| Message ID | 20130312225151.2faeb395@kryten (mailing list archive) |
|---|---|
| State | Accepted, archived |
| Headers | show |
Hi, On Tue, Mar 12, 2013 at 10:51:51PM +1100, Anton Blanchard wrote: > Commit 5ac47f7a6efb (powerpc: Relocate prom_init.c on 64bit) made > prom_init.c position independent by manually relocating its entries > in the TOC. > > We get the address of the TOC entries with the __prom_init_toc_start > linker symbol. If __prom_init_toc_start ends up as an entry in the > TOC then we need to add an offset to get the current address. This is > the case for older toolchains. > > On the other hand, if we have a newer toolchain that supports > -mcmodel=medium then __prom_init_toc_start will be created by a > relative offset from r2 (the TOC pointer). Since r2 has already been > relocated, nothing more needs to be done. Adding an offset in this > case is wrong and Aaro Koskinen and Alexander Graf have noticed noticed > G5 and OpenBIOS breakage. > > Alan Modra suggested we just use r2 to get at the TOC which is simpler > and works with both old and new toolchains. > > Reported-by: Alexander Graf <agraf@suse.de> > Signed-off-by: Anton Blanchard <anton@samba.org> > --- > > Thanks Aaro for reporting this, and Alexander for an initial > fix. This tested ok for me with both a new and an old toolchain, > but would appreciate if you could double check it fixes your issues > too. This seems to fix the issue also for me, so: Tested-by: Aaro Koskinen <aaro.koskinen@iki.fi> Thanks, A. > diff --git a/arch/powerpc/kernel/prom_init.c b/arch/powerpc/kernel/prom_init.c > index 7f7fb7f..13f8d16 100644 > --- a/arch/powerpc/kernel/prom_init.c > +++ b/arch/powerpc/kernel/prom_init.c > @@ -2832,11 +2832,13 @@ static void unreloc_toc(void) > { > } > #else > -static void __reloc_toc(void *tocstart, unsigned long offset, > - unsigned long nr_entries) > +static void __reloc_toc(unsigned long offset, unsigned long nr_entries) > { > unsigned long i; > - unsigned long *toc_entry = (unsigned long *)tocstart; > + unsigned long *toc_entry; > + > + /* Get the start of the TOC by using r2 directly. */ > + asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry)); > > for (i = 0; i < nr_entries; i++) { > *toc_entry = *toc_entry + offset; > @@ -2850,8 +2852,7 @@ static void reloc_toc(void) > unsigned long nr_entries = > (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long); > > - /* Need to add offset to get at __prom_init_toc_start */ > - __reloc_toc(__prom_init_toc_start + offset, offset, nr_entries); > + __reloc_toc(offset, nr_entries); > > mb(); > } > @@ -2864,8 +2865,7 @@ static void unreloc_toc(void) > > mb(); > > - /* __prom_init_toc_start has been relocated, no need to add offset */ > - __reloc_toc(__prom_init_toc_start, -offset, nr_entries); > + __reloc_toc(-offset, nr_entries); > } > #endif > #endif > _______________________________________________ > Linuxppc-dev mailing list > Linuxppc-dev@lists.ozlabs.org > https://lists.ozlabs.org/listinfo/linuxppc-dev
On 12.03.2013, at 12:51, Anton Blanchard wrote: > > Commit 5ac47f7a6efb (powerpc: Relocate prom_init.c on 64bit) made > prom_init.c position independent by manually relocating its entries > in the TOC. > > We get the address of the TOC entries with the __prom_init_toc_start > linker symbol. If __prom_init_toc_start ends up as an entry in the > TOC then we need to add an offset to get the current address. This is > the case for older toolchains. > > On the other hand, if we have a newer toolchain that supports > -mcmodel=medium then __prom_init_toc_start will be created by a > relative offset from r2 (the TOC pointer). Since r2 has already been > relocated, nothing more needs to be done. Adding an offset in this > case is wrong and Aaro Koskinen and Alexander Graf have noticed noticed > G5 and OpenBIOS breakage. > > Alan Modra suggested we just use r2 to get at the TOC which is simpler > and works with both old and new toolchains. > > Reported-by: Alexander Graf <agraf@suse.de> > Signed-off-by: Anton Blanchard <anton@samba.org> This fixes my G5 again as well. Acked-by: Alexander Graf <agraf@suse.de> Please make sure this patch goes into 3.9. Alex > --- > > Thanks Aaro for reporting this, and Alexander for an initial > fix. This tested ok for me with both a new and an old toolchain, > but would appreciate if you could double check it fixes your issues > too. > > diff --git a/arch/powerpc/kernel/prom_init.c b/arch/powerpc/kernel/prom_init.c > index 7f7fb7f..13f8d16 100644 > --- a/arch/powerpc/kernel/prom_init.c > +++ b/arch/powerpc/kernel/prom_init.c > @@ -2832,11 +2832,13 @@ static void unreloc_toc(void) > { > } > #else > -static void __reloc_toc(void *tocstart, unsigned long offset, > - unsigned long nr_entries) > +static void __reloc_toc(unsigned long offset, unsigned long nr_entries) > { > unsigned long i; > - unsigned long *toc_entry = (unsigned long *)tocstart; > + unsigned long *toc_entry; > + > + /* Get the start of the TOC by using r2 directly. */ > + asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry)); > > for (i = 0; i < nr_entries; i++) { > *toc_entry = *toc_entry + offset; > @@ -2850,8 +2852,7 @@ static void reloc_toc(void) > unsigned long nr_entries = > (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long); > > - /* Need to add offset to get at __prom_init_toc_start */ > - __reloc_toc(__prom_init_toc_start + offset, offset, nr_entries); > + __reloc_toc(offset, nr_entries); > > mb(); > } > @@ -2864,8 +2865,7 @@ static void unreloc_toc(void) > > mb(); > > - /* __prom_init_toc_start has been relocated, no need to add offset */ > - __reloc_toc(__prom_init_toc_start, -offset, nr_entries); > + __reloc_toc(-offset, nr_entries); > } > #endif > #endif
diff --git a/arch/powerpc/kernel/prom_init.c b/arch/powerpc/kernel/prom_init.c index 7f7fb7f..13f8d16 100644 --- a/arch/powerpc/kernel/prom_init.c +++ b/arch/powerpc/kernel/prom_init.c @@ -2832,11 +2832,13 @@ static void unreloc_toc(void) { } #else -static void __reloc_toc(void *tocstart, unsigned long offset, - unsigned long nr_entries) +static void __reloc_toc(unsigned long offset, unsigned long nr_entries) { unsigned long i; - unsigned long *toc_entry = (unsigned long *)tocstart; + unsigned long *toc_entry; + + /* Get the start of the TOC by using r2 directly. */ + asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry)); for (i = 0; i < nr_entries; i++) { *toc_entry = *toc_entry + offset; @@ -2850,8 +2852,7 @@ static void reloc_toc(void) unsigned long nr_entries = (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long); - /* Need to add offset to get at __prom_init_toc_start */ - __reloc_toc(__prom_init_toc_start + offset, offset, nr_entries); + __reloc_toc(offset, nr_entries); mb(); } @@ -2864,8 +2865,7 @@ static void unreloc_toc(void) mb(); - /* __prom_init_toc_start has been relocated, no need to add offset */ - __reloc_toc(__prom_init_toc_start, -offset, nr_entries); + __reloc_toc(-offset, nr_entries); } #endif #endif
Commit 5ac47f7a6efb (powerpc: Relocate prom_init.c on 64bit) made prom_init.c position independent by manually relocating its entries in the TOC. We get the address of the TOC entries with the __prom_init_toc_start linker symbol. If __prom_init_toc_start ends up as an entry in the TOC then we need to add an offset to get the current address. This is the case for older toolchains. On the other hand, if we have a newer toolchain that supports -mcmodel=medium then __prom_init_toc_start will be created by a relative offset from r2 (the TOC pointer). Since r2 has already been relocated, nothing more needs to be done. Adding an offset in this case is wrong and Aaro Koskinen and Alexander Graf have noticed noticed G5 and OpenBIOS breakage. Alan Modra suggested we just use r2 to get at the TOC which is simpler and works with both old and new toolchains. Reported-by: Alexander Graf <agraf@suse.de> Signed-off-by: Anton Blanchard <anton@samba.org> --- Thanks Aaro for reporting this, and Alexander for an initial fix. This tested ok for me with both a new and an old toolchain, but would appreciate if you could double check it fixes your issues too.