| Message ID | CAFEAcA_-S8z6ZMr8TMPvkcTyyCSMOj1DExVsY4JDQjRXDU5WGw@mail.gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Aug 15, 2013 at 05:45:42PM +0100, Peter Maydell wrote: > On 13 August 2013 04:40, Guenter Roeck <linux@roeck-us.net> wrote: > > Patch tested and working with qemu 1.5.2, using the configuration file > > from the yocto project. Patch applied on top of kernel version 3.11-rc5. > > OK, I tested this on PB926+PCI backplane hardware, and it is > definitely better than current mainline, in that the test USB > card that I have no longer causes the kernel to generate this sort of > backtrace: > Do you mean my patch fixes the traceback below as a side effect ? Would be great ... it would be one more reason to get it applied. > ehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver > ehci-pci: EHCI PCI platform driver > ehci-pci 0000:00:1e.2: EHCI Host Controller > ehci-pci 0000:00:1e.2: new USB bus registered, assigned bus number 1 > ehci-pci 0000:00:1e.2: irq 91, io mem 0x50002000 > ehci-pci 0000:00:1e.2: USB 2.0 started, EHCI 1.00 > usb usb1: New USB device found, idVendor=1d6b, idProduct=0002 > usb usb1: New USB device strings: Mfr=3, Product=2, SerialNumber=1 > usb usb1: Product: EHCI Host Controller > usb usb1: Manufacturer: Linux 3.10.0+ ehci_hcd > usb usb1: SerialNumber: 0000:00:1e.2 > hub 1-0:1.0: USB hub found > hub 1-0:1.0: 3 ports detected > ohci_hcd: USB 1.1 'Open' Host Controller (OHCI) Driver > ohci-pci: OHCI PCI platform driver > ohci-pci 0000:00:1e.0: OHCI PCI host controller > ohci-pci 0000:00:1e.0: new USB bus registered, assigned bus number 2 > irq 93: nobody cared (try booting with the "irqpoll" option) > CPU: 0 PID: 1 Comm: swapper Not tainted 3.10.0+ #9 > [<c00187d4>] (unwind_backtrace+0x0/0xf0) from [<c0015b48>] > (show_stack+0x10/0x14) > [<c0015b48>] (show_stack+0x10/0x14) from [<c0054cac>] > (__report_bad_irq+0x24/0xb8) > [<c0054cac>] (__report_bad_irq+0x24/0xb8) from [<c005511c>] > (note_interrupt+0x1cc/0x230) > [<c005511c>] (note_interrupt+0x1cc/0x230) from [<c005363c>] > (handle_irq_event_percpu+0xac/0x1c4) > [<c005363c>] (handle_irq_event_percpu+0xac/0x1c4) from [<c005377c>] > (handle_irq_event+0x28/0x38) > [<c005377c>] (handle_irq_event+0x28/0x38) from [<c00558c8>] > (handle_level_irq+0x80/0xd4) > [<c00558c8>] (handle_level_irq+0x80/0xd4) from [<c0052fb4>] > (generic_handle_irq+0x2c/0x40) > [<c0052fb4>] (generic_handle_irq+0x2c/0x40) from [<c016bf38>] > (fpga_irq_handle+0x3c/0x50) > [<c016bf38>] (fpga_irq_handle+0x3c/0x50) from [<c0052fb4>] > (generic_handle_irq+0x2c/0x40) > [<c0052fb4>] (generic_handle_irq+0x2c/0x40) from [<c0013e8c>] > (handle_IRQ+0x30/0x84) > [<c0013e8c>] (handle_IRQ+0x30/0x84) from [<c0008750>] (vic_handle_irq+0x5c/0x9c) > [<c0008750>] (vic_handle_irq+0x5c/0x9c) from [<c00165c0>] (__irq_svc+0x40/0x4c) > Exception stack(0xc7829cc8 to 0xc7829d10) > 9cc0: 00000001 0000000a 00000100 20000013 00000002 00000024 > 9ce0: c7828000 c0476980 3fb96c1c c0443950 c04693c0 00000001 c0456a50 c7829d10 > 9d00: c0025f38 c0025fa8 20000013 ffffffff > [<c00165c0>] (__irq_svc+0x40/0x4c) from [<c0025fa8>] (__do_softirq+0x80/0x1b4) > [<c0025fa8>] (__do_softirq+0x80/0x1b4) from [<c00263a4>] (irq_exit+0x54/0x90) > [<c00263a4>] (irq_exit+0x54/0x90) from [<c0013e90>] (handle_IRQ+0x34/0x84) > [<c0013e90>] (handle_IRQ+0x34/0x84) from [<c0008750>] (vic_handle_irq+0x5c/0x9c) > [<c0008750>] (vic_handle_irq+0x5c/0x9c) from [<c00165c0>] (__irq_svc+0x40/0x4c) > Exception stack(0xc7829d80 to 0xc7829dc8) > 9d80: 00000000 0000005d 20000000 00000000 c79bb7e0 c0446990 60000013 c79c0000 > 9da0: 0000005d 00000000 c04469c4 00000001 00000000 c7829dc8 c00555c8 c0054460 > 9dc0: 40000013 ffffffff > [<c00165c0>] (__irq_svc+0x40/0x4c) from [<c0054460>] (__setup_irq+0x1f4/0x3f0) > [<c0054460>] (__setup_irq+0x1f4/0x3f0) from [<c00548ac>] > (request_threaded_irq+0xb4/0x138) > [<c00548ac>] (request_threaded_irq+0xb4/0x138) from [<c01fa7a8>] > (usb_add_hcd+0x4f0/0x6f0) > [<c01fa7a8>] (usb_add_hcd+0x4f0/0x6f0) from [<c0206bdc>] > (usb_hcd_pci_probe+0x200/0x36c) > [<c0206bdc>] (usb_hcd_pci_probe+0x200/0x36c) from [<c017467c>] > (pci_device_probe+0x68/0x90) > [<c017467c>] (pci_device_probe+0x68/0x90) from [<c01b3a80>] > (driver_probe_device+0x78/0x200) > [<c01b3a80>] (driver_probe_device+0x78/0x200) from [<c01b3c94>] > (__driver_attach+0x8c/0x90) > [<c01b3c94>] (__driver_attach+0x8c/0x90) from [<c01b2288>] > (bus_for_each_dev+0x58/0x88) > [<c01b2288>] (bus_for_each_dev+0x58/0x88) from [<c01b29e8>] > (bus_add_driver+0xd8/0x220) > [<c01b29e8>] (bus_add_driver+0xd8/0x220) from [<c01b4258>] > (driver_register+0x78/0x144) > [<c01b4258>] (driver_register+0x78/0x144) from [<c0410820>] > (do_one_initcall+0x94/0x154) > [<c0410820>] (do_one_initcall+0x94/0x154) from [<c04109cc>] > (kernel_init_freeable+0xec/0x1b0) > [<c04109cc>] (kernel_init_freeable+0xec/0x1b0) from [<c0323104>] > (kernel_init+0x8/0xe4) > [<c0323104>] (kernel_init+0x8/0xe4) from [<c00130b0>] (ret_from_fork+0x14/0x24) > handlers: > [<c01f7e48>] usb_hcd_irq > Disabling IRQ #93 > > However it still doesn't seem to reliably detect the USB harddisk > plugged into the card, so I think there may be further issues, possibly > some subset of those Arnd identified and fixed with this patch: > http://permalink.gmane.org/gmane.linux.ports.arm.kernel/93397 > Does it get better if you apply Arnd's patch ? > so I'd like to continue testing. > > The other thing this patch should (IMHO) have is the > line in pci_versatile_setup() which tells QEMU that the > kernel really does expect hardware-like behaviour: > > --- a/arch/arm/mach-versatile/pci.c > +++ b/arch/arm/mach-versatile/pci.c > @@ -295,6 +295,19 @@ int __init pci_versatile_setup(int nr, struct > pci_sys_data *sys) > __raw_writel(PHYS_OFFSET, local_pci_cfg_base + PCI_BASE_ADDRESS_2); > > /* > + * For many years the kernel and QEMU were symbiotically buggy > + * in that they both assumed the same broken IRQ mapping. > + * QEMU therefore attempts to auto-detect old broken kernels > + * so that they still work on newer QEMU as they did on old > + * QEMU. Since we now use the correct (ie matching-hardware) > + * IRQ mapping we write a definitely different value to a > + * PCI_INTERRUPT_LINE register to tell QEMU that we expect > + * real hardware behaviour and it need not be backwards > + * compatible for us. This write is harmless on real hardware. > + */ > + __raw_writel(0, VERSATILE_PCI_VIRT_BASE+PCI_INTERRUPT_LINE); > + > + /* > * Do not to map Versatile FPGA PCI device into memory space > */ > pci_slot_ignore |= (1 << myslot); > > I appreciate that QEMU-specific code in the kernel is a bit ugly, but > (a) 99.9% of the people running this code are going to be running it on QEMU > (b) the bugs in this area have been extremely long-standing in both > the kernel and QEMU and so there are a lot of legacy kernel images > out there that worked just fine with QEMU > (c) it's rather less code than I have in QEMU as the QEMU-side part > of this backwards-compatibility autodetection. > > (Without this line QEMU will guess whether the kernel is broken > or not and will get it right most but not necessarily all of the time.) > Might make sense, but I think it should be a separate patch. If I understand correctly, my patch fixes the SCSI problem. Is that correct ? If so, can we get it applied to mainline ? Thanks, Guenter
On 15 August 2013 18:54, Guenter Roeck <linux@roeck-us.net> wrote: > On Thu, Aug 15, 2013 at 05:45:42PM +0100, Peter Maydell wrote: >> On 13 August 2013 04:40, Guenter Roeck <linux@roeck-us.net> wrote: >> > Patch tested and working with qemu 1.5.2, using the configuration file >> > from the yocto project. Patch applied on top of kernel version 3.11-rc5. >> >> OK, I tested this on PB926+PCI backplane hardware, and it is >> definitely better than current mainline, in that the test USB >> card that I have no longer causes the kernel to generate this sort of >> backtrace: >> > Do you mean my patch fixes the traceback below as a side effect ? > Would be great ... it would be one more reason to get it applied. Yes, exactly -- the kernel currently has the wrong irq mapping, which causes the traceback (ie h/w asserts irq 93 but the kernel is listening on something else). That the patch fixes this confirms that it is the behaviour of hardware as well as of QEMU. >> However it still doesn't seem to reliably detect the USB harddisk >> plugged into the card, so I think there may be further issues, possibly >> some subset of those Arnd identified and fixed with this patch: >> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/93397 >> > Does it get better if you apply Arnd's patch ? Arnd's patch is ancient and won't apply as is (due to intervening changes and also because some of the things it fixes were fixed in later patches); I'm currently trying to extract the relevant parts. If you want you can confirm that I/O port PCI access is broken on QEMU too -- disable CONFIG_SCSI_SYM53C8XX_MMIO so the driver uses PCI IO rather than MMIO and you'll see QEMU's SCSI device doesn't work any more. >> so I'd like to continue testing. >> >> The other thing this patch should (IMHO) have is the >> line in pci_versatile_setup() which tells QEMU that the >> kernel really does expect hardware-like behaviour: >> (Without this line QEMU will guess whether the kernel is broken >> or not and will get it right most but not necessarily all of the time.) >> > Might make sense, but I think it should be a separate patch. It needs to go in the same patch, because a kernel with the fixed irq remapping must also tell QEMU it is fixed; if you split the two then at the point between the two patches the kernel is broken for bisection purposes. > If I understand correctly, my patch fixes the SCSI problem. > Is that correct ? If so, can we get it applied to mainline ? I'd rather get to a point where I have the hardware definitely completely working first. There's no real hurry, this has been broken for months and months. -- PMM
On Thu, Aug 15, 2013 at 07:05:22PM +0100, Peter Maydell wrote: > On 15 August 2013 18:54, Guenter Roeck <linux@roeck-us.net> wrote: > > On Thu, Aug 15, 2013 at 05:45:42PM +0100, Peter Maydell wrote: > >> On 13 August 2013 04:40, Guenter Roeck <linux@roeck-us.net> wrote: > >> > Patch tested and working with qemu 1.5.2, using the configuration file > >> > from the yocto project. Patch applied on top of kernel version 3.11-rc5. > >> > >> OK, I tested this on PB926+PCI backplane hardware, and it is > >> definitely better than current mainline, in that the test USB > >> card that I have no longer causes the kernel to generate this sort of > >> backtrace: > >> > > Do you mean my patch fixes the traceback below as a side effect ? > > Would be great ... it would be one more reason to get it applied. > > Yes, exactly -- the kernel currently has the wrong irq mapping, > which causes the traceback (ie h/w asserts irq 93 but the kernel > is listening on something else). That the patch fixes this confirms > that it is the behaviour of hardware as well as of QEMU. > > >> However it still doesn't seem to reliably detect the USB harddisk > >> plugged into the card, so I think there may be further issues, possibly > >> some subset of those Arnd identified and fixed with this patch: > >> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/93397 > >> > > Does it get better if you apply Arnd's patch ? > > Arnd's patch is ancient and won't apply as is (due to intervening > changes and also because some of the things it fixes were fixed > in later patches); I'm currently trying to extract the relevant parts. > > If you want you can confirm that I/O port PCI access is broken on > QEMU too -- disable CONFIG_SCSI_SYM53C8XX_MMIO so > the driver uses PCI IO rather than MMIO and you'll see QEMU's > SCSI device doesn't work any more. > > >> so I'd like to continue testing. > >> > >> The other thing this patch should (IMHO) have is the > >> line in pci_versatile_setup() which tells QEMU that the > >> kernel really does expect hardware-like behaviour: > > >> (Without this line QEMU will guess whether the kernel is broken > >> or not and will get it right most but not necessarily all of the time.) > >> > > Might make sense, but I think it should be a separate patch. > > It needs to go in the same patch, because a kernel with the fixed > irq remapping must also tell QEMU it is fixed; if you split the > two then at the point between the two patches the kernel is > broken for bisection purposes. > > > If I understand correctly, my patch fixes the SCSI problem. > > Is that correct ? If so, can we get it applied to mainline ? > > I'd rather get to a point where I have the hardware definitely > completely working first. There's no real hurry, this has been > broken for months and months. > Ok with me, if it doesn't get lost. Until it gets fixed, arm status on 3.10 kernels will show as "failed" for qemu test runs. Thanks, Guenter
On Thu, Aug 15, 2013 at 07:05:22PM +0100, Peter Maydell wrote: > On 15 August 2013 18:54, Guenter Roeck <linux@roeck-us.net> wrote: > > On Thu, Aug 15, 2013 at 05:45:42PM +0100, Peter Maydell wrote: > >> On 13 August 2013 04:40, Guenter Roeck <linux@roeck-us.net> wrote: > >> > Patch tested and working with qemu 1.5.2, using the configuration file > >> > from the yocto project. Patch applied on top of kernel version 3.11-rc5. > >> > >> OK, I tested this on PB926+PCI backplane hardware, and it is > >> definitely better than current mainline, in that the test USB > >> card that I have no longer causes the kernel to generate this sort of > >> backtrace: > >> > > Do you mean my patch fixes the traceback below as a side effect ? > > Would be great ... it would be one more reason to get it applied. > > Yes, exactly -- the kernel currently has the wrong irq mapping, > which causes the traceback (ie h/w asserts irq 93 but the kernel > is listening on something else). That the patch fixes this confirms > that it is the behaviour of hardware as well as of QEMU. > > >> However it still doesn't seem to reliably detect the USB harddisk > >> plugged into the card, so I think there may be further issues, possibly > >> some subset of those Arnd identified and fixed with this patch: > >> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/93397 > >> > > Does it get better if you apply Arnd's patch ? > > Arnd's patch is ancient and won't apply as is (due to intervening > changes and also because some of the things it fixes were fixed > in later patches); I'm currently trying to extract the relevant parts. > > If you want you can confirm that I/O port PCI access is broken on > QEMU too -- disable CONFIG_SCSI_SYM53C8XX_MMIO so > the driver uses PCI IO rather than MMIO and you'll see QEMU's > SCSI device doesn't work any more. > > >> so I'd like to continue testing. > >> > >> The other thing this patch should (IMHO) have is the > >> line in pci_versatile_setup() which tells QEMU that the > >> kernel really does expect hardware-like behaviour: > > >> (Without this line QEMU will guess whether the kernel is broken > >> or not and will get it right most but not necessarily all of the time.) > >> > > Might make sense, but I think it should be a separate patch. > > It needs to go in the same patch, because a kernel with the fixed > irq remapping must also tell QEMU it is fixed; if you split the > two then at the point between the two patches the kernel is > broken for bisection purposes. > Thinking about it - is that really true ? My image with the patch applied works just fine under qemu 1.5.2, and unless I am missing something it won't work with qemu 1.4 anyway. So what exactly is broken ? Thanks, Guenter
On 15 August 2013 21:50, Guenter Roeck <linux@roeck-us.net> wrote: > On Thu, Aug 15, 2013 at 07:05:22PM +0100, Peter Maydell wrote: >> It needs to go in the same patch, because a kernel with the fixed >> irq remapping must also tell QEMU it is fixed; if you split the >> two then at the point between the two patches the kernel is >> broken for bisection purposes. >> > Thinking about it - is that really true ? My image with the > patch applied works just fine under qemu 1.5.2, and unless > I am missing something it won't work with qemu 1.4 anyway. > So what exactly is broken ? You're OK unless the kernel happens to pick the same interrupt number to write to PCI_INTERRUPT_LINE as one of the previous broken kernel versions did (in which case QEMU will incorrectly assume you're a broken kernel). This can't happen with the way the kernel is currently picking interrupt numbers (ie with a straightforward relationship between h/w irqs and values written), but as I understand from Arnd there is a plan to move to a different approach ("sparse irqs") at which point this won't hold: http://lists.gnu.org/archive/html/qemu-devel/2013-03/msg04579.html So it's better for the kernel to make sure it gets the behaviour it wants rather than getting unpleasant surprises later. -- PMM
On 08/15/2013 02:49 PM, Peter Maydell wrote: > On 15 August 2013 21:50, Guenter Roeck <linux@roeck-us.net> wrote: >> On Thu, Aug 15, 2013 at 07:05:22PM +0100, Peter Maydell wrote: >>> It needs to go in the same patch, because a kernel with the fixed >>> irq remapping must also tell QEMU it is fixed; if you split the >>> two then at the point between the two patches the kernel is >>> broken for bisection purposes. >>> >> Thinking about it - is that really true ? My image with the >> patch applied works just fine under qemu 1.5.2, and unless >> I am missing something it won't work with qemu 1.4 anyway. >> So what exactly is broken ? > > You're OK unless the kernel happens to pick the same interrupt > number to write to PCI_INTERRUPT_LINE as one of the previous > broken kernel versions did (in which case QEMU will incorrectly > assume you're a broken kernel). This can't happen with the way > the kernel is currently picking interrupt numbers (ie with a > straightforward relationship between h/w irqs and values written), > but as I understand from Arnd there is a plan to move to a > different approach ("sparse irqs") at which point this won't hold: > http://lists.gnu.org/archive/html/qemu-devel/2013-03/msg04579.html > So it's better for the kernel to make sure it gets the > behaviour it wants rather than getting unpleasant surprises > later. > But doesn't that mean that there is _currently_ no problem ? If so, we can introduce the additional code when the problem really shows up. Being Preemptive is good, but if it is not really needed today I would rather have today's problems resolved and bother about tomorrow's when they show up. Guenter
On 15 August 2013 23:18, Guenter Roeck <linux@roeck-us.net> wrote: > But doesn't that mean that there is _currently_ no problem ? If so, > we can introduce the additional code when the problem really shows up. > Being Preemptive is good, but if it is not really needed today > I would rather have today's problems resolved and bother about tomorrow's > when they show up. Conceptually the two parts go together: rely on correct irq routing, tell qemu we rely on correct irq routing. It's only one extra line... -- PMM
On 08/15/2013 03:23 PM, Peter Maydell wrote: > On 15 August 2013 23:18, Guenter Roeck <linux@roeck-us.net> wrote: >> But doesn't that mean that there is _currently_ no problem ? If so, >> we can introduce the additional code when the problem really shows up. >> Being Preemptive is good, but if it is not really needed today >> I would rather have today's problems resolved and bother about tomorrow's >> when they show up. > > Conceptually the two parts go together: rely on correct > irq routing, tell qemu we rely on correct irq routing. > It's only one extra line... > Ok if Russel accepts it ... Guenter
On Thu, Aug 15, 2013 at 11:23:58PM +0100, Peter Maydell wrote: > On 15 August 2013 23:18, Guenter Roeck <linux@roeck-us.net> wrote: > > But doesn't that mean that there is _currently_ no problem ? If so, > > we can introduce the additional code when the problem really shows up. > > Being Preemptive is good, but if it is not really needed today > > I would rather have today's problems resolved and bother about tomorrow's > > when they show up. > > Conceptually the two parts go together: rely on correct > irq routing, tell qemu we rely on correct irq routing. > It's only one extra line... > Possibly, but the lack of progress suggests that by tying both parts together we might get neither accepted. Old saying - surgery successful, patient dead. Guenter
--- a/arch/arm/mach-versatile/pci.c +++ b/arch/arm/mach-versatile/pci.c @@ -295,6 +295,19 @@ int __init pci_versatile_setup(int nr, struct pci_sys_data *sys) __raw_writel(PHYS_OFFSET, local_pci_cfg_base + PCI_BASE_ADDRESS_2); /* + * For many years the kernel and QEMU were symbiotically buggy + * in that they both assumed the same broken IRQ mapping. + * QEMU therefore attempts to auto-detect old broken kernels + * so that they still work on newer QEMU as they did on old + * QEMU. Since we now use the correct (ie matching-hardware) + * IRQ mapping we write a definitely different value to a + * PCI_INTERRUPT_LINE register to tell QEMU that we expect + * real hardware behaviour and it need not be backwards + * compatible for us. This write is harmless on real hardware. + */ + __raw_writel(0, VERSATILE_PCI_VIRT_BASE+PCI_INTERRUPT_LINE); + + /* * Do not to map Versatile FPGA PCI device into memory space */