| Message ID | 20240509164905.41016-1-bcfradella@proton.me (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
| Series | p2sb: Don't init until unassigned resources have been assigned. | expand |
[cc += Shin'ichiro, Klara, Andy, Danil] On Thu, May 09, 2024 at 04:49:34PM +0000, bcfradella@proton.me wrote: > From: Ben Fradella <bfradell@netapp.com> > > The P2SB could get an invalid BAR from the BIOS, and that won't be fixed > up until pcibios_assign_resources(), which is an fs_initcall(). > > - Move p2sb_fs_init() to an fs_initcall_sync(). This is still early > enough to avoid a race with any dependent drivers. > > - Add a check for IORESOURCE_UNSET in p2sb_valid_resource() to catch > unset BARs going forward. > > - Return error values from p2sb_fs_init() so that the 'initcall_debug' > cmdline arg provides useful data. > > Signed-off-by: Ben Fradella <bfradell@netapp.com> > --- > drivers/platform/x86/p2sb.c | 29 +++++++++++++++-------------- > 1 file changed, 15 insertions(+), 14 deletions(-) > > diff --git a/drivers/platform/x86/p2sb.c b/drivers/platform/x86/p2sb.c > index 3d66e1d4eb1f..1938a3ef9480 100644 > --- a/drivers/platform/x86/p2sb.c > +++ b/drivers/platform/x86/p2sb.c > @@ -56,12 +56,9 @@ static int p2sb_get_devfn(unsigned int *devfn) > return 0; > } > > -static bool p2sb_valid_resource(struct resource *res) > +static bool p2sb_valid_resource(const struct resource *res) > { > - if (res->flags) > - return true; > - > - return false; > + return res->flags & ~IORESOURCE_UNSET; > } > > /* Copy resource from the first BAR of the device in question */ > @@ -220,16 +217,20 @@ EXPORT_SYMBOL_GPL(p2sb_bar); > > static int __init p2sb_fs_init(void) > { > - p2sb_cache_resources(); > - return 0; > + return p2sb_cache_resources(); > } > > /* > - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can > - * not be locked in sysfs pci bus rescan path because of deadlock. To > - * avoid the deadlock, access to P2SB devices with the lock at an early > - * step in kernel initialization and cache required resources. This > - * should happen after subsys_initcall which initializes PCI subsystem > - * and before device_initcall which requires P2SB resources. > + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path > + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock > + * at an early step in kernel initialization and cache required resources. > + * > + * We want to run as early as possible. If the P2SB was assigned a bad BAR, > + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of > + * initcall dependencies looks something like this: > + * > + * ... > + * subsys_initcall (pci_subsys_init) > + * fs_initcall (pcibios_assign_resources) > */ > -fs_initcall(p2sb_fs_init); > +fs_initcall_sync(p2sb_fs_init); > -- > 2.43.0
On Thu, May 09, 2024 at 07:04:32PM +0200, Lukas Wunner wrote: > [cc += Shin'ichiro, Klara, Andy, Danil] Thank you! > On Thu, May 09, 2024 at 04:49:34PM +0000, bcfradella@proton.me wrote: > > From: Ben Fradella <bfradell@netapp.com> > > > > The P2SB could get an invalid BAR from the BIOS, and that won't be fixed > > up until pcibios_assign_resources(), which is an fs_initcall(). > > > > - Move p2sb_fs_init() to an fs_initcall_sync(). This is still early > > enough to avoid a race with any dependent drivers. > > > > - Add a check for IORESOURCE_UNSET in p2sb_valid_resource() to catch > > unset BARs going forward. > > > > - Return error values from p2sb_fs_init() so that the 'initcall_debug' > > cmdline arg provides useful data. ... > > /* > > - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can > > - * not be locked in sysfs pci bus rescan path because of deadlock. To > > - * avoid the deadlock, access to P2SB devices with the lock at an early > > - * step in kernel initialization and cache required resources. This > > - * should happen after subsys_initcall which initializes PCI subsystem > > - * and before device_initcall which requires P2SB resources. > > + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path PCI bus > > + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock > > + * at an early step in kernel initialization and cache required resources. > > + * > > + * We want to run as early as possible. If the P2SB was assigned a bad BAR, > > + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of > > + * initcall dependencies looks something like this: > > + * > > + * ... > > + * subsys_initcall (pci_subsys_init) > > + * fs_initcall (pcibios_assign_resources) > > */ This makes sense to me Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> I assume others will conduct the proper review and testing.
On 2024-05-10 16:50, Andy Shevchenko wrote: > On Thu, May 09, 2024 at 07:04:32PM +0200, Lukas Wunner wrote: >> [cc += Shin'ichiro, Klara, Andy, Danil] > > Thank you! > >> On Thu, May 09, 2024 at 04:49:34PM +0000, bcfradella@proton.me wrote: >>> From: Ben Fradella <bfradell@netapp.com> >>> >>> The P2SB could get an invalid BAR from the BIOS, and that won't be fixed >>> up until pcibios_assign_resources(), which is an fs_initcall(). >>> >>> - Move p2sb_fs_init() to an fs_initcall_sync(). This is still early >>> enough to avoid a race with any dependent drivers. >>> >>> - Add a check for IORESOURCE_UNSET in p2sb_valid_resource() to catch >>> unset BARs going forward. >>> >>> - Return error values from p2sb_fs_init() so that the 'initcall_debug' >>> cmdline arg provides useful data. > > ... > >>> /* >>> - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can >>> - * not be locked in sysfs pci bus rescan path because of deadlock. To >>> - * avoid the deadlock, access to P2SB devices with the lock at an early >>> - * step in kernel initialization and cache required resources. This >>> - * should happen after subsys_initcall which initializes PCI subsystem >>> - * and before device_initcall which requires P2SB resources. >>> + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path > > PCI bus > >>> + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock >>> + * at an early step in kernel initialization and cache required resources. >>> + * >>> + * We want to run as early as possible. If the P2SB was assigned a bad BAR, >>> + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of >>> + * initcall dependencies looks something like this: >>> + * >>> + * ... >>> + * subsys_initcall (pci_subsys_init) >>> + * fs_initcall (pcibios_assign_resources) >>> */ > > This makes sense to me > Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> > > I assume others will conduct the proper review and testing. > This patch did not introduce any new issues on my previously problematic system. Tested-by: Klara Modin <klarasmodin@gmail.com>
On May 10, 2024 / 21:22, Klara Modin wrote: > On 2024-05-10 16:50, Andy Shevchenko wrote: > > On Thu, May 09, 2024 at 07:04:32PM +0200, Lukas Wunner wrote: > > > [cc += Shin'ichiro, Klara, Andy, Danil] > > > > Thank you! > > > > > On Thu, May 09, 2024 at 04:49:34PM +0000, bcfradella@proton.me wrote: > > > > From: Ben Fradella <bfradell@netapp.com> > > > > > > > > The P2SB could get an invalid BAR from the BIOS, and that won't be fixed > > > > up until pcibios_assign_resources(), which is an fs_initcall(). > > > > > > > > - Move p2sb_fs_init() to an fs_initcall_sync(). This is still early > > > > enough to avoid a race with any dependent drivers. > > > > > > > > - Add a check for IORESOURCE_UNSET in p2sb_valid_resource() to catch > > > > unset BARs going forward. > > > > > > > > - Return error values from p2sb_fs_init() so that the 'initcall_debug' > > > > cmdline arg provides useful data. > > > > ... > > > > > > /* > > > > - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can > > > > - * not be locked in sysfs pci bus rescan path because of deadlock. To > > > > - * avoid the deadlock, access to P2SB devices with the lock at an early > > > > - * step in kernel initialization and cache required resources. This > > > > - * should happen after subsys_initcall which initializes PCI subsystem > > > > - * and before device_initcall which requires P2SB resources. > > > > + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path > > > > PCI bus > > > > > > + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock > > > > + * at an early step in kernel initialization and cache required resources. > > > > + * > > > > + * We want to run as early as possible. If the P2SB was assigned a bad BAR, > > > > + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of > > > > + * initcall dependencies looks something like this: > > > > + * > > > > + * ... > > > > + * subsys_initcall (pci_subsys_init) > > > > + * fs_initcall (pcibios_assign_resources) > > > > */ > > > > This makes sense to me > > Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> > > > > I assume others will conduct the proper review and testing. > > > > This patch did not introduce any new issues on my previously problematic > system. > > Tested-by: Klara Modin <klarasmodin@gmail.com> The changes look reasonable and good to me. I also confirmed that the patch does not affect on my use case using two my test machines. Reviewed-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Hi, On 5/9/24 6:49 PM, bcfradella@proton.me wrote: > From: Ben Fradella <bfradell@netapp.com> > > The P2SB could get an invalid BAR from the BIOS, and that won't be fixed > up until pcibios_assign_resources(), which is an fs_initcall(). > > - Move p2sb_fs_init() to an fs_initcall_sync(). This is still early > enough to avoid a race with any dependent drivers. > > - Add a check for IORESOURCE_UNSET in p2sb_valid_resource() to catch > unset BARs going forward. > > - Return error values from p2sb_fs_init() so that the 'initcall_debug' > cmdline arg provides useful data. > > Signed-off-by: Ben Fradella <bfradell@netapp.com> Thank you for your patch, I've applied this patch to my review-hans branch: https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/log/?h=review-hans Note it will show up in my review-hans branch once I've pushed my local branch there, which might take a while. Once I've run some tests on this branch the patches there will be added to the platform-drivers-x86/for-next branch and eventually will be included in the pdx86 pull-request to Linus for the next merge-window. Regards, Hans > --- > drivers/platform/x86/p2sb.c | 29 +++++++++++++++-------------- > 1 file changed, 15 insertions(+), 14 deletions(-) > > diff --git a/drivers/platform/x86/p2sb.c b/drivers/platform/x86/p2sb.c > index 3d66e1d4eb1f..1938a3ef9480 100644 > --- a/drivers/platform/x86/p2sb.c > +++ b/drivers/platform/x86/p2sb.c > @@ -56,12 +56,9 @@ static int p2sb_get_devfn(unsigned int *devfn) > return 0; > } > > -static bool p2sb_valid_resource(struct resource *res) > +static bool p2sb_valid_resource(const struct resource *res) > { > - if (res->flags) > - return true; > - > - return false; > + return res->flags & ~IORESOURCE_UNSET; > } > > /* Copy resource from the first BAR of the device in question */ > @@ -220,16 +217,20 @@ EXPORT_SYMBOL_GPL(p2sb_bar); > > static int __init p2sb_fs_init(void) > { > - p2sb_cache_resources(); > - return 0; > + return p2sb_cache_resources(); > } > > /* > - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can > - * not be locked in sysfs pci bus rescan path because of deadlock. To > - * avoid the deadlock, access to P2SB devices with the lock at an early > - * step in kernel initialization and cache required resources. This > - * should happen after subsys_initcall which initializes PCI subsystem > - * and before device_initcall which requires P2SB resources. > + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path > + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock > + * at an early step in kernel initialization and cache required resources. > + * > + * We want to run as early as possible. If the P2SB was assigned a bad BAR, > + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of > + * initcall dependencies looks something like this: > + * > + * ... > + * subsys_initcall (pci_subsys_init) > + * fs_initcall (pcibios_assign_resources) > */ > -fs_initcall(p2sb_fs_init); > +fs_initcall_sync(p2sb_fs_init);
> On 2024-05-13 07:03, hdegoede@redhat.com wrote: > > Thank you for your patch, I've applied this patch to my review-hans branch: https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/log/?h=review-hans > > Note it will show up in my review-hans branch once I've pushed my local branch there, which might take a while. > > Once I've run some tests on this branch the patches there will be added to the platform-drivers-x86/for-next branch and eventually will be included in the pdx86 pull-request to Linus for the next merge-window. > >Regards, > >Hans Thank you!
diff --git a/drivers/platform/x86/p2sb.c b/drivers/platform/x86/p2sb.c index 3d66e1d4eb1f..1938a3ef9480 100644 --- a/drivers/platform/x86/p2sb.c +++ b/drivers/platform/x86/p2sb.c @@ -56,12 +56,9 @@ static int p2sb_get_devfn(unsigned int *devfn) return 0; } -static bool p2sb_valid_resource(struct resource *res) +static bool p2sb_valid_resource(const struct resource *res) { - if (res->flags) - return true; - - return false; + return res->flags & ~IORESOURCE_UNSET; } /* Copy resource from the first BAR of the device in question */ @@ -220,16 +217,20 @@ EXPORT_SYMBOL_GPL(p2sb_bar); static int __init p2sb_fs_init(void) { - p2sb_cache_resources(); - return 0; + return p2sb_cache_resources(); } /* - * pci_rescan_remove_lock to avoid access to unhidden P2SB devices can - * not be locked in sysfs pci bus rescan path because of deadlock. To - * avoid the deadlock, access to P2SB devices with the lock at an early - * step in kernel initialization and cache required resources. This - * should happen after subsys_initcall which initializes PCI subsystem - * and before device_initcall which requires P2SB resources. + * pci_rescan_remove_lock() can not be locked in sysfs pci bus rescan path + * because of deadlock. To avoid the deadlock, access P2SB devices with the lock + * at an early step in kernel initialization and cache required resources. + * + * We want to run as early as possible. If the P2SB was assigned a bad BAR, + * we'll need to wait on pcibios_assign_resources() to fix it. So, our list of + * initcall dependencies looks something like this: + * + * ... + * subsys_initcall (pci_subsys_init) + * fs_initcall (pcibios_assign_resources) */ -fs_initcall(p2sb_fs_init); +fs_initcall_sync(p2sb_fs_init);