| Message ID | 20240313125711.20651-1-piliu@redhat.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | arm64: mm: Use __create_pgd_mapping_locked() in | expand |
Hello Pingfan, On Wed, 13 Mar 2024 at 13:57, Pingfan Liu <piliu@redhat.com> wrote: > > Hi everybody, I tried this stuff again. Tried what again? Frankly, I have no idea what the purpose of this patch series is, and this is v1. Could you please explain? Also, the early arm64 startup code is changing substantially - please refer to https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git/log/?h=for-next/stage1-lpa2 for details. > Last time when I tried this, > Catalin raised concern about the intrumentation, and Ard doubted this > way due to alignement issue with mmu-off. > > Last time, the alignment issue looked unsoluable and I gave up. But > nowadays, when I looked at it, I think it is partially resovable. (for > detail, please see the commit log in [PATCH 08/10] arm64: mm: Enforce > memory alignment in mmu_head) > > Overall, at this very early stage, the using of C routines faces three > challenge: > PIC > instrumentation > alignment > > [2/10] resolves instrumentation issue > > [3/10] makes mmu_head self-contained and prevent the outside > PIC/ instrumentation/ alignment issues from seeping in. And check the code PIC. > > [PATCH 08/10] explains the alignement issue, in theory, it can be > checked and resolved. And in this patch, it is partially resolved. > > > Cc: Ard Biesheuvel <ardb@kernel.org> > Cc: Catalin Marinas <catalin.marinas@arm.com> > Cc: Will Deacon <will@kernel.org> > Cc: Mark Rutland <mark.rutland@arm.com> > To: linux-arm-kernel@lists.infradead.org > --- > > Pingfan Liu (10): > arm64: mm: Split out routines for code reuse > arm64: mm: Introduce mmu_head routines without instrumentation > arm64: mm: Use if-conditon to truncate external dependency > arm64: head: Enable __create_pgd_mapping() to handle pgtable's paddr > arm64: mm: Force early mapping aligned on SWAPPER_BLOCK_SIZE > arm64: mm: Handle scope beyond the capacity of kernel pgtable in > mmu_head_create_pgd_mapping() > arm64: mm: Introduce head_pool routines to enable pgtabl allocation > arm64: mm: Enforce memory alignment in mmu_head > arm64: head: Use __create_pgd_mapping_locked() to serve the creation > of pgtable > arm64: head: Clean up unneeded routines > > arch/arm64/include/asm/kernel-pgtable.h | 1 + > arch/arm64/include/asm/mmu.h | 4 + > arch/arm64/include/asm/pgtable.h | 11 +- > arch/arm64/kernel/head.S | 314 +++++++----------------- > arch/arm64/mm/Makefile | 22 +- > arch/arm64/mm/mmu.c | 289 +--------------------- > arch/arm64/mm/mmu_head.c | 134 ++++++++++ > arch/arm64/mm/mmu_inc.c | 292 ++++++++++++++++++++++ > 8 files changed, 558 insertions(+), 509 deletions(-) > create mode 100644 arch/arm64/mm/mmu_head.c > create mode 100644 arch/arm64/mm/mmu_inc.c > > -- > 2.41.0 >
On Wed, Mar 13, 2024 at 9:05 PM Ard Biesheuvel <ardb@kernel.org> wrote: > > Hello Pingfan, > > On Wed, 13 Mar 2024 at 13:57, Pingfan Liu <piliu@redhat.com> wrote: > > > > Hi everybody, I tried this stuff again. > > Tried what again? Frankly, I have no idea what the purpose of this > patch series is, and this is v1. > Sorry that I should paste the original link for the history: https://lore.kernel.org/all/20210531084540.78546-1-kernelfans@gmail.com/ > Could you please explain? > It is about calling the C routine of __create_pgd_mapping() at the stage of mmu-off. > Also, the early arm64 startup code is changing substantially - please refer to > > https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git/log/?h=for-next/stage1-lpa2 > > for details. > Oh, it seems that most of the ideas in my series have been implemented. I will dive it for more detail. Thank you very much. Regards, Pingfan > > Last time when I tried this, > > Catalin raised concern about the intrumentation, and Ard doubted this > > way due to alignement issue with mmu-off. > > > > Last time, the alignment issue looked unsoluable and I gave up. But > > nowadays, when I looked at it, I think it is partially resovable. (for > > detail, please see the commit log in [PATCH 08/10] arm64: mm: Enforce > > memory alignment in mmu_head) > > > > Overall, at this very early stage, the using of C routines faces three > > challenge: > > PIC > > instrumentation > > alignment > > > > [2/10] resolves instrumentation issue > > > > [3/10] makes mmu_head self-contained and prevent the outside > > PIC/ instrumentation/ alignment issues from seeping in. And check the code PIC. > > > > [PATCH 08/10] explains the alignement issue, in theory, it can be > > checked and resolved. And in this patch, it is partially resolved. > > > > > > Cc: Ard Biesheuvel <ardb@kernel.org> > > Cc: Catalin Marinas <catalin.marinas@arm.com> > > Cc: Will Deacon <will@kernel.org> > > Cc: Mark Rutland <mark.rutland@arm.com> > > To: linux-arm-kernel@lists.infradead.org > > --- > > > > Pingfan Liu (10): > > arm64: mm: Split out routines for code reuse > > arm64: mm: Introduce mmu_head routines without instrumentation > > arm64: mm: Use if-conditon to truncate external dependency > > arm64: head: Enable __create_pgd_mapping() to handle pgtable's paddr > > arm64: mm: Force early mapping aligned on SWAPPER_BLOCK_SIZE > > arm64: mm: Handle scope beyond the capacity of kernel pgtable in > > mmu_head_create_pgd_mapping() > > arm64: mm: Introduce head_pool routines to enable pgtabl allocation > > arm64: mm: Enforce memory alignment in mmu_head > > arm64: head: Use __create_pgd_mapping_locked() to serve the creation > > of pgtable > > arm64: head: Clean up unneeded routines > > > > arch/arm64/include/asm/kernel-pgtable.h | 1 + > > arch/arm64/include/asm/mmu.h | 4 + > > arch/arm64/include/asm/pgtable.h | 11 +- > > arch/arm64/kernel/head.S | 314 +++++++----------------- > > arch/arm64/mm/Makefile | 22 +- > > arch/arm64/mm/mmu.c | 289 +--------------------- > > arch/arm64/mm/mmu_head.c | 134 ++++++++++ > > arch/arm64/mm/mmu_inc.c | 292 ++++++++++++++++++++++ > > 8 files changed, 558 insertions(+), 509 deletions(-) > > create mode 100644 arch/arm64/mm/mmu_head.c > > create mode 100644 arch/arm64/mm/mmu_inc.c > > > > -- > > 2.41.0 > > >
On Thu, Mar 14, 2024 at 10:54:12AM +0800, Pingfan Liu wrote: > On Wed, Mar 13, 2024 at 9:05 PM Ard Biesheuvel <ardb@kernel.org> wrote: > > Also, the early arm64 startup code is changing substantially - please refer to > > > > https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git/log/?h=for-next/stage1-lpa2 > > > > for details. > > Oh, it seems that most of the ideas in my series have been > implemented. I will dive it for more detail. Yeah, better post after 6.9-rc1 when this stuff will land.
Hi everybody, I tried this stuff again. Last time when I tried this, Catalin raised concern about the intrumentation, and Ard doubted this way due to alignement issue with mmu-off. Last time, the alignment issue looked unsoluable and I gave up. But nowadays, when I looked at it, I think it is partially resovable. (for detail, please see the commit log in [PATCH 08/10] arm64: mm: Enforce memory alignment in mmu_head) Overall, at this very early stage, the using of C routines faces three challenge: PIC instrumentation alignment [2/10] resolves instrumentation issue [3/10] makes mmu_head self-contained and prevent the outside PIC/ instrumentation/ alignment issues from seeping in. And check the code PIC. [PATCH 08/10] explains the alignement issue, in theory, it can be checked and resolved. And in this patch, it is partially resolved. Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Will Deacon <will@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> To: linux-arm-kernel@lists.infradead.org --- Pingfan Liu (10): arm64: mm: Split out routines for code reuse arm64: mm: Introduce mmu_head routines without instrumentation arm64: mm: Use if-conditon to truncate external dependency arm64: head: Enable __create_pgd_mapping() to handle pgtable's paddr arm64: mm: Force early mapping aligned on SWAPPER_BLOCK_SIZE arm64: mm: Handle scope beyond the capacity of kernel pgtable in mmu_head_create_pgd_mapping() arm64: mm: Introduce head_pool routines to enable pgtabl allocation arm64: mm: Enforce memory alignment in mmu_head arm64: head: Use __create_pgd_mapping_locked() to serve the creation of pgtable arm64: head: Clean up unneeded routines arch/arm64/include/asm/kernel-pgtable.h | 1 + arch/arm64/include/asm/mmu.h | 4 + arch/arm64/include/asm/pgtable.h | 11 +- arch/arm64/kernel/head.S | 314 +++++++----------------- arch/arm64/mm/Makefile | 22 +- arch/arm64/mm/mmu.c | 289 +--------------------- arch/arm64/mm/mmu_head.c | 134 ++++++++++ arch/arm64/mm/mmu_inc.c | 292 ++++++++++++++++++++++ 8 files changed, 558 insertions(+), 509 deletions(-) create mode 100644 arch/arm64/mm/mmu_head.c create mode 100644 arch/arm64/mm/mmu_inc.c