| Message ID | 1585907888-12554-1-git-send-email-bmeng.cn@gmail.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | serial: ns16550: Fix ordering of getting base address | expand |
On Fri, Apr 03, 2020 at 02:58:08AM -0700, Bin Meng wrote: > Currently the driver gets ns16550 base address in the driver > probe() routine, which may potentially break any ns16550 wrapper > driver that does additional initialization before calling > ns16550_serial_probe(). > > Things are complicated that we need consider ns16550 devices on > both simple-bus and PCI bus. To fix the issue we move the base > address assignment for simple-bus ns16550 device back to the > ofdata_to_platdata(), and assign base address for PCI ns16550 > device in ns16550_serial_probe(). > > This is still not perfect. Ideally if any PCI bus based ns16550 > wrapper driver tries to access plat->base before calling probe(), > it is subject to break. Thank you. I have tested it and it fixes my case. Tested-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> > Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from > ofdata_to_platdata() to probe()") I believe this should be one line. > Reported-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> > Signed-off-by: Bin Meng <bmeng.cn at gmail.com> > > --- > > drivers/serial/ns16550.c | 51 +++++++++++++++++++++++------------------------- > 1 file changed, 24 insertions(+), 27 deletions(-) > > diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c > index c1b303f..5e3cd1c 100644 > --- a/drivers/serial/ns16550.c > +++ b/drivers/serial/ns16550.c > @@ -479,39 +479,24 @@ static int ns16550_serial_getinfo(struct udevice *dev, > return 0; > } > > -#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) > -static int ns1655_serial_set_base_addr(struct udevice *dev) > -{ > - fdt_addr_t addr; > - struct ns16550_platdata *plat; > - > - plat = dev_get_platdata(dev); > - > - addr = dev_read_addr_pci(dev); > - if (addr == FDT_ADDR_T_NONE) > - return -EINVAL; > - > -#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > - plat->base = addr; > -#else > - plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); > -#endif > - > - return 0; > -} > -#endif > - > int ns16550_serial_probe(struct udevice *dev) > { > + struct ns16550_platdata *plat = dev->platdata; > struct NS16550 *const com_port = dev_get_priv(dev); > struct reset_ctl_bulk reset_bulk; > + fdt_addr_t addr; > int ret; > > -#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) > - ret = ns1655_serial_set_base_addr(dev); > - if (ret) > - return ret; > -#endif > + /* > + * If we are on PCI bus, either directly attached to a PCI root port, > + * or via a PCI bridge, assign platdata->base before probing hardware. > + */ > + if (device_is_on_pci_bus(dev)) { > + addr = devfdt_get_addr_pci(dev); > + if (addr == FDT_ADDR_T_NONE) > + return -EINVAL; > + plat->base = addr; > + } > > ret = reset_get_bulk(dev, &reset_bulk); > if (!ret) > @@ -535,9 +520,21 @@ int ns16550_serial_ofdata_to_platdata(struct udevice *dev) > { > struct ns16550_platdata *plat = dev->platdata; > const u32 port_type = dev_get_driver_data(dev); > + fdt_addr_t addr; > struct clk clk; > int err; > > + addr = dev_read_addr(dev); > + if (addr != FDT_ADDR_T_NONE) { > +#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > + plat->base = addr; > +#else > + plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); > +#endif > + } else if (!device_is_on_pci_bus(dev)) { > + return -EINVAL; > + } > + > plat->reg_offset = dev_read_u32_default(dev, "reg-offset", 0); > plat->reg_shift = dev_read_u32_default(dev, "reg-shift", 0); > plat->reg_width = dev_read_u32_default(dev, "reg-io-width", 1); > -- > 2.7.4 >
Hi Andy, On Fri, Apr 3, 2020 at 6:05 PM Andy Shevchenko <andriy.shevchenko at linux.intel.com> wrote: > > On Fri, Apr 03, 2020 at 02:58:08AM -0700, Bin Meng wrote: > > Currently the driver gets ns16550 base address in the driver > > probe() routine, which may potentially break any ns16550 wrapper > > driver that does additional initialization before calling > > ns16550_serial_probe(). > > > > Things are complicated that we need consider ns16550 devices on > > both simple-bus and PCI bus. To fix the issue we move the base > > address assignment for simple-bus ns16550 device back to the > > ofdata_to_platdata(), and assign base address for PCI ns16550 > > device in ns16550_serial_probe(). > > > > This is still not perfect. Ideally if any PCI bus based ns16550 > > wrapper driver tries to access plat->base before calling probe(), > > it is subject to break. > > Thank you. > I have tested it and it fixes my case. > > Tested-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> Thanks for testing! > > > > Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from > > ofdata_to_platdata() to probe()") > > I believe this should be one line. I thought putting it in one line violates the 80 characters per column rule. And checkpatch.pl does not complain about this, so I think it's fine :) > > > Reported-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> > > Signed-off-by: Bin Meng <bmeng.cn at gmail.com> > > > > --- > > > > drivers/serial/ns16550.c | 51 +++++++++++++++++++++++------------------------- > > 1 file changed, 24 insertions(+), 27 deletions(-) > > Regards, Bin
On Fri, Apr 03, 2020 at 06:16:51PM +0800, Bin Meng wrote: > On Fri, Apr 3, 2020 at 6:05 PM Andy Shevchenko > <andriy.shevchenko at linux.intel.com> wrote: > > On Fri, Apr 03, 2020 at 02:58:08AM -0700, Bin Meng wrote: ... > > > Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from > > > ofdata_to_platdata() to probe()") > > > > I believe this should be one line. > > I thought putting it in one line violates the 80 characters per column > rule. And checkpatch.pl does not complain about this, so I think it's > fine :) Breaking tags to two or more lines actually breaks the scripts around this stuff. JFYI, in Linux kernel it was now in documentation (not to break Fixes, etc to two or more lines).
Hi Bin, Thanks for taking care of this! -----"Bin Meng" <bmeng.cn at gmail.com> schrieb: ----- >An: "Simon Glass" <sjg at chromium.org>, "Tom Rini" ><trini at konsulko.com>, "Andy Shevchenko" ><andriy.shevchenko at linux.intel.com>, "Wolfgang Wallner" ><wolfgang.wallner at br-automation.com>, "Chee Hong Ang" ><chee.hong.ang at intel.com>, "U-Boot Mailing List" ><u-boot at lists.denx.de> >Von: "Bin Meng" <bmeng.cn at gmail.com> >Datum: 03.04.2020 11:58 >Betreff: [PATCH] serial: ns16550: Fix ordering of getting base >address > >Currently the driver gets ns16550 base address in the driver >probe() routine, which may potentially break any ns16550 wrapper >driver that does additional initialization before calling >ns16550_serial_probe(). > >Things are complicated that we need consider ns16550 devices on >both simple-bus and PCI bus. To fix the issue we move the base >address assignment for simple-bus ns16550 device back to the >ofdata_to_platdata(), and assign base address for PCI ns16550 >device in ns16550_serial_probe(). > >This is still not perfect. Ideally if any PCI bus based ns16550 >wrapper driver tries to access plat->base before calling probe(), >it is subject to break. > >Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from >ofdata_to_platdata() to probe()") >Reported-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> >Signed-off-by: Bin Meng <bmeng.cn at gmail.com> > >--- > > drivers/serial/ns16550.c | 51 >+++++++++++++++++++++++------------------------- > 1 file changed, 24 insertions(+), 27 deletions(-) > >diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c >index c1b303f..5e3cd1c 100644 >--- a/drivers/serial/ns16550.c >+++ b/drivers/serial/ns16550.c >@@ -479,39 +479,24 @@ static int ns16550_serial_getinfo(struct >udevice *dev, > return 0; > } > >-#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) >-static int ns1655_serial_set_base_addr(struct udevice *dev) >-{ >- fdt_addr_t addr; >- struct ns16550_platdata *plat; >- >- plat = dev_get_platdata(dev); >- >- addr = dev_read_addr_pci(dev); >- if (addr == FDT_ADDR_T_NONE) >- return -EINVAL; >- >-#ifdef CONFIG_SYS_NS16550_PORT_MAPPED >- plat->base = addr; >-#else >- plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); >-#endif >- >- return 0; >-} >-#endif >- > int ns16550_serial_probe(struct udevice *dev) > { >+ struct ns16550_platdata *plat = dev->platdata; > struct NS16550 *const com_port = dev_get_priv(dev); > struct reset_ctl_bulk reset_bulk; >+ fdt_addr_t addr; > int ret; > >-#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) >- ret = ns1655_serial_set_base_addr(dev); >- if (ret) >- return ret; >-#endif >+ /* >+ * If we are on PCI bus, either directly attached to a PCI root >port, >+ * or via a PCI bridge, assign platdata->base before probing >hardware. >+ */ >+ if (device_is_on_pci_bus(dev)) { >+ addr = devfdt_get_addr_pci(dev); >+ if (addr == FDT_ADDR_T_NONE) >+ return -EINVAL; >+ plat->base = addr; The assignment here to plat->base does not distinguish any more based on CONFIG_SYS_NS16550_PORT_MAPPED. Is it not necessary any more in this case? >+ } > > ret = reset_get_bulk(dev, &reset_bulk); > if (!ret) >@@ -535,9 +520,21 @@ int ns16550_serial_ofdata_to_platdata(struct >udevice *dev) > { > struct ns16550_platdata *plat = dev->platdata; > const u32 port_type = dev_get_driver_data(dev); >+ fdt_addr_t addr; > struct clk clk; > int err; > >+ addr = dev_read_addr(dev); >+ if (addr != FDT_ADDR_T_NONE) { >+#ifdef CONFIG_SYS_NS16550_PORT_MAPPED >+ plat->base = addr; >+#else >+ plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); >+#endif >+ } else if (!device_is_on_pci_bus(dev)) { >+ return -EINVAL; >+ } >+ > plat->reg_offset = dev_read_u32_default(dev, "reg-offset", 0); > plat->reg_shift = dev_read_u32_default(dev, "reg-shift", 0); > plat->reg_width = dev_read_u32_default(dev, "reg-io-width", 1); >-- >2.7.4 > > Tested-by: Wolfgang Wallner <wolfgang.wallner at br-automation.com> Tested on an Intel Apollo Lake based board booting with Coreboot and using U-Boot as payload. Reviewed-by: Wolfgang Wallner <wolfgang.wallner at br-automation.com>
Hi Wolfgang, On Fri, Apr 3, 2020 at 7:47 PM Wolfgang Wallner <wolfgang.wallner at br-automation.com> wrote: > > Hi Bin, > > Thanks for taking care of this! > > -----"Bin Meng" <bmeng.cn at gmail.com> schrieb: ----- > > >An: "Simon Glass" <sjg at chromium.org>, "Tom Rini" > ><trini at konsulko.com>, "Andy Shevchenko" > ><andriy.shevchenko at linux.intel.com>, "Wolfgang Wallner" > ><wolfgang.wallner at br-automation.com>, "Chee Hong Ang" > ><chee.hong.ang at intel.com>, "U-Boot Mailing List" > ><u-boot at lists.denx.de> > >Von: "Bin Meng" <bmeng.cn at gmail.com> > >Datum: 03.04.2020 11:58 > >Betreff: [PATCH] serial: ns16550: Fix ordering of getting base > >address > > > >Currently the driver gets ns16550 base address in the driver > >probe() routine, which may potentially break any ns16550 wrapper > >driver that does additional initialization before calling > >ns16550_serial_probe(). > > > >Things are complicated that we need consider ns16550 devices on > >both simple-bus and PCI bus. To fix the issue we move the base > >address assignment for simple-bus ns16550 device back to the > >ofdata_to_platdata(), and assign base address for PCI ns16550 > >device in ns16550_serial_probe(). > > > >This is still not perfect. Ideally if any PCI bus based ns16550 > >wrapper driver tries to access plat->base before calling probe(), > >it is subject to break. > > > >Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from > >ofdata_to_platdata() to probe()") > >Reported-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> > >Signed-off-by: Bin Meng <bmeng.cn at gmail.com> > > > >--- > > > > drivers/serial/ns16550.c | 51 > >+++++++++++++++++++++++------------------------- > > 1 file changed, 24 insertions(+), 27 deletions(-) > > > >diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c > >index c1b303f..5e3cd1c 100644 > >--- a/drivers/serial/ns16550.c > >+++ b/drivers/serial/ns16550.c > >@@ -479,39 +479,24 @@ static int ns16550_serial_getinfo(struct > >udevice *dev, > > return 0; > > } > > > >-#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) > >-static int ns1655_serial_set_base_addr(struct udevice *dev) > >-{ > >- fdt_addr_t addr; > >- struct ns16550_platdata *plat; > >- > >- plat = dev_get_platdata(dev); > >- > >- addr = dev_read_addr_pci(dev); > >- if (addr == FDT_ADDR_T_NONE) > >- return -EINVAL; > >- > >-#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > >- plat->base = addr; > >-#else > >- plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); > >-#endif > >- > >- return 0; > >-} > >-#endif > >- > > int ns16550_serial_probe(struct udevice *dev) > > { > >+ struct ns16550_platdata *plat = dev->platdata; > > struct NS16550 *const com_port = dev_get_priv(dev); > > struct reset_ctl_bulk reset_bulk; > >+ fdt_addr_t addr; > > int ret; > > > >-#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) > >- ret = ns1655_serial_set_base_addr(dev); > >- if (ret) > >- return ret; > >-#endif > >+ /* > >+ * If we are on PCI bus, either directly attached to a PCI root > >port, > >+ * or via a PCI bridge, assign platdata->base before probing > >hardware. > >+ */ > >+ if (device_is_on_pci_bus(dev)) { > >+ addr = devfdt_get_addr_pci(dev); > >+ if (addr == FDT_ADDR_T_NONE) > >+ return -EINVAL; > >+ plat->base = addr; > > The assignment here to plat->base does not distinguish any more based on > CONFIG_SYS_NS16550_PORT_MAPPED. Is it not necessary any more in this case? > Yep, you are right. Both two cases should be handled consistently. I will spin a v2. > >+ } > > > > ret = reset_get_bulk(dev, &reset_bulk); > > if (!ret) > >@@ -535,9 +520,21 @@ int ns16550_serial_ofdata_to_platdata(struct > >udevice *dev) > > { > > struct ns16550_platdata *plat = dev->platdata; > > const u32 port_type = dev_get_driver_data(dev); > >+ fdt_addr_t addr; > > struct clk clk; > > int err; > > > >+ addr = dev_read_addr(dev); > >+ if (addr != FDT_ADDR_T_NONE) { > >+#ifdef CONFIG_SYS_NS16550_PORT_MAPPED > >+ plat->base = addr; > >+#else > >+ plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); > >+#endif > >+ } else if (!device_is_on_pci_bus(dev)) { > >+ return -EINVAL; > >+ } > >+ > > plat->reg_offset = dev_read_u32_default(dev, "reg-offset", 0); > > plat->reg_shift = dev_read_u32_default(dev, "reg-shift", 0); > > plat->reg_width = dev_read_u32_default(dev, "reg-io-width", 1); > >-- > >2.7.4 > > > > > > Tested-by: Wolfgang Wallner <wolfgang.wallner at br-automation.com> > Tested on an Intel Apollo Lake based board booting with Coreboot and using > U-Boot as payload. > Thanks for testing! > Reviewed-by: Wolfgang Wallner <wolfgang.wallner at br-automation.com> > Regards, Bin
diff --git a/drivers/serial/ns16550.c b/drivers/serial/ns16550.c index c1b303f..5e3cd1c 100644 --- a/drivers/serial/ns16550.c +++ b/drivers/serial/ns16550.c @@ -479,39 +479,24 @@ static int ns16550_serial_getinfo(struct udevice *dev, return 0; } -#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) -static int ns1655_serial_set_base_addr(struct udevice *dev) -{ - fdt_addr_t addr; - struct ns16550_platdata *plat; - - plat = dev_get_platdata(dev); - - addr = dev_read_addr_pci(dev); - if (addr == FDT_ADDR_T_NONE) - return -EINVAL; - -#ifdef CONFIG_SYS_NS16550_PORT_MAPPED - plat->base = addr; -#else - plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); -#endif - - return 0; -} -#endif - int ns16550_serial_probe(struct udevice *dev) { + struct ns16550_platdata *plat = dev->platdata; struct NS16550 *const com_port = dev_get_priv(dev); struct reset_ctl_bulk reset_bulk; + fdt_addr_t addr; int ret; -#if CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA) - ret = ns1655_serial_set_base_addr(dev); - if (ret) - return ret; -#endif + /* + * If we are on PCI bus, either directly attached to a PCI root port, + * or via a PCI bridge, assign platdata->base before probing hardware. + */ + if (device_is_on_pci_bus(dev)) { + addr = devfdt_get_addr_pci(dev); + if (addr == FDT_ADDR_T_NONE) + return -EINVAL; + plat->base = addr; + } ret = reset_get_bulk(dev, &reset_bulk); if (!ret) @@ -535,9 +520,21 @@ int ns16550_serial_ofdata_to_platdata(struct udevice *dev) { struct ns16550_platdata *plat = dev->platdata; const u32 port_type = dev_get_driver_data(dev); + fdt_addr_t addr; struct clk clk; int err; + addr = dev_read_addr(dev); + if (addr != FDT_ADDR_T_NONE) { +#ifdef CONFIG_SYS_NS16550_PORT_MAPPED + plat->base = addr; +#else + plat->base = (unsigned long)map_physmem(addr, 0, MAP_NOCACHE); +#endif + } else if (!device_is_on_pci_bus(dev)) { + return -EINVAL; + } + plat->reg_offset = dev_read_u32_default(dev, "reg-offset", 0); plat->reg_shift = dev_read_u32_default(dev, "reg-shift", 0); plat->reg_width = dev_read_u32_default(dev, "reg-io-width", 1);
Currently the driver gets ns16550 base address in the driver probe() routine, which may potentially break any ns16550 wrapper driver that does additional initialization before calling ns16550_serial_probe(). Things are complicated that we need consider ns16550 devices on both simple-bus and PCI bus. To fix the issue we move the base address assignment for simple-bus ns16550 device back to the ofdata_to_platdata(), and assign base address for PCI ns16550 device in ns16550_serial_probe(). This is still not perfect. Ideally if any PCI bus based ns16550 wrapper driver tries to access plat->base before calling probe(), it is subject to break. Fixes: 720f9e1fdb0c9 ("serial: ns16550: Move PCI access from ofdata_to_platdata() to probe()") Reported-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com> Signed-off-by: Bin Meng <bmeng.cn at gmail.com> --- drivers/serial/ns16550.c | 51 +++++++++++++++++++++++------------------------- 1 file changed, 24 insertions(+), 27 deletions(-)