| Message ID | 20230612172805.681179-1-eric.devolder@oracle.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | refactor Kconfig to consolidate KEXEC and CRASH options | expand |
My apologies, but this patch series is 13 patches, not 21. The last patch is "PATCH v1 13/21 sh/kexec: refactor for kernel/Kconfig.kexec" I'll correct for v2. eric On 6/12/23 12:27, Eric DeVolder wrote: > The Kconfig is refactored to consolidate KEXEC and CRASH options from > various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. > > The Kconfig.kexec is now a submenu titled "Kexec and crash features" > located under "General Setup". > > The following options are impacted: > > - KEXEC > - KEXEC_FILE > - KEXEC_SIG > - KEXEC_SIG_FORCE > - KEXEC_BZIMAGE_VERIFY_SIG > - KEXEC_JUMP > - CRASH_DUMP > > Over time, these options have been copied between Kconfig files and > are very similar to one another, but with slight differences. > > The following architectures are impacted by the refactor (because of > use of one or more KEXEC/CRASH options): > > - arm > - arm64 > - ia64 > - loongarch > - m68k > - mips > - parisc > - powerpc > - riscv > - s390 > - sh > - x86 > > More information: > > In the patch series "crash: Kernel handling of CPU and memory hot > un/plug" > > https://lore.kernel.org/lkml/20230503224145.7405-1-eric.devolder@oracle.com/ > > the new kernel feature introduces the config option CRASH_HOTPLUG. > > In reviewing, Thomas Gleixner requested that the new config option > not be placed in x86 Kconfig. Rather the option needs a generic/common > home. To Thomas' point, the KEXEC and CRASH options have largely been > duplicated in the various arch/<arch>/Kconfig files, with minor > differences. This kind of proliferation is to be avoid/stopped. > > https://lore.kernel.org/lkml/875y91yv63.ffs@tglx/ > > To that end, I have refactored the arch Kconfigs so as to consolidate > the various KEXEC and CRASH options. Generally speaking, this work has > the following themes: > > - KEXEC and CRASH options are moved into new file kernel/Kconfig.kexec > - These items from arch/Kconfig: > CRASH_CORE KEXEC_CORE KEXEC_ELF HAVE_IMA_KEXEC > - These items from arch/x86/Kconfig form the common options: > KEXEC KEXEC_FILE KEXEC_SIG KEXEC_SIG_FORCE > KEXEC_BZIMAGE_VERIFY_SIG KEXEC_JUMP CRASH_DUMP > - The crash hotplug series appends CRASH_HOTPLUG to Kconfig.kexec > NOTE: PHYSICAL_START could be argued to be included in this series. > - The Kconfig.kexec is now a submenu titled "Kexec and crash features" > - The Kconfig.kexec is now listed in "General Setup" submenu from > init/Kconfig > - To control the main common options, new options ARCH_HAS_KEXEC, > ARCH_HAS_KEXEC_FILE and ARCH_HAS_CRASH_DUMP are introduced. > NOTE: I went with ARCH_HAS_ due to the existing ARCH_HAS_KEXEC_PURGATORY. > - To account for the slight differences, new options ARCH_SUPPORTS_KEXEC, > ARCH_SUPPORTS_KEXEC_FILE and ARCH_SUPPORTS_CRASH_DUMP are used to > elicit the same side effects as the original arch/<arch>/Kconfig > files for KEXEC and CRASH options. > NOTE: I'm open to a better name than 'ARCH_SUPPORTS', perhaps > ARCH_CUSTOMIZE ? > > An example, 'make menuconfig' illustrating the submenu: > > > General setup > Kexec and crash features > [*] Enable kexec system call > [*] Enable kexec file based system call > [*] Verify kernel signature during kexec_file_load() syscall > [ ] Require a valid signature in kexec_file_load() syscall > [ ] Enable bzImage signature verification support > [*] kexec jump > [*] kernel crash dumps > [*] Update the crash elfcorehdr on system configuration changes > > The three main options are KEXEC, KEXEC_FILE and CRASH_DUMP. In the > process of consolidating these options, I encountered slight differences > in the coding of these options in several of the architectures. As a > result, I settled on the following solution: > > - Each of three main options has a 'depends on ARCH_HAS_<option>' > statement: ARCH_HAS_KEXEC, ARCH_HAS_KEXEC_FILE, ARCH_HAS_CRASH_DUMP. > > For example, the KEXEC_FILE option has a 'depends on > ARCH_HAS_KEXEC_FILE' statement. > > - The boolean ARCH_HAS_<option> in effect allows the arch to determine > when the feature is allowed. Archs which don't have the feature > simply do not provide the corresponding ARCH_HAS_<option>. > For each arch, where there previously were KEXEC and/or CRASH > options, these have been replaced with the corresponding boolean > ARCH_HAS_<option>, and an appropriate def_bool statement. > > For example, if the arch supports KEXEC_FILE, then the > ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the > KEXEC_FILE option to be available. > > If the arch has a 'depends on' statement in its original coding > of the option, then that expression becomes part of the def_bool > expression. For example, arm64 had: > > config KEXEC > depends on PM_SLEEP_SMP > > and in this solution, this converts to: > > config ARCH_HAS_KEXEC > def_bool PM_SLEEP_SMP > > > - In order to account for the differences in the config coding for > the three common options, the ARCH_SUPPORTS_<option> is used. > This options has a 'depends on <option>' statement to couple it > to the main option, and from there can insert the differences > from the common option and the arch original coding of that option. > > For example, a few archs enable CRYPTO and CRYTPO_SHA256 for > KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and > 'select CRYPTO' and 'select CRYPTO_SHA256' statements. > > Illustrating the option relationships: > > For KEXEC: > ARCH_HAS_KEXEC <- KEXEC <- ARCH_SUPPORTS_KEXEC > > KEXEC # in Kconfig.kexec > ARCH_HAS_KEXEC # in arch/<arch>/Kconfig, as needed > ARCH_SUPPORTS_KEXEC # in arch/<arch>/Kconfig, as needed > > > For KEXEC_FILE: > ARCH_HAS_KEXEC_FILE <- KEXEC_FILE <- ARCH_SUPPORTS_KEXEC_FILE > > KEXEC_FILE # in Kconfig.kexec > ARCH_HAS_KEXEC_FILE # in arch/<arch>/Kconfig, as needed > ARCH_SUPPORTS_KEXEC_FILE # in arch/<arch>/Kconfig, as needed > > > For CRASH: > ARCH_HAS_CRASH_DUMP <- CRASH_DUMP <- ARCH_SUPPORTS_CRASH_DUMP > > CRASH_DUMP # in Kconfig.kexec > ARCH_HAS_CRASH_DUMP # in arch/<arch>/Kconfig, as needed > ARCH_SUPPORTS_CRASH_DUMP # in arch/<arch>/Kconfig, as needed > > To summarize, the ARCH_HAS_<option> permits the <option> to be > enabled, and the ARCH_SUPPORTS_<option> handles side effects (ie. > select statements). > > Examples: > A few examples to show the new strategy in action: > > ===== x86 (minus the help section) ===== > Original: > config KEXEC > bool "kexec system call" > select KEXEC_CORE > > config KEXEC_FILE > bool "kexec file based system call" > select KEXEC_CORE > select HAVE_IMA_KEXEC if IMA > depends on X86_64 > depends on CRYPTO=y > depends on CRYPTO_SHA256=y > > config ARCH_HAS_KEXEC_PURGATORY > def_bool KEXEC_FILE > > config KEXEC_SIG > bool "Verify kernel signature during kexec_file_load() syscall" > depends on KEXEC_FILE > > config KEXEC_SIG_FORCE > bool "Require a valid signature in kexec_file_load() syscall" > depends on KEXEC_SIG > > config KEXEC_BZIMAGE_VERIFY_SIG > bool "Enable bzImage signature verification support" > depends on KEXEC_SIG > depends on SIGNED_PE_FILE_VERIFICATION > select SYSTEM_TRUSTED_KEYRING > > config CRASH_DUMP > bool "kernel crash dumps" > depends on X86_64 || (X86_32 && HIGHMEM) > > config KEXEC_JUMP > bool "kexec jump" > depends on KEXEC && HIBERNATION > help > > becomes... > New: > config ARCH_HAS_KEXEC > def_bool y > > config ARCH_HAS_KEXEC_FILE > def_bool X86_64 && CRYPTO && CRYPTO_SHA256 > > config ARCH_SUPPORTS_KEXEC_FILE > def_bool y > depends on KEXEC_FILE > select HAVE_IMA_KEXEC if IMA > > config ARCH_HAS_KEXEC_PURGATORY > def_bool KEXEC_FILE > > config ARCH_HAS_KEXEC_JUMP > def_bool y > > config ARCH_HAS_CRASH_DUMP > def_bool X86_64 || (X86_32 && HIGHMEM) > > > ===== powerpc (minus the help section) ===== > Original: > config KEXEC > bool "kexec system call" > depends on PPC_BOOK3S || PPC_E500 || (44x && !SMP) > select KEXEC_CORE > > config KEXEC_FILE > bool "kexec file based system call" > select KEXEC_CORE > select HAVE_IMA_KEXEC if IMA > select KEXEC_ELF > depends on PPC64 > depends on CRYPTO=y > depends on CRYPTO_SHA256=y > > config ARCH_HAS_KEXEC_PURGATORY > def_bool KEXEC_FILE > > config CRASH_DUMP > bool "Build a dump capture kernel" > depends on PPC64 || PPC_BOOK3S_32 || PPC_85xx || (44x && !SMP) > select RELOCATABLE if PPC64 || 44x || PPC_85xx > > becomes... > New: > config ARCH_HAS_KEXEC > def_bool PPC_BOOK3S || PPC_E500 || (44x && !SMP) > > config ARCH_HAS_KEXEC_FILE > def_bool PPC64 && CRYPTO && CRYPTO_SHA256 > > config ARCH_HAS_KEXEC_PURGATORY > def_bool KEXEC_FILE > > config ARCH_SUPPORTS_KEXEC_FILE > def_bool y > depends on KEXEC_FILE > select KEXEC_ELF > select HAVE_IMA_KEXEC if IMA > > config ARCH_HAS_CRASH_DUMP > def_bool PPC64 || PPC_BOOK3S_32 || PPC_85xx || (44x && !SMP) > > config ARCH_SUPPORTS_CRASH_DUMP > def_bool y > depends on CRASH_DUMP > select RELOCATABLE if PPC64 || 44x || PPC_85xx > > > Testing Approach and Results > > There are 388 config files in the arch/<arch>/configs directories. > For each of these config files, a .config is generated both before and > after this Kconfig series, and checked for equivalence. This approach > allows for a rather rapid check of all architectures and a wide > variety of configs wrt/ KEXEC and CRASH, and avoids requiring > compiling for all architectures and running kernels and run-time > testing. > > As such, I developed the following script to compare the before and > after of 'make olddefconfig'. The new symbols introduced by this > series are filtered out, but otherwise the config files are PASS > only if they were equivalent, and FAIL otherwise. > > The script performs the test by doing the following: > > # Obtain the "golden" .config output for given config file > # Reset test sandbox > git checkout master > git branch -D test_Kconfig > git checkout -B test_Kconfig master > make distclean > # Write out updated config > cp -f <config file> .config > make ARCH=<arch> olddefconfig > # Track each item in .config, LHSB is "golden" > scoreboard .config > > # Obtain the "changed" .config output for given config file > # Reset test sandbox > make distclean > # Apply this Kconfig series > git am <this Kconfig series> > # Write out updated config > cp -f <config file> .config > make ARCH=<arch> olddefconfig > # Track each item in .config, RHSB is "changed" > scoreboard .config > > # Determine test result > # Filter-out new symbols introduced by this series > # Filter-out symbol=n which not in either scoreboard > # Compare LHSB "golden" and RHSB "changed" scoreboards and issue PASS/FAIL > > The script was instrumental during the refactoring of Kconfig as it > continually revealed problems. The end result being that the solution > presented in this series passes all configs as checked by the script. > > Regards, > eric > > > --- > v1: 12jun2023 > - Initial > - Based on 6.4.0-rc6 > > --- > Eric DeVolder (21): > kexec: consolidate kexec and crash options into kernel/Kconfig.kexec > x86/kexec: refactor for kernel/Kconfig.kexec > arm/kexec: refactor for kernel/Kconfig.kexec > ia64/kexec: refactor for kernel/Kconfig.kexec > arm64/kexec: refactor for kernel/Kconfig.kexec > loongarch/kexec: refactor for kernel/Kconfig.kexec > m68k/kexec: refactor for kernel/Kconfig.kexec > mips/kexec: refactor for kernel/Kconfig.kexec > parisc/kexec: refactor for kernel/Kconfig.kexec > powerpc/kexec: refactor for kernel/Kconfig.kexec > riscv/kexec: refactor for kernel/Kconfig.kexec > s390/kexec: refactor for kernel/Kconfig.kexec > sh/kexec: refactor for kernel/Kconfig.kexec > crash: move a few code bits to setup support of crash hotplug > crash: add generic infrastructure for crash hotplug support > kexec: exclude elfcorehdr from the segment digest > crash: memory and CPU hotplug sysfs attributes > x86/crash: add x86 crash hotplug support > crash: hotplug support for kexec_load() > crash: change crash_prepare_elf64_headers() to for_each_possible_cpu() > x86/crash: optimize CPU changes > > .../admin-guide/mm/memory-hotplug.rst | 8 + > Documentation/core-api/cpu_hotplug.rst | 18 + > arch/Kconfig | 13 - > arch/arm/Kconfig | 29 +- > arch/arm64/Kconfig | 61 +-- > arch/ia64/Kconfig | 28 +- > arch/loongarch/Kconfig | 26 +- > arch/m68k/Kconfig | 19 +- > arch/mips/Kconfig | 32 +- > arch/parisc/Kconfig | 34 +- > arch/powerpc/Kconfig | 55 +-- > arch/riscv/Kconfig | 48 +-- > arch/s390/Kconfig | 65 +--- > arch/sh/Kconfig | 46 +-- > arch/x86/Kconfig | 90 +---- > arch/x86/include/asm/kexec.h | 18 + > arch/x86/kernel/crash.c | 140 ++++++- > drivers/base/cpu.c | 14 + > drivers/base/memory.c | 13 + > include/linux/crash_core.h | 9 + > include/linux/kexec.h | 63 +++- > include/uapi/linux/kexec.h | 1 + > init/Kconfig | 2 + > kernel/Kconfig.kexec | 134 +++++++ > kernel/crash_core.c | 355 ++++++++++++++++++ > kernel/kexec.c | 5 + > kernel/kexec_core.c | 6 + > kernel/kexec_file.c | 187 +-------- > kernel/ksysfs.c | 15 + > 29 files changed, 900 insertions(+), 634 deletions(-) > create mode 100644 kernel/Kconfig.kexec >
On Mon, Jun 12, 2023 at 01:27:52PM -0400, Eric DeVolder wrote: > The Kconfig is refactored to consolidate KEXEC and CRASH options from > various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. This looks very nice! > [...] > - The boolean ARCH_HAS_<option> in effect allows the arch to determine > when the feature is allowed. Archs which don't have the feature > simply do not provide the corresponding ARCH_HAS_<option>. > For each arch, where there previously were KEXEC and/or CRASH > options, these have been replaced with the corresponding boolean > ARCH_HAS_<option>, and an appropriate def_bool statement. > > For example, if the arch supports KEXEC_FILE, then the > ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the > KEXEC_FILE option to be available. > > If the arch has a 'depends on' statement in its original coding > of the option, then that expression becomes part of the def_bool > expression. For example, arm64 had: > > config KEXEC > depends on PM_SLEEP_SMP > > and in this solution, this converts to: > > config ARCH_HAS_KEXEC > def_bool PM_SLEEP_SMP > > > - In order to account for the differences in the config coding for > the three common options, the ARCH_SUPPORTS_<option> is used. > This options has a 'depends on <option>' statement to couple it > to the main option, and from there can insert the differences > from the common option and the arch original coding of that option. > > For example, a few archs enable CRYPTO and CRYTPO_SHA256 for > KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and > 'select CRYPTO' and 'select CRYPTO_SHA256' statements. Naming nit: "HAS" and "SUPPORTS" feel very similar, and looking at existing configs, "ARCH_SUPPORTS_..." is already used for doing this kind of bare "bool" management. e.g. see ARCH_SUPPORTS_INT128 It looks like you need to split "depends" and "select" so the options can be chosen separately from the "selectable" configs. How about naming this ARCH_SELECTS_<option>, since that's what it's there for? -Kees
On 6/13/23 15:21, Kees Cook wrote: > On Mon, Jun 12, 2023 at 01:27:52PM -0400, Eric DeVolder wrote: >> The Kconfig is refactored to consolidate KEXEC and CRASH options from >> various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. > > This looks very nice! > Thank you Kees! >> [...] >> - The boolean ARCH_HAS_<option> in effect allows the arch to determine >> when the feature is allowed. Archs which don't have the feature >> simply do not provide the corresponding ARCH_HAS_<option>. >> For each arch, where there previously were KEXEC and/or CRASH >> options, these have been replaced with the corresponding boolean >> ARCH_HAS_<option>, and an appropriate def_bool statement. >> >> For example, if the arch supports KEXEC_FILE, then the >> ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the >> KEXEC_FILE option to be available. >> >> If the arch has a 'depends on' statement in its original coding >> of the option, then that expression becomes part of the def_bool >> expression. For example, arm64 had: >> >> config KEXEC >> depends on PM_SLEEP_SMP >> >> and in this solution, this converts to: >> >> config ARCH_HAS_KEXEC >> def_bool PM_SLEEP_SMP >> >> >> - In order to account for the differences in the config coding for >> the three common options, the ARCH_SUPPORTS_<option> is used. >> This options has a 'depends on <option>' statement to couple it >> to the main option, and from there can insert the differences >> from the common option and the arch original coding of that option. >> >> For example, a few archs enable CRYPTO and CRYTPO_SHA256 for >> KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and >> 'select CRYPTO' and 'select CRYPTO_SHA256' statements. > > Naming nit: "HAS" and "SUPPORTS" feel very similar, and looking at > existing configs, "ARCH_SUPPORTS_..." is already used for doing this > kind of bare "bool" management. e.g. see ARCH_SUPPORTS_INT128 > > It looks like you need to split "depends" and "select" so the options > can be chosen separately from the "selectable" configs. > > How about naming this ARCH_SELECTS_<option>, since that's what it's > there for? > I'm OK with this. Let's see if others agree? Thank you! eric > -Kees >
Eric DeVolder <eric.devolder@oracle.com> writes: > On 6/13/23 15:21, Kees Cook wrote: >> On Mon, Jun 12, 2023 at 01:27:52PM -0400, Eric DeVolder wrote: >>> The Kconfig is refactored to consolidate KEXEC and CRASH options from >>> various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. >> >> This looks very nice! >> > Thank you Kees! > >>> [...] >>> - The boolean ARCH_HAS_<option> in effect allows the arch to determine >>> when the feature is allowed. Archs which don't have the feature >>> simply do not provide the corresponding ARCH_HAS_<option>. >>> For each arch, where there previously were KEXEC and/or CRASH >>> options, these have been replaced with the corresponding boolean >>> ARCH_HAS_<option>, and an appropriate def_bool statement. >>> >>> For example, if the arch supports KEXEC_FILE, then the >>> ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the >>> KEXEC_FILE option to be available. >>> >>> If the arch has a 'depends on' statement in its original coding >>> of the option, then that expression becomes part of the def_bool >>> expression. For example, arm64 had: >>> >>> config KEXEC >>> depends on PM_SLEEP_SMP >>> >>> and in this solution, this converts to: >>> >>> config ARCH_HAS_KEXEC >>> def_bool PM_SLEEP_SMP >>> >>> >>> - In order to account for the differences in the config coding for >>> the three common options, the ARCH_SUPPORTS_<option> is used. >>> This options has a 'depends on <option>' statement to couple it >>> to the main option, and from there can insert the differences >>> from the common option and the arch original coding of that option. >>> >>> For example, a few archs enable CRYPTO and CRYTPO_SHA256 for >>> KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and >>> 'select CRYPTO' and 'select CRYPTO_SHA256' statements. >> >> Naming nit: "HAS" and "SUPPORTS" feel very similar, and looking at >> existing configs, "ARCH_SUPPORTS_..." is already used for doing this >> kind of bare "bool" management. e.g. see ARCH_SUPPORTS_INT128 >> >> It looks like you need to split "depends" and "select" so the options >> can be chosen separately from the "selectable" configs. >> >> How about naming this ARCH_SELECTS_<option>, since that's what it's >> there for? >> > I'm OK with this. Let's see if others agree? Yeah please rename one or both of them. At a glance the difference between HAS and SUPPORTS is very non-obvious. I like Kees' suggestion to use ARCH_SUPPORTS and ARCH_SELECTS. cheers
On 6/14/23 22:26, Michael Ellerman wrote: > Eric DeVolder <eric.devolder@oracle.com> writes: >> On 6/13/23 15:21, Kees Cook wrote: >>> On Mon, Jun 12, 2023 at 01:27:52PM -0400, Eric DeVolder wrote: >>>> The Kconfig is refactored to consolidate KEXEC and CRASH options from >>>> various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. >>> >>> This looks very nice! >>> >> Thank you Kees! >> >>>> [...] >>>> - The boolean ARCH_HAS_<option> in effect allows the arch to determine >>>> when the feature is allowed. Archs which don't have the feature >>>> simply do not provide the corresponding ARCH_HAS_<option>. >>>> For each arch, where there previously were KEXEC and/or CRASH >>>> options, these have been replaced with the corresponding boolean >>>> ARCH_HAS_<option>, and an appropriate def_bool statement. >>>> >>>> For example, if the arch supports KEXEC_FILE, then the >>>> ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the >>>> KEXEC_FILE option to be available. >>>> >>>> If the arch has a 'depends on' statement in its original coding >>>> of the option, then that expression becomes part of the def_bool >>>> expression. For example, arm64 had: >>>> >>>> config KEXEC >>>> depends on PM_SLEEP_SMP >>>> >>>> and in this solution, this converts to: >>>> >>>> config ARCH_HAS_KEXEC >>>> def_bool PM_SLEEP_SMP >>>> >>>> >>>> - In order to account for the differences in the config coding for >>>> the three common options, the ARCH_SUPPORTS_<option> is used. >>>> This options has a 'depends on <option>' statement to couple it >>>> to the main option, and from there can insert the differences >>>> from the common option and the arch original coding of that option. >>>> >>>> For example, a few archs enable CRYPTO and CRYTPO_SHA256 for >>>> KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and >>>> 'select CRYPTO' and 'select CRYPTO_SHA256' statements. >>> >>> Naming nit: "HAS" and "SUPPORTS" feel very similar, and looking at >>> existing configs, "ARCH_SUPPORTS_..." is already used for doing this >>> kind of bare "bool" management. e.g. see ARCH_SUPPORTS_INT128 >>> >>> It looks like you need to split "depends" and "select" so the options >>> can be chosen separately from the "selectable" configs. >>> >>> How about naming this ARCH_SELECTS_<option>, since that's what it's >>> there for? >>> >> I'm OK with this. Let's see if others agree? > > Yeah please rename one or both of them. At a glance the difference > between HAS and SUPPORTS is very non-obvious. > > I like Kees' suggestion to use ARCH_SUPPORTS and ARCH_SELECTS. > > cheers Michael, ok thanks! eric
The Kconfig is refactored to consolidate KEXEC and CRASH options from various arch/<arch>/Kconfig files into new file kernel/Kconfig.kexec. The Kconfig.kexec is now a submenu titled "Kexec and crash features" located under "General Setup". The following options are impacted: - KEXEC - KEXEC_FILE - KEXEC_SIG - KEXEC_SIG_FORCE - KEXEC_BZIMAGE_VERIFY_SIG - KEXEC_JUMP - CRASH_DUMP Over time, these options have been copied between Kconfig files and are very similar to one another, but with slight differences. The following architectures are impacted by the refactor (because of use of one or more KEXEC/CRASH options): - arm - arm64 - ia64 - loongarch - m68k - mips - parisc - powerpc - riscv - s390 - sh - x86 More information: In the patch series "crash: Kernel handling of CPU and memory hot un/plug" https://lore.kernel.org/lkml/20230503224145.7405-1-eric.devolder@oracle.com/ the new kernel feature introduces the config option CRASH_HOTPLUG. In reviewing, Thomas Gleixner requested that the new config option not be placed in x86 Kconfig. Rather the option needs a generic/common home. To Thomas' point, the KEXEC and CRASH options have largely been duplicated in the various arch/<arch>/Kconfig files, with minor differences. This kind of proliferation is to be avoid/stopped. https://lore.kernel.org/lkml/875y91yv63.ffs@tglx/ To that end, I have refactored the arch Kconfigs so as to consolidate the various KEXEC and CRASH options. Generally speaking, this work has the following themes: - KEXEC and CRASH options are moved into new file kernel/Kconfig.kexec - These items from arch/Kconfig: CRASH_CORE KEXEC_CORE KEXEC_ELF HAVE_IMA_KEXEC - These items from arch/x86/Kconfig form the common options: KEXEC KEXEC_FILE KEXEC_SIG KEXEC_SIG_FORCE KEXEC_BZIMAGE_VERIFY_SIG KEXEC_JUMP CRASH_DUMP - The crash hotplug series appends CRASH_HOTPLUG to Kconfig.kexec NOTE: PHYSICAL_START could be argued to be included in this series. - The Kconfig.kexec is now a submenu titled "Kexec and crash features" - The Kconfig.kexec is now listed in "General Setup" submenu from init/Kconfig - To control the main common options, new options ARCH_HAS_KEXEC, ARCH_HAS_KEXEC_FILE and ARCH_HAS_CRASH_DUMP are introduced. NOTE: I went with ARCH_HAS_ due to the existing ARCH_HAS_KEXEC_PURGATORY. - To account for the slight differences, new options ARCH_SUPPORTS_KEXEC, ARCH_SUPPORTS_KEXEC_FILE and ARCH_SUPPORTS_CRASH_DUMP are used to elicit the same side effects as the original arch/<arch>/Kconfig files for KEXEC and CRASH options. NOTE: I'm open to a better name than 'ARCH_SUPPORTS', perhaps ARCH_CUSTOMIZE ? An example, 'make menuconfig' illustrating the submenu: > General setup > Kexec and crash features [*] Enable kexec system call [*] Enable kexec file based system call [*] Verify kernel signature during kexec_file_load() syscall [ ] Require a valid signature in kexec_file_load() syscall [ ] Enable bzImage signature verification support [*] kexec jump [*] kernel crash dumps [*] Update the crash elfcorehdr on system configuration changes The three main options are KEXEC, KEXEC_FILE and CRASH_DUMP. In the process of consolidating these options, I encountered slight differences in the coding of these options in several of the architectures. As a result, I settled on the following solution: - Each of three main options has a 'depends on ARCH_HAS_<option>' statement: ARCH_HAS_KEXEC, ARCH_HAS_KEXEC_FILE, ARCH_HAS_CRASH_DUMP. For example, the KEXEC_FILE option has a 'depends on ARCH_HAS_KEXEC_FILE' statement. - The boolean ARCH_HAS_<option> in effect allows the arch to determine when the feature is allowed. Archs which don't have the feature simply do not provide the corresponding ARCH_HAS_<option>. For each arch, where there previously were KEXEC and/or CRASH options, these have been replaced with the corresponding boolean ARCH_HAS_<option>, and an appropriate def_bool statement. For example, if the arch supports KEXEC_FILE, then the ARCH_HAS_KEXEC_FILE simply has a 'def_bool y'. This permits the KEXEC_FILE option to be available. If the arch has a 'depends on' statement in its original coding of the option, then that expression becomes part of the def_bool expression. For example, arm64 had: config KEXEC depends on PM_SLEEP_SMP and in this solution, this converts to: config ARCH_HAS_KEXEC def_bool PM_SLEEP_SMP - In order to account for the differences in the config coding for the three common options, the ARCH_SUPPORTS_<option> is used. This options has a 'depends on <option>' statement to couple it to the main option, and from there can insert the differences from the common option and the arch original coding of that option. For example, a few archs enable CRYPTO and CRYTPO_SHA256 for KEXEC_FILE. These require a ARCH_SUPPORTS_KEXEC_FILE and 'select CRYPTO' and 'select CRYPTO_SHA256' statements. Illustrating the option relationships: For KEXEC: ARCH_HAS_KEXEC <- KEXEC <- ARCH_SUPPORTS_KEXEC KEXEC # in Kconfig.kexec ARCH_HAS_KEXEC # in arch/<arch>/Kconfig, as needed ARCH_SUPPORTS_KEXEC # in arch/<arch>/Kconfig, as needed For KEXEC_FILE: ARCH_HAS_KEXEC_FILE <- KEXEC_FILE <- ARCH_SUPPORTS_KEXEC_FILE KEXEC_FILE # in Kconfig.kexec ARCH_HAS_KEXEC_FILE # in arch/<arch>/Kconfig, as needed ARCH_SUPPORTS_KEXEC_FILE # in arch/<arch>/Kconfig, as needed For CRASH: ARCH_HAS_CRASH_DUMP <- CRASH_DUMP <- ARCH_SUPPORTS_CRASH_DUMP CRASH_DUMP # in Kconfig.kexec ARCH_HAS_CRASH_DUMP # in arch/<arch>/Kconfig, as needed ARCH_SUPPORTS_CRASH_DUMP # in arch/<arch>/Kconfig, as needed To summarize, the ARCH_HAS_<option> permits the <option> to be enabled, and the ARCH_SUPPORTS_<option> handles side effects (ie. select statements). Examples: A few examples to show the new strategy in action: ===== x86 (minus the help section) ===== Original: config KEXEC bool "kexec system call" select KEXEC_CORE config KEXEC_FILE bool "kexec file based system call" select KEXEC_CORE select HAVE_IMA_KEXEC if IMA depends on X86_64 depends on CRYPTO=y depends on CRYPTO_SHA256=y config ARCH_HAS_KEXEC_PURGATORY def_bool KEXEC_FILE config KEXEC_SIG bool "Verify kernel signature during kexec_file_load() syscall" depends on KEXEC_FILE config KEXEC_SIG_FORCE bool "Require a valid signature in kexec_file_load() syscall" depends on KEXEC_SIG config KEXEC_BZIMAGE_VERIFY_SIG bool "Enable bzImage signature verification support" depends on KEXEC_SIG depends on SIGNED_PE_FILE_VERIFICATION select SYSTEM_TRUSTED_KEYRING config CRASH_DUMP bool "kernel crash dumps" depends on X86_64 || (X86_32 && HIGHMEM) config KEXEC_JUMP bool "kexec jump" depends on KEXEC && HIBERNATION help becomes... New: config ARCH_HAS_KEXEC def_bool y config ARCH_HAS_KEXEC_FILE def_bool X86_64 && CRYPTO && CRYPTO_SHA256 config ARCH_SUPPORTS_KEXEC_FILE def_bool y depends on KEXEC_FILE select HAVE_IMA_KEXEC if IMA config ARCH_HAS_KEXEC_PURGATORY def_bool KEXEC_FILE config ARCH_HAS_KEXEC_JUMP def_bool y config ARCH_HAS_CRASH_DUMP def_bool X86_64 || (X86_32 && HIGHMEM) ===== powerpc (minus the help section) ===== Original: config KEXEC bool "kexec system call" depends on PPC_BOOK3S || PPC_E500 || (44x && !SMP) select KEXEC_CORE config KEXEC_FILE bool "kexec file based system call" select KEXEC_CORE select HAVE_IMA_KEXEC if IMA select KEXEC_ELF depends on PPC64 depends on CRYPTO=y depends on CRYPTO_SHA256=y config ARCH_HAS_KEXEC_PURGATORY def_bool KEXEC_FILE config CRASH_DUMP bool "Build a dump capture kernel" depends on PPC64 || PPC_BOOK3S_32 || PPC_85xx || (44x && !SMP) select RELOCATABLE if PPC64 || 44x || PPC_85xx becomes... New: config ARCH_HAS_KEXEC def_bool PPC_BOOK3S || PPC_E500 || (44x && !SMP) config ARCH_HAS_KEXEC_FILE def_bool PPC64 && CRYPTO && CRYPTO_SHA256 config ARCH_HAS_KEXEC_PURGATORY def_bool KEXEC_FILE config ARCH_SUPPORTS_KEXEC_FILE def_bool y depends on KEXEC_FILE select KEXEC_ELF select HAVE_IMA_KEXEC if IMA config ARCH_HAS_CRASH_DUMP def_bool PPC64 || PPC_BOOK3S_32 || PPC_85xx || (44x && !SMP) config ARCH_SUPPORTS_CRASH_DUMP def_bool y depends on CRASH_DUMP select RELOCATABLE if PPC64 || 44x || PPC_85xx Testing Approach and Results There are 388 config files in the arch/<arch>/configs directories. For each of these config files, a .config is generated both before and after this Kconfig series, and checked for equivalence. This approach allows for a rather rapid check of all architectures and a wide variety of configs wrt/ KEXEC and CRASH, and avoids requiring compiling for all architectures and running kernels and run-time testing. As such, I developed the following script to compare the before and after of 'make olddefconfig'. The new symbols introduced by this series are filtered out, but otherwise the config files are PASS only if they were equivalent, and FAIL otherwise. The script performs the test by doing the following: # Obtain the "golden" .config output for given config file # Reset test sandbox git checkout master git branch -D test_Kconfig git checkout -B test_Kconfig master make distclean # Write out updated config cp -f <config file> .config make ARCH=<arch> olddefconfig # Track each item in .config, LHSB is "golden" scoreboard .config # Obtain the "changed" .config output for given config file # Reset test sandbox make distclean # Apply this Kconfig series git am <this Kconfig series> # Write out updated config cp -f <config file> .config make ARCH=<arch> olddefconfig # Track each item in .config, RHSB is "changed" scoreboard .config # Determine test result # Filter-out new symbols introduced by this series # Filter-out symbol=n which not in either scoreboard # Compare LHSB "golden" and RHSB "changed" scoreboards and issue PASS/FAIL The script was instrumental during the refactoring of Kconfig as it continually revealed problems. The end result being that the solution presented in this series passes all configs as checked by the script. Regards, eric --- v1: 12jun2023 - Initial - Based on 6.4.0-rc6 --- Eric DeVolder (21): kexec: consolidate kexec and crash options into kernel/Kconfig.kexec x86/kexec: refactor for kernel/Kconfig.kexec arm/kexec: refactor for kernel/Kconfig.kexec ia64/kexec: refactor for kernel/Kconfig.kexec arm64/kexec: refactor for kernel/Kconfig.kexec loongarch/kexec: refactor for kernel/Kconfig.kexec m68k/kexec: refactor for kernel/Kconfig.kexec mips/kexec: refactor for kernel/Kconfig.kexec parisc/kexec: refactor for kernel/Kconfig.kexec powerpc/kexec: refactor for kernel/Kconfig.kexec riscv/kexec: refactor for kernel/Kconfig.kexec s390/kexec: refactor for kernel/Kconfig.kexec sh/kexec: refactor for kernel/Kconfig.kexec crash: move a few code bits to setup support of crash hotplug crash: add generic infrastructure for crash hotplug support kexec: exclude elfcorehdr from the segment digest crash: memory and CPU hotplug sysfs attributes x86/crash: add x86 crash hotplug support crash: hotplug support for kexec_load() crash: change crash_prepare_elf64_headers() to for_each_possible_cpu() x86/crash: optimize CPU changes .../admin-guide/mm/memory-hotplug.rst | 8 + Documentation/core-api/cpu_hotplug.rst | 18 + arch/Kconfig | 13 - arch/arm/Kconfig | 29 +- arch/arm64/Kconfig | 61 +-- arch/ia64/Kconfig | 28 +- arch/loongarch/Kconfig | 26 +- arch/m68k/Kconfig | 19 +- arch/mips/Kconfig | 32 +- arch/parisc/Kconfig | 34 +- arch/powerpc/Kconfig | 55 +-- arch/riscv/Kconfig | 48 +-- arch/s390/Kconfig | 65 +--- arch/sh/Kconfig | 46 +-- arch/x86/Kconfig | 90 +---- arch/x86/include/asm/kexec.h | 18 + arch/x86/kernel/crash.c | 140 ++++++- drivers/base/cpu.c | 14 + drivers/base/memory.c | 13 + include/linux/crash_core.h | 9 + include/linux/kexec.h | 63 +++- include/uapi/linux/kexec.h | 1 + init/Kconfig | 2 + kernel/Kconfig.kexec | 134 +++++++ kernel/crash_core.c | 355 ++++++++++++++++++ kernel/kexec.c | 5 + kernel/kexec_core.c | 6 + kernel/kexec_file.c | 187 +-------- kernel/ksysfs.c | 15 + 29 files changed, 900 insertions(+), 634 deletions(-) create mode 100644 kernel/Kconfig.kexec