializing and is
> > working. The backing device is an optional zram-based feature.
> > I think it can be created later.
>
> backing_dev_store() can't be called on an initialized device,
> that's what init_done() at the very beginning of backing_dev_store()
> is supposed to ensure:

We have modified the logic here to allow backing_dev_store() to be 
called multiple times. That's why the call trace comes in. On the other 
hand, I realized that I should show the upstream's call stack instead
of my custom feature next time. Thank you for your toleration.

> ...
> 	down_write(&zram->init_lock);
> 	if (init_done(zram)) {
> 		pr_info("Can't setup backing device for initialized device\n");
> 		err = -EBUSY;
> 		goto out;
> 	}
> ...