- 29 Sep, 2016 8 commits
-
-
Balbir Singh authored
When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region. This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation. The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region. I've tested the patches lightly at my end. The full solution requires migration of THP pages in the CMA region. That work will be done incrementally on top of this. Signed-off-by:
Balbir Singh <bsingharora@gmail.com> Acked-by:
Alexey Kardashevskiy <aik@ozlabs.ru> [mpe: Merged via powerpc tree as that's where the changes are] Signed-off-by:
Michael Ellerman <mpe@ellerman.id.au>
-
Gavin Shan authored
This supports PCI surprise hotplug. The design is highlighted as below: * The PCI slot's surprise hotplug capability is exposed through device node property "ibm,slot-surprise-pluggable", meaning PCI surprise hotplug will be disabled if skiboot doesn't support it yet. * The interrupt because of presence or link state change is raised on surprise hotplug event. One event is allocated and queued to the PCI slot for workqueue to pick it up and process in serialized fashion. The code flow for surprise hotplug is same to that for managed hotplug except: the affected PEs are put into frozen state to avoid unexpected EEH error reporting in surprise hot remove path. Signed-off-by:Gavin Shan <gwshan@linux.vnet.ibm.com> Signed-off-by:
-
Gavin Shan authored
This removes likely() and unlikely() in pnv_php.c as the code isn't running in hot path. Those macros to affect CPU's branch stream don't help a lot for performance. I used them to identify the cases are likely or unlikely to happen. No logical changes introduced. Signed-off-by:
-
Gavin Shan authored
This unfreezes PE when it's initialized because the PE might be put into frozen state in the last hot remove path. It's not harmful to do so if the PE is already in unfrozen state. Signed-off-by:
-
Gavin Shan authored
This exports eeh_pe_state_mark(). It will be used to mark the surprise hot removed PE as isolated to avoid unexpected EEH error reporting in surprise remove path. Signed-off-by:
-
Gavin Shan authored
This exports @confirm_error_lock so that eeh_serialize_{lock, unlock}() can be used to freeze the affected PE in PCI surprise hot remove path. Signed-off-by: -
Gavin Shan authored
Function eeh_pe_set_option() is used to apply the requested options (enable, disable, unfreeze) in EEH virtualization path. The semantics of this function isn't complete until freezing is supported. This allows to freeze the indicated PE. The new semantics is going to be used in PCI surprise hot remove path, to freeze removed PCI devices (PE) to avoid unexpected EEH error reporting. Signed-off-by:
-
Gavin Shan authored
When issuing PHB reset, OPAL API opal_pci_poll() is called to drive the state machine in OPAL forward. However, we needn't always call the function under some circumstances like reset deassert. This avoids calling opal_pci_poll() when OPAL_SUCCESS is returned from opal_pci_reset(). Except the overhead introduced by additional one unnecessary OPAL call, I didn't run into real issue because of this. Reported-by:
Pridhiviraj Paidipeddi <ppaiddipe@in.ibm.com> Signed-off-by:
-
- 28 Sep, 2016 6 commits
-
-
Oliver O'Halloran authored
This patch adds an option to use XZ compression for the kernel image. Currently this is only enabled for 64-bit Book3S targets, which is roughly equivalent to the platforms that use the kernel's zImage wrapper, and that have been tested. The bulk of the 32-bit platforms and 64-bit BookE use uboot images, which relies on uboot implementing XZ. In future we can enable XZ support for those targets once someone has tested it. Signed-off-by:
Oliver O'Halloran <oohall@gmail.com> Signed-off-by:
-
Oliver O'Halloran authored
This modifies the wrapper script so that the -Z option takes an argument to specify the compression type. It can either be 'gz', 'xz' or 'none'. The legazy --no-gzip and -z options are still supported and will set the compression to none and gzip respectively, but they are not documented. Only XZ -6 is used for compression rather than XZ -9. Using compression levels higher than 6 requires the decompressor to build a large (64MB) dictionary when decompressing and some environments cannot satisfy such large allocations (e.g. POWER 6 LPAR partition firmware). Signed-off-by:
-
Oliver O'Halloran authored
This code is no longer used and can be removed. Signed-off-by:
-
Oliver O'Halloran authored
Currently the powerpc boot wrapper has its own wrapper around zlib to handle decompressing gzipped kernels. The kernel decompressor library functions now provide a generic interface that can be used in the pre-boot environment. This allows boot wrappers to easily support different compression algorithms. This patch converts the wrapper to use this new API, but does not add support for using new algorithms. Signed-off-by:
-
Oliver O'Halloran authored
Most architectures allow the compression algorithm used to produced the vmlinuz image to be selected as a kernel config option. In preperation for supporting algorithms other than gzip in the powerpc boot wrapper the makefile needs to be modified to use these config options. Signed-off-by:
-
Oliver O'Halloran authored
The powerpc boot wrapper is potentially compiled with a separate toolchain and/or toolchain flags than the rest of the kernel. The usual case is a 64-bit big endian kernel builds a 32-bit big endian wrapper. The main problem with this is that the wrapper does not have access to the kernel headers (without a lot of gross hacks). To get around this the required headers are copied into the build directory via several sed scripts which rewrite problematic includes. This patch moves these fixups out of the makefile into a separate .sed script file to clean up makefile slightly. Signed-off-by:
-
- 23 Sep, 2016 1 commit
-
-
Cyril Bur authored
It might be nice to compile selftests against older kernels and headers but which may not have HWCAP2. Signed-off-by:
Cyril Bur <cyrilbur@gmail.com> Signed-off-by:
-
- 22 Sep, 2016 15 commits
-
-
Rui Teng authored
The same logic appears twice and should probably be pulled out into a function. Suggested-by:
Rui Teng <rui.teng@linux.vnet.ibm.com> [mpe: Rename to tm_flush_hash_page() and move comment into the function] Signed-off-by:
-
Andrew Donnellan authored
Signed-off-by:
Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by:
-
Christophe Leroy authored
CLR_TOP32() is defined as blank. Last useful instance of CLR_TOP32() was removed by commit 40ef8cbc ("powerpc: Get 64-bit configs to compile with ARCH=powerpc") in 2005. Signed-off-by:
Christophe Leroy <christophe.leroy@c-s.fr> Signed-off-by:
-
Christophe Leroy authored
On some CPUs like the 8xx, _PAGE_RW hence _PAGE_WRITE is defined as 0 and _PAGE_RO has to be set when a page is not writable _PAGE_RO is defined by default in pte-common.h, however BOOK3S/64 doesn't include that file so _PAGE_RO has to be defined explicitly in book3s/64/pgtable.h Fixes: a7b9f671 ("powerpc32: adds handling of _PAGE_RO") Signed-off-by:
Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Signed-off-by:
-
Russell Currey authored
In eeh_handle_special_event(), eeh_pe_bus_get() is called before calling eeh_report_failure() on every device under a PE. If a PE was missing a bus for some reason, the error would occur before reporting failure, even though eeh_report_failure() doesn't require a bus. Fix this by moving the bus retrieval and error check after the eeh_report_failure() calls. Signed-off-by:
Russell Currey <ruscur@russell.cc> Reviewed-by:
-
Russell Currey authored
When the PE used in pnv_eeh_reset() is that of a VF, pnv_eeh_reset_vf_pe() is used. Unlike the other reset functions called in pnv_eeh_reset(), the VF reset doesn't require a bus, and if a bus was missing the function would error out before resetting the VF PE. To avoid this, reorder the VF reset function to occur before finding and checking the bus. Signed-off-by:
-
Russell Currey authored
eeh_pe_bus_get() can return NULL if a PCI bus isn't found for a given PE. Some callers don't check this, and can cause a null pointer dereference under certain circumstances. Fix this by checking NULL everywhere eeh_pe_bus_get() is called. Fixes: 8a6b1bc7 ("powerpc/eeh: EEH core to handle special event") Cc: stable@vger.kernel.org # v3.11+ Signed-off-by:
-
Nicholas Piggin authored
When we originally added the ability to split the exception vectors from the kernel (commit 1f6a93e4 ("powerpc: Make it possible to move the interrupt handlers away from the kernel" 2008-09-15)), the LOAD_HANDLER() macro used an addi instruction to compute the offset of the common handler from the kernel base address. Using addi meant the handler had to be within 32K of the kernel base address, due to the addi instruction taking a signed immediate value. That necessitated creating a trampoline for the system call handler, because system_call_common (in entry64.S) is not linked within 32K of the kernel base address. Later in commit 61e2390e ("powerpc: Make load_hander handle upto 64k offset" 2012-11-15) we changed LOAD_HANDLER to take a 64K offset, by changing it to use ori. Although system_call_common is not in head_64.S or exceptions-64s.S, it is included in head-y, which causes it to be linked early in the kernel text, so in practice it ends up below 64K. Additionally if it can't be placed below 64K the linker will fail to build with a "relocation truncated to fit" error. So remove the trampoline. Newer toolchains are able to work out that the ori in LOAD_HANDLER only takes a 16 bit offset, and so they generate a 16 bit relocation. Older toolchains (binutils 2.22 at least) are not so smart, so we have to add the @l annotation to tell the assembler to generate a 16 bit relocation. Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> Signed-off-by:
-
Nicholas Piggin authored
The 0xf80 hv_facility_unavailable trampoline branches to the 0xf60 handler. This works because they both do the same thing, but it should be fixed. Signed-off-by:
-
Russell Currey authored
On EEH events the kernel will print a dump of relevant registers. If EEH is unavailable (i.e. CONFIG_EEH is disabled, a new platform doesn't have EEH support, etc) this information isn't readily available. Add a new debugfs handler to trigger a PHB register dump, so that this information can be made available on demand. Signed-off-by:
-
Benjamin Herrenschmidt authored
The only difference is now the TCE table check which doesn't need to be ifdef'ed out, it will basically do nothing on BookE (it is only useful for ancient IBM machines). Signed-off-by:
Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by:
-
Benjamin Herrenschmidt authored
Currently we turn the MMU off after copying the image, and we make sure there is no overlap between the hash table and the target pages in that case. That doesn't work for Radix however. In that case, the page tables are scattered and we can't really enforce that the target of the image isn't overlapping one of them. So instead, let's turn the MMU off before copying the image in radix mode. Thankfully, in radix mode, even under a hypervisor, we know we don't have the same kind of RMA limitations that hash mode has. While at it, also turn the MMU off early when using hash in non-LPAR mode, that way we can get rid of the collision check completely. Signed-off-by:
-
Aneesh Kumar K.V authored
Signed-off-by:
-
Benjamin Herrenschmidt authored
Just using the hash ops won't work anymore since radix will have NULL in there. Instead create an mmu_cleanup_all() function which will do the right thing based on the MMU mode. For Radix, for now I clear UPRT and the PTCR, effectively switching back to Radix with no partition table setup. Currently set it to NULL on BookE thought it might be a good idea to wipe the TLB there (Scott ?) Signed-off-by:
-
Benjamin Herrenschmidt authored
With Radix, it can be NULL even on !BOOKE these days so replace the ifdef with a NULL check which is cleaner anyway. Signed-off-by:
-
- 20 Sep, 2016 10 commits
-
-
Michael Ellerman authored
NO_IRQ has been == 0 on powerpc for just over ten years (since commit 0ebfff14 ("[POWERPC] Add new interrupt mapping core and change platforms to use it")). It's also 0 on most other arches. Although it's fairly harmless, every now and then it causes confusion when a driver is built on powerpc and another arch which doesn't define NO_IRQ. There's at least 6 definitions of NO_IRQ in drivers/, at least some of which are to work around that problem. So we'd like to remove it. This is fairly trivial in the arch code, we just convert: if (irq == NO_IRQ) to if (!irq) if (irq != NO_IRQ) to if (irq) irq = NO_IRQ; to irq = 0; return NO_IRQ; to return 0; And a few other odd cases as well. At least for now we keep the #define NO_IRQ, because there is driver code that uses NO_IRQ and the fixes to remove those will go via other trees. Note we also change some occurrences in PPC sound drivers, drivers/ps3, and drivers/macintosh. Signed-off-by:
-
Michael Neuling authored
Fred has taken over the cxl maintenance I was doing. This updates the MAINTAINERS file to reflect this. It also removes a duplicate entry in the files covered and adds an entry for the CXL PCI code in arch/powerpc. Signed-off-by:
Michael Neuling <mikey@neuling.org> Reviewed-by:
Frederic Barrat <fbarrat@linux.vnet.ibm.com> Signed-off-by:
-
Andrew Donnellan authored
If we fail to allocate work, we don't end up using hp_errlog_copy. Free it in the error path. Signed-off-by:
Nathan Fontenot <nfont@linux.vnet.ibm.com> Signed-off-by:
-
Pan Xinhui authored
When we merge two contiguous partitions whose signatures are marked NVRAM_SIG_FREE, We need update prev's length and checksum, then write it to nvram, not cur's. So lets fix this mistake now. Also use memset instead of strncpy to set the partition's name. It's more readable if we want to fill up with duplicate chars . Fixes: fa2b4e54 ("powerpc/nvram: Improve partition removal") Signed-off-by:
Pan Xinhui <xinhui.pan@linux.vnet.ibm.com> Signed-off-by:
-
Pan Xinhui authored
If kmemdup fails, We need kfree *buff* first then return -ENOMEM. Otherwise there is a memory leak. Signed-off-by:
-
Nicholas Piggin authored
mtmsrd with L=1 only affects MSR_EE and MSR_RI bits, and we always know what state those bits are, so the kernel MSR does not need to be loaded when modifying them. mtmsrd is often in the critical execution path, so avoiding dependency on even L1 load is noticable. On a POWER8 this saves about 3 cycles from the syscall path, and possibly a few from other exception returns (not measured). Signed-off-by:
-
Nicholas Piggin authored
The mflr r10 instruction was left over from when the code used LR to branch to system_call_entry from the exception handler. That was changed by commit 6a404806 ("powerpc: Avoid link stack corruption in MMU on syscall entry path") to use the count register. The value is never used now, so mflr can be removed, and r10 can be used for storage rather than spilling to the SPR scratch register. The scratch register spill causes a long pipeline stall due to the SPR read after write. This change brings getppid syscall cost from 406 to 376 cycles on POWER8. getppid for non-relocatable case is 371 cycles. Signed-off-by:
-
Aneesh Kumar K.V authored
For hugetlb to work with 4K page size, we need MAX_ORDER to be 13 or more. When switching from a 64K page size to 4K linux page size using make oldconfig, we end up with a CONFIG_FORCE_MAX_ZONEORDER value of 9. This results in a 16M hugepage beiing considered as a gigantic huge page which in turn results in failure to setup hugepages if gigantic hugepage support is not enabled. This also results in kernel crash with 4K radix configuration. We hit the below BUG_ON on radix: kernel BUG at mm/huge_memory.c:364! Oops: Exception in kernel mode, sig: 5 [#1] SMP NR_CPUS=2048 NUMA PowerNV CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.8.0-rc1-00006-gbae9cc6 #1 task: c0000000f1af8000 task.stack: c0000000f1aec000 NIP: c000000000c5fa0c LR: c000000000c5f9d8 CTR: c000000000c5f9a4 REGS: c0000000f1aef920 TRAP: 0700 Not tainted (4.8.0-rc1-00006-gbae9cc6) MSR: 9000000102029033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE,TM[E]> CR: 24000844 XER: 00000000 CFAR: c000000000c5f9e0 SOFTE: 1 .... NIP [c000000000c5fa0c] hugepage_init+0x68/0x238 LR [c000000000c5f9d8] hugepage_init+0x34/0x238 Fixes: a7ee5395 ("powerpc/Kconfig: Update config option based on page size") Cc: stable@vger.kernel.org # v4.7+ Reported-by:
Santhosh <santhog4@linux.vnet.ibm.com> Signed-off-by:
-
Nicholas Piggin authored
Add some documentation for the 64-bit syscall ABI, which doesn't seem to be documented elsewhere. This attempts to document existing practice. The only small discrepancy is glibc clobbers not quite matching the kernel (e.g., xer, some vsyscalls trash cr1 whereas glibc only clobbers cr0). These will be resolved after this document is merged. Signed-off-by:
-
Nicholas Piggin authored
The HMI (Hypervisor Maintenance Interrupt) is defined by the architecture to be higher priority than other maskable interrupts, so replay it first, as a best-effort to replay according to hardware priorities. Signed-off-by:
-
