| Message ID | 20210604070633.32363-1-wangkefeng.wang@huawei.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | init_mm: cleanup ARCH's text/data/brk setup code | expand |
Hello Kefeng, On Fri, Jun 04, 2021 at 03:06:18PM +0800, Kefeng Wang wrote: > Add setup_initial_init_mm() helper, then use it > to cleanup the text, data and brk setup code. > > v2: > - change argument from "char *" to "void *" setup_initial_init_mm() > suggested by Geert Uytterhoeven > - use NULL instead of (void *)0 on h8300 and m68k > - collect ACKs > > Cc: linux-snps-arc@lists.infradead.org > Cc: linux-arm-kernel@lists.infradead.org > Cc: linux-csky@vger.kernel.org > Cc: uclinux-h8-devel@lists.sourceforge.jp > Cc: linux-m68k@lists.linux-m68k.org > Cc: openrisc@lists.librecores.org > Cc: linuxppc-dev@lists.ozlabs.org > Cc: linux-riscv@lists.infradead.org > Cc: linux-sh@vger.kernel.org > Cc: linux-s390@vger.kernel.org > Kefeng Wang (15): > mm: add setup_initial_init_mm() helper > arc: convert to setup_initial_init_mm() > arm: convert to setup_initial_init_mm() > arm64: convert to setup_initial_init_mm() > csky: convert to setup_initial_init_mm() > h8300: convert to setup_initial_init_mm() > m68k: convert to setup_initial_init_mm() > nds32: convert to setup_initial_init_mm() > nios2: convert to setup_initial_init_mm() > openrisc: convert to setup_initial_init_mm() > powerpc: convert to setup_initial_init_mm() > riscv: convert to setup_initial_init_mm() > s390: convert to setup_initial_init_mm() > sh: convert to setup_initial_init_mm() > x86: convert to setup_initial_init_mm() I might be missing something, but AFAIU the init_mm.start_code and other fields are not used really early so the new setup_initial_init_mm() function can be called in the generic code outside setup_arch(), e.g in mm_init(). > arch/arc/mm/init.c | 5 +---- > arch/arm/kernel/setup.c | 5 +---- > arch/arm64/kernel/setup.c | 5 +---- > arch/csky/kernel/setup.c | 5 +---- > arch/h8300/kernel/setup.c | 5 +---- > arch/m68k/kernel/setup_mm.c | 5 +---- > arch/m68k/kernel/setup_no.c | 5 +---- > arch/nds32/kernel/setup.c | 5 +---- > arch/nios2/kernel/setup.c | 5 +---- > arch/openrisc/kernel/setup.c | 5 +---- > arch/powerpc/kernel/setup-common.c | 5 +---- > arch/riscv/kernel/setup.c | 5 +---- > arch/s390/kernel/setup.c | 5 +---- > arch/sh/kernel/setup.c | 5 +---- > arch/x86/kernel/setup.c | 5 +---- > include/linux/mm_types.h | 8 ++++++++ > 16 files changed, 23 insertions(+), 60 deletions(-) > > -- > 2.26.2 > > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv
On 2021/6/7 5:29, Mike Rapoport wrote: > Hello Kefeng, > > On Fri, Jun 04, 2021 at 03:06:18PM +0800, Kefeng Wang wrote: >> Add setup_initial_init_mm() helper, then use it >> to cleanup the text, data and brk setup code. >> >> v2: >> - change argument from "char *" to "void *" setup_initial_init_mm() >> suggested by Geert Uytterhoeven >> - use NULL instead of (void *)0 on h8300 and m68k >> - collect ACKs >> >> Cc: linux-snps-arc@lists.infradead.org >> Cc: linux-arm-kernel@lists.infradead.org >> Cc: linux-csky@vger.kernel.org >> Cc: uclinux-h8-devel@lists.sourceforge.jp >> Cc: linux-m68k@lists.linux-m68k.org >> Cc: openrisc@lists.librecores.org >> Cc: linuxppc-dev@lists.ozlabs.org >> Cc: linux-riscv@lists.infradead.org >> Cc: linux-sh@vger.kernel.org >> Cc: linux-s390@vger.kernel.org >> Kefeng Wang (15): >> mm: add setup_initial_init_mm() helper >> arc: convert to setup_initial_init_mm() >> arm: convert to setup_initial_init_mm() >> arm64: convert to setup_initial_init_mm() >> csky: convert to setup_initial_init_mm() >> h8300: convert to setup_initial_init_mm() >> m68k: convert to setup_initial_init_mm() >> nds32: convert to setup_initial_init_mm() >> nios2: convert to setup_initial_init_mm() >> openrisc: convert to setup_initial_init_mm() >> powerpc: convert to setup_initial_init_mm() >> riscv: convert to setup_initial_init_mm() >> s390: convert to setup_initial_init_mm() >> sh: convert to setup_initial_init_mm() >> x86: convert to setup_initial_init_mm() > I might be missing something, but AFAIU the init_mm.start_code and other > fields are not used really early so the new setup_initial_init_mm() > function can be called in the generic code outside setup_arch(), e.g in > mm_init(). Hi Mike， each architecture has their own value, not the same, eg m68K and h8300, also the name of the text/code/brk is different in some arch, so I keep unchanged. > >> arch/arc/mm/init.c | 5 +---- >> arch/arm/kernel/setup.c | 5 +---- >> arch/arm64/kernel/setup.c | 5 +---- >> arch/csky/kernel/setup.c | 5 +---- >> arch/h8300/kernel/setup.c | 5 +---- >> arch/m68k/kernel/setup_mm.c | 5 +---- >> arch/m68k/kernel/setup_no.c | 5 +---- >> arch/nds32/kernel/setup.c | 5 +---- >> arch/nios2/kernel/setup.c | 5 +---- >> arch/openrisc/kernel/setup.c | 5 +---- >> arch/powerpc/kernel/setup-common.c | 5 +---- >> arch/riscv/kernel/setup.c | 5 +---- >> arch/s390/kernel/setup.c | 5 +---- >> arch/sh/kernel/setup.c | 5 +---- >> arch/x86/kernel/setup.c | 5 +---- >> include/linux/mm_types.h | 8 ++++++++ >> 16 files changed, 23 insertions(+), 60 deletions(-) >> >> -- >> 2.26.2 >> >> >> _______________________________________________ >> linux-riscv mailing list >> linux-riscv@lists.infradead.org >> http://lists.infradead.org/mailman/listinfo/linux-riscv
Hi Kefeng, Le 07/06/2021 à 02:55, Kefeng Wang a écrit : > > On 2021/6/7 5:29, Mike Rapoport wrote: >> Hello Kefeng, >> >> On Fri, Jun 04, 2021 at 03:06:18PM +0800, Kefeng Wang wrote: >>> Add setup_initial_init_mm() helper, then use it >>> to cleanup the text, data and brk setup code. >>> >>> v2: >>> - change argument from "char *" to "void *" setup_initial_init_mm() >>> suggested by Geert Uytterhoeven >>> - use NULL instead of (void *)0 on h8300 and m68k >>> - collect ACKs >>> >>> Cc: linux-snps-arc@lists.infradead.org >>> Cc: linux-arm-kernel@lists.infradead.org >>> Cc: linux-csky@vger.kernel.org >>> Cc: uclinux-h8-devel@lists.sourceforge.jp >>> Cc: linux-m68k@lists.linux-m68k.org >>> Cc: openrisc@lists.librecores.org >>> Cc: linuxppc-dev@lists.ozlabs.org >>> Cc: linux-riscv@lists.infradead.org >>> Cc: linux-sh@vger.kernel.org >>> Cc: linux-s390@vger.kernel.org >>> Kefeng Wang (15): >>> mm: add setup_initial_init_mm() helper >>> arc: convert to setup_initial_init_mm() >>> arm: convert to setup_initial_init_mm() >>> arm64: convert to setup_initial_init_mm() >>> csky: convert to setup_initial_init_mm() >>> h8300: convert to setup_initial_init_mm() >>> m68k: convert to setup_initial_init_mm() >>> nds32: convert to setup_initial_init_mm() >>> nios2: convert to setup_initial_init_mm() >>> openrisc: convert to setup_initial_init_mm() >>> powerpc: convert to setup_initial_init_mm() >>> riscv: convert to setup_initial_init_mm() >>> s390: convert to setup_initial_init_mm() >>> sh: convert to setup_initial_init_mm() >>> x86: convert to setup_initial_init_mm() >> I might be missing something, but AFAIU the init_mm.start_code and other >> fields are not used really early so the new setup_initial_init_mm() >> function can be called in the generic code outside setup_arch(), e.g in >> mm_init(). > > Hi Mike， each architecture has their own value, not the same, eg m68K and > > h8300, also the name of the text/code/brk is different in some arch, so I keep > > unchanged. What you could do is to define a __weak function that architectures can override and call that function from mm_init() as suggested by Mike, Something like: void __weak setup_initial_init_mm(void) { init_mm.start_code = (unsigned long)_stext; init_mm.end_code = (unsigned long)_etext; init_mm.end_data = (unsigned long)_edata; init_mm.brk = (unsigned long)_end; } Then only the few architecture that are different would override it. I see a few archictectures are usigne PAGE_OFFSET to set .start_code, but it is likely that this is equivalent to _stext. Christophe
On 2021/6/7 13:48, Christophe Leroy wrote: > Hi Kefeng, > > Le 07/06/2021 à 02:55, Kefeng Wang a écrit : >> >> On 2021/6/7 5:29, Mike Rapoport wrote: >>> Hello Kefeng, >>> >>> On Fri, Jun 04, 2021 at 03:06:18PM +0800, Kefeng Wang wrote: >>>> Add setup_initial_init_mm() helper, then use it >>>> to cleanup the text, data and brk setup code. >>>> >>>> v2: >>>> - change argument from "char *" to "void *" setup_initial_init_mm() >>>> suggested by Geert Uytterhoeven >>>> - use NULL instead of (void *)0 on h8300 and m68k >>>> - collect ACKs >>>> >>>> Cc: linux-snps-arc@lists.infradead.org >>>> Cc: linux-arm-kernel@lists.infradead.org >>>> Cc: linux-csky@vger.kernel.org >>>> Cc: uclinux-h8-devel@lists.sourceforge.jp >>>> Cc: linux-m68k@lists.linux-m68k.org >>>> Cc: openrisc@lists.librecores.org >>>> Cc: linuxppc-dev@lists.ozlabs.org >>>> Cc: linux-riscv@lists.infradead.org >>>> Cc: linux-sh@vger.kernel.org >>>> Cc: linux-s390@vger.kernel.org >>>> Kefeng Wang (15): >>>> mm: add setup_initial_init_mm() helper >>>> arc: convert to setup_initial_init_mm() >>>> arm: convert to setup_initial_init_mm() >>>> arm64: convert to setup_initial_init_mm() >>>> csky: convert to setup_initial_init_mm() >>>> h8300: convert to setup_initial_init_mm() >>>> m68k: convert to setup_initial_init_mm() >>>> nds32: convert to setup_initial_init_mm() >>>> nios2: convert to setup_initial_init_mm() >>>> openrisc: convert to setup_initial_init_mm() >>>> powerpc: convert to setup_initial_init_mm() >>>> riscv: convert to setup_initial_init_mm() >>>> s390: convert to setup_initial_init_mm() >>>> sh: convert to setup_initial_init_mm() >>>> x86: convert to setup_initial_init_mm() >>> I might be missing something, but AFAIU the init_mm.start_code and >>> other >>> fields are not used really early so the new setup_initial_init_mm() >>> function can be called in the generic code outside setup_arch(), e.g in >>> mm_init(). >> >> Hi Mike， each architecture has their own value, not the same, eg m68K >> and >> >> h8300, also the name of the text/code/brk is different in some arch, >> so I keep >> >> unchanged. > > What you could do is to define a __weak function that architectures > can override and call that function from mm_init() as suggested by Mike, > > Something like: > > void __weak setup_initial_init_mm(void) > { > init_mm.start_code = (unsigned long)_stext; > init_mm.end_code = (unsigned long)_etext; > init_mm.end_data = (unsigned long)_edata; > init_mm.brk = (unsigned long)_end; > } > > Then only the few architecture that are different would override it. > > I see a few archictectures are usigne PAGE_OFFSET to set .start_code, > but it is likely that this is equivalent to _stext. Yes, the __weak function is option, but the change is only covered 14 archs, there are 7 other archs（alpha hexagon ia64 microblaze mips sparc um xtensa）without related setup code. Also like x86, it has own brk , maybe there are some other different in some arch, so I think let's keep unchanged for now, thanks. > > Christophe > . >
On Mon, Jun 07, 2021 at 07:48:54AM +0200, Christophe Leroy wrote: > Hi Kefeng, > > What you could do is to define a __weak function that architectures can > override and call that function from mm_init() as suggested by Mike, The problem with weak functions is that they bloat the kernel. Each time a weak function is overriden, it becomes dead unreachable code within the kernel image. At some point we're probabily going to have to enable -ffunction-sections to (hopefully) allow the dead code to be discarded.