| Message ID | 160402498564.4173389.2743697400148832021.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | x86/mm: Fix phys_to_target_node() export | expand |
On 10/29/20 7:29 PM, Dan Williams wrote: > The core-mm has a default __weak implementation of phys_to_target_node() > when the architecture does not override it. That symbol is exported > for modules. However, while the export in mm/memory_hotplug.c exported > the symbol in the configuration cases of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=y > > ...and: > > CONFIG_NUMA_KEEP_MEMINFO=n > CONFIG_MEMORY_HOTPLUG=y > > ...it failed to export the symbol in the case of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=n > > Always export the symbol from the CONFIG_NUMA_KEEP_MEMINFO section of > arch/x86/mm/numa.c, and teach mm/memory_hotplug.c to optionally export > in case arch/x86/mm/numa.c has already performed the export. > > The dependency on NUMA_KEEP_MEMINFO for DEV_DAX_HMEM_DEVICES is invalid > now that the symbol is properly exported in all combinations of > CONFIG_NUMA_KEEP_MEMINFO and CONFIG_MEMORY_HOTPLUG. Note that in the > CONFIG_NUMA=n case no export is needed since their is a dummy static > inline implementation of phys_to_target_node() in that case. > > Reported-by: Randy Dunlap <rdunlap@infradead.org> > Reported-by: Thomas Gleixner <tglx@linutronix.de> > Reported-by: kernel test robot <lkp@intel.com> > Fixes: a035b6bf863e ("mm/memory_hotplug: introduce default phys_to_target_node() implementation") > Cc: Joao Martins <joao.m.martins@oracle.com> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: x86@kernel.org > Cc: Vishal Verma <vishal.l.verma@intel.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> Looks good. Thanks. Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested > --- > arch/x86/mm/numa.c | 1 + > drivers/dax/Kconfig | 1 - > mm/memory_hotplug.c | 5 +++++ > 3 files changed, 6 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c > index 44148691d78b..e025947f19e0 100644 > --- a/arch/x86/mm/numa.c > +++ b/arch/x86/mm/numa.c > @@ -938,6 +938,7 @@ int phys_to_target_node(phys_addr_t start) > > return meminfo_to_nid(&numa_reserved_meminfo, start); > } > +EXPORT_SYMBOL_GPL(phys_to_target_node); > > int memory_add_physaddr_to_nid(u64 start) > { > diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig > index 567428e10b7b..d2834c2cfa10 100644 > --- a/drivers/dax/Kconfig > +++ b/drivers/dax/Kconfig > @@ -50,7 +50,6 @@ config DEV_DAX_HMEM > Say M if unsure. > > config DEV_DAX_HMEM_DEVICES > - depends on NUMA_KEEP_MEMINFO # for phys_to_target_node() > depends on DEV_DAX_HMEM && DAX=y > def_bool y > > diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c > index b44d4c7ba73b..ed326b489674 100644 > --- a/mm/memory_hotplug.c > +++ b/mm/memory_hotplug.c > @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) > start); > return 0; > } > + > +/* If the arch did not export a strong symbol, export the weak one. */ > +#ifndef CONFIG_NUMA_KEEP_MEMINFO > EXPORT_SYMBOL_GPL(phys_to_target_node); > #endif > > +#endif > + > /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ > static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, > unsigned long start_pfn, > >
On 30.10.20 03:29, Dan Williams wrote: > The core-mm has a default __weak implementation of phys_to_target_node() > when the architecture does not override it. That symbol is exported > for modules. However, while the export in mm/memory_hotplug.c exported > the symbol in the configuration cases of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=y > > ...and: > > CONFIG_NUMA_KEEP_MEMINFO=n > CONFIG_MEMORY_HOTPLUG=y > > ...it failed to export the symbol in the case of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=n > > Always export the symbol from the CONFIG_NUMA_KEEP_MEMINFO section of > arch/x86/mm/numa.c, and teach mm/memory_hotplug.c to optionally export > in case arch/x86/mm/numa.c has already performed the export. > > The dependency on NUMA_KEEP_MEMINFO for DEV_DAX_HMEM_DEVICES is invalid > now that the symbol is properly exported in all combinations of > CONFIG_NUMA_KEEP_MEMINFO and CONFIG_MEMORY_HOTPLUG. Note that in the > CONFIG_NUMA=n case no export is needed since their is a dummy static > inline implementation of phys_to_target_node() in that case. > > Reported-by: Randy Dunlap <rdunlap@infradead.org> > Reported-by: Thomas Gleixner <tglx@linutronix.de> > Reported-by: kernel test robot <lkp@intel.com> > Fixes: a035b6bf863e ("mm/memory_hotplug: introduce default phys_to_target_node() implementation") > Cc: Joao Martins <joao.m.martins@oracle.com> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: x86@kernel.org > Cc: Vishal Verma <vishal.l.verma@intel.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > arch/x86/mm/numa.c | 1 + > drivers/dax/Kconfig | 1 - > mm/memory_hotplug.c | 5 +++++ > 3 files changed, 6 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c > index 44148691d78b..e025947f19e0 100644 > --- a/arch/x86/mm/numa.c > +++ b/arch/x86/mm/numa.c > @@ -938,6 +938,7 @@ int phys_to_target_node(phys_addr_t start) > > return meminfo_to_nid(&numa_reserved_meminfo, start); > } > +EXPORT_SYMBOL_GPL(phys_to_target_node); > > int memory_add_physaddr_to_nid(u64 start) > { > diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig > index 567428e10b7b..d2834c2cfa10 100644 > --- a/drivers/dax/Kconfig > +++ b/drivers/dax/Kconfig > @@ -50,7 +50,6 @@ config DEV_DAX_HMEM > Say M if unsure. > > config DEV_DAX_HMEM_DEVICES > - depends on NUMA_KEEP_MEMINFO # for phys_to_target_node() > depends on DEV_DAX_HMEM && DAX=y > def_bool y > > diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c > index b44d4c7ba73b..ed326b489674 100644 > --- a/mm/memory_hotplug.c > +++ b/mm/memory_hotplug.c > @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) > start); > return 0; > } > + > +/* If the arch did not export a strong symbol, export the weak one. */ > +#ifndef CONFIG_NUMA_KEEP_MEMINFO > EXPORT_SYMBOL_GPL(phys_to_target_node); > #endif > > +#endif > + > /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ > static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, > unsigned long start_pfn, > > Reviewed-by: David Hildenbrand <david@redhat.com>
On Thu, Oct 29, 2020 at 7:29 PM Dan Williams <dan.j.williams@intel.com> wrote: > > The core-mm has a default __weak implementation of phys_to_target_node() > when the architecture does not override it. That symbol is exported > for modules. However, while the export in mm/memory_hotplug.c exported > the symbol in the configuration cases of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=y > > ...and: > > CONFIG_NUMA_KEEP_MEMINFO=n > CONFIG_MEMORY_HOTPLUG=y > > ...it failed to export the symbol in the case of: > > CONFIG_NUMA_KEEP_MEMINFO=y > CONFIG_MEMORY_HOTPLUG=n > > Always export the symbol from the CONFIG_NUMA_KEEP_MEMINFO section of > arch/x86/mm/numa.c, and teach mm/memory_hotplug.c to optionally export > in case arch/x86/mm/numa.c has already performed the export. > > The dependency on NUMA_KEEP_MEMINFO for DEV_DAX_HMEM_DEVICES is invalid > now that the symbol is properly exported in all combinations of > CONFIG_NUMA_KEEP_MEMINFO and CONFIG_MEMORY_HOTPLUG. Note that in the > CONFIG_NUMA=n case no export is needed since their is a dummy static > inline implementation of phys_to_target_node() in that case. > > Reported-by: Randy Dunlap <rdunlap@infradead.org> > Reported-by: Thomas Gleixner <tglx@linutronix.de> > Reported-by: kernel test robot <lkp@intel.com> > Fixes: a035b6bf863e ("mm/memory_hotplug: introduce default phys_to_target_node() implementation") > Cc: Joao Martins <joao.m.martins@oracle.com> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: x86@kernel.org > Cc: Vishal Verma <vishal.l.verma@intel.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> Thomas, do you want to ack this so Andrew can pick it up, or I can take it through as a device-dax update, but either way the diffstat warrants x86 + mm acks.
On Fri, Oct 30 2020 at 18:54, Dan Williams wrote: > Thomas, do you want to ack this so Andrew can pick it up, or I can > take it through as a device-dax update, but either way the diffstat > warrants x86 + mm acks. It's butt ugly but I couldn't come up with anything better right away. So, FWIW: Acked-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Thomas Gleixner <tglx@linutronix.de>
On Fri, Oct 30, 2020 at 8:02 PM Thomas Gleixner <tglx@linutronix.de> wrote: > > On Fri, Oct 30 2020 at 18:54, Dan Williams wrote: > > Thomas, do you want to ack this so Andrew can pick it up, or I can > > take it through as a device-dax update, but either way the diffstat > > warrants x86 + mm acks. > > It's butt ugly but I couldn't come up with anything better right > away. So, FWIW: > > Acked-by: Thomas Gleixner <tglx@linutronix.de> > Tested-by: Thomas Gleixner <tglx@linutronix.de> Almost-threw-up-in-my-mouth-a-little-bit-by?
On Thu, Oct 29, 2020 at 07:29:45PM -0700, Dan Williams wrote: > The core-mm has a default __weak implementation of phys_to_target_node() > when the architecture does not override it. That symbol is exported > for modules. However, while the export in mm/memory_hotplug.c exported > the symbol in the configuration cases of: Which just means that we should never export weak symbols. So instead of hacking around this introduce a symbol that indicates that the architecture impements phys_to_target_node, and don't defined it at all in common code for that case. > --- a/mm/memory_hotplug.c > +++ b/mm/memory_hotplug.c > @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) > start); > return 0; > } > + > +/* If the arch did not export a strong symbol, export the weak one. */ > +#ifndef CONFIG_NUMA_KEEP_MEMINFO > EXPORT_SYMBOL_GPL(phys_to_target_node); > #endif > > +#endif i.e. move the ifdef to include the actual phys_to_target_node definition, and remove the __weak from it here.
On Sat, Oct 31, 2020 at 2:10 AM Christoph Hellwig <hch@infradead.org> wrote: > > On Thu, Oct 29, 2020 at 07:29:45PM -0700, Dan Williams wrote: > > The core-mm has a default __weak implementation of phys_to_target_node() > > when the architecture does not override it. That symbol is exported > > for modules. However, while the export in mm/memory_hotplug.c exported > > the symbol in the configuration cases of: > > Which just means that we should never export weak symbols. So instead > of hacking around this introduce a symbol that indicates that the > architecture impements phys_to_target_node, and don't defined it at all > in common code for that case. So I agree with this, but it made me realize that the way memory_add_physaddr_to_nid() was defined as an exported weak symbol is similarly broken. > > --- a/mm/memory_hotplug.c > > +++ b/mm/memory_hotplug.c > > @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) > > start); > > return 0; > > } > > + > > +/* If the arch did not export a strong symbol, export the weak one. */ > > +#ifndef CONFIG_NUMA_KEEP_MEMINFO > > EXPORT_SYMBOL_GPL(phys_to_target_node); > > #endif > > > > +#endif > > i.e. move the ifdef to include the actual phys_to_target_node > definition, and remove the __weak from it here. The trick is finding an arch common way to pick up the presence of the phys_to_target_node() override, and it still has the wart of ifdefery in C code. I went a bit deeper and moved all the fallback routines to linux/numa.h and the overrides in all archs that care to asm/sparsemem.h. Note that asm/sparsemem.h was not my first choice, but it happened to be where powerpc was already defining its phys-addr-to-node-id infrastructure, and my first choice header, asm/numa.h, is not universally available. The attached patch is going through some kbuild-robot exposure to make sure I did not break anything else.
On 11/2/20 3:52 PM, Dan Williams wrote: > On Sat, Oct 31, 2020 at 2:10 AM Christoph Hellwig <hch@infradead.org> wrote: >> >> On Thu, Oct 29, 2020 at 07:29:45PM -0700, Dan Williams wrote: >>> The core-mm has a default __weak implementation of phys_to_target_node() >>> when the architecture does not override it. That symbol is exported >>> for modules. However, while the export in mm/memory_hotplug.c exported >>> the symbol in the configuration cases of: >> >> Which just means that we should never export weak symbols. So instead >> of hacking around this introduce a symbol that indicates that the >> architecture impements phys_to_target_node, and don't defined it at all >> in common code for that case. > > So I agree with this, but it made me realize that the way > memory_add_physaddr_to_nid() was defined as an exported weak symbol is > similarly broken. > >>> --- a/mm/memory_hotplug.c >>> +++ b/mm/memory_hotplug.c >>> @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) >>> start); >>> return 0; >>> } >>> + >>> +/* If the arch did not export a strong symbol, export the weak one. */ >>> +#ifndef CONFIG_NUMA_KEEP_MEMINFO >>> EXPORT_SYMBOL_GPL(phys_to_target_node); >>> #endif >>> >>> +#endif >> >> i.e. move the ifdef to include the actual phys_to_target_node >> definition, and remove the __weak from it here. > > The trick is finding an arch common way to pick up the presence of the > phys_to_target_node() override, and it still has the wart of ifdefery > in C code. > > I went a bit deeper and moved all the fallback routines to > linux/numa.h and the overrides in all archs that care to > asm/sparsemem.h. Note that asm/sparsemem.h was not my first choice, > but it happened to be where powerpc was already defining its > phys-addr-to-node-id infrastructure, and my first choice header, > asm/numa.h, is not universally available. > > The attached patch is going through some kbuild-robot exposure to make > sure I did not break anything else. > Works for me. Thanks. Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
This version looks sensible to me:
Reviewed-by: Christoph Hellwig <hch@lst.de>
On Mon, Nov 02 2020 at 15:52, Dan Williams wrote: > On Sat, Oct 31, 2020 at 2:10 AM Christoph Hellwig <hch@infradead.org> wrote: > The dependency on NUMA_KEEP_MEMINFO for DEV_DAX_HMEM_DEVICES is invalid > now that the symbol is properly exported / stubbed in all combinations > of CONFIG_NUMA_KEEP_MEMINFO and CONFIG_MEMORY_HOTPLUG. > > Reported-by: Randy Dunlap <rdunlap@infradead.org> > Reported-by: Thomas Gleixner <tglx@linutronix.de> > Reported-by: kernel test robot <lkp@intel.com> > Reported-by: Christoph Hellwig <hch@infradead.org> > Fixes: a035b6bf863e ("mm/memory_hotplug: introduce default phys_to_target_node() implementation") > Cc: Joao Martins <joao.m.martins@oracle.com> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: x86@kernel.org > Cc: Tony Luck <tony.luck@intel.com> > Cc: Fenghua Yu <fenghua.yu@intel.com> > Cc: Michael Ellerman <mpe@ellerman.id.au> > Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> > Cc: Paul Mackerras <paulus@samba.org> > Cc: Vishal Verma <vishal.l.verma@intel.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> Tested-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
On Mon, 2 Nov 2020 15:52:39 -0800 Dan Williams <dan.j.williams@intel.com> wrote: > The attached patch is going through some kbuild-robot exposure to make > sure I did not break anything else. I'll duck this for now - please send it along formally if/when testing is successful.
On Tue, Nov 3, 2020 at 5:38 PM Andrew Morton <akpm@linux-foundation.org> wrote: > > On Mon, 2 Nov 2020 15:52:39 -0800 Dan Williams <dan.j.williams@intel.com> wrote: > > > The attached patch is going through some kbuild-robot exposure to make > > sure I did not break anything else. > > I'll duck this for now - please send it along formally if/when testing > is successful. Yeah, the robots are angry, some reworks needed.
diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c index 44148691d78b..e025947f19e0 100644 --- a/arch/x86/mm/numa.c +++ b/arch/x86/mm/numa.c @@ -938,6 +938,7 @@ int phys_to_target_node(phys_addr_t start) return meminfo_to_nid(&numa_reserved_meminfo, start); } +EXPORT_SYMBOL_GPL(phys_to_target_node); int memory_add_physaddr_to_nid(u64 start) { diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig index 567428e10b7b..d2834c2cfa10 100644 --- a/drivers/dax/Kconfig +++ b/drivers/dax/Kconfig @@ -50,7 +50,6 @@ config DEV_DAX_HMEM Say M if unsure. config DEV_DAX_HMEM_DEVICES - depends on NUMA_KEEP_MEMINFO # for phys_to_target_node() depends on DEV_DAX_HMEM && DAX=y def_bool y diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index b44d4c7ba73b..ed326b489674 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -365,9 +365,14 @@ int __weak phys_to_target_node(u64 start) start); return 0; } + +/* If the arch did not export a strong symbol, export the weak one. */ +#ifndef CONFIG_NUMA_KEEP_MEMINFO EXPORT_SYMBOL_GPL(phys_to_target_node); #endif +#endif + /* find the smallest valid pfn in the range [start_pfn, end_pfn) */ static unsigned long find_smallest_section_pfn(int nid, struct zone *zone, unsigned long start_pfn,
The core-mm has a default __weak implementation of phys_to_target_node() when the architecture does not override it. That symbol is exported for modules. However, while the export in mm/memory_hotplug.c exported the symbol in the configuration cases of: CONFIG_NUMA_KEEP_MEMINFO=y CONFIG_MEMORY_HOTPLUG=y ...and: CONFIG_NUMA_KEEP_MEMINFO=n CONFIG_MEMORY_HOTPLUG=y ...it failed to export the symbol in the case of: CONFIG_NUMA_KEEP_MEMINFO=y CONFIG_MEMORY_HOTPLUG=n Always export the symbol from the CONFIG_NUMA_KEEP_MEMINFO section of arch/x86/mm/numa.c, and teach mm/memory_hotplug.c to optionally export in case arch/x86/mm/numa.c has already performed the export. The dependency on NUMA_KEEP_MEMINFO for DEV_DAX_HMEM_DEVICES is invalid now that the symbol is properly exported in all combinations of CONFIG_NUMA_KEEP_MEMINFO and CONFIG_MEMORY_HOTPLUG. Note that in the CONFIG_NUMA=n case no export is needed since their is a dummy static inline implementation of phys_to_target_node() in that case. Reported-by: Randy Dunlap <rdunlap@infradead.org> Reported-by: Thomas Gleixner <tglx@linutronix.de> Reported-by: kernel test robot <lkp@intel.com> Fixes: a035b6bf863e ("mm/memory_hotplug: introduce default phys_to_target_node() implementation") Cc: Joao Martins <joao.m.martins@oracle.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: x86@kernel.org Cc: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- arch/x86/mm/numa.c | 1 + drivers/dax/Kconfig | 1 - mm/memory_hotplug.c | 5 +++++ 3 files changed, 6 insertions(+), 1 deletion(-)