NetBSD Problem Report #58027

From www@netbsd.org  Sun Mar 10 09:57:33 2024
Return-Path: <www@netbsd.org>
Received: from mail.netbsd.org (mail.netbsd.org [199.233.217.200])
	(using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits))
	(Client CN "mail.NetBSD.org", Issuer "mail.NetBSD.org CA" (not verified))
	by mollari.NetBSD.org (Postfix) with ESMTPS id 99B4F1A923C
	for <gnats-bugs@gnats.NetBSD.org>; Sun, 10 Mar 2024 09:57:33 +0000 (UTC)
Message-Id: <20240310095337.317EF1A923F@mollari.NetBSD.org>
Date: Sun, 10 Mar 2024 09:53:37 +0000 (UTC)
From: mrg@eterna23.net
Reply-To: mrg@eterna23.net
To: gnats-bugs@NetBSD.org
Subject: ehci failed attach stops suspend/resume
X-Send-Pr-Version: www-1.0

>Number:         58027
>Category:       kern
>Synopsis:       ehci failed attach stops suspend/resume
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Mar 10 10:00:01 +0000 2024
>Last-Modified:  Mon Mar 11 11:00:01 +0000 2024
>Originator:     matthew green
>Release:        -current
>Organization:
>Environment:
amd64
>Description:
if something in the ehci_pci attach fails, it never registers any pmf
handlers, so the system is unable to suspend.
>How-To-Repeat:
try to suspend a system that has a:

ehci0 at pci1 dev 0 function 4: Realtek Semiconductor product 816d (rev. 0x0e)
ehci0: 32-bit DMA
ehci0: interrupting at msix2 vec 0
ehci0: pre-2.0 USB rev, device ignored

>Fix:
https://www.netbsd.org/~mrg/ehci-detach.v4.diff

>Audit-Trail:
From: Taylor R Campbell <riastradh@NetBSD.org>
To: mrg@NetBSD.org
Cc: gnats-bugs@NetBSD.org, netbsd-bugs@NetBSD.org
Subject: Re: kern/58027: ehci failed attach stops suspend/resume
Date: Mon, 11 Mar 2024 10:57:27 +0000

 >  	if (sc->sc_ih) {
 >  		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
 > +		pci_intr_release(sc->sc_pc, sc->sc_pihp, 1);
 >  		sc->sc_ih = NULL;
 >  	}

 This part seems wrong -- I think it should be a separate branch:

  	if (sc->sc_ih) {
  		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
  		sc->sc_ih = NULL;
  	}
 	if (sc->sc_pihp) {
 		pci_intr_release(sc->sc_pc, sc->sc_pihp, 1);
 		sc->sc_pihp = NULL;
 	}

 and then we can delete some more redundant code in ehci_pci_attach's
 pci_intr_establish_xname error branch.

 > @@ -262,6 +263,7 @@ ehci_pci_attach(device_t parent, device_
 >  	int err = ehci_init(&sc->sc);
 >  	if (err) {
 >  		aprint_error_dev(self, "init failed, error=%d\n", err);
 > +		ehci_release_ownership(&sc->sc, sc->sc_pc, sc->sc_tag);
 >  		goto fail;
 >  	}
 > ...
 >  	if (sc->sc.sc_size) {
 > -		ehci_release_ownership(&sc->sc, sc->sc_pc, sc->sc_tag);
 >  		bus_space_unmap(sc->sc.iot, sc->sc.ioh, sc->sc.sc_size);
 >  		sc->sc.sc_size = 0;
 >  	}
 > ...
 > @@ -302,27 +317,17 @@ ehci_pci_detach(device_t self, int flags
 > ...
 > -	if (sc->sc.sc_size) {
 >  		ehci_release_ownership(&sc->sc, sc->sc_pc, sc->sc_tag);
 > -		bus_space_unmap(sc->sc.iot, sc->sc.ioh, sc->sc.sc_size);
 > -		sc->sc.sc_size = 0;
 >  	}

 Why did the ehci_release_ownership part move?  I think we should
 generally have less logic in the `if (err) goto fail' branches
 themselves, and more logic shared in the common cleanup path.

Home	PR Database Search