linux-kernel - Re: [PATCH] USB: serial: mos7720: defer state restore to a workqueue

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]

Date:   Mon, 16 Nov 2020 18:09:18 +0100
From:   Johan Hovold <johan@...nel.org>
To:     Davidlohr Bueso <dave@...olabs.net>
Cc:     Johan Hovold <johan@...nel.org>, linux-usb@...r.kernel.org,
        linux-kernel@...r.kernel.org, Davidlohr Bueso <dbueso@...e.de>
Subject: Re: [PATCH] USB: serial: mos7720: defer state restore to a workqueue

On Fri, Nov 13, 2020 at 08:27:25PM -0800, Davidlohr Bueso wrote:
> The parallel port restore operation currently defers writes
> to a tasklet, if it sees a locked disconnect mutex. The
> driver goes to a lot of trouble to ensure writes happen
> in a non-blocking context, but things can be greatly
> simplified if it's done in regular process context and
> this is not a system performance critical path. As such,
> instead of doing the async state restore writes in irq
> context, use a workqueue and just do regular synchronous
> writes.
> 
> In addition to the cleanup, this also imposes less on the
> overall system as tasklets have been deprecated because
> of it's BH implications, potentially blocking a higher
> priority task from running. We also get rid of hacks
> such as trylocking a mutex in irq, something which does
> not play nice with priority boosting in PREEMPT_RT.
> 
> Signed-off-by: Davidlohr Bueso <dbueso@...e.de>

> -/*
> - * This is the the common top part of all parallel port callback operations that
> + * This is the common top part of all parallel port callback operations that
>   * send synchronous messages to the device.  This implements convoluted locking
>   * that avoids two scenarios: (1) a port operation is called after usbserial
>   * has called our release function, at which point struct mos7715_parport has

This is an unrelated change, but ok.

> @@ -641,10 +488,9 @@ static void parport_mos7715_restore_state(struct parport *pp,
> 	}
> 	mos_parport->shadowDCR = s->u.pc.ctr;
> 	mos_parport->shadowECR = s->u.pc.ecr;
> -	write_parport_reg_nonblock(mos_parport, MOS7720_DCR,
> -				   mos_parport->shadowDCR);
> -	write_parport_reg_nonblock(mos_parport, MOS7720_ECR,
> -				   mos_parport->shadowECR);
> +
> +	/* defer synchronous writes outside of irq */

This one isn't called in interrupt context, but with interrupts
disabled and a spin lock held.

> +	schedule_work(&mos_parport->work);
> 	spin_unlock(&release_lock);
>  }

> @@ -1869,8 +1712,6 @@ static void mos7720_release(struct usb_serial *serial)
> 
> 	if (le16_to_cpu(serial->dev->descriptor.idProduct)
> 	    == MOSCHIP_DEVICE_ID_7715) {
> -		struct urbtracker *urbtrack;
> -		unsigned long flags;
> 		struct mos7715_parport *mos_parport =
> 			usb_get_serial_data(serial);
> 
> @@ -1883,21 +1724,17 @@ static void mos7720_release(struct usb_serial *serial)
> 		if (mos_parport->msg_pending)
> 			wait_for_completion_timeout(&mos_parport->syncmsg_compl,
> 					    msecs_to_jiffies(MOS_WDR_TIMEOUT));
> +		/*
> +		 * If delayed work is currently scheduled, wait for it to
> +		 * complete. This also implies barriers that ensure the
> +		 * below serial clearing is not hoisted above the ->work.
> +		 */
> +		cancel_work_sync(&mos_parport->work);

As I mentioned, this needs to be done *after* deregistering the port or
you could theoretically end up with the work item being requeued.

Yes, the same applies for the "synchronous" requests, but that's a
preexisting issue.

> 		parport_remove_port(mos_parport->pp);
> 		usb_set_serial_data(serial, NULL);
> 		mos_parport->serial = NULL;

Johan