Message ID | 4867f9385a2.1e1ca6c1@auth.smtp.1and1.co.uk (mailing list archive) |
---|---|
State | Accepted |
Headers | show |
Darren Stevens <darren@stevens-zone.net> writes: > Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: > vmalloc abstraction in preparation for radix) turned kernel memory > and IO addresses from #defined constants to variables initialised > at runtime. > > On PA6T systems the setup_arch machine call initialises the onboard > PCI-e root-ports, and uses pci_io_base to do this, which is now before > its value has been set resulting in a panic right after 'booting > linux via __start()' > > Move the pci_io_base initialisation to the same place as vmalloc > ranges are set (hash__early_init_mmu()/radix__early_init_mmu()) > > Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de> > Signed-off-by: Darren Stevens <darren@stevens-zone.net> Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> > > --- > > Tested on my AmigaOneX1000, I don't have access to a refence board system, > and our developer with one is on honeymoon. > > I am hoping to follow this patch with others to reduce the size of the nemo > patch we apply, eventually I'd like to see the patch gone, but that is a > very big job. > > Kind regards > Darren > diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h > index 88a5eca..ab84c89 100644 > --- a/arch/powerpc/include/asm/book3s/64/pgtable.h > +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h > @@ -230,6 +230,7 @@ extern unsigned long __kernel_virt_size; > #define KERN_VIRT_SIZE __kernel_virt_size > extern struct page *vmemmap; > extern unsigned long ioremap_bot; > +extern unsigned long pci_io_base; > #endif /* __ASSEMBLY__ */ > > #include <asm/book3s/64/hash.h> > diff --git a/arch/powerpc/kernel/pci_64.c b/arch/powerpc/kernel/pci_64.c > index 3759df5..a5ae49a 100644 > --- a/arch/powerpc/kernel/pci_64.c > +++ b/arch/powerpc/kernel/pci_64.c > @@ -47,7 +47,6 @@ static int __init pcibios_init(void) > > printk(KERN_INFO "PCI: Probing PCI hardware\n"); > > - pci_io_base = ISA_IO_BASE; > /* For now, override phys_mem_access_prot. If we need it,g > * later, we may move that initialization to each ppc_md > */ > diff --git a/arch/powerpc/mm/hash_utils_64.c b/arch/powerpc/mm/hash_utils_64.c > index 5b22ba0..b5b5fe6 100644 > --- a/arch/powerpc/mm/hash_utils_64.c > +++ b/arch/powerpc/mm/hash_utils_64.c > @@ -922,6 +922,8 @@ void __init hash__early_init_mmu(void) > vmemmap = (struct page *)H_VMEMMAP_BASE; > ioremap_bot = IOREMAP_BASE; > > + pci_io_base = ISA_IO_BASE; > + > /* Initialize the MMU Hash table and create the linear mapping > * of memory. Has to be done before SLB initialization as this is > * currently where the page size encoding is obtained. > diff --git a/arch/powerpc/mm/pgtable-radix.c b/arch/powerpc/mm/pgtable-radix.c > index e58707d..095fbfa 100644 > --- a/arch/powerpc/mm/pgtable-radix.c > +++ b/arch/powerpc/mm/pgtable-radix.c > @@ -328,6 +328,9 @@ void __init radix__early_init_mmu(void) > __vmalloc_end = RADIX_VMALLOC_END; > vmemmap = (struct page *)RADIX_VMEMMAP_BASE; > ioremap_bot = IOREMAP_BASE; > + > + pci_io_base = ISA_IO_BASE; > + > /* > * For now radix also use the same frag size > */
Hi, On Wed, Jun 29, 2016 at 1:06 PM, Darren Stevens <darren@stevens-zone.net> wrote: > > Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: > vmalloc abstraction in preparation for radix) turned kernel memory > and IO addresses from #defined constants to variables initialised > at runtime. > > On PA6T systems the setup_arch machine call initialises the onboard > PCI-e root-ports, and uses pci_io_base to do this, which is now before > its value has been set resulting in a panic right after 'booting > linux via __start()' I don't see the panic here on a Chitra, at least not on recent -next. What config are you building with when you see this? Please cc me on PA Semi related patches if you want my attention. I still have hardware, in fact I boot every -next release on it through automation. -Olof
Olof Johansson <olof@lixom.net> writes: > Hi, > > On Wed, Jun 29, 2016 at 1:06 PM, Darren Stevens <darren@stevens-zone.net> wrote: >> >> Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: >> vmalloc abstraction in preparation for radix) turned kernel memory >> and IO addresses from #defined constants to variables initialised >> at runtime. >> >> On PA6T systems the setup_arch machine call initialises the onboard >> PCI-e root-ports, and uses pci_io_base to do this, which is now before >> its value has been set resulting in a panic right after 'booting >> linux via __start()' > > I don't see the panic here on a Chitra, at least not on recent -next. > > What config are you building with when you see this? > > Please cc me on PA Semi related patches if you want my attention. I > still have hardware, in fact I boot every -next release on it through > automation. > You can find more details in the below thread. https://lkml.kernel.org/r/8B4C4AB7-5C17-4992-935A-361153472793@xenosoft.de -aneesh
Hello Olof On 29/06/2016, Olof Johansson wrote: > I don't see the panic here on a Chitra, at least not on recent -next. > > What config are you building with when you see this? If you are not seeing this on the Chitra, it means that our Nemo patch is causing the problem. I don't know where, as we haven't patched the pas_pci_init function, or any function it directly calls, although there are patches in pasemi/pci.c to enable support for the SB600. The best solution would be for us to add the patch for setup.c I found to our nemo patch. Sorry for wasting everyones time. > Please cc me on PA Semi related patches if you want my attention. I > still have hardware, in fact I boot every -next release on it through > automation. Didn't know that, I will do so in future, just be aware that all of my patches will be in relation to the X1000 as that is the only PA-Semi machine I have here. If you have the time, there is something you can check for us - we've had to add 'PPC_PASEMI_IOMMU_DMA_FORCE' to our config since 4.4 to get the onboard gig-e ethernet to work. We need PPC_PASEMI_IOMMU for the SB600, or we are stuck with a kernel that will only work with 2GRam Regards Darren
On Wed, 2016-29-06 at 20:06:28 UTC, Darren Stevens wrote: > Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: > vmalloc abstraction in preparation for radix) turned kernel memory > and IO addresses from #defined constants to variables initialised > at runtime. > > On PA6T systems the setup_arch machine call initialises the onboard > PCI-e root-ports, and uses pci_io_base to do this, which is now before > its value has been set resulting in a panic right after 'booting > linux via __start()' > > Move the pci_io_base initialisation to the same place as vmalloc > ranges are set (hash__early_init_mmu()/radix__early_init_mmu()) > > Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de> > Signed-off-by: Darren Stevens <darren@stevens-zone.net> > Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Applied to powerpc fixes, thanks. https://git.kernel.org/powerpc/c/bfa37087aa04e45f56c41142df cheers
Many thanks! - Christian On 04 July 2016 at 07:51 AM, Michael Ellerman wrote: > On Wed, 2016-29-06 at 20:06:28 UTC, Darren Stevens wrote: >> Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: >> vmalloc abstraction in preparation for radix) turned kernel memory >> and IO addresses from #defined constants to variables initialised >> at runtime. >> >> On PA6T systems the setup_arch machine call initialises the onboard >> PCI-e root-ports, and uses pci_io_base to do this, which is now before >> its value has been set resulting in a panic right after 'booting >> linux via __start()' >> >> Move the pci_io_base initialisation to the same place as vmalloc >> ranges are set (hash__early_init_mmu()/radix__early_init_mmu()) >> >> Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de> >> Signed-off-by: Darren Stevens <darren@stevens-zone.net> >> Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> > Applied to powerpc fixes, thanks. > > https://git.kernel.org/powerpc/c/bfa37087aa04e45f56c41142df > > cheers >
diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h index 88a5eca..ab84c89 100644 --- a/arch/powerpc/include/asm/book3s/64/pgtable.h +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h @@ -230,6 +230,7 @@ extern unsigned long __kernel_virt_size; #define KERN_VIRT_SIZE __kernel_virt_size extern struct page *vmemmap; extern unsigned long ioremap_bot; +extern unsigned long pci_io_base; #endif /* __ASSEMBLY__ */ #include <asm/book3s/64/hash.h> diff --git a/arch/powerpc/kernel/pci_64.c b/arch/powerpc/kernel/pci_64.c index 3759df5..a5ae49a 100644 --- a/arch/powerpc/kernel/pci_64.c +++ b/arch/powerpc/kernel/pci_64.c @@ -47,7 +47,6 @@ static int __init pcibios_init(void) printk(KERN_INFO "PCI: Probing PCI hardware\n"); - pci_io_base = ISA_IO_BASE; /* For now, override phys_mem_access_prot. If we need it,g * later, we may move that initialization to each ppc_md */ diff --git a/arch/powerpc/mm/hash_utils_64.c b/arch/powerpc/mm/hash_utils_64.c index 5b22ba0..b5b5fe6 100644 --- a/arch/powerpc/mm/hash_utils_64.c +++ b/arch/powerpc/mm/hash_utils_64.c @@ -922,6 +922,8 @@ void __init hash__early_init_mmu(void) vmemmap = (struct page *)H_VMEMMAP_BASE; ioremap_bot = IOREMAP_BASE; + pci_io_base = ISA_IO_BASE; + /* Initialize the MMU Hash table and create the linear mapping * of memory. Has to be done before SLB initialization as this is * currently where the page size encoding is obtained. diff --git a/arch/powerpc/mm/pgtable-radix.c b/arch/powerpc/mm/pgtable-radix.c index e58707d..095fbfa 100644 --- a/arch/powerpc/mm/pgtable-radix.c +++ b/arch/powerpc/mm/pgtable-radix.c @@ -328,6 +328,9 @@ void __init radix__early_init_mmu(void) __vmalloc_end = RADIX_VMALLOC_END; vmemmap = (struct page *)RADIX_VMEMMAP_BASE; ioremap_bot = IOREMAP_BASE; + + pci_io_base = ISA_IO_BASE; + /* * For now radix also use the same frag size */
Commit:d6a9996e84ac4beb7713e9485f4563e100a9b03e (powerpc/mm: vmalloc abstraction in preparation for radix) turned kernel memory and IO addresses from #defined constants to variables initialised at runtime. On PA6T systems the setup_arch machine call initialises the onboard PCI-e root-ports, and uses pci_io_base to do this, which is now before its value has been set resulting in a panic right after 'booting linux via __start()' Move the pci_io_base initialisation to the same place as vmalloc ranges are set (hash__early_init_mmu()/radix__early_init_mmu()) Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de> Signed-off-by: Darren Stevens <darren@stevens-zone.net> --- Tested on my AmigaOneX1000, I don't have access to a refence board system, and our developer with one is on honeymoon. I am hoping to follow this patch with others to reduce the size of the nemo patch we apply, eventually I'd like to see the patch gone, but that is a very big job. Kind regards Darren