lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d78a1704-31dc-4eaa-8189-e3aba51d3fe2@kernel.dk>
Date: Mon, 6 Oct 2025 06:48:52 -0600
From: Jens Axboe <axboe@...nel.dk>
To: John Paul Adrian Glaubitz <glaubitz@...sik.fu-berlin.de>,
 linux-block@...r.kernel.org, linux-kernel@...r.kernel.org
Cc: Andreas Larsson <andreas@...sler.com>,
 Anthony Yznaga <anthony.yznaga@...cle.com>, Sam James <sam@...too.org>,
 "David S . Miller" <davem@...emloft.net>,
 Michael Karcher <kernel@...rcher.dialup.fu-berlin.de>,
 sparclinux@...r.kernel.org
Subject: Re: [PATCH v2] Revert "sunvdc: Do not spin in an infinite loop when
 vio_ldc_send() returns EAGAIN"

On 10/6/25 4:02 AM, John Paul Adrian Glaubitz wrote:
> In a11f6ca9aef9 ("sunvdc: Do not spin in an infinite loop when vio_ldc_send()
> returns EAGAIN"), a maximum retry count was added to __vdc_tx_trigger().
> 
> After this change, several users reported disk I/O errors when running Linux
> inside a logical domain on Solaris 11.4:
> 
> [19095.192532] sunvdc: vdc_tx_trigger() failure, err=-11
> [19095.192605] I/O error, dev vdiskc, sector 368208928 op 0x1:(WRITE) flags 0x1000 phys_seg 2 prio class 2
> [19095.205681] XFS (vdiskc1): metadata I/O error in "xfs_buf_ioend+0x28c/0x600 [xfs]" at daddr 0x15f26420 len 32 error 5
> [19432.043471] sunvdc: vdc_tx_trigger() failure, err=-11
> [19432.043529] I/O error, dev vdiskc, sector 3732568 op 0x1:(WRITE) flags 0x1000 phys_seg 1 prio class 2
> [19432.058821] sunvdc: vdc_tx_trigger() failure, err=-11
> [19432.058843] I/O error, dev vdiskc, sector 3736256 op 0x1:(WRITE) flags 0x1000 phys_seg 4 prio class 2
> [19432.074109] sunvdc: vdc_tx_trigger() failure, err=-11
> [19432.074128] I/O error, dev vdiskc, sector 3736512 op 0x1:(WRITE) flags 0x1000 phys_seg 4 prio class 2
> [19432.089425] sunvdc: vdc_tx_trigger() failure, err=-11
> [19432.089443] I/O error, dev vdiskc, sector 3737024 op 0x1:(WRITE) flags 0x1000 phys_seg 1 prio class 2
> [19432.100964] XFS (vdiskc1): metadata I/O error in "xfs_buf_ioend+0x28c/0x600 [xfs]" at daddr 0x38ec58 len 8 error 5
> 
> Since this change seems to have only been justified by reading the code which
> becomes evident by the reference to adddc32d6fde ("sunvnet: Do not spin in an
> infinite loop when vio_ldc_send() returns EAGAIN") in the commit message, it
> can be safely assumed that the change was neither properly tested nor motivated
> by any actual bug reports.
> 
> Thus, let's revert this change to address the disk I/O errors above.
> 
> Signed-off-by: John Paul Adrian Glaubitz <glaubitz@...sik.fu-berlin.de>
> ---
> Changes since v1:
> - Rephrase commit message
> ---
>  drivers/block/sunvdc.c | 5 -----
>  1 file changed, 5 deletions(-)
> 
> diff --git a/drivers/block/sunvdc.c b/drivers/block/sunvdc.c
> index 282f81616a78..f56023c2b033 100644
> --- a/drivers/block/sunvdc.c
> +++ b/drivers/block/sunvdc.c
> @@ -45,8 +45,6 @@ MODULE_VERSION(DRV_MODULE_VERSION);
>  #define WAITING_FOR_GEN_CMD	0x04
>  #define WAITING_FOR_ANY		-1
>  
> -#define	VDC_MAX_RETRIES	10
> -
>  static struct workqueue_struct *sunvdc_wq;
>  
>  struct vdc_req_entry {
> @@ -437,7 +435,6 @@ static int __vdc_tx_trigger(struct vdc_port *port)
>  		.end_idx		= dr->prod,
>  	};
>  	int err, delay;
> -	int retries = 0;
>  
>  	hdr.seq = dr->snd_nxt;
>  	delay = 1;
> @@ -450,8 +447,6 @@ static int __vdc_tx_trigger(struct vdc_port *port)
>  		udelay(delay);
>  		if ((delay <<= 1) > 128)
>  			delay = 128;
> -		if (retries++ > VDC_MAX_RETRIES)
> -			break;
>  	} while (err == -EAGAIN);
>  
>  	if (err == -ENOTCONN)

When you apply this patch and things work, how many times does it
generally spin where it would've failed before? It's a bit unnerving to
have a never ending spin loop for this, with udelay spinning in between
as well. Looking at vio_ldc_send() as well, that spins for potentially
1000 loops of 1usec each, which would be 1ms. With the current limit of
10 retries, the driver would end up doing udelays of:

1
2
4
8
16
32
64
128
128
128

which is 511 usec on top, for 10.5ms in total spinning time before
giving up. That is kind of mind boggling, that's an eternity.

Not that it's _really_ that important as this is a pretty niche driver,
but still pretty ugly... Does it work reliably with a limit of 100
spins? If things get truly stuck, spinning forever in that loop is not
going to help. In any case, your patch should have

Cc: stable@...r.kernel.org
Fixes: a11f6ca9aef9 ("sunvdc: Do not spin in an infinite loop when vio_ldc_send() returns EAGAIN")

tags added.

-- 
Jens Axboe

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ