Message ID | alpine.LRH.2.02.2005061122440.16395@file01.intranet.prod.int.rdu2.redhat.com |
---|---|
State | Not Applicable |
Headers | show |
Series | None | expand |
On Wed, May 06, 2020 at 11:29:29AM -0400, Mikulas Patocka wrote: > > > On Wed, 6 May 2020, Greg Kroah-Hartman wrote: > > > > Index: linux-stable/drivers/tty/serial/8250/8250_port.c > > > =================================================================== > > > --- linux-stable.orig/drivers/tty/serial/8250/8250_port.c 2020-05-06 08:25:19.000000000 +0200 > > > +++ linux-stable/drivers/tty/serial/8250/8250_port.c 2020-05-06 09:04:17.000000000 +0200 > > > @@ -442,6 +442,10 @@ static unsigned int mem32be_serial_in(st > > > > > > static unsigned int io_serial_in(struct uart_port *p, int offset) > > > { > > > +#ifdef CONFIG_ALPHA > > > +/* we need a small delay, the Alpha Avanti chipset locks up with back-to-back accesses */ > > > + ndelay(300); > > > +#endif > > > > We really do not like #ifdef in .c files, especially ones that cause a > > coding style violation :) > > > > Why can't you do this as a quirk for this specific chipset? You should > > tie it to the serial port hardware type, not to the CPU type. > > > > thanks, > > > > greg k-h > > Do you want this patch? It enables the delay based on the specific PCI-ISA > bridge. > > There is still "#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI)" because > if we want to reference a variable defined by the PCI subsystem in the > arch/alpha tree, we must do it conditionally. If you want to get rid of > these #ifs, please describe how. > > Mikulas > > > From: Mikulas Patocka <mpatocka@redhat.com> > > The patch 92d7223a74235054f2aa7227d207d9c57f84dca0 ("alpha: io: reorder > barriers to guarantee writeX() and iowriteX() ordering #2") broke boot on > the Alpha Avanti platform. > > The patch changes timing between accesses to the ISA bus, in particular, > it reduces the time between "write" access and a subsequent "read" access. > > This causes lock-up when accessing the real time clock and serial ports. > > This patch fixes the serial ports by adding a small delay before the "inb" > instruction. > > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> > Fixes: 92d7223a7423 ("alpha: io: reorder barriers to guarantee writeX() and iowriteX() ordering #2") > Cc: stable@vger.kernel.org # v4.17+ > > --- > arch/alpha/include/asm/pci.h | 2 ++ > arch/alpha/kernel/pci.c | 4 ++++ > drivers/tty/serial/8250/8250_core.c | 7 +++++++ > drivers/tty/serial/8250/8250_port.c | 3 +++ > include/linux/serial_core.h | 1 + > 5 files changed, 17 insertions(+) > > Index: linux-stable/include/linux/serial_core.h > =================================================================== > --- linux-stable.orig/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > +++ linux-stable/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > @@ -154,6 +154,7 @@ struct uart_port { > > /* quirks must be updated while holding port mutex */ > #define UPQ_NO_TXEN_TEST BIT(0) > +#define UPQ_DELAY_BEFORE_READ BIT(1) > > unsigned int read_status_mask; /* driver specific */ > unsigned int ignore_status_mask; /* driver specific */ > Index: linux-stable/drivers/tty/serial/8250/8250_core.c > =================================================================== > --- linux-stable.orig/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > +++ linux-stable/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > @@ -37,6 +37,9 @@ > #ifdef CONFIG_SPARC > #include <linux/sunserialcore.h> > #endif > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > +#include <linux/pci.h> > +#endif Don't need the #if for this. > > #include <asm/irq.h> > > @@ -490,6 +493,10 @@ static void univ8250_rsa_support(struct > static inline void serial8250_apply_quirks(struct uart_8250_port *up) > { > up->port.quirks |= skip_txen_test ? UPQ_NO_TXEN_TEST : 0; > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > + if (alpha_serial_port_needs_delay) > + up->port.quirks |= UPQ_DELAY_BEFORE_READ; > +#endif Why is a #define needed here? You can do this same type of change without any #ifdefs in any .c files and that would be much nicer. thanks, greg k-h
On Wed, 6 May 2020, Greg Kroah-Hartman wrote: > On Wed, May 06, 2020 at 11:29:29AM -0400, Mikulas Patocka wrote: > > > > > > On Wed, 6 May 2020, Greg Kroah-Hartman wrote: > > > > > > Index: linux-stable/drivers/tty/serial/8250/8250_port.c > > > > =================================================================== > > > > --- linux-stable.orig/drivers/tty/serial/8250/8250_port.c 2020-05-06 08:25:19.000000000 +0200 > > > > +++ linux-stable/drivers/tty/serial/8250/8250_port.c 2020-05-06 09:04:17.000000000 +0200 > > > > @@ -442,6 +442,10 @@ static unsigned int mem32be_serial_in(st > > > > > > > > static unsigned int io_serial_in(struct uart_port *p, int offset) > > > > { > > > > +#ifdef CONFIG_ALPHA > > > > +/* we need a small delay, the Alpha Avanti chipset locks up with back-to-back accesses */ > > > > + ndelay(300); > > > > +#endif > > > > > > We really do not like #ifdef in .c files, especially ones that cause a > > > coding style violation :) > > > > > > Why can't you do this as a quirk for this specific chipset? You should > > > tie it to the serial port hardware type, not to the CPU type. > > > > > > thanks, > > > > > > greg k-h > > > > Do you want this patch? It enables the delay based on the specific PCI-ISA > > bridge. > > > > There is still "#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI)" because > > if we want to reference a variable defined by the PCI subsystem in the > > arch/alpha tree, we must do it conditionally. If you want to get rid of > > these #ifs, please describe how. > > > > Mikulas > > > > > > From: Mikulas Patocka <mpatocka@redhat.com> > > > > The patch 92d7223a74235054f2aa7227d207d9c57f84dca0 ("alpha: io: reorder > > barriers to guarantee writeX() and iowriteX() ordering #2") broke boot on > > the Alpha Avanti platform. > > > > The patch changes timing between accesses to the ISA bus, in particular, > > it reduces the time between "write" access and a subsequent "read" access. > > > > This causes lock-up when accessing the real time clock and serial ports. > > > > This patch fixes the serial ports by adding a small delay before the "inb" > > instruction. > > > > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> > > Fixes: 92d7223a7423 ("alpha: io: reorder barriers to guarantee writeX() and iowriteX() ordering #2") > > Cc: stable@vger.kernel.org # v4.17+ > > > > --- > > arch/alpha/include/asm/pci.h | 2 ++ > > arch/alpha/kernel/pci.c | 4 ++++ > > drivers/tty/serial/8250/8250_core.c | 7 +++++++ > > drivers/tty/serial/8250/8250_port.c | 3 +++ > > include/linux/serial_core.h | 1 + > > 5 files changed, 17 insertions(+) > > > > Index: linux-stable/include/linux/serial_core.h > > =================================================================== > > --- linux-stable.orig/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > > +++ linux-stable/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > > @@ -154,6 +154,7 @@ struct uart_port { > > > > /* quirks must be updated while holding port mutex */ > > #define UPQ_NO_TXEN_TEST BIT(0) > > +#define UPQ_DELAY_BEFORE_READ BIT(1) > > > > unsigned int read_status_mask; /* driver specific */ > > unsigned int ignore_status_mask; /* driver specific */ > > Index: linux-stable/drivers/tty/serial/8250/8250_core.c > > =================================================================== > > --- linux-stable.orig/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > > +++ linux-stable/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > > @@ -37,6 +37,9 @@ > > #ifdef CONFIG_SPARC > > #include <linux/sunserialcore.h> > > #endif > > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > > +#include <linux/pci.h> > > +#endif > > Don't need the #if for this. You're right. > > #include <asm/irq.h> > > > > @@ -490,6 +493,10 @@ static void univ8250_rsa_support(struct > > static inline void serial8250_apply_quirks(struct uart_8250_port *up) > > { > > up->port.quirks |= skip_txen_test ? UPQ_NO_TXEN_TEST : 0; > > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > > + if (alpha_serial_port_needs_delay) > > + up->port.quirks |= UPQ_DELAY_BEFORE_READ; > > +#endif > > Why is a #define needed here? > > You can do this same type of change without any #ifdefs in any .c files > and that would be much nicer. Because alpha_serial_port_needs_delay is defined only on Alpha - so we need to guard accesses to it with #ifdef CONFIG_ALPHA - otherwise the kernel wouldn't link on non-Alpha platforms. Should I make it defined for all architectures? Mikulas
On Wed, May 06, 2020 at 11:57:08AM -0400, Mikulas Patocka wrote: > > > On Wed, 6 May 2020, Greg Kroah-Hartman wrote: > > > On Wed, May 06, 2020 at 11:29:29AM -0400, Mikulas Patocka wrote: > > > > > > > > > On Wed, 6 May 2020, Greg Kroah-Hartman wrote: > > > > > > > > Index: linux-stable/drivers/tty/serial/8250/8250_port.c > > > > > =================================================================== > > > > > --- linux-stable.orig/drivers/tty/serial/8250/8250_port.c 2020-05-06 08:25:19.000000000 +0200 > > > > > +++ linux-stable/drivers/tty/serial/8250/8250_port.c 2020-05-06 09:04:17.000000000 +0200 > > > > > @@ -442,6 +442,10 @@ static unsigned int mem32be_serial_in(st > > > > > > > > > > static unsigned int io_serial_in(struct uart_port *p, int offset) > > > > > { > > > > > +#ifdef CONFIG_ALPHA > > > > > +/* we need a small delay, the Alpha Avanti chipset locks up with back-to-back accesses */ > > > > > + ndelay(300); > > > > > +#endif > > > > > > > > We really do not like #ifdef in .c files, especially ones that cause a > > > > coding style violation :) > > > > > > > > Why can't you do this as a quirk for this specific chipset? You should > > > > tie it to the serial port hardware type, not to the CPU type. > > > > > > > > thanks, > > > > > > > > greg k-h > > > > > > Do you want this patch? It enables the delay based on the specific PCI-ISA > > > bridge. > > > > > > There is still "#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI)" because > > > if we want to reference a variable defined by the PCI subsystem in the > > > arch/alpha tree, we must do it conditionally. If you want to get rid of > > > these #ifs, please describe how. > > > > > > Mikulas > > > > > > > > > From: Mikulas Patocka <mpatocka@redhat.com> > > > > > > The patch 92d7223a74235054f2aa7227d207d9c57f84dca0 ("alpha: io: reorder > > > barriers to guarantee writeX() and iowriteX() ordering #2") broke boot on > > > the Alpha Avanti platform. > > > > > > The patch changes timing between accesses to the ISA bus, in particular, > > > it reduces the time between "write" access and a subsequent "read" access. > > > > > > This causes lock-up when accessing the real time clock and serial ports. > > > > > > This patch fixes the serial ports by adding a small delay before the "inb" > > > instruction. > > > > > > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> > > > Fixes: 92d7223a7423 ("alpha: io: reorder barriers to guarantee writeX() and iowriteX() ordering #2") > > > Cc: stable@vger.kernel.org # v4.17+ > > > > > > --- > > > arch/alpha/include/asm/pci.h | 2 ++ > > > arch/alpha/kernel/pci.c | 4 ++++ > > > drivers/tty/serial/8250/8250_core.c | 7 +++++++ > > > drivers/tty/serial/8250/8250_port.c | 3 +++ > > > include/linux/serial_core.h | 1 + > > > 5 files changed, 17 insertions(+) > > > > > > Index: linux-stable/include/linux/serial_core.h > > > =================================================================== > > > --- linux-stable.orig/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > > > +++ linux-stable/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 > > > @@ -154,6 +154,7 @@ struct uart_port { > > > > > > /* quirks must be updated while holding port mutex */ > > > #define UPQ_NO_TXEN_TEST BIT(0) > > > +#define UPQ_DELAY_BEFORE_READ BIT(1) > > > > > > unsigned int read_status_mask; /* driver specific */ > > > unsigned int ignore_status_mask; /* driver specific */ > > > Index: linux-stable/drivers/tty/serial/8250/8250_core.c > > > =================================================================== > > > --- linux-stable.orig/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > > > +++ linux-stable/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 > > > @@ -37,6 +37,9 @@ > > > #ifdef CONFIG_SPARC > > > #include <linux/sunserialcore.h> > > > #endif > > > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > > > +#include <linux/pci.h> > > > +#endif > > > > Don't need the #if for this. > > You're right. > > > > #include <asm/irq.h> > > > > > > @@ -490,6 +493,10 @@ static void univ8250_rsa_support(struct > > > static inline void serial8250_apply_quirks(struct uart_8250_port *up) > > > { > > > up->port.quirks |= skip_txen_test ? UPQ_NO_TXEN_TEST : 0; > > > +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) > > > + if (alpha_serial_port_needs_delay) > > > + up->port.quirks |= UPQ_DELAY_BEFORE_READ; > > > +#endif > > > > Why is a #define needed here? > > > > You can do this same type of change without any #ifdefs in any .c files > > and that would be much nicer. > > Because alpha_serial_port_needs_delay is defined only on Alpha - so we > need to guard accesses to it with #ifdef CONFIG_ALPHA - otherwise the > kernel wouldn't link on non-Alpha platforms. > > Should I make it defined for all architectures? Yes, it's not the first time we have had to do things like this :) But, there is no other way to detect this based on hardware signatures/types instead? That is usually the best way to do it, right? thanks, greg k-h
Index: linux-stable/include/linux/serial_core.h =================================================================== --- linux-stable.orig/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 +++ linux-stable/include/linux/serial_core.h 2020-05-06 17:16:28.000000000 +0200 @@ -154,6 +154,7 @@ struct uart_port { /* quirks must be updated while holding port mutex */ #define UPQ_NO_TXEN_TEST BIT(0) +#define UPQ_DELAY_BEFORE_READ BIT(1) unsigned int read_status_mask; /* driver specific */ unsigned int ignore_status_mask; /* driver specific */ Index: linux-stable/drivers/tty/serial/8250/8250_core.c =================================================================== --- linux-stable.orig/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 +++ linux-stable/drivers/tty/serial/8250/8250_core.c 2020-05-06 17:16:28.000000000 +0200 @@ -37,6 +37,9 @@ #ifdef CONFIG_SPARC #include <linux/sunserialcore.h> #endif +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) +#include <linux/pci.h> +#endif #include <asm/irq.h> @@ -490,6 +493,10 @@ static void univ8250_rsa_support(struct static inline void serial8250_apply_quirks(struct uart_8250_port *up) { up->port.quirks |= skip_txen_test ? UPQ_NO_TXEN_TEST : 0; +#if defined(CONFIG_ALPHA) && defined(CONFIG_PCI) + if (alpha_serial_port_needs_delay) + up->port.quirks |= UPQ_DELAY_BEFORE_READ; +#endif } static void __init serial8250_isa_init_ports(void) Index: linux-stable/arch/alpha/include/asm/pci.h =================================================================== --- linux-stable.orig/arch/alpha/include/asm/pci.h 2020-05-06 17:16:28.000000000 +0200 +++ linux-stable/arch/alpha/include/asm/pci.h 2020-05-06 17:16:28.000000000 +0200 @@ -97,4 +97,6 @@ extern void pci_adjust_legacy_attr(struc extern int pci_create_resource_files(struct pci_dev *dev); extern void pci_remove_resource_files(struct pci_dev *dev); +extern int alpha_serial_port_needs_delay; + #endif /* __ALPHA_PCI_H */ Index: linux-stable/arch/alpha/kernel/pci.c =================================================================== --- linux-stable.orig/arch/alpha/kernel/pci.c 2020-05-06 17:16:28.000000000 +0200 +++ linux-stable/arch/alpha/kernel/pci.c 2020-05-06 17:16:28.000000000 +0200 @@ -61,9 +61,13 @@ struct pci_controller *pci_isa_hose; * Quirks. */ +int alpha_serial_port_needs_delay = 0; +EXPORT_SYMBOL(alpha_serial_port_needs_delay); + static void quirk_isa_bridge(struct pci_dev *dev) { dev->class = PCI_CLASS_BRIDGE_ISA << 8; + alpha_serial_port_needs_delay = 1; } DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82378, quirk_isa_bridge); Index: linux-stable/drivers/tty/serial/8250/8250_port.c =================================================================== --- linux-stable.orig/drivers/tty/serial/8250/8250_port.c 2020-05-06 17:16:28.000000000 +0200 +++ linux-stable/drivers/tty/serial/8250/8250_port.c 2020-05-06 17:16:32.000000000 +0200 @@ -442,6 +442,9 @@ static unsigned int mem32be_serial_in(st static unsigned int io_serial_in(struct uart_port *p, int offset) { + if (unlikely(p->quirks & UPQ_DELAY_BEFORE_READ)) + ndelay(300); + offset = offset << p->regshift; return inb(p->iobase + offset); }