When the kdump service is loaded, if a CPU or memory is hot
un/plugged, the crash elfcorehdr, which describes the CPUs
and memory in the system, must also be updated, else the resulting
vmcore is inaccurate (eg. missing either CPU context or memory
regions).
The current solution utilizes udev to initiate an unload-then-reload
of the kdump image (e. kernel, initrd, boot_params, puratory and
elfcorehdr) by the userspace kexec utility. In previous posts I have
outlined the significant performance problems related to offloading
this activity to userspace.
This patchset introduces a generic crash hot un/plug handler that
registers with the CPU and memory notifiers. Upon CPU or memory
changes, this generic handler is invoked and performs important
housekeeping, for example obtaining the appropriate lock, and then
invokes an architecture specific handler to do the appropriate
updates.
In the case of x86_64, the arch specific handler generates a new
elfcorehdr, and overwrites the old one in memory. No involvement
with userspace needed.
To realize the benefits/test this patchset, one must make a couple
of minor changes to userspace:
- Prevent udev from updating kdump crash kernel on hot un/plug changes.
Add the following as the first lines to the udev rule file
/usr/lib/udev/rules.d/98-kexec.rules:
# The kernel handles updates to crash elfcorehdr for cpu and memory changes
SUBSYSTEM=="cpu", ATTRS{crash_hotplug}=="1", GOTO="kdump_reload_end"
SUBSYSTEM=="memory", ATTRS{crash_hotplug}=="1", GOTO="kdump_reload_end"
These lines will cause cpu and memory hot un/plug events to be
skipped within this rule file, if the kernel has these changes
enabled.
- Change to the kexec_file_load for loading the kdump kernel:
Eg. on RHEL: in /usr/bin/kdumpctl, change to:
standard_kexec_args="-p -d -s"
which adds the -s to select kexec_file_load syscall.
This patchset supports kexec_load with a modified kexec userspace
utility, and a working changeset to the kexec userspace utility
is provided here (and to use, the above change to standard_kexec_args
would be, for example, to append --hotplug instead of -s).
diff --git a/kexec/arch/i386/crashdump-x86.c b/kexec/arch/i386/crashdump-x86.c
index 9826f6d..4ed395a 100644
--- a/kexec/arch/i386/crashdump-x86.c
+++ b/kexec/arch/i386/crashdump-x86.c
@@ -48,6 +48,7 @@
#include <x86/x86-linux.h>
extern struct arch_options_t arch_options;
+extern int do_hotplug;
static int get_kernel_page_offset(struct kexec_info *UNUSED(info),
struct crash_elf_info *elf_info)
@@ -975,6 +976,14 @@ int load_crashdump_segments(struct kexec_info *info, char* mod_cmdline,
} else {
memsz = bufsz;
}
+
+ /* If hotplug support enabled, use larger size to accomodate changes */
+ if (do_hotplug) {
+ long int nr_cpus = get_nr_cpus();
+ memsz = (nr_cpus + CRASH_MAX_MEMORY_RANGES) * sizeof(Elf64_Phdr);
+ }
+
+ info->elfcorehdr =
elfcorehdr = add_buffer(info, tmp, bufsz, memsz, align, min_base,
max_addr, -1);
dbgprintf("Created elf header segment at 0x%lx\n", elfcorehdr);
diff --git a/kexec/crashdump-elf.c b/kexec/crashdump-elf.c
index b8bb686..5e29f7a 100644
--- a/kexec/crashdump-elf.c
+++ b/kexec/crashdump-elf.c
@@ -43,11 +43,7 @@ int FUNC(struct kexec_info *info,
int (*get_note_info)(int cpu, uint64_t *addr, uint64_t *len);
long int count_cpu;
- if (xen_present())
- nr_cpus = xen_get_nr_phys_cpus();
- else
- nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
-
+ nr_cpus = get_nr_cpus();
if (nr_cpus < 0) {
return -1;
}
diff --git a/kexec/crashdump.h b/kexec/crashdump.h
index 18bd691..28d3278 100644
--- a/kexec/crashdump.h
+++ b/kexec/crashdump.h
@@ -57,7 +57,6 @@ unsigned long phys_to_virt(struct crash_elf_info *elf_info,
unsigned long long paddr);
unsigned long xen_architecture(struct crash_elf_info *elf_info);
-int xen_get_nr_phys_cpus(void);
int xen_get_note(int cpu, uint64_t *addr, uint64_t *len);
int xen_get_crashkernel_region(uint64_t *start, uint64_t *end);
diff --git a/kexec/kexec-xen.h b/kexec/kexec-xen.h
index 70fb576..f54a2dd 100644
--- a/kexec/kexec-xen.h
+++ b/kexec/kexec-xen.h
@@ -83,5 +83,6 @@ extern int __xc_interface_close(xc_interface *xch);
#endif
int xen_get_kexec_range(int range, uint64_t *start, uint64_t *end);
+int xen_get_nr_phys_cpus(void);
#endif /* KEXEC_XEN_H */
diff --git a/kexec/kexec.c b/kexec/kexec.c
index 829a6ea..3668b73 100644
--- a/kexec/kexec.c
+++ b/kexec/kexec.c
@@ -58,6 +58,7 @@
unsigned long long mem_min = 0;
unsigned long long mem_max = ULONG_MAX;
+int do_hotplug = 0;
static unsigned long kexec_flags = 0;
/* Flags for kexec file (fd) based syscall */
static unsigned long kexec_file_flags = 0;
@@ -489,6 +490,17 @@ static int add_backup_segments(struct kexec_info *info,
return 0;
}
+long int get_nr_cpus(void)
+{
+ long int nr_cpus;
+
+ if (xen_present())
+ nr_cpus = xen_get_nr_phys_cpus();
+ else
+ nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
+ return nr_cpus;
+}
+
static char *slurp_fd(int fd, const char *filename, off_t size, off_t *nread)
{
char *buf;
@@ -672,6 +684,14 @@ static void update_purgatory(struct kexec_info *info)
if (info->segment[i].mem == (void *)info->rhdr.rel_addr) {
continue;
}
+
+ /* Don't include elfcorehdr in the checksum, if hotplug
+ * support enabled.
+ */
+ if (do_hotplug && (info->segment[i].mem == (void *)info->elfcorehdr)) {
+ continue;
+ }
+
sha256_update(&ctx, info->segment[i].buf,
info->segment[i].bufsz);
nullsz = info->segment[i].memsz - info->segment[i].bufsz;
@@ -1565,6 +1585,9 @@ int main(int argc, char *argv[])
case OPT_PRINT_CKR_SIZE:
print_crashkernel_region_size();
return 0;
+ case OPT_HOTPLUG:
+ do_hotplug = 1;
+ break;
default:
break;
}
diff --git a/kexec/kexec.h b/kexec/kexec.h
index 0f97a97..b0428cc 100644
--- a/kexec/kexec.h
+++ b/kexec/kexec.h
@@ -169,6 +169,7 @@ struct kexec_info {
int command_line_len;
int skip_checks;
+ unsigned long elfcorehdr;
};
struct arch_map_entry {
@@ -231,7 +232,8 @@ extern int file_types;
#define OPT_PRINT_CKR_SIZE 262
#define OPT_LOAD_LIVE_UPDATE 263
#define OPT_EXEC_LIVE_UPDATE 264
-#define OPT_MAX 265
+#define OPT_HOTPLUG 265
+#define OPT_MAX 266
#define KEXEC_OPTIONS \
{ "help", 0, 0, OPT_HELP }, \
{ "version", 0, 0, OPT_VERSION }, \
@@ -258,6 +260,7 @@ extern int file_types;
{ "debug", 0, 0, OPT_DEBUG }, \
{ "status", 0, 0, OPT_STATUS }, \
{ "print-ckr-size", 0, 0, OPT_PRINT_CKR_SIZE }, \
+ { "hotplug", 0, 0, OPT_HOTPLUG }, \
#define KEXEC_OPT_STR "h?vdfixyluet:pscaS"
@@ -290,6 +293,8 @@ extern unsigned long add_buffer_phys_virt(struct kexec_info *info,
int buf_end, int phys);
extern void arch_reuse_initrd(void);
+extern long int get_nr_cpus(void);
+
extern int ifdown(void);
extern char purgatory[];
Regards,
eric
---
v12: 9sep2022
- Rebased onto 6.0-rc4
- Addressed some minor formatting items, per Baoquan
v11: 26aug2022
https://lkml.org/lkml/2022/8/26/963
- Rebased onto 6.0-rc2
- Redid the rework of __weak to use asm/kexec.h, per Baoquan
- Reworked some comments and minor items, per Baoquan
v10: 21jul2022
https://lkml.org/lkml/2022/7/21/1007
- Rebased to 5.19.0-rc7
- Per Sourabh, corrected build issue with arch_un/map_crash_pages()
for architectures not supporting this feature.
- Per David Hildebrand, removed the WARN_ONCE() altogether.
- Per David Hansen, converted to use of kmap_local_page().
- Per Baoquan He, replaced use of __weak with the kexec technique.
v9: 13jun2022
https://lkml.org/lkml/2022/6/13/3382
- Rebased to 5.18.0
- Per Sourabh, moved crash_prepare_elf64_headers() into common
crash_core.c to avoid compile issues with kexec_load only path.
- Per David Hildebrand, replaced mutex_trylock() with mutex_lock().
- Changed the __weak arch_crash_handle_hotplug_event() to utilize
WARN_ONCE() instead of WARN(). Fix some formatting issues.
- Per Sourabh, introduced sysfs attribute crash_hotplug for memory
and CPUs; for use by userspace (udev) to determine if the kernel
performs crash hot un/plug support.
- Per Sourabh, moved the code detecting the elfcorehdr segment from
arch/x86 into crash_core:handle_hotplug_event() so both kexec_load
and kexec_file_load can benefit.
- Updated userspace kexec-tools kexec utility to reflect change to
using CRASH_MAX_MEMORY_RANGES and get_nr_cpus().
- Updated the new proposed udev rules to reflect using the sysfs
attributes crash_hotplug.
v8: 5may2022
https://lkml.org/lkml/2022/5/5/1133
- Per Borislav Petkov, eliminated CONFIG_CRASH_HOTPLUG in favor
of CONFIG_HOTPLUG_CPU || CONFIG_MEMORY_HOTPLUG, ie a new define
is not needed. Also use of IS_ENABLED() rather than #ifdef's.
Renamed crash_hotplug_handler() to handle_hotplug_event().
And other corrections.
- Per Baoquan, minimized the parameters to the arch_crash_
handle_hotplug_event() to hp_action and cpu.
- Introduce KEXEC_CRASH_HP_INVALID_CPU definition, per Baoquan.
- Per Sourabh Jain, renamed and repurposed CRASH_HOTPLUG_ELFCOREHDR_SZ
to CONFIG_CRASH_MAX_MEMORY_RANGES, mirroring kexec-tools change
by David Hildebrand. Folded this patch into the x86
kexec_file_load support patch.
v7: 13apr2022
https://lkml.org/lkml/2022/4/13/850
- Resolved parameter usage to crash_hotplug_handler(), per Baoquan.
v6: 1apr2022
https://lkml.org/lkml/2022/4/1/1203
- Reword commit messages and some comment cleanup per Baoquan.
- Changed elf_index to elfcorehdr_index for clarity.
- Minor code changes per Baoquan.
v5: 3mar2022
https://lkml.org/lkml/2022/3/3/674
- Reworded description of CRASH_HOTPLUG_ELFCOREHDR_SZ, per
David Hildenbrand.
- Refactored slightly a few patches per Baoquan recommendation.
v4: 9feb2022
https://lkml.org/lkml/2022/2/9/1406
- Refactored patches per Baoquan suggestsions.
- A few corrections, per Baoquan.
v3: 10jan2022
https://lkml.org/lkml/2022/1/10/1212
- Rebasing per Baoquan He request.
- Changed memory notifier per David Hildenbrand.
- Providing example kexec userspace change in cover letter.
RFC v2: 7dec2021
https://lkml.org/lkml/2021/12/7/1088
- Acting upon Baoquan He suggestion of removing elfcorehdr from
the purgatory list of segments, removed purgatory code from
patchset, and it is signficiantly simpler now.
RFC v1: 18nov2021
https://lkml.org/lkml/2021/11/18/845
- working patchset demonstrating kernel handling of hotplug
updates to x86 elfcorehdr for kexec_file_load
RFC: 14dec2020
https://lkml.org/lkml/2020/12/14/532
- proposed concept of allowing kernel to handle hotplug update
of elfcorehdr
---
Eric DeVolder (7):
crash: move crash_prepare_elf64_headers
crash: prototype change for crash_prepare_elf64_headers
crash: add generic infrastructure for crash hotplug support
kexec: exclude elfcorehdr from the segment digest
kexec: exclude hot remove cpu from elfcorehdr notes
crash: memory and cpu hotplug sysfs attributes
x86/crash: Add x86 crash hotplug support
.../admin-guide/mm/memory-hotplug.rst | 8 +
Documentation/core-api/cpu_hotplug.rst | 18 ++
arch/arm64/kernel/machine_kexec_file.c | 6 +-
arch/powerpc/kexec/file_load_64.c | 2 +-
arch/riscv/kernel/elf_kexec.c | 7 +-
arch/x86/Kconfig | 11 +
arch/x86/include/asm/kexec.h | 20 ++
arch/x86/kernel/crash.c | 104 +++++++-
drivers/base/cpu.c | 14 +
drivers/base/memory.c | 13 +
include/linux/crash_core.h | 8 +
include/linux/kexec.h | 41 ++-
kernel/crash_core.c | 249 ++++++++++++++++++
kernel/kexec_file.c | 105 +-------
14 files changed, 497 insertions(+), 109 deletions(-)
--
2.31.1
For x86_64, when CPU or memory is hot un/plugged, the crash
elfcorehdr, which describes the CPUs and memory in the system,
must also be updated.
When loading the crash kernel via kexec_load or kexec_file_load,
the elfcorehdr is identified at run time in
crash_core:handle_hotplug_event().
To update the elfcorehdr for x86_64, a new elfcorehdr must be
generated from the available CPUs and memory. The new elfcorehdr
is prepared into a buffer, and then installed over the top of
the existing elfcorehdr.
In the patch 'kexec: exclude elfcorehdr from the segment digest'
the need to update purgatory due to the change in elfcorehdr was
eliminated. As a result, no changes to purgatory or boot_params
(as the elfcorehdr= kernel command line parameter pointer
remains unchanged and correct) are needed, just elfcorehdr.
To accommodate a growing number of resources via hotplug, the
elfcorehdr segment must be sufficiently large enough to accommodate
changes, see the CRASH_MAX_MEMORY_RANGES configure item.
With this change, crash hotplug for kexec_file_load syscall
is supported. The kexec_load is also supported, but also
requires a corresponding change to userspace kexec-tools.
Signed-off-by: Eric DeVolder <[email protected]>
Acked-by: Baoquan He <[email protected]>
---
arch/x86/Kconfig | 11 ++++
arch/x86/include/asm/kexec.h | 20 +++++++
arch/x86/kernel/crash.c | 102 +++++++++++++++++++++++++++++++++++
3 files changed, 133 insertions(+)
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index f9920f1341c8..cdfc9b2fdf98 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -2056,6 +2056,17 @@ config CRASH_DUMP
(CONFIG_RELOCATABLE=y).
For more details see Documentation/admin-guide/kdump/kdump.rst
+config CRASH_MAX_MEMORY_RANGES
+ depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
+ int
+ default 32768
+ help
+ For the kexec_file_load path, specify the maximum number of
+ memory regions, eg. as represented by the 'System RAM' entries
+ in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
+ This value is combined with NR_CPUS and multiplied by Elf64_Phdr
+ size to determine the final buffer size.
+
config KEXEC_JUMP
bool "kexec jump"
depends on KEXEC && HIBERNATION
diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
index a3760ca796aa..432073385b2d 100644
--- a/arch/x86/include/asm/kexec.h
+++ b/arch/x86/include/asm/kexec.h
@@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
extern void kdump_nmi_shootdown_cpus(void);
+void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
+#define arch_map_crash_pages arch_map_crash_pages
+
+void arch_unmap_crash_pages(void **ptr);
+#define arch_unmap_crash_pages arch_unmap_crash_pages
+
+void arch_crash_handle_hotplug_event(struct kimage *image,
+ unsigned int hp_action);
+#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
+
+#ifdef CONFIG_HOTPLUG_CPU
+static inline int crash_hotplug_cpu_support(void) { return 1; }
+#define crash_hotplug_cpu_support crash_hotplug_cpu_support
+#endif
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+static inline int crash_hotplug_memory_support(void) { return 1; }
+#define crash_hotplug_memory_support crash_hotplug_memory_support
+#endif
+
#endif /* __ASSEMBLY__ */
#endif /* _ASM_X86_KEXEC_H */
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index 9ceb93c176a6..8fc7d678ac72 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -25,6 +25,7 @@
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/memblock.h>
+#include <linux/highmem.h>
#include <asm/processor.h>
#include <asm/hardirq.h>
@@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
image->elf_headers = kbuf.buffer;
image->elf_headers_sz = kbuf.bufsz;
+#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
+ /* Ensure elfcorehdr segment large enough for hotplug changes */
+ kbuf.memsz =
+ (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
+ sizeof(Elf64_Phdr);
+ /* Mark as usable to crash kernel, else crash kernel fails on boot */
+ image->elf_headers_sz = kbuf.memsz;
+ image->elfcorehdr_index = image->nr_segments;
+ image->elfcorehdr_index_valid = true;
+#else
kbuf.memsz = kbuf.bufsz;
+#endif
kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
ret = kexec_add_buffer(&kbuf);
@@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
return ret;
}
#endif /* CONFIG_KEXEC_FILE */
+
+#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
+/*
+ * NOTE: The addresses and sizes passed to this routine have
+ * already been fully aligned on page boundaries. There is no
+ * need for massaging the address or size.
+ */
+void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
+{
+ void *ptr = NULL;
+
+ if (size > 0) {
+ struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
+
+ ptr = kmap_local_page(page);
+ }
+
+ return ptr;
+}
+
+void arch_unmap_crash_pages(void **ptr)
+{
+ if (ptr) {
+ if (*ptr)
+ kunmap_local(*ptr);
+ *ptr = NULL;
+ }
+}
+
+/**
+ * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
+ * @image: the active struct kimage
+ * @hp_action: the hot un/plug action being handled
+ *
+ * To accurately reflect hot un/plug changes, the new elfcorehdr
+ * is prepared in a kernel buffer, and then it is written on top
+ * of the existing/old elfcorehdr.
+ */
+void arch_crash_handle_hotplug_event(struct kimage *image,
+ unsigned int hp_action)
+{
+ struct kexec_segment *ksegment;
+ unsigned char *ptr = NULL;
+ unsigned long elfsz = 0;
+ void *elfbuf = NULL;
+ unsigned long mem, memsz;
+
+ /*
+ * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
+ */
+ ksegment = &image->segment[image->elfcorehdr_index];
+ mem = ksegment->mem;
+ memsz = ksegment->memsz;
+
+ /*
+ * Create the new elfcorehdr reflecting the changes to CPU and/or
+ * memory resources.
+ */
+ if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
+ pr_err("crash hp: unable to prepare elfcore headers");
+ goto out;
+ }
+ if (elfsz > memsz) {
+ pr_err("crash hp: update elfcorehdr elfsz %lu > memsz %lu",
+ elfsz, memsz);
+ goto out;
+ }
+
+ /*
+ * At this point, we are all but assured of success.
+ * Copy new elfcorehdr into destination.
+ */
+ ptr = arch_map_crash_pages(mem, memsz);
+ if (ptr) {
+ /*
+ * Temporarily invalidate the crash image while the
+ * elfcorehdr is updated.
+ */
+ xchg(&kexec_crash_image, NULL);
+ memcpy_flushcache((void *)ptr, elfbuf, elfsz);
+ xchg(&kexec_crash_image, image);
+ }
+ arch_unmap_crash_pages((void **)&ptr);
+ pr_debug("crash hp: re-loaded elfcorehdr at 0x%lx\n", mem);
+
+out:
+ if (elfbuf)
+ vfree(elfbuf);
+}
+#endif
--
2.31.1
This introduces the crash_hotplug attribute for memory and CPUs
for use by userspace. This change directly facilitates the udev
rule for managing userspace re-loading of the crash kernel upon
hot un/plug changes.
For memory, this changeset introduces the crash_hotplug attribute
to the /sys/devices/system/memory directory. For example:
# udevadm info --attribute-walk /sys/devices/system/memory/memory81
looking at device '/devices/system/memory/memory81':
KERNEL=="memory81"
SUBSYSTEM=="memory"
DRIVER==""
ATTR{online}=="1"
ATTR{phys_device}=="0"
ATTR{phys_index}=="00000051"
ATTR{removable}=="1"
ATTR{state}=="online"
ATTR{valid_zones}=="Movable"
looking at parent device '/devices/system/memory':
KERNELS=="memory"
SUBSYSTEMS==""
DRIVERS==""
ATTRS{auto_online_blocks}=="offline"
ATTRS{block_size_bytes}=="8000000"
ATTRS{crash_hotplug}=="1"
For CPUs, this changeset introduces the crash_hotplug attribute
to the /sys/devices/system/cpu directory. For example:
# udevadm info --attribute-walk /sys/devices/system/cpu/cpu0
looking at device '/devices/system/cpu/cpu0':
KERNEL=="cpu0"
SUBSYSTEM=="cpu"
DRIVER=="processor"
ATTR{crash_notes}=="277c38600"
ATTR{crash_notes_size}=="368"
ATTR{online}=="1"
looking at parent device '/devices/system/cpu':
KERNELS=="cpu"
SUBSYSTEMS==""
DRIVERS==""
ATTRS{crash_hotplug}=="1"
ATTRS{isolated}==""
ATTRS{kernel_max}=="8191"
ATTRS{nohz_full}==" (null)"
ATTRS{offline}=="4-7"
ATTRS{online}=="0-3"
ATTRS{possible}=="0-7"
ATTRS{present}=="0-3"
With these sysfs attributes in place, it is possible to efficiently
instruct the udev rule to skip crash kernel reloading.
For example, the following is the proposed udev rule change for RHEL
system 98-kexec.rules (as the first lines of the rule file):
# The kernel handles updates to crash elfcorehdr for cpu and memory changes
SUBSYSTEM=="cpu", ATTRS{crash_hotplug}=="1", GOTO="kdump_reload_end"
SUBSYSTEM=="memory", ATTRS{crash_hotplug}=="1", GOTO="kdump_reload_end"
When examined in the context of 98-kexec.rules, the above change
tests if crash_hotplug is set, and if so, it skips the userspace
initiated unload-then-reload of the crash kernel.
Cpu and memory checks are separated in accordance with
CONFIG_HOTPLUG_CPU and CONFIG_MEMORY_HOTPLUG kernel config options.
If an architecture supports, for example, memory hotplug but not
CPU hotplug, then the /sys/devices/system/memory/crash_hotplug
attribute file is present, but the /sys/devices/system/cpu/crash_hotplug
attribute file will NOT be present. Thus the udev rule will skip
userspace processing of memory hot un/plug events, but the udev
rule will fail for CPU events, thus allowing userspace to process
cpu hot un/plug events (ie the unload-then-reload of the kdump
capture kernel).
Signed-off-by: Eric DeVolder <[email protected]>
Acked-by: Baoquan He <[email protected]>
---
.../admin-guide/mm/memory-hotplug.rst | 8 ++++++++
Documentation/core-api/cpu_hotplug.rst | 18 ++++++++++++++++++
drivers/base/cpu.c | 14 ++++++++++++++
drivers/base/memory.c | 13 +++++++++++++
include/linux/kexec.h | 8 ++++++++
5 files changed, 61 insertions(+)
diff --git a/Documentation/admin-guide/mm/memory-hotplug.rst b/Documentation/admin-guide/mm/memory-hotplug.rst
index a3c9e8ad8fa0..15fd1751a63c 100644
--- a/Documentation/admin-guide/mm/memory-hotplug.rst
+++ b/Documentation/admin-guide/mm/memory-hotplug.rst
@@ -293,6 +293,14 @@ The following files are currently defined:
Availability depends on the CONFIG_ARCH_MEMORY_PROBE
kernel configuration option.
``uevent`` read-write: generic udev file for device subsystems.
+``crash_hotplug`` read-only: when changes to the system memory map
+ occur due to hot un/plug of memory, this file contains
+ '1' if the kernel updates the kdump capture kernel memory
+ map itself (via elfcorehdr), or '0' if userspace must update
+ the kdump capture kernel memory map.
+
+ Availability depends on the CONFIG_MEMORY_HOTPLUG kernel
+ configuration option.
====================== =========================================================
.. note::
diff --git a/Documentation/core-api/cpu_hotplug.rst b/Documentation/core-api/cpu_hotplug.rst
index c6f4ba2fb32d..13e33d098645 100644
--- a/Documentation/core-api/cpu_hotplug.rst
+++ b/Documentation/core-api/cpu_hotplug.rst
@@ -750,6 +750,24 @@ will receive all events. A script like::
can process the event further.
+When changes to the CPUs in the system occur, the sysfs file
+/sys/devices/system/cpu/crash_hotplug contains '1' if the kernel
+updates the kdump capture kernel list of CPUs itself (via elfcorehdr),
+or '0' if userspace must update the kdump capture kernel list of CPUs.
+
+The availability depends on the CONFIG_HOTPLUG_CPU kernel configuration
+option.
+
+To skip userspace processing of CPU hot un/plug events for kdump
+(ie the unload-then-reload to obtain a current list of CPUs), this sysfs
+file can be used in a udev rule as follows:
+
+ SUBSYSTEM=="cpu", ATTRS{crash_hotplug}=="1", GOTO="kdump_reload_end"
+
+For a cpu hot un/plug event, if the architecture supports kernel updates
+of the elfcorehdr (which contains the list of CPUs), then the rule skips
+the unload-then-reload of the kdump capture kernel.
+
Kernel Inline Documentations Reference
======================================
diff --git a/drivers/base/cpu.c b/drivers/base/cpu.c
index 4c98849577d4..fedbf87f9d13 100644
--- a/drivers/base/cpu.c
+++ b/drivers/base/cpu.c
@@ -293,6 +293,17 @@ static ssize_t print_cpus_nohz_full(struct device *dev,
static DEVICE_ATTR(nohz_full, 0444, print_cpus_nohz_full, NULL);
#endif
+#ifdef CONFIG_HOTPLUG_CPU
+#include <linux/kexec.h>
+static ssize_t crash_hotplug_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buf)
+{
+ return sprintf(buf, "%d\n", crash_hotplug_cpu_support());
+}
+static DEVICE_ATTR_ADMIN_RO(crash_hotplug);
+#endif
+
static void cpu_device_release(struct device *dev)
{
/*
@@ -469,6 +480,9 @@ static struct attribute *cpu_root_attrs[] = {
#ifdef CONFIG_NO_HZ_FULL
&dev_attr_nohz_full.attr,
#endif
+#ifdef CONFIG_HOTPLUG_CPU
+ &dev_attr_crash_hotplug.attr,
+#endif
#ifdef CONFIG_GENERIC_CPU_AUTOPROBE
&dev_attr_modalias.attr,
#endif
diff --git a/drivers/base/memory.c b/drivers/base/memory.c
index bc60c9cd3230..b754918c3dac 100644
--- a/drivers/base/memory.c
+++ b/drivers/base/memory.c
@@ -483,6 +483,16 @@ static ssize_t auto_online_blocks_store(struct device *dev,
static DEVICE_ATTR_RW(auto_online_blocks);
+#ifdef CONFIG_MEMORY_HOTPLUG
+#include <linux/kexec.h>
+static ssize_t crash_hotplug_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ return sprintf(buf, "%d\n", crash_hotplug_memory_support());
+}
+static DEVICE_ATTR_RO(crash_hotplug);
+#endif
+
/*
* Some architectures will have custom drivers to do this, and
* will not need to do it from userspace. The fake hot-add code
@@ -887,6 +897,9 @@ static struct attribute *memory_root_attrs[] = {
&dev_attr_block_size_bytes.attr,
&dev_attr_auto_online_blocks.attr,
+#ifdef CONFIG_MEMORY_HOTPLUG
+ &dev_attr_crash_hotplug.attr,
+#endif
NULL
};
diff --git a/include/linux/kexec.h b/include/linux/kexec.h
index 9597b41136ec..a48577a36fb8 100644
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -529,6 +529,14 @@ static inline void arch_crash_handle_hotplug_event(struct kimage *image,
}
#endif
+#ifndef crash_hotplug_cpu_support
+static inline int crash_hotplug_cpu_support(void) { return 0; }
+#endif
+
+#ifndef crash_hotplug_memory_support
+static inline int crash_hotplug_memory_support(void) { return 0; }
+#endif
+
#else /* !CONFIG_KEXEC_CORE */
struct pt_regs;
struct task_struct;
--
2.31.1
On 09/09/22 at 05:05pm, Eric DeVolder wrote:
> When the kdump service is loaded, if a CPU or memory is hot
> un/plugged, the crash elfcorehdr, which describes the CPUs
> and memory in the system, must also be updated, else the resulting
> vmcore is inaccurate (eg. missing either CPU context or memory
> regions).
>
> The current solution utilizes udev to initiate an unload-then-reload
> of the kdump image (e. kernel, initrd, boot_params, puratory and
> elfcorehdr) by the userspace kexec utility. In previous posts I have
> outlined the significant performance problems related to offloading
> this activity to userspace.
>
> This patchset introduces a generic crash hot un/plug handler that
> registers with the CPU and memory notifiers. Upon CPU or memory
> changes, this generic handler is invoked and performs important
> housekeeping, for example obtaining the appropriate lock, and then
> invokes an architecture specific handler to do the appropriate
> updates.
>
> In the case of x86_64, the arch specific handler generates a new
> elfcorehdr, and overwrites the old one in memory. No involvement
> with userspace needed.
Thank a lot for all the effort, Eric.
Hi Boris, Andrew,
This version looks good to me. It introduces a framework for kdump
to react mem/cpu hotplug eveut and add x86 handler. Should this go
to x86 tree or mm tree? Please check what else we need do to fix or
improve.
Thanks
Baoquan
On Fri, Sep 09, 2022 at 05:05:09PM -0400, Eric DeVolder wrote:
> For x86_64, when CPU or memory is hot un/plugged, the crash
> elfcorehdr, which describes the CPUs and memory in the system,
> must also be updated.
>
> When loading the crash kernel via kexec_load or kexec_file_load,
Please end function names with parentheses. Check the whole patch pls.
> the elfcorehdr is identified at run time in
> crash_core:handle_hotplug_event().
>
> To update the elfcorehdr for x86_64, a new elfcorehdr must be
> generated from the available CPUs and memory. The new elfcorehdr
> is prepared into a buffer, and then installed over the top of
> the existing elfcorehdr.
>
> In the patch 'kexec: exclude elfcorehdr from the segment digest'
> the need to update purgatory due to the change in elfcorehdr was
> eliminated. As a result, no changes to purgatory or boot_params
> (as the elfcorehdr= kernel command line parameter pointer
> remains unchanged and correct) are needed, just elfcorehdr.
>
> To accommodate a growing number of resources via hotplug, the
> elfcorehdr segment must be sufficiently large enough to accommodate
> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>
> With this change, crash hotplug for kexec_file_load syscall
> is supported.
Redundant sentence.
> The kexec_load is also supported, but also
> requires a corresponding change to userspace kexec-tools.
Ditto.
> Signed-off-by: Eric DeVolder <[email protected]>
> Acked-by: Baoquan He <[email protected]>
> ---
> arch/x86/Kconfig | 11 ++++
> arch/x86/include/asm/kexec.h | 20 +++++++
> arch/x86/kernel/crash.c | 102 +++++++++++++++++++++++++++++++++++
> 3 files changed, 133 insertions(+)
>
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index f9920f1341c8..cdfc9b2fdf98 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
> (CONFIG_RELOCATABLE=y).
> For more details see Documentation/admin-guide/kdump/kdump.rst
>
> +config CRASH_MAX_MEMORY_RANGES
> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> + int
> + default 32768
> + help
> + For the kexec_file_load path, specify the maximum number of
> + memory regions, eg. as represented by the 'System RAM' entries
> + in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
> + This value is combined with NR_CPUS and multiplied by Elf64_Phdr
> + size to determine the final buffer size.
If I'm purely a user, I'm left wondering how to determine what to
specify. Do you have a guidance text somewhere you can point to from
here?
> +
> config KEXEC_JUMP
> bool "kexec jump"
> depends on KEXEC && HIBERNATION
> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
> index a3760ca796aa..432073385b2d 100644
> --- a/arch/x86/include/asm/kexec.h
> +++ b/arch/x86/include/asm/kexec.h
> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
> extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
> extern void kdump_nmi_shootdown_cpus(void);
>
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
> +#define arch_map_crash_pages arch_map_crash_pages
> +
> +void arch_unmap_crash_pages(void **ptr);
> +#define arch_unmap_crash_pages arch_unmap_crash_pages
> +
> +void arch_crash_handle_hotplug_event(struct kimage *image,
> + unsigned int hp_action);
> +#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
> +
> +#ifdef CONFIG_HOTPLUG_CPU
> +static inline int crash_hotplug_cpu_support(void) { return 1; }
> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
> +#endif
> +
> +#ifdef CONFIG_MEMORY_HOTPLUG
> +static inline int crash_hotplug_memory_support(void) { return 1; }
> +#define crash_hotplug_memory_support crash_hotplug_memory_support
> +#endif
> +
> #endif /* __ASSEMBLY__ */
>
> #endif /* _ASM_X86_KEXEC_H */
> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
> index 9ceb93c176a6..8fc7d678ac72 100644
> --- a/arch/x86/kernel/crash.c
> +++ b/arch/x86/kernel/crash.c
> @@ -25,6 +25,7 @@
> #include <linux/slab.h>
> #include <linux/vmalloc.h>
> #include <linux/memblock.h>
> +#include <linux/highmem.h>
>
> #include <asm/processor.h>
> #include <asm/hardirq.h>
> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
> image->elf_headers = kbuf.buffer;
> image->elf_headers_sz = kbuf.bufsz;
>
> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> + /* Ensure elfcorehdr segment large enough for hotplug changes */
> + kbuf.memsz =
> + (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
> + sizeof(Elf64_Phdr);
kbuf.memsz = CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
kbuf.memsz *= sizeof(Elf64_Phdr);
looks more readable to me.
> + /* Mark as usable to crash kernel, else crash kernel fails on boot */
> + image->elf_headers_sz = kbuf.memsz;
> + image->elfcorehdr_index = image->nr_segments;
> + image->elfcorehdr_index_valid = true;
> +#else
> kbuf.memsz = kbuf.bufsz;
Do that initialization at the top where you declare kbuf and get rid of
the #else branch.
> +#endif
> kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
> kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
> ret = kexec_add_buffer(&kbuf);
> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
> return ret;
> }
> #endif /* CONFIG_KEXEC_FILE */
> +
> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
This ugly ifdeffery is still here. Why don't you have stubs for the
!defined() cases in the header so that you can drop those here?
> +/*
> + * NOTE: The addresses and sizes passed to this routine have
> + * already been fully aligned on page boundaries. There is no
> + * need for massaging the address or size.
> + */
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
> +{
> + void *ptr = NULL;
> +
> + if (size > 0) {
> + struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
> +
> + ptr = kmap_local_page(page);
> + }
> +
> + return ptr;
> +}
if (size > 0)
return kmap_local_page(pfn_to_page(paddr >> PAGE_SHIFT));
else
return NULL;
That's it.
> +
> +void arch_unmap_crash_pages(void **ptr)
> +{
> + if (ptr) {
> + if (*ptr)
> + kunmap_local(*ptr);
> + *ptr = NULL;
> + }
Oh wow, this is just nuts. Why does it have to pass in a pointer to
pointer which you have to carefully check twice? And why is it a void
**?
And why are those called arch_ if all I see is the x86 variants? Are
there gonna be other arches? And even if, why can't the other arches do
kmap_local_page() too?
> +}
> +
> +/**
> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
> + * @image: the active struct kimage
> + * @hp_action: the hot un/plug action being handled
> + *
> + * To accurately reflect hot un/plug changes, the new elfcorehdr
> + * is prepared in a kernel buffer, and then it is written on top
> + * of the existing/old elfcorehdr.
> + */
> +void arch_crash_handle_hotplug_event(struct kimage *image,
> + unsigned int hp_action)
Align arguments on the opening brace.
> +{
> + struct kexec_segment *ksegment;
> + unsigned char *ptr = NULL;
> + unsigned long elfsz = 0;
> + void *elfbuf = NULL;
> + unsigned long mem, memsz;
Please sort function local variables declaration in a reverse christmas
tree order:
<type A> longest_variable_name;
<type B> shorter_var_name;
<type C> even_shorter;
<type D> i;
> +
> + /*
> + * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
Elfcorehdr_index_valid??
> + */
> + ksegment = &image->segment[image->elfcorehdr_index];
> + mem = ksegment->mem;
> + memsz = ksegment->memsz;
> +
> + /*
> + * Create the new elfcorehdr reflecting the changes to CPU and/or
> + * memory resources.
> + */
> + if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
> + pr_err("crash hp: unable to prepare elfcore headers");
^^^^^^^^
this thing is done with pr_fmt(). Grep the tree for examples.
> + goto out;
> + }
The three lines above reading ksegment need to be here, where the test
is done.
> + if (elfsz > memsz) {
> + pr_err("crash hp: update elfcorehdr elfsz %lu > memsz %lu",
> + elfsz, memsz);
> + goto out;
> + }
> +
> + /*
> + * At this point, we are all but assured of success.
Who is "we"?
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
Boris,
Thanks for the feedback! Inline responses below.
eric
On 9/12/22 01:52, Borislav Petkov wrote:
> On Fri, Sep 09, 2022 at 05:05:09PM -0400, Eric DeVolder wrote:
>> For x86_64, when CPU or memory is hot un/plugged, the crash
>> elfcorehdr, which describes the CPUs and memory in the system,
>> must also be updated.
>>
>> When loading the crash kernel via kexec_load or kexec_file_load,
>
> Please end function names with parentheses. Check the whole patch pls.
Done.
>
>> the elfcorehdr is identified at run time in
>> crash_core:handle_hotplug_event().
>>
>> To update the elfcorehdr for x86_64, a new elfcorehdr must be
>> generated from the available CPUs and memory. The new elfcorehdr
>> is prepared into a buffer, and then installed over the top of
>> the existing elfcorehdr.
>>
>> In the patch 'kexec: exclude elfcorehdr from the segment digest'
>> the need to update purgatory due to the change in elfcorehdr was
>> eliminated. As a result, no changes to purgatory or boot_params
>> (as the elfcorehdr= kernel command line parameter pointer
>> remains unchanged and correct) are needed, just elfcorehdr.
>>
>> To accommodate a growing number of resources via hotplug, the
>> elfcorehdr segment must be sufficiently large enough to accommodate
>> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>>
>> With this change, crash hotplug for kexec_file_load syscall
>> is supported.
>
> Redundant sentence.
Removed.
>
>> The kexec_load is also supported, but also
>> requires a corresponding change to userspace kexec-tools.
>
> Ditto.
Removed.
>
>> Signed-off-by: Eric DeVolder <[email protected]>
>> Acked-by: Baoquan He <[email protected]>
>> ---
>> arch/x86/Kconfig | 11 ++++
>> arch/x86/include/asm/kexec.h | 20 +++++++
>> arch/x86/kernel/crash.c | 102 +++++++++++++++++++++++++++++++++++
>> 3 files changed, 133 insertions(+)
>>
>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>> index f9920f1341c8..cdfc9b2fdf98 100644
>> --- a/arch/x86/Kconfig
>> +++ b/arch/x86/Kconfig
>> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
>> (CONFIG_RELOCATABLE=y).
>> For more details see Documentation/admin-guide/kdump/kdump.rst
>>
>> +config CRASH_MAX_MEMORY_RANGES
>> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
>> + int
>> + default 32768
>> + help
>> + For the kexec_file_load path, specify the maximum number of
>> + memory regions, eg. as represented by the 'System RAM' entries
>> + in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
>> + This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>> + size to determine the final buffer size.
>
> If I'm purely a user, I'm left wondering how to determine what to
> specify. Do you have a guidance text somewhere you can point to from
> here?
This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
David points out that terminology is tricky here due to differing behaviors.
And perhaps that is your point in asking for guidance text. It can be
complicated, but it all comes down to System RAM entries.
I could perhaps offer an overly simplified example such that for 1GiB block
size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow for 32TiB
of memory?
>
>> +
>> config KEXEC_JUMP
>> bool "kexec jump"
>> depends on KEXEC && HIBERNATION
>> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
>> index a3760ca796aa..432073385b2d 100644
>> --- a/arch/x86/include/asm/kexec.h
>> +++ b/arch/x86/include/asm/kexec.h
>> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
>> extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
>> extern void kdump_nmi_shootdown_cpus(void);
>>
>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
>> +#define arch_map_crash_pages arch_map_crash_pages
>> +
>> +void arch_unmap_crash_pages(void **ptr);
>> +#define arch_unmap_crash_pages arch_unmap_crash_pages
>> +
>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>> + unsigned int hp_action);
>> +#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
>> +
>> +#ifdef CONFIG_HOTPLUG_CPU
>> +static inline int crash_hotplug_cpu_support(void) { return 1; }
>> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
>> +#endif
>> +
>> +#ifdef CONFIG_MEMORY_HOTPLUG
>> +static inline int crash_hotplug_memory_support(void) { return 1; }
>> +#define crash_hotplug_memory_support crash_hotplug_memory_support
>> +#endif
>> +
>> #endif /* __ASSEMBLY__ */
>>
>> #endif /* _ASM_X86_KEXEC_H */
>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>> index 9ceb93c176a6..8fc7d678ac72 100644
>> --- a/arch/x86/kernel/crash.c
>> +++ b/arch/x86/kernel/crash.c
>> @@ -25,6 +25,7 @@
>> #include <linux/slab.h>
>> #include <linux/vmalloc.h>
>> #include <linux/memblock.h>
>> +#include <linux/highmem.h>
>>
>> #include <asm/processor.h>
>> #include <asm/hardirq.h>
>> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
>> image->elf_headers = kbuf.buffer;
>> image->elf_headers_sz = kbuf.bufsz;
>>
>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>> + /* Ensure elfcorehdr segment large enough for hotplug changes */
>> + kbuf.memsz =
>> + (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
>> + sizeof(Elf64_Phdr);
>
>
> kbuf.memsz = CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
> kbuf.memsz *= sizeof(Elf64_Phdr);
>
> looks more readable to me.
Done.
>
>
>> + /* Mark as usable to crash kernel, else crash kernel fails on boot */
>> + image->elf_headers_sz = kbuf.memsz;
>> + image->elfcorehdr_index = image->nr_segments;
>> + image->elfcorehdr_index_valid = true;
>> +#else
>> kbuf.memsz = kbuf.bufsz;
>
> Do that initialization at the top where you declare kbuf and get rid of
> the #else branch.
The kbuf.bufsz value is obtained via a call to prepare_elf_headers(); I can not initialize it at its
declaration.
>
>> +#endif
>> kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>> kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>> ret = kexec_add_buffer(&kbuf);
>> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
>> return ret;
>> }
>> #endif /* CONFIG_KEXEC_FILE */
>> +
>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>
> This ugly ifdeffery is still here. Why don't you have stubs for the
> !defined() cases in the header so that you can drop those here?
>
I'm at a loss as to what to do differently here. You've raised this issue before and I went back and
looked at the suggestions then and I don't see how that applies to this situation. How is this
situation different than the #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
I've included a copy of the current state of this section below for additional markup.
>> +/*
>> + * NOTE: The addresses and sizes passed to this routine have
>> + * already been fully aligned on page boundaries. There is no
>> + * need for massaging the address or size.
>> + */
>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
>> +{
>> + void *ptr = NULL;
>> +
>> + if (size > 0) {
>> + struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
>> +
>> + ptr = kmap_local_page(page);
>> + }
>> +
>> + return ptr;
>> +}
>
> if (size > 0)
> return kmap_local_page(pfn_to_page(paddr >> PAGE_SHIFT));
> else
> return NULL;
>
> That's it.
Done.
>
>> +
>> +void arch_unmap_crash_pages(void **ptr)
>> +{
>> + if (ptr) {
>> + if (*ptr)
>> + kunmap_local(*ptr);
>> + *ptr = NULL;
>> + }
>
> Oh wow, this is just nuts. Why does it have to pass in a pointer to
> pointer which you have to carefully check twice? And why is it a void
> **?
A long time ago this made sense, but it no longer makes sense. I've corrected this.
>
> And why are those called arch_ if all I see is the x86 variants? Are
> there gonna be other arches? And even if, why can't the other arches do
> kmap_local_page() too?
Currently there is a concurrent effort for PPC support by Sourabh Jain, and in that effort
arch_map_crash_pages() is using __va(paddr).
I do not know the nuances between kmap_local_page() and __va() to answer the question.
If kmap_local_page() works for all archs, then I'm happy to drop these arch_ variants
and use it directly.
>
>> +}
>> +
>> +/**
>> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
>> + * @image: the active struct kimage
>> + * @hp_action: the hot un/plug action being handled
>> + *
>> + * To accurately reflect hot un/plug changes, the new elfcorehdr
>> + * is prepared in a kernel buffer, and then it is written on top
>> + * of the existing/old elfcorehdr.
>> + */
>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>> + unsigned int hp_action)
>
> Align arguments on the opening brace.
Done.
>
>> +{
>> + struct kexec_segment *ksegment;
>> + unsigned char *ptr = NULL;
>> + unsigned long elfsz = 0;
>> + void *elfbuf = NULL;
>> + unsigned long mem, memsz;
>
> Please sort function local variables declaration in a reverse christmas
> tree order:
>
> <type A> longest_variable_name;
> <type B> shorter_var_name;
> <type C> even_shorter;
> <type D> i;
>
Done.
>> +
>> + /*
>> + * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
>
> Elfcorehdr_index_valid??
Comment reworked.
>
>> + */
>> + ksegment = &image->segment[image->elfcorehdr_index];
>> + mem = ksegment->mem;
>> + memsz = ksegment->memsz;
>> +
>> + /*
>> + * Create the new elfcorehdr reflecting the changes to CPU and/or
>> + * memory resources.
>> + */
>> + if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
>> + pr_err("crash hp: unable to prepare elfcore headers");
> ^^^^^^^^
>
> this thing is done with pr_fmt(). Grep the tree for examples.
Done, thanks for pointing that out.
>
>> + goto out;
>> + }
>
> The three lines above reading ksegment need to be here, where the test
> is done.
Done.
>
>> + if (elfsz > memsz) {
>> + pr_err("crash hp: update elfcorehdr elfsz %lu > memsz %lu",
>> + elfsz, memsz);
>> + goto out;
>> + }
>> +
>> + /*
>> + * At this point, we are all but assured of success.
>
> Who is "we"?
>
Comment reworked.
Here is a copy of the current state of this code, for determining how to address the question above.
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
#undef pr_fmt
#define pr_fmt(fmt) "crash hp: " fmt
/*
* NOTE: The addresses and sizes passed to this routine have
* already been fully aligned on page boundaries. There is no
* need for massaging the address or size.
*/
void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
{
if (size > 0)
return kmap_local_page(pfn_to_page(paddr >> PAGE_SHIFT));
else
return NULL;
}
void arch_unmap_crash_pages(void *ptr)
{
if (ptr)
kunmap_local(ptr);
}
/**
* arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
* @image: the active struct kimage
* @hp_action: the hot un/plug action being handled
*
* To accurately reflect hot un/plug changes, the new elfcorehdr
* is prepared in a kernel buffer, and then it is written on top
* of the existing/old elfcorehdr.
*/
void arch_crash_handle_hotplug_event(struct kimage *image,
unsigned int hp_action)
{
unsigned long mem, memsz;
unsigned long elfsz = 0;
void *elfbuf = NULL;
void *ptr;
/*
* Create the new elfcorehdr reflecting the changes to CPU and/or
* memory resources.
*/
if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
pr_err("unable to prepare elfcore headers");
goto out;
}
/*
* Obtain address and size of the elfcorehdr segment, and
* check it against the new elfcorehdr buffer.
*/
mem = image->segment[image->elfcorehdr_index].mem;
memsz = image->segment[image->elfcorehdr_index].memsz;
if (elfsz > memsz) {
pr_err("update elfcorehdr elfsz %lu > memsz %lu",
elfsz, memsz);
goto out;
}
/*
* Copy new elfcorehdr over the old elfcorehdr at destination.
*/
ptr = arch_map_crash_pages(mem, memsz);
if (ptr) {
/*
* Temporarily invalidate the crash image while the
* elfcorehdr is updated.
*/
xchg(&kexec_crash_image, NULL);
memcpy_flushcache(ptr, elfbuf, elfsz);
xchg(&kexec_crash_image, image);
}
arch_unmap_crash_pages(ptr);
pr_debug("re-loaded elfcorehdr at 0x%lx\n", mem);
out:
if (elfbuf)
vfree(elfbuf);
}
#endif
On 10/09/22 02:35, Eric DeVolder wrote:
> For x86_64, when CPU or memory is hot un/plugged, the crash
> elfcorehdr, which describes the CPUs and memory in the system,
> must also be updated.
>
> When loading the crash kernel via kexec_load or kexec_file_load,
> the elfcorehdr is identified at run time in
> crash_core:handle_hotplug_event().
>
> To update the elfcorehdr for x86_64, a new elfcorehdr must be
> generated from the available CPUs and memory. The new elfcorehdr
> is prepared into a buffer, and then installed over the top of
> the existing elfcorehdr.
>
> In the patch 'kexec: exclude elfcorehdr from the segment digest'
> the need to update purgatory due to the change in elfcorehdr was
> eliminated. As a result, no changes to purgatory or boot_params
> (as the elfcorehdr= kernel command line parameter pointer
> remains unchanged and correct) are needed, just elfcorehdr.
>
> To accommodate a growing number of resources via hotplug, the
> elfcorehdr segment must be sufficiently large enough to accommodate
> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>
> With this change, crash hotplug for kexec_file_load syscall
> is supported. The kexec_load is also supported, but also
> requires a corresponding change to userspace kexec-tools.
>
> Signed-off-by: Eric DeVolder <[email protected]>
> Acked-by: Baoquan He <[email protected]>
> ---
> arch/x86/Kconfig | 11 ++++
> arch/x86/include/asm/kexec.h | 20 +++++++
> arch/x86/kernel/crash.c | 102 +++++++++++++++++++++++++++++++++++
> 3 files changed, 133 insertions(+)
>
> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> index f9920f1341c8..cdfc9b2fdf98 100644
> --- a/arch/x86/Kconfig
> +++ b/arch/x86/Kconfig
> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
> (CONFIG_RELOCATABLE=y).
> For more details see Documentation/admin-guide/kdump/kdump.rst
>
> +config CRASH_MAX_MEMORY_RANGES
> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> + int
> + default 32768
> + help
> + For the kexec_file_load path, specify the maximum number of
> + memory regions, eg. as represented by the 'System RAM' entries
> + in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
> + This value is combined with NR_CPUS and multiplied by Elf64_Phdr
> + size to determine the final buffer size.
> +
> config KEXEC_JUMP
> bool "kexec jump"
> depends on KEXEC && HIBERNATION
> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
> index a3760ca796aa..432073385b2d 100644
> --- a/arch/x86/include/asm/kexec.h
> +++ b/arch/x86/include/asm/kexec.h
> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
> extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
> extern void kdump_nmi_shootdown_cpus(void);
>
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
> +#define arch_map_crash_pages arch_map_crash_pages
> +
> +void arch_unmap_crash_pages(void **ptr);
> +#define arch_unmap_crash_pages arch_unmap_crash_pages
> +
> +void arch_crash_handle_hotplug_event(struct kimage *image,
> + unsigned int hp_action);
> +#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
> +
> +#ifdef CONFIG_HOTPLUG_CPU
> +static inline int crash_hotplug_cpu_support(void) { return 1; }
> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
> +#endif
> +
> +#ifdef CONFIG_MEMORY_HOTPLUG
> +static inline int crash_hotplug_memory_support(void) { return 1; }
> +#define crash_hotplug_memory_support crash_hotplug_memory_support
> +#endif
> +
> #endif /* __ASSEMBLY__ */
>
> #endif /* _ASM_X86_KEXEC_H */
> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
> index 9ceb93c176a6..8fc7d678ac72 100644
> --- a/arch/x86/kernel/crash.c
> +++ b/arch/x86/kernel/crash.c
> @@ -25,6 +25,7 @@
> #include <linux/slab.h>
> #include <linux/vmalloc.h>
> #include <linux/memblock.h>
> +#include <linux/highmem.h>
>
> #include <asm/processor.h>
> #include <asm/hardirq.h>
> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
> image->elf_headers = kbuf.buffer;
> image->elf_headers_sz = kbuf.bufsz;
>
> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> + /* Ensure elfcorehdr segment large enough for hotplug changes */
> + kbuf.memsz =
> + (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
> + sizeof(Elf64_Phdr);
> + /* Mark as usable to crash kernel, else crash kernel fails on boot */
> + image->elf_headers_sz = kbuf.memsz;
> + image->elfcorehdr_index = image->nr_segments;
> + image->elfcorehdr_index_valid = true;
> +#else
> kbuf.memsz = kbuf.bufsz;
> +#endif
> kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
> kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
> ret = kexec_add_buffer(&kbuf);
> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
> return ret;
> }
> #endif /* CONFIG_KEXEC_FILE */
> +
> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> +/*
> + * NOTE: The addresses and sizes passed to this routine have
> + * already been fully aligned on page boundaries. There is no
> + * need for massaging the address or size.
> + */
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
> +{
> + void *ptr = NULL;
> +
> + if (size > 0) {
> + struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
> +
> + ptr = kmap_local_page(page);
> + }
> +
> + return ptr;
> +}
> +
> +void arch_unmap_crash_pages(void **ptr)
> +{
> + if (ptr) {
> + if (*ptr)
> + kunmap_local(*ptr);
> + *ptr = NULL;
> + }
> +}
> +
> +/**
> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
> + * @image: the active struct kimage
> + * @hp_action: the hot un/plug action being handled
> + *
> + * To accurately reflect hot un/plug changes, the new elfcorehdr
> + * is prepared in a kernel buffer, and then it is written on top
> + * of the existing/old elfcorehdr.
> + */
> +void arch_crash_handle_hotplug_event(struct kimage *image,
> + unsigned int hp_action)
> +{
> + struct kexec_segment *ksegment;
> + unsigned char *ptr = NULL;
> + unsigned long elfsz = 0;
> + void *elfbuf = NULL;
> + unsigned long mem, memsz;
> +
> + /*
> + * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
> + */
> + ksegment = &image->segment[image->elfcorehdr_index];
> + mem = ksegment->mem;
> + memsz = ksegment->memsz;
> +
> + /*
> + * Create the new elfcorehdr reflecting the changes to CPU and/or
> + * memory resources.
> + */
> + if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
> + pr_err("crash hp: unable to prepare elfcore headers");
> + goto out;
On PowerPC, while preparing the elf core header the memblock structure
is used to prepare program header for memory regions of elfcorehdr.
Since the above arch specific hotplug handler gets invoked when memory
is marked offline (MEM_OFFLINE) which is before memblock structure gets
updated so on PowerPC the above handler may not work for memory hotplug
case.
Just wondering which data structure is used to get the list of memory
regions while preparing program header for memory regions of elfcorehdr
on other architectures?
Thanks,
Sourabh Jain
Boris,
I've a few questions for you below. With your responses, I am hopeful we can finish this series soon!
Thanks,
eric
On 9/13/22 14:12, Eric DeVolder wrote:
> Boris,
> Thanks for the feedback! Inline responses below.
> eric
>
> On 9/12/22 01:52, Borislav Petkov wrote:
>> On Fri, Sep 09, 2022 at 05:05:09PM -0400, Eric DeVolder wrote:
>>> For x86_64, when CPU or memory is hot un/plugged, the crash
>>> elfcorehdr, which describes the CPUs and memory in the system,
>>> must also be updated.
>>>
>>> When loading the crash kernel via kexec_load or kexec_file_load,
>>
>> Please end function names with parentheses. Check the whole patch pls.
> Done.
>
>>
>>> the elfcorehdr is identified at run time in
>>> crash_core:handle_hotplug_event().
>>>
>>> To update the elfcorehdr for x86_64, a new elfcorehdr must be
>>> generated from the available CPUs and memory. The new elfcorehdr
>>> is prepared into a buffer, and then installed over the top of
>>> the existing elfcorehdr.
>>>
>>> In the patch 'kexec: exclude elfcorehdr from the segment digest'
>>> the need to update purgatory due to the change in elfcorehdr was
>>> eliminated. As a result, no changes to purgatory or boot_params
>>> (as the elfcorehdr= kernel command line parameter pointer
>>> remains unchanged and correct) are needed, just elfcorehdr.
>>>
>>> To accommodate a growing number of resources via hotplug, the
>>> elfcorehdr segment must be sufficiently large enough to accommodate
>>> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>>>
>>> With this change, crash hotplug for kexec_file_load syscall
>>> is supported.
>>
>> Redundant sentence.
> Removed.
>
>>
>>> The kexec_load is also supported, but also
>>> requires a corresponding change to userspace kexec-tools.
>>
>> Ditto.
> Removed.
>
>>
>>> Signed-off-by: Eric DeVolder <[email protected]>
>>> Acked-by: Baoquan He <[email protected]>
>>> ---
>>>  arch/x86/Kconfig            | 11 ++++
>>> Â arch/x86/include/asm/kexec.h |Â 20 +++++++
>>>  arch/x86/kernel/crash.c     | 102 +++++++++++++++++++++++++++++++++++
>>> Â 3 files changed, 133 insertions(+)
>>>
>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>> index f9920f1341c8..cdfc9b2fdf98 100644
>>> --- a/arch/x86/Kconfig
>>> +++ b/arch/x86/Kconfig
>>> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
>>> Â Â Â Â Â Â Â (CONFIG_RELOCATABLE=y).
>>> Â Â Â Â Â Â Â For more details see Documentation/admin-guide/kdump/kdump.rst
>>> +config CRASH_MAX_MEMORY_RANGES
>>> +Â Â Â depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
>>> +Â Â Â int
>>> +Â Â Â default 32768
>>> +Â Â Â help
>>> +Â Â Â Â Â For the kexec_file_load path, specify the maximum number of
>>> +Â Â Â Â Â memory regions, eg. as represented by the 'System RAM' entries
>>> +Â Â Â Â Â in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
>>> +Â Â Â Â Â This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>>> +Â Â Â Â Â size to determine the final buffer size.
>>
>> If I'm purely a user, I'm left wondering how to determine what to
>> specify. Do you have a guidance text somewhere you can point to from
>> here?
>
> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
> David points out that terminology is tricky here due to differing behaviors.
> And perhaps that is your point in asking for guidance text. It can be
> complicated, but it all comes down to System RAM entries.
>
> I could perhaps offer an overly simplified example such that for 1GiB block
> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow for 32TiB
> of memory?
>
>>
>>> +
>>> Â config KEXEC_JUMP
>>> Â Â Â Â Â bool "kexec jump"
>>> Â Â Â Â Â depends on KEXEC && HIBERNATION
>>> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
>>> index a3760ca796aa..432073385b2d 100644
>>> --- a/arch/x86/include/asm/kexec.h
>>> +++ b/arch/x86/include/asm/kexec.h
>>> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
>>> Â extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
>>> Â extern void kdump_nmi_shootdown_cpus(void);
>>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
>>> +#define arch_map_crash_pages arch_map_crash_pages
>>> +
>>> +void arch_unmap_crash_pages(void **ptr);
>>> +#define arch_unmap_crash_pages arch_unmap_crash_pages
>>> +
>>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>>> +Â Â Â Â Â Â Â unsigned int hp_action);
>>> +#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
>>> +
>>> +#ifdef CONFIG_HOTPLUG_CPU
>>> +static inline int crash_hotplug_cpu_support(void) { return 1; }
>>> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
>>> +#endif
>>> +
>>> +#ifdef CONFIG_MEMORY_HOTPLUG
>>> +static inline int crash_hotplug_memory_support(void) { return 1; }
>>> +#define crash_hotplug_memory_support crash_hotplug_memory_support
>>> +#endif
>>> +
>>> Â #endif /* __ASSEMBLY__ */
>>> Â #endif /* _ASM_X86_KEXEC_H */
>>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>>> index 9ceb93c176a6..8fc7d678ac72 100644
>>> --- a/arch/x86/kernel/crash.c
>>> +++ b/arch/x86/kernel/crash.c
>>> @@ -25,6 +25,7 @@
>>> Â #include <linux/slab.h>
>>> Â #include <linux/vmalloc.h>
>>> Â #include <linux/memblock.h>
>>> +#include <linux/highmem.h>
>>> Â #include <asm/processor.h>
>>> Â #include <asm/hardirq.h>
>>> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â image->elf_headers = kbuf.buffer;
>>> Â Â Â Â Â image->elf_headers_sz = kbuf.bufsz;
>>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>> +Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>>> +Â Â Â kbuf.memsz =
>>> +Â Â Â Â Â Â Â (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
>>> +Â Â Â Â Â Â Â Â Â Â Â sizeof(Elf64_Phdr);
>>
>>
>>     kbuf.memsz = CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
>> Â Â Â Â kbuf.memsz *= sizeof(Elf64_Phdr);
>>
>> looks more readable to me.
> Done.
>
>>
>>
>>> +Â Â Â /* Mark as usable to crash kernel, else crash kernel fails on boot */
>>> +Â Â Â image->elf_headers_sz = kbuf.memsz;
>>> +Â Â Â image->elfcorehdr_index = image->nr_segments;
>>> +Â Â Â image->elfcorehdr_index_valid = true;
>>> +#else
>>> Â Â Â Â Â kbuf.memsz = kbuf.bufsz;
>>
>> Do that initialization at the top where you declare kbuf and get rid of
>> the #else branch.
> The kbuf.bufsz value is obtained via a call to prepare_elf_headers(); I can not initialize it at its
> declaration.
>
>>
>>> +#endif
>>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>>> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â return ret;
>>> Â }
>>> Â #endif /* CONFIG_KEXEC_FILE */
>>> +
>>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>
>> This ugly ifdeffery is still here. Why don't you have stubs for the
>> !defined() cases in the header so that you can drop those here?
>>
>
> I'm at a loss as to what to do differently here. You've raised this issue before and I went back and
> looked at the suggestions then and I don't see how that applies to this situation. How is this
> situation different than the #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
>
> I've included a copy of the current state of this section below for additional markup.
>
>>> +/*
>>> + * NOTE: The addresses and sizes passed to this routine have
>>> + * already been fully aligned on page boundaries. There is no
>>> + * need for massaging the address or size.
>>> + */
>>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
>>> +{
>>> +Â Â Â void *ptr = NULL;
>>> +
>>> +Â Â Â if (size > 0) {
>>> +Â Â Â Â Â Â Â struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
>>> +
>>> +Â Â Â Â Â Â Â ptr = kmap_local_page(page);
>>> +Â Â Â }
>>> +
>>> +Â Â Â return ptr;
>>> +}
>>
>> Â Â Â Â if (size > 0)
>> Â Â Â Â Â Â Â return kmap_local_page(pfn_to_page(paddr >> PAGE_SHIFT));
>> Â Â Â Â else
>> Â Â Â Â Â Â Â return NULL;
>>
>> That's it.
> Done.
>
>>
>>> +
>>> +void arch_unmap_crash_pages(void **ptr)
>>> +{
>>> +Â Â Â if (ptr) {
>>> +Â Â Â Â Â Â Â if (*ptr)
>>> +Â Â Â Â Â Â Â Â Â Â Â kunmap_local(*ptr);
>>> +Â Â Â Â Â Â Â *ptr = NULL;
>>> +Â Â Â }
>>
>> Oh wow, this is just nuts. Why does it have to pass in a pointer to
>> pointer which you have to carefully check twice? And why is it a void
>> **?
> A long time ago this made sense, but it no longer makes sense. I've corrected this.
>
>>
>> And why are those called arch_ if all I see is the x86 variants? Are
>> there gonna be other arches? And even if, why can't the other arches do
>> kmap_local_page() too?
> Currently there is a concurrent effort for PPC support by Sourabh Jain, and in that effort
> arch_map_crash_pages() is using __va(paddr).
>
> I do not know the nuances between kmap_local_page() and __va() to answer the question.
>
> If kmap_local_page() works for all archs, then I'm happy to drop these arch_ variants
> and use it directly.
>
>>
>>> +}
>>> +
>>> +/**
>>> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
>>> + * @image: the active struct kimage
>>> + * @hp_action: the hot un/plug action being handled
>>> + *
>>> + * To accurately reflect hot un/plug changes, the new elfcorehdr
>>> + * is prepared in a kernel buffer, and then it is written on top
>>> + * of the existing/old elfcorehdr.
>>> + */
>>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>>> +Â Â Â unsigned int hp_action)
>>
>> Align arguments on the opening brace.
> Done.
>
>>
>>> +{
>>> +Â Â Â struct kexec_segment *ksegment;
>>> +Â Â Â unsigned char *ptr = NULL;
>>> +Â Â Â unsigned long elfsz = 0;
>>> +Â Â Â void *elfbuf = NULL;
>>> +Â Â Â unsigned long mem, memsz;
>>
>> Please sort function local variables declaration in a reverse christmas
>> tree order:
>>
>> Â Â Â Â <type A> longest_variable_name;
>> Â Â Â Â <type B> shorter_var_name;
>> Â Â Â Â <type C> even_shorter;
>> Â Â Â Â <type D> i;
>>
> Done.
>
>>> +
>>> +Â Â Â /*
>>> +Â Â Â Â * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
>>
>> Elfcorehdr_index_valid??
> Comment reworked.
>
>
>>
>>> +Â Â Â Â */
>>> +Â Â Â ksegment = &image->segment[image->elfcorehdr_index];
>>> +Â Â Â mem = ksegment->mem;
>>> +Â Â Â memsz = ksegment->memsz;
>>> +
>>> +Â Â Â /*
>>> +Â Â Â Â * Create the new elfcorehdr reflecting the changes to CPU and/or
>>> +Â Â Â Â * memory resources.
>>> +Â Â Â Â */
>>> +Â Â Â if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
>>> +Â Â Â Â Â Â Â pr_err("crash hp: unable to prepare elfcore headers");
>> Â Â Â Â Â Â Â Â Â Â Â ^^^^^^^^
>>
>> this thing is done with pr_fmt(). Grep the tree for examples.
> Done, thanks for pointing that out.
>
>>
>>> +Â Â Â Â Â Â Â goto out;
>>> +Â Â Â }
>>
>> The three lines above reading ksegment need to be here, where the test
>> is done.
> Done.
>
>>
>>> +Â Â Â if (elfsz > memsz) {
>>> +Â Â Â Â Â Â Â pr_err("crash hp: update elfcorehdr elfsz %lu > memsz %lu",
>>> +Â Â Â Â Â Â Â Â Â Â Â elfsz, memsz);
>>> +Â Â Â Â Â Â Â goto out;
>>> +Â Â Â }
>>> +
>>> +Â Â Â /*
>>> +Â Â Â Â * At this point, we are all but assured of success.
>>
>> Who is "we"?
>>
> Comment reworked.
>
>
> Here is a copy of the current state of this code, for determining how to address the question above.
>
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>
> #undef pr_fmt
> #define pr_fmt(fmt) "crash hp: " fmt
>
> /*
> Â * NOTE: The addresses and sizes passed to this routine have
> Â * already been fully aligned on page boundaries. There is no
> Â * need for massaging the address or size.
> Â */
> void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
> {
> Â Â Â Â Â Â Â if (size > 0)
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â return kmap_local_page(pfn_to_page(paddr >> PAGE_SHIFT));
> Â Â Â Â Â Â Â else
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â return NULL;
> }
>
> void arch_unmap_crash_pages(void *ptr)
> {
> Â Â Â Â Â Â Â if (ptr)
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â kunmap_local(ptr);
> }
>
> /**
> Â * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
> Â * @image: the active struct kimage
> Â * @hp_action: the hot un/plug action being handled
> Â *
> Â * To accurately reflect hot un/plug changes, the new elfcorehdr
> Â * is prepared in a kernel buffer, and then it is written on top
> Â * of the existing/old elfcorehdr.
> Â */
> void arch_crash_handle_hotplug_event(struct kimage *image,
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â unsigned int hp_action)
> {
> Â Â Â Â Â Â Â unsigned long mem, memsz;
> Â Â Â Â Â Â Â unsigned long elfsz = 0;
> Â Â Â Â Â Â Â void *elfbuf = NULL;
> Â Â Â Â Â Â Â void *ptr;
>
> Â Â Â Â Â Â Â /*
> Â Â Â Â Â Â Â Â * Create the new elfcorehdr reflecting the changes to CPU and/or
> Â Â Â Â Â Â Â Â * memory resources.
> Â Â Â Â Â Â Â Â */
> Â Â Â Â Â Â Â if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â pr_err("unable to prepare elfcore headers");
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â goto out;
> Â Â Â Â Â Â Â }
>
> Â Â Â Â Â Â Â /*
> Â Â Â Â Â Â Â Â * Obtain address and size of the elfcorehdr segment, and
> Â Â Â Â Â Â Â Â * check it against the new elfcorehdr buffer.
> Â Â Â Â Â Â Â Â */
> Â Â Â Â Â Â Â mem = image->segment[image->elfcorehdr_index].mem;
> Â Â Â Â Â Â Â memsz = image->segment[image->elfcorehdr_index].memsz;
> Â Â Â Â Â Â Â if (elfsz > memsz) {
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â pr_err("update elfcorehdr elfsz %lu > memsz %lu",
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â elfsz, memsz);
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â goto out;
> Â Â Â Â Â Â Â }
>
> Â Â Â Â Â Â Â /*
> Â Â Â Â Â Â Â Â * Copy new elfcorehdr over the old elfcorehdr at destination.
> Â Â Â Â Â Â Â Â */
> Â Â Â Â Â Â Â ptr = arch_map_crash_pages(mem, memsz);
> Â Â Â Â Â Â Â if (ptr) {
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â /*
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â * Temporarily invalidate the crash image while the
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â * elfcorehdr is updated.
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â */
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â xchg(&kexec_crash_image, NULL);
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â memcpy_flushcache(ptr, elfbuf, elfsz);
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â xchg(&kexec_crash_image, image);
> Â Â Â Â Â Â Â }
> Â Â Â Â Â Â Â arch_unmap_crash_pages(ptr);
> Â Â Â Â Â Â Â pr_debug("re-loaded elfcorehdr at 0x%lx\n", mem);
>
> out:
> Â Â Â Â Â Â Â if (elfbuf)
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â vfree(elfbuf);
> }
> #endif
>
On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
Please do not use lkml.org to refer to lkml messages. We have a
perfectly fine archival system at lore.kernel.org. You simply do
https://lore.kernel.org/r/<Message-ID>
when you want to point to a previous mail.
> David points out that terminology is tricky here due to differing behaviors.
> And perhaps that is your point in asking for guidance text. It can be
> complicated
Which means you need an explanation how to use this even more.
And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
something you discover from the hardware?
Your help text talks about System RAM entries in /proc/iomem which means
that those entries are present somewhere in the kernel and you can read
them out and do the proper calculations dynamically instead of doing the
static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
> , but it all comes down to System RAM entries.
>
> I could perhaps offer an overly simplified example such that for 1GiB block
> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow for 32TiB
> of memory?
Yes, and stick it somewhere in Documentation/admin-guide/kdump/ and
refer to it in that help text so that people can find it and read how to
use your new option.
> The kbuf.bufsz value is obtained via a call to prepare_elf_headers(); I can
> not initialize it at its declaration.
Sorry, I meant this:
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index 8fc7d678ac72..ee6fd9f1b2b9 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
if (ret)
return ret;
- image->elf_headers = kbuf.buffer;
- image->elf_headers_sz = kbuf.bufsz;
+ image->elf_headers = kbuf.buffer;
+ image->elf_headers_sz = kbuf.bufsz;
+ kbuf.memsz = kbuf.bufsz;
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
/* Ensure elfcorehdr segment large enough for hotplug changes */
@@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
image->elf_headers_sz = kbuf.memsz;
image->elfcorehdr_index = image->nr_segments;
image->elfcorehdr_index_valid = true;
-#else
- kbuf.memsz = kbuf.bufsz;
#endif
+
kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
ret = kexec_add_buffer(&kbuf);
> I'm at a loss as to what to do differently here. You've raised this issue
> before and I went back and looked at the suggestions then and I don't see
> how that applies to this situation. How is this situation different than the
> #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
See the diff at the end. I'm not saying this is how you should do it
but it should give you a better idea. The logic being, the functions
in the .c file don't really need ifdeffery around them - you're adding
1-2 functions and crash.c is not that big - so they can be built in
unconditionally. You'd need the ifdeffery *in the header only* when
crash.c is not being built.
But I've done it with ifdeffery in the .c file now because yes, the
kexec code is a minefield of ifdeffery. Hell, there's ifdeffery even in
the headers for structs. Ifdeffery you don't really need. Someone should
clean that up and simplify this immensely.
> Currently there is a concurrent effort for PPC support by Sourabh
> Jain, and in that effort arch_map_crash_pages() is using __va(paddr).
Why?
> I do not know the nuances between kmap_local_page() and __va() to
> answer the question.
kmap_local_page() is a generic interface and it should work on any arch.
And it is documented even:
$ git grep kmap_local_page Documentation/
> If kmap_local_page() works for all archs, then I'm happy to drop these
> arch_ variants and use it directly.
Yes, pls do.
---
diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
index 432073385b2d..b73c9628cd85 100644
--- a/arch/x86/include/asm/kexec.h
+++ b/arch/x86/include/asm/kexec.h
@@ -205,6 +205,17 @@ void *arch_kexec_kernel_image_load(struct kimage *image);
int arch_kimage_file_post_load_cleanup(struct kimage *image);
#define arch_kimage_file_post_load_cleanup arch_kimage_file_post_load_cleanup
+
+#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
+void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
+void arch_unmap_crash_pages(void **ptr);
+void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action);
+#else
+void *arch_map_crash_pages(unsigned long paddr, unsigned long size) { return NULL; }
+void arch_unmap_crash_pages(void **ptr) { }
+void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
+#endif
+
#endif
#endif
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index 8fc7d678ac72..a526c893abe8 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
if (ret)
return ret;
- image->elf_headers = kbuf.buffer;
- image->elf_headers_sz = kbuf.bufsz;
+ image->elf_headers = kbuf.buffer;
+ image->elf_headers_sz = kbuf.bufsz;
+ kbuf.memsz = kbuf.bufsz;
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
/* Ensure elfcorehdr segment large enough for hotplug changes */
@@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
image->elf_headers_sz = kbuf.memsz;
image->elfcorehdr_index = image->nr_segments;
image->elfcorehdr_index_valid = true;
-#else
- kbuf.memsz = kbuf.bufsz;
#endif
+
kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
ret = kexec_add_buffer(&kbuf);
@@ -425,7 +425,8 @@ int crash_load_segments(struct kimage *image)
}
#endif /* CONFIG_KEXEC_FILE */
-#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
+#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
+
/*
* NOTE: The addresses and sizes passed to this routine have
* already been fully aligned on page boundaries. There is no
@@ -462,8 +463,7 @@ void arch_unmap_crash_pages(void **ptr)
* is prepared in a kernel buffer, and then it is written on top
* of the existing/old elfcorehdr.
*/
-void arch_crash_handle_hotplug_event(struct kimage *image,
- unsigned int hp_action)
+void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action)
{
struct kexec_segment *ksegment;
unsigned char *ptr = NULL;
@@ -513,4 +513,5 @@ void arch_crash_handle_hotplug_event(struct kimage *image,
if (elfbuf)
vfree(elfbuf);
}
-#endif
+
+#endif /* CONFIG_CRASH_MAX_MEMORY_RANGES */
diff --git a/include/linux/kexec.h b/include/linux/kexec.h
index a48577a36fb8..0f79ad4c4f80 100644
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -27,6 +27,19 @@ extern struct resource crashk_res;
extern struct resource crashk_low_res;
extern note_buf_t __percpu *crash_notes;
+/* Alignment required for elf header segment */
+#define ELF_CORE_HEADER_ALIGN 4096
+
+struct crash_mem_range {
+ u64 start, end;
+};
+
+struct crash_mem {
+ unsigned int max_nr_ranges;
+ unsigned int nr_ranges;
+ struct crash_mem_range ranges[];
+};
+
#ifdef CONFIG_KEXEC_CORE
#include <linux/list.h>
#include <linux/compat.h>
@@ -237,19 +250,6 @@ static inline int arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
}
#endif
-/* Alignment required for elf header segment */
-#define ELF_CORE_HEADER_ALIGN 4096
-
-struct crash_mem_range {
- u64 start, end;
-};
-
-struct crash_mem {
- unsigned int max_nr_ranges;
- unsigned int nr_ranges;
- struct crash_mem_range ranges[];
-};
-
extern int crash_exclude_mem_range(struct crash_mem *mem,
unsigned long long mstart,
unsigned long long mend);
diff --git a/kernel/crash_core.c b/kernel/crash_core.c
index 5bc5159d9cb1..f6b5d835f826 100644
--- a/kernel/crash_core.c
+++ b/kernel/crash_core.c
@@ -622,6 +622,15 @@ static int __init crash_save_vmcoreinfo_init(void)
subsys_initcall(crash_save_vmcoreinfo_init);
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
+
+void __weak *arch_map_crash_pages(unsigned long paddr, unsigned long size)
+{
+ return NULL;
+}
+
+void __weak arch_unmap_crash_pages(void **ptr) { }
+void __weak arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
+
/*
* To accurately reflect hot un/plug changes, the elfcorehdr (which
* is passed to the crash kernel via the elfcorehdr= parameter)
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Wed, Sep 28, 2022 at 06:07:24PM +0200, Borislav Petkov wrote:
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> /* Ensure elfcorehdr segment large enough for hotplug changes */
> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
> image->elf_headers_sz = kbuf.memsz;
> image->elfcorehdr_index = image->nr_segments;
> image->elfcorehdr_index_valid = true;
And that ifdeffery above can be made more readable too:
diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
index a526c893abe8..7aab6e942761 100644
--- a/arch/x86/kernel/crash.c
+++ b/arch/x86/kernel/crash.c
@@ -399,16 +399,15 @@ int crash_load_segments(struct kimage *image)
image->elf_headers_sz = kbuf.bufsz;
kbuf.memsz = kbuf.bufsz;
-#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
/* Ensure elfcorehdr segment large enough for hotplug changes */
- kbuf.memsz =
- (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
- sizeof(Elf64_Phdr);
- /* Mark as usable to crash kernel, else crash kernel fails on boot */
- image->elf_headers_sz = kbuf.memsz;
- image->elfcorehdr_index = image->nr_segments;
- image->elfcorehdr_index_valid = true;
-#endif
+ if (IS_ENABLED(CONFIG_CRASH_MAX_MEMORY_RANGES)) {
+ kbuf.memsz = (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) * sizeof(Elf64_Phdr);
+
+ /* Mark as usable to crash kernel, else crash kernel fails on boot */
+ image->elf_headers_sz = kbuf.memsz;
+ image->elfcorehdr_index = image->nr_segments;
+ image->elfcorehdr_index_valid = true;
+ }
kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 9/28/22 11:07, Borislav Petkov wrote:
> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
>
> Please do not use lkml.org to refer to lkml messages. We have a
> perfectly fine archival system at lore.kernel.org. You simply do
>
> https://lore.kernel.org/r/<Message-ID>
>
> when you want to point to a previous mail.
ok, thanks for pointing that out to me.
>
>> David points out that terminology is tricky here due to differing behaviors.
>> And perhaps that is your point in asking for guidance text. It can be
>> complicated
>
> Which means you need an explanation how to use this even more.
>
> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
> something you discover from the hardware?
No, is the short answer.
>
> Your help text talks about System RAM entries in /proc/iomem which means
> that those entries are present somewhere in the kernel and you can read
> them out and do the proper calculations dynamically instead of doing the
> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
The intent is to compute the max size buffer needed to contain a maximum populated elfcorehdr, which
is primarily based on the number of CPUs and memory regions. Thus far I (and others involved) have
not found a kernel method to determine the maximum number of memory regions possible (if you are
aware of one, please let me know!). Thus CONFIG_CRASH_MAX_MEMORY_RANGES was born (rather borrowed
from kexec-tools).
So no dynamic computation is possible, yet.
>
>> , but it all comes down to System RAM entries.
>>
>> I could perhaps offer an overly simplified example such that for 1GiB block
>> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow for 32TiB
>> of memory?
>
> Yes, and stick it somewhere in Documentation/admin-guide/kdump/ and
> refer to it in that help text so that people can find it and read how to
> use your new option.
>
ok
>> The kbuf.bufsz value is obtained via a call to prepare_elf_headers(); I can
>> not initialize it at its declaration.
>
> Sorry, I meant this:
>
> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
> index 8fc7d678ac72..ee6fd9f1b2b9 100644
> --- a/arch/x86/kernel/crash.c
> +++ b/arch/x86/kernel/crash.c
> @@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
> if (ret)
> return ret;
>
> - image->elf_headers = kbuf.buffer;
> - image->elf_headers_sz = kbuf.bufsz;
> + image->elf_headers = kbuf.buffer;
> + image->elf_headers_sz = kbuf.bufsz;
> + kbuf.memsz = kbuf.bufsz;
>
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> /* Ensure elfcorehdr segment large enough for hotplug changes */
> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
> image->elf_headers_sz = kbuf.memsz;
> image->elfcorehdr_index = image->nr_segments;
> image->elfcorehdr_index_valid = true;
> -#else
> - kbuf.memsz = kbuf.bufsz;
> #endif
> +
> kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
> kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
> ret = kexec_add_buffer(&kbuf);
>
ok
>> I'm at a loss as to what to do differently here. You've raised this issue
>> before and I went back and looked at the suggestions then and I don't see
>> how that applies to this situation. How is this situation different than the
>> #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
>
> See the diff at the end. I'm not saying this is how you should do it
> but it should give you a better idea. The logic being, the functions
> in the .c file don't really need ifdeffery around them - you're adding
> 1-2 functions and crash.c is not that big - so they can be built in
> unconditionally. You'd need the ifdeffery *in the header only* when
> crash.c is not being built.
ok; I've overlooked that scenario.
>
> But I've done it with ifdeffery in the .c file now because yes, the
> kexec code is a minefield of ifdeffery. Hell, there's ifdeffery even in
> the headers for structs. Ifdeffery you don't really need. Someone should
> clean that up and simplify this immensely.
ok
>
>> Currently there is a concurrent effort for PPC support by Sourabh
>> Jain, and in that effort arch_map_crash_pages() is using __va(paddr).
>
> Why?
>
>> I do not know the nuances between kmap_local_page() and __va() to
>> answer the question.
>
> kmap_local_page() is a generic interface and it should work on any arch.
>
> And it is documented even:
>
> $ git grep kmap_local_page Documentation/
>
>> If kmap_local_page() works for all archs, then I'm happy to drop these
>> arch_ variants and use it directly.
>
> Yes, pls do.
I'll check with Sourabh to see if PPC can work with kmap_local_page().
>
> ---
>
> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
> index 432073385b2d..b73c9628cd85 100644
> --- a/arch/x86/include/asm/kexec.h
> +++ b/arch/x86/include/asm/kexec.h
> @@ -205,6 +205,17 @@ void *arch_kexec_kernel_image_load(struct kimage *image);
>
> int arch_kimage_file_post_load_cleanup(struct kimage *image);
> #define arch_kimage_file_post_load_cleanup arch_kimage_file_post_load_cleanup
> +
> +#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
So I think the use of CONFIG_CRASH_MAX_MEMORY_RANGES is not correct; it still needs to be based on
the cpu or memory hotplug options.
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
> +void arch_unmap_crash_pages(void **ptr);
> +void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action);
> +#else
> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size) { return NULL; }
> +void arch_unmap_crash_pages(void **ptr) { }
> +void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
> +#endif
> +
> #endif
> #endif
>
> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
> index 8fc7d678ac72..a526c893abe8 100644
> --- a/arch/x86/kernel/crash.c
> +++ b/arch/x86/kernel/crash.c
> @@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
> if (ret)
> return ret;
>
> - image->elf_headers = kbuf.buffer;
> - image->elf_headers_sz = kbuf.bufsz;
> + image->elf_headers = kbuf.buffer;
> + image->elf_headers_sz = kbuf.bufsz;
> + kbuf.memsz = kbuf.bufsz;
>
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> /* Ensure elfcorehdr segment large enough for hotplug changes */
> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
> image->elf_headers_sz = kbuf.memsz;
> image->elfcorehdr_index = image->nr_segments;
> image->elfcorehdr_index_valid = true;
> -#else
> - kbuf.memsz = kbuf.bufsz;
> #endif
> +
> kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
> kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
> ret = kexec_add_buffer(&kbuf);
> @@ -425,7 +425,8 @@ int crash_load_segments(struct kimage *image)
> }
> #endif /* CONFIG_KEXEC_FILE */
>
> -#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> +#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
Again, I don't think CONFIG_CRASH_MAX_MEMORY_RANGES makes sense, at all.
> +
> /*
> * NOTE: The addresses and sizes passed to this routine have
> * already been fully aligned on page boundaries. There is no
> @@ -462,8 +463,7 @@ void arch_unmap_crash_pages(void **ptr)
> * is prepared in a kernel buffer, and then it is written on top
> * of the existing/old elfcorehdr.
> */
> -void arch_crash_handle_hotplug_event(struct kimage *image,
> - unsigned int hp_action)
> +void arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action)
> {
> struct kexec_segment *ksegment;
> unsigned char *ptr = NULL;
> @@ -513,4 +513,5 @@ void arch_crash_handle_hotplug_event(struct kimage *image,
> if (elfbuf)
> vfree(elfbuf);
> }
> -#endif
> +
> +#endif /* CONFIG_CRASH_MAX_MEMORY_RANGES */
> diff --git a/include/linux/kexec.h b/include/linux/kexec.h
> index a48577a36fb8..0f79ad4c4f80 100644
> --- a/include/linux/kexec.h
> +++ b/include/linux/kexec.h
> @@ -27,6 +27,19 @@ extern struct resource crashk_res;
> extern struct resource crashk_low_res;
> extern note_buf_t __percpu *crash_notes;
>
> +/* Alignment required for elf header segment */
> +#define ELF_CORE_HEADER_ALIGN 4096
> +
> +struct crash_mem_range {
> + u64 start, end;
> +};
> +
> +struct crash_mem {
> + unsigned int max_nr_ranges;
> + unsigned int nr_ranges;
> + struct crash_mem_range ranges[];
> +};
> +
> #ifdef CONFIG_KEXEC_CORE
> #include <linux/list.h>
> #include <linux/compat.h>
> @@ -237,19 +250,6 @@ static inline int arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
> }
> #endif
>
> -/* Alignment required for elf header segment */
> -#define ELF_CORE_HEADER_ALIGN 4096
> -
> -struct crash_mem_range {
> - u64 start, end;
> -};
> -
> -struct crash_mem {
> - unsigned int max_nr_ranges;
> - unsigned int nr_ranges;
> - struct crash_mem_range ranges[];
> -};
> -
> extern int crash_exclude_mem_range(struct crash_mem *mem,
> unsigned long long mstart,
> unsigned long long mend);
> diff --git a/kernel/crash_core.c b/kernel/crash_core.c
> index 5bc5159d9cb1..f6b5d835f826 100644
> --- a/kernel/crash_core.c
> +++ b/kernel/crash_core.c
> @@ -622,6 +622,15 @@ static int __init crash_save_vmcoreinfo_init(void)
> subsys_initcall(crash_save_vmcoreinfo_init);
>
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> +
> +void __weak *arch_map_crash_pages(unsigned long paddr, unsigned long size)
> +{
> + return NULL;
> +}
> +
> +void __weak arch_unmap_crash_pages(void **ptr) { }
> +void __weak arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
> +
I was asked by Baoquan He to eliminate the use of __weak, which I did. I followed the technique used
by other kexec infrastructure.
> /*
> * To accurately reflect hot un/plug changes, the elfcorehdr (which
> * is passed to the crash kernel via the elfcorehdr= parameter)
>
On Fri, Sep 30, 2022 at 10:36:49AM -0500, Eric DeVolder wrote:
> > Your help text talks about System RAM entries in /proc/iomem which means
> > that those entries are present somewhere in the kernel and you can read
> > them out and do the proper calculations dynamically instead of doing the
> > static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>
> The intent is to compute the max size buffer needed to contain a maximum
> populated elfcorehdr, which is primarily based on the number of CPUs and
> memory regions. Thus far I (and others involved) have not found a kernel
> method to determine the maximum number of memory regions possible (if you
> are aware of one, please let me know!). Thus CONFIG_CRASH_MAX_MEMORY_RANGES
> was born (rather borrowed from kexec-tools).
Let's ask some mm folks.
mm folks, is there a way to enumerate all the memory regions a machine
has?
It looks to me like register_memory_resource() in mm/memory_hotplug.c
does register the resource so there should be a way to count that list
of resources or at least maintain a count somewhere so that kexec/crash
code can know how big its elfcodehdr buffer should be instead of doing a
clumsy Kconfig item where people would need to guess...
Hmm.
> > +#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
> So I think the use of CONFIG_CRASH_MAX_MEMORY_RANGES is not correct; it
> still needs to be based on the cpu or memory hotplug options.
You're kidding, right?
+config CRASH_MAX_MEMORY_RANGES
+ depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > @@ -622,6 +622,15 @@ static int __init crash_save_vmcoreinfo_init(void)
> > subsys_initcall(crash_save_vmcoreinfo_init);
> > #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
> > +
> > +void __weak *arch_map_crash_pages(unsigned long paddr, unsigned long size)
> > +{
> > + return NULL;
> > +}
> > +
> > +void __weak arch_unmap_crash_pages(void **ptr) { }
> > +void __weak arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
> > +
> I was asked by Baoquan He to eliminate the use of __weak
Because?
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 9/30/22 11:50, Borislav Petkov wrote:
> On Fri, Sep 30, 2022 at 10:36:49AM -0500, Eric DeVolder wrote:
>>> Your help text talks about System RAM entries in /proc/iomem which means
>>> that those entries are present somewhere in the kernel and you can read
>>> them out and do the proper calculations dynamically instead of doing the
>>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>>
>> The intent is to compute the max size buffer needed to contain a maximum
>> populated elfcorehdr, which is primarily based on the number of CPUs and
>> memory regions. Thus far I (and others involved) have not found a kernel
>> method to determine the maximum number of memory regions possible (if you
>> are aware of one, please let me know!). Thus CONFIG_CRASH_MAX_MEMORY_RANGES
>> was born (rather borrowed from kexec-tools).
>
> Let's ask some mm folks.
>
> mm folks, is there a way to enumerate all the memory regions a machine
> has?
>
> It looks to me like register_memory_resource() in mm/memory_hotplug.c
> does register the resource so there should be a way to count that list
> of resources or at least maintain a count somewhere so that kexec/crash
> code can know how big its elfcodehdr buffer should be instead of doing a
> clumsy Kconfig item where people would need to guess...
>
> Hmm.
>
There is of course a way to enumerate the memory regions in use on the machine, that is not what
this code needs. In order to compute the maximum buffer size needed (this buffer size is computed
once), the count of the maximum number of memory regions possible (even if not currently in use) is
what is needed.
>>> +#ifdef CONFIG_CRASH_MAX_MEMORY_RANGES
>> So I think the use of CONFIG_CRASH_MAX_MEMORY_RANGES is not correct; it
>> still needs to be based on the cpu or memory hotplug options.
>
> You're kidding, right?
>
> +config CRASH_MAX_MEMORY_RANGES
> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> ^^^^^^^^^^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Oh, that would be an error of haste on my part. This should be:
depends on CRASH_DUMP && MEMORY_HOTPLUG
>
>>> @@ -622,6 +622,15 @@ static int __init crash_save_vmcoreinfo_init(void)
>>> subsys_initcall(crash_save_vmcoreinfo_init);
>>> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>> +
>>> +void __weak *arch_map_crash_pages(unsigned long paddr, unsigned long size)
>>> +{
>>> + return NULL;
>>> +}
>>> +
>>> +void __weak arch_unmap_crash_pages(void **ptr) { }
>>> +void __weak arch_crash_handle_hotplug_event(struct kimage *image, unsigned int hp_action) { }
>>> +
>> I was asked by Baoquan He to eliminate the use of __weak
>
> Because?
>
Baoquan pointed me to:
https://lore.kernel.org/lkml/[email protected]/T/
Thanks,
eric
On Fri, Sep 30, 2022 at 12:11:26PM -0500, Eric DeVolder wrote:
> There is of course a way to enumerate the memory regions in use on the
> machine, that is not what this code needs. In order to compute the maximum
> buffer size needed (this buffer size is computed once), the count of the
> maximum number of memory regions possible (even if not currently in use) is
> what is needed.
Isn't that max number documented somewhere in memory hotplug docs?
Because then you don't need that Kconfig item either. Imagine you're a
distro kernel distributor and you want crash to work on all machines
your kernel works.
So you go and set that number to max. And that would be the 99% of the
kernel configs out there.
Which means, you can just set it to max without a Kconfig item.
> Oh, that would be an error of haste on my part. This should be:
> depends on CRASH_DUMP && MEMORY_HOTPLUG
You need a Kconfig item which enables all this gunk as MEMORY_HOTPLUG is
not a omnipresent feature. And that Kconfig item should depend on the
other Kconfig items of the technology you need.
> Baoquan pointed me to:
>
> https://lore.kernel.org/lkml/[email protected]/T/
In that thread says:
"- arch_kexec_apply_relocations_add() is only overridden by x86 and s390.
Retain the function prototype for those and move the weak
implementation into the header as a static inline for other
architectures."
So yes, that's even better.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 30/09/22 21:06, Eric DeVolder wrote:
>
>
> On 9/28/22 11:07, Borislav Petkov wrote:
>> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>>> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
>>
>> Please do not use lkml.org to refer to lkml messages. We have a
>> perfectly fine archival system at lore.kernel.org. You simply do
>>
>> https://lore.kernel.org/r/<Message-ID>
>>
>> when you want to point to a previous mail.
>
> ok, thanks for pointing that out to me.
>>
>>> David points out that terminology is tricky here due to differing
>>> behaviors.
>>> And perhaps that is your point in asking for guidance text. It can be
>>> complicated
>>
>> Which means you need an explanation how to use this even more.
>>
>> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
>> something you discover from the hardware?
>
> No, is the short answer.
>
>>
>> Your help text talks about System RAM entries in /proc/iomem which means
>> that those entries are present somewhere in the kernel and you can read
>> them out and do the proper calculations dynamically instead of doing the
>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>
> The intent is to compute the max size buffer needed to contain a
> maximum populated elfcorehdr, which is primarily based on the number
> of CPUs and memory regions. Thus far I (and others involved) have not
> found a kernel method to determine the maximum number of memory
> regions possible (if you are aware of one, please let me know!). Thus
> CONFIG_CRASH_MAX_MEMORY_RANGES was born (rather borrowed from
> kexec-tools).
>
> So no dynamic computation is possible, yet.
Hello Eric,
How about allocating buffer space for max program header possible in a
elfcorehdr?
mage->elf_headers_sz = kbuf.memsz = PN_XNUM * sizeof(Elf64_Phdr);
PN_XNUM is part of linux/elf.h (include/uapi/linux/elf.h).
Refer below link for more details:
https://man7.org/linux/man-pages/man5/elf.5.html
Thanks,
Sourabh Jain
On 30/09/22 21:06, Eric DeVolder wrote:
>
>
> On 9/28/22 11:07, Borislav Petkov wrote:
>> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>>> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
>>
>> Please do not use lkml.org to refer to lkml messages. We have a
>> perfectly fine archival system at lore.kernel.org. You simply do
>>
>> https://lore.kernel.org/r/<Message-ID>
>>
>> when you want to point to a previous mail.
>
> ok, thanks for pointing that out to me.
>>
>>> David points out that terminology is tricky here due to differing
>>> behaviors.
>>> And perhaps that is your point in asking for guidance text. It can be
>>> complicated
>>
>> Which means you need an explanation how to use this even more.
>>
>> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
>> something you discover from the hardware?
>
> No, is the short answer.
>
>>
>> Your help text talks about System RAM entries in /proc/iomem which means
>> that those entries are present somewhere in the kernel and you can read
>> them out and do the proper calculations dynamically instead of doing the
>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>
> The intent is to compute the max size buffer needed to contain a
> maximum populated elfcorehdr, which is primarily based on the number
> of CPUs and memory regions. Thus far I (and others involved) have not
> found a kernel method to determine the maximum number of memory
> regions possible (if you are aware of one, please let me know!). Thus
> CONFIG_CRASH_MAX_MEMORY_RANGES was born (rather borrowed from
> kexec-tools).
>
> So no dynamic computation is possible, yet.
>
>>
>>> , but it all comes down to System RAM entries.
>>>
>>> I could perhaps offer an overly simplified example such that for
>>> 1GiB block
>>> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow
>>> for 32TiB
>>> of memory?
>>
>> Yes, and stick it somewhere in Documentation/admin-guide/kdump/ and
>> refer to it in that help text so that people can find it and read how to
>> use your new option.
>>
> ok
>
>>> The kbuf.bufsz value is obtained via a call to
>>> prepare_elf_headers(); I can
>>> not initialize it at its declaration.
>>
>> Sorry, I meant this:
>>
>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>> index 8fc7d678ac72..ee6fd9f1b2b9 100644
>> --- a/arch/x86/kernel/crash.c
>> +++ b/arch/x86/kernel/crash.c
>> @@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
>> Â Â Â Â Â if (ret)
>> Â Â Â Â Â Â Â Â Â return ret;
>> Â -Â Â Â image->elf_headers = kbuf.buffer;
>> -Â Â Â image->elf_headers_sz = kbuf.bufsz;
>> +   image->elf_headers   = kbuf.buffer;
>> +   image->elf_headers_sz   = kbuf.bufsz;
>> +   kbuf.memsz       = kbuf.bufsz;
>> Â Â #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>> Â Â Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
>> Â Â Â Â Â image->elf_headers_sz = kbuf.memsz;
>> Â Â Â Â Â image->elfcorehdr_index = image->nr_segments;
>> Â Â Â Â Â image->elfcorehdr_index_valid = true;
>> -#else
>> -Â Â Â kbuf.memsz = kbuf.bufsz;
>> Â #endif
>> +
>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>>
> ok
>
>>> I'm at a loss as to what to do differently here. You've raised this
>>> issue
>>> before and I went back and looked at the suggestions then and I
>>> don't see
>>> how that applies to this situation. How is this situation different
>>> than the
>>> #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
>>
>> See the diff at the end. I'm not saying this is how you should do it
>> but it should give you a better idea. The logic being, the functions
>> in the .c file don't really need ifdeffery around them - you're adding
>> 1-2 functions and crash.c is not that big - so they can be built in
>> unconditionally. You'd need the ifdeffery *in the header only* when
>> crash.c is not being built.
> ok; I've overlooked that scenario.
>>
>> But I've done it with ifdeffery in the .c file now because yes, the
>> kexec code is a minefield of ifdeffery. Hell, there's ifdeffery even in
>> the headers for structs. Ifdeffery you don't really need. Someone should
>> clean that up and simplify this immensely.
>
> ok
>
>>
>>> Currently there is a concurrent effort for PPC support by Sourabh
>>> Jain, and in that effort arch_map_crash_pages() is using __va(paddr).
>>
>> Why?
>>
>>> I do not know the nuances between kmap_local_page() and __va() to
>>> answer the question.
>>
>> kmap_local_page() is a generic interface and it should work on any arch.
>>
>> And it is documented even:
>>
>> $ git grep kmap_local_page Documentation/
>>
>>> If kmap_local_page() works for all archs, then I'm happy to drop these
>>> arch_ variants and use it directly.
>>
>> Yes, pls do.
>
> I'll check with Sourabh to see if PPC can work with kmap_local_page().
I think kmap_local_page do support on PowerPC. But can you explain why
we need this
function here, aren't the reserve memory already available to use?
Thanks,
Sourabh Jain
On 9/19/22 02:06, Sourabh Jain wrote:
>
> On 10/09/22 02:35, Eric DeVolder wrote:
>> For x86_64, when CPU or memory is hot un/plugged, the crash
>> elfcorehdr, which describes the CPUs and memory in the system,
>> must also be updated.
>>
>> When loading the crash kernel via kexec_load or kexec_file_load,
>> the elfcorehdr is identified at run time in
>> crash_core:handle_hotplug_event().
>>
>> To update the elfcorehdr for x86_64, a new elfcorehdr must be
>> generated from the available CPUs and memory. The new elfcorehdr
>> is prepared into a buffer, and then installed over the top of
>> the existing elfcorehdr.
>>
>> In the patch 'kexec: exclude elfcorehdr from the segment digest'
>> the need to update purgatory due to the change in elfcorehdr was
>> eliminated. As a result, no changes to purgatory or boot_params
>> (as the elfcorehdr= kernel command line parameter pointer
>> remains unchanged and correct) are needed, just elfcorehdr.
>>
>> To accommodate a growing number of resources via hotplug, the
>> elfcorehdr segment must be sufficiently large enough to accommodate
>> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>>
>> With this change, crash hotplug for kexec_file_load syscall
>> is supported. The kexec_load is also supported, but also
>> requires a corresponding change to userspace kexec-tools.
>>
>> Signed-off-by: Eric DeVolder <[email protected]>
>> Acked-by: Baoquan He <[email protected]>
>> ---
>>  arch/x86/Kconfig            | 11 ++++
>> Â arch/x86/include/asm/kexec.h |Â 20 +++++++
>>  arch/x86/kernel/crash.c     | 102 +++++++++++++++++++++++++++++++++++
>> Â 3 files changed, 133 insertions(+)
>>
>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>> index f9920f1341c8..cdfc9b2fdf98 100644
>> --- a/arch/x86/Kconfig
>> +++ b/arch/x86/Kconfig
>> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
>> Â Â Â Â Â Â Â (CONFIG_RELOCATABLE=y).
>> Â Â Â Â Â Â Â For more details see Documentation/admin-guide/kdump/kdump.rst
>> +config CRASH_MAX_MEMORY_RANGES
>> +Â Â Â depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
>> +Â Â Â int
>> +Â Â Â default 32768
>> +Â Â Â help
>> +Â Â Â Â Â For the kexec_file_load path, specify the maximum number of
>> +Â Â Â Â Â memory regions, eg. as represented by the 'System RAM' entries
>> +Â Â Â Â Â in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
>> +Â Â Â Â Â This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>> +Â Â Â Â Â size to determine the final buffer size.
>> +
>> Â config KEXEC_JUMP
>> Â Â Â Â Â bool "kexec jump"
>> Â Â Â Â Â depends on KEXEC && HIBERNATION
>> diff --git a/arch/x86/include/asm/kexec.h b/arch/x86/include/asm/kexec.h
>> index a3760ca796aa..432073385b2d 100644
>> --- a/arch/x86/include/asm/kexec.h
>> +++ b/arch/x86/include/asm/kexec.h
>> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
>> Â extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
>> Â extern void kdump_nmi_shootdown_cpus(void);
>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
>> +#define arch_map_crash_pages arch_map_crash_pages
>> +
>> +void arch_unmap_crash_pages(void **ptr);
>> +#define arch_unmap_crash_pages arch_unmap_crash_pages
>> +
>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>> +Â Â Â Â Â Â Â unsigned int hp_action);
>> +#define arch_crash_handle_hotplug_event arch_crash_handle_hotplug_event
>> +
>> +#ifdef CONFIG_HOTPLUG_CPU
>> +static inline int crash_hotplug_cpu_support(void) { return 1; }
>> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
>> +#endif
>> +
>> +#ifdef CONFIG_MEMORY_HOTPLUG
>> +static inline int crash_hotplug_memory_support(void) { return 1; }
>> +#define crash_hotplug_memory_support crash_hotplug_memory_support
>> +#endif
>> +
>> Â #endif /* __ASSEMBLY__ */
>> Â #endif /* _ASM_X86_KEXEC_H */
>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>> index 9ceb93c176a6..8fc7d678ac72 100644
>> --- a/arch/x86/kernel/crash.c
>> +++ b/arch/x86/kernel/crash.c
>> @@ -25,6 +25,7 @@
>> Â #include <linux/slab.h>
>> Â #include <linux/vmalloc.h>
>> Â #include <linux/memblock.h>
>> +#include <linux/highmem.h>
>> Â #include <asm/processor.h>
>> Â #include <asm/hardirq.h>
>> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
>> Â Â Â Â Â image->elf_headers = kbuf.buffer;
>> Â Â Â Â Â image->elf_headers_sz = kbuf.bufsz;
>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>> +Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>> +Â Â Â kbuf.memsz =
>> +Â Â Â Â Â Â Â (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
>> +Â Â Â Â Â Â Â Â Â Â Â sizeof(Elf64_Phdr);
>> +Â Â Â /* Mark as usable to crash kernel, else crash kernel fails on boot */
>> +Â Â Â image->elf_headers_sz = kbuf.memsz;
>> +Â Â Â image->elfcorehdr_index = image->nr_segments;
>> +Â Â Â image->elfcorehdr_index_valid = true;
>> +#else
>> Â Â Â Â Â kbuf.memsz = kbuf.bufsz;
>> +#endif
>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
>> Â Â Â Â Â return ret;
>> Â }
>> Â #endif /* CONFIG_KEXEC_FILE */
>> +
>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>> +/*
>> + * NOTE: The addresses and sizes passed to this routine have
>> + * already been fully aligned on page boundaries. There is no
>> + * need for massaging the address or size.
>> + */
>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
>> +{
>> +Â Â Â void *ptr = NULL;
>> +
>> +Â Â Â if (size > 0) {
>> +Â Â Â Â Â Â Â struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
>> +
>> +Â Â Â Â Â Â Â ptr = kmap_local_page(page);
>> +Â Â Â }
>> +
>> +Â Â Â return ptr;
>> +}
>> +
>> +void arch_unmap_crash_pages(void **ptr)
>> +{
>> +Â Â Â if (ptr) {
>> +Â Â Â Â Â Â Â if (*ptr)
>> +Â Â Â Â Â Â Â Â Â Â Â kunmap_local(*ptr);
>> +Â Â Â Â Â Â Â *ptr = NULL;
>> +Â Â Â }
>> +}
>> +
>> +/**
>> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
>> + * @image: the active struct kimage
>> + * @hp_action: the hot un/plug action being handled
>> + *
>> + * To accurately reflect hot un/plug changes, the new elfcorehdr
>> + * is prepared in a kernel buffer, and then it is written on top
>> + * of the existing/old elfcorehdr.
>> + */
>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>> +Â Â Â unsigned int hp_action)
>> +{
>> +Â Â Â struct kexec_segment *ksegment;
>> +Â Â Â unsigned char *ptr = NULL;
>> +Â Â Â unsigned long elfsz = 0;
>> +Â Â Â void *elfbuf = NULL;
>> +Â Â Â unsigned long mem, memsz;
>> +
>> +Â Â Â /*
>> +Â Â Â Â * Elfcorehdr_index_valid checked in crash_core:handle_hotplug_event()
>> +Â Â Â Â */
>> +Â Â Â ksegment = &image->segment[image->elfcorehdr_index];
>> +Â Â Â mem = ksegment->mem;
>> +Â Â Â memsz = ksegment->memsz;
>> +
>> +Â Â Â /*
>> +Â Â Â Â * Create the new elfcorehdr reflecting the changes to CPU and/or
>> +Â Â Â Â * memory resources.
>> +Â Â Â Â */
>> +Â Â Â if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
>> +Â Â Â Â Â Â Â pr_err("crash hp: unable to prepare elfcore headers");
>> +Â Â Â Â Â Â Â goto out;
>
> On PowerPC, while preparing the elf core header the memblock structure is used to prepare program
> header for memory regions of elfcorehdr. Since the above arch specific hotplug handler gets invoked
> when memory is marked offline (MEM_OFFLINE) which is before memblock structure gets updated so on
> PowerPC the above handler may not work for memory hotplug case.
>
> Just wondering which data structure is used to get the list of memory regions while preparing
> program header for memory regions of elfcorehdr on other architectures?
>
> Thanks,
> Sourabh Jain
I think your request to report the memory block address in comments of patch 3/7 "crash: add generic
infrastructure" cover this scenario now.
Thanks,
eric
On 10/4/22 02:03, Sourabh Jain wrote:
>
> On 30/09/22 21:06, Eric DeVolder wrote:
>>
>>
>> On 9/28/22 11:07, Borislav Petkov wrote:
>>> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>>>> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
>>>
>>> Please do not use lkml.org to refer to lkml messages. We have a
>>> perfectly fine archival system at lore.kernel.org. You simply do
>>>
>>> https://lore.kernel.org/r/<Message-ID>
>>>
>>> when you want to point to a previous mail.
>>
>> ok, thanks for pointing that out to me.
>>>
>>>> David points out that terminology is tricky here due to differing behaviors.
>>>> And perhaps that is your point in asking for guidance text. It can be
>>>> complicated
>>>
>>> Which means you need an explanation how to use this even more.
>>>
>>> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
>>> something you discover from the hardware?
>>
>> No, is the short answer.
>>
>>>
>>> Your help text talks about System RAM entries in /proc/iomem which means
>>> that those entries are present somewhere in the kernel and you can read
>>> them out and do the proper calculations dynamically instead of doing the
>>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>>
>> The intent is to compute the max size buffer needed to contain a maximum populated elfcorehdr,
>> which is primarily based on the number of CPUs and memory regions. Thus far I (and others
>> involved) have not found a kernel method to determine the maximum number of memory regions
>> possible (if you are aware of one, please let me know!). Thus CONFIG_CRASH_MAX_MEMORY_RANGES was
>> born (rather borrowed from kexec-tools).
>>
>> So no dynamic computation is possible, yet.
>
> Hello Eric,
>
> How about allocating buffer space for max program header possible in a elfcorehdr?
>
> mage->elf_headers_sz = kbuf.memsz = PN_XNUM * sizeof(Elf64_Phdr);
>
> PN_XNUM is part of linux/elf.h (include/uapi/linux/elf.h).
>
> Refer below link for more details:
> https://man7.org/linux/man-pages/man5/elf.5.html
>
> Thanks,
> Sourabh Jain
>
Well, that is an idea. I'm not sure it is the answer yet, but if I do compute
a value, then that value needs to be checked against PN_XNUM so it still results
in a valid elfcorehdr.
Thanks,
eric
On 10/4/22 04:10, Sourabh Jain wrote:
>
> On 30/09/22 21:06, Eric DeVolder wrote:
>>
>>
>> On 9/28/22 11:07, Borislav Petkov wrote:
>>> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>>>> This topic was discussed previously https://lkml.org/lkml/2022/3/3/372.
>>>
>>> Please do not use lkml.org to refer to lkml messages. We have a
>>> perfectly fine archival system at lore.kernel.org. You simply do
>>>
>>> https://lore.kernel.org/r/<Message-ID>
>>>
>>> when you want to point to a previous mail.
>>
>> ok, thanks for pointing that out to me.
>>>
>>>> David points out that terminology is tricky here due to differing behaviors.
>>>> And perhaps that is your point in asking for guidance text. It can be
>>>> complicated
>>>
>>> Which means you need an explanation how to use this even more.
>>>
>>> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
>>> something you discover from the hardware?
>>
>> No, is the short answer.
>>
>>>
>>> Your help text talks about System RAM entries in /proc/iomem which means
>>> that those entries are present somewhere in the kernel and you can read
>>> them out and do the proper calculations dynamically instead of doing the
>>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>>
>> The intent is to compute the max size buffer needed to contain a maximum populated elfcorehdr,
>> which is primarily based on the number of CPUs and memory regions. Thus far I (and others
>> involved) have not found a kernel method to determine the maximum number of memory regions
>> possible (if you are aware of one, please let me know!). Thus CONFIG_CRASH_MAX_MEMORY_RANGES was
>> born (rather borrowed from kexec-tools).
>>
>> So no dynamic computation is possible, yet.
>>
>>>
>>>> , but it all comes down to System RAM entries.
>>>>
>>>> I could perhaps offer an overly simplified example such that for 1GiB block
>>>> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would allow for 32TiB
>>>> of memory?
>>>
>>> Yes, and stick it somewhere in Documentation/admin-guide/kdump/ and
>>> refer to it in that help text so that people can find it and read how to
>>> use your new option.
>>>
>> ok
>>
>>>> The kbuf.bufsz value is obtained via a call to prepare_elf_headers(); I can
>>>> not initialize it at its declaration.
>>>
>>> Sorry, I meant this:
>>>
>>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>>> index 8fc7d678ac72..ee6fd9f1b2b9 100644
>>> --- a/arch/x86/kernel/crash.c
>>> +++ b/arch/x86/kernel/crash.c
>>> @@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â if (ret)
>>> Â Â Â Â Â Â Â Â Â return ret;
>>> Â -Â Â Â image->elf_headers = kbuf.buffer;
>>> -Â Â Â image->elf_headers_sz = kbuf.bufsz;
>>> +   image->elf_headers   = kbuf.buffer;
>>> +   image->elf_headers_sz   = kbuf.bufsz;
>>> +   kbuf.memsz       = kbuf.bufsz;
>>> Â Â #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>> Â Â Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>>> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â image->elf_headers_sz = kbuf.memsz;
>>> Â Â Â Â Â image->elfcorehdr_index = image->nr_segments;
>>> Â Â Â Â Â image->elfcorehdr_index_valid = true;
>>> -#else
>>> -Â Â Â kbuf.memsz = kbuf.bufsz;
>>> Â #endif
>>> +
>>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>>>
>> ok
>>
>>>> I'm at a loss as to what to do differently here. You've raised this issue
>>>> before and I went back and looked at the suggestions then and I don't see
>>>> how that applies to this situation. How is this situation different than the
>>>> #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
>>>
>>> See the diff at the end. I'm not saying this is how you should do it
>>> but it should give you a better idea. The logic being, the functions
>>> in the .c file don't really need ifdeffery around them - you're adding
>>> 1-2 functions and crash.c is not that big - so they can be built in
>>> unconditionally. You'd need the ifdeffery *in the header only* when
>>> crash.c is not being built.
>> ok; I've overlooked that scenario.
>>>
>>> But I've done it with ifdeffery in the .c file now because yes, the
>>> kexec code is a minefield of ifdeffery. Hell, there's ifdeffery even in
>>> the headers for structs. Ifdeffery you don't really need. Someone should
>>> clean that up and simplify this immensely.
>>
>> ok
>>
>>>
>>>> Currently there is a concurrent effort for PPC support by Sourabh
>>>> Jain, and in that effort arch_map_crash_pages() is using __va(paddr).
>>>
>>> Why?
>>>
>>>> I do not know the nuances between kmap_local_page() and __va() to
>>>> answer the question.
>>>
>>> kmap_local_page() is a generic interface and it should work on any arch.
>>>
>>> And it is documented even:
>>>
>>> $ git grep kmap_local_page Documentation/
>>>
>>>> If kmap_local_page() works for all archs, then I'm happy to drop these
>>>> arch_ variants and use it directly.
>>>
>>> Yes, pls do.
>>
>> I'll check with Sourabh to see if PPC can work with kmap_local_page().
> I think kmap_local_page do support on PowerPC. But can you explain why we need this
> function here, aren't the reserve memory already available to use?
On x86, attempts to access the elfcorehdr without mapping it did not work (resulted
in a fault).
Let me know if using kmap_local_page() in place of __va() in arch_map_crash_pages().
If it does, then I can eliminate arch_un/map_crash_pages() and use kmap_local_page()
directly.
Thanks,
eric
>
> Thanks,
> Sourabh Jain
On 09/30/22 at 07:40pm, Borislav Petkov wrote:
> On Fri, Sep 30, 2022 at 12:11:26PM -0500, Eric DeVolder wrote:
> > There is of course a way to enumerate the memory regions in use on the
> > machine, that is not what this code needs. In order to compute the maximum
> > buffer size needed (this buffer size is computed once), the count of the
> > maximum number of memory regions possible (even if not currently in use) is
> > what is needed.
>
> Isn't that max number documented somewhere in memory hotplug docs?
Memory hptlug is not limited by a certin or a max number of memory
regions, but limited by how large of the linear mapping range which
physical can be mapped into.
E.g on x86_64, with 4-level page tables, it has 64TB linear mapping
range by default. On principle, we can add 64TB of phisical memory
into system altogether from booting and memory hotplug. While with
KASLR enabled, it has 10TB of linear mapping range by default, see
CONFIG_RANDOMIZE_MEMORY_PHYSICAL_PADDING. Means there's only 10TB
phisical memory being allowed to be added into system.
For the Kconfig CRASH_MAX_MEMORY_RANGES Eric added, it's meaningful to
me to set a fixed value which is enough in reality. For extreme testing
with special purpose, it could be broken easily, people need decide by
self whether the CONFIG_CRASH_MAX_MEMORY_RANGES is enlarged or not.
E.g on x86_64, we make a system with memory smaller than 64G, this will
cause the memory block size being probed as 256M. Then we hot added many
Tera Bytes of physical memory every second memory block after bootup with
a shell shell script. It could be easier to manipulate this with virtiomem.
Please see function probe_memory_block_size() on x86_64 about the memory
block size probing. However, I don't think on real system, this kind of
system could really exist, with a tiny memory booted up, a huge memory
hot added sparsely.
>
> Because then you don't need that Kconfig item either. Imagine you're a
> distro kernel distributor and you want crash to work on all machines
> your kernel works.
>
> So you go and set that number to max. And that would be the 99% of the
> kernel configs out there.
>
> Which means, you can just set it to max without a Kconfig item.
>
> > Oh, that would be an error of haste on my part. This should be:
> > depends on CRASH_DUMP && MEMORY_HOTPLUG
>
> You need a Kconfig item which enables all this gunk as MEMORY_HOTPLUG is
> not a omnipresent feature. And that Kconfig item should depend on the
> other Kconfig items of the technology you need.
>
> > Baoquan pointed me to:
> >
> > https://lore.kernel.org/lkml/[email protected]/T/
>
> In that thread says:
>
> "- arch_kexec_apply_relocations_add() is only overridden by x86 and s390.
> Retain the function prototype for those and move the weak
> implementation into the header as a static inline for other
> architectures."
>
> So yes, that's even better.
>
> --
> Regards/Gruss,
> Boris.
>
> https://people.kernel.org/tglx/notes-about-netiquette
>
On 08/10/22 01:30, Eric DeVolder wrote:
>
>
> On 10/4/22 04:10, Sourabh Jain wrote:
>>
>> On 30/09/22 21:06, Eric DeVolder wrote:
>>>
>>>
>>> On 9/28/22 11:07, Borislav Petkov wrote:
>>>> On Tue, Sep 13, 2022 at 02:12:31PM -0500, Eric DeVolder wrote:
>>>>> This topic was discussed previously
>>>>> https://lkml.org/lkml/2022/3/3/372.
>>>>
>>>> Please do not use lkml.org to refer to lkml messages. We have a
>>>> perfectly fine archival system at lore.kernel.org. You simply do
>>>>
>>>> https://lore.kernel.org/r/<Message-ID>
>>>>
>>>> when you want to point to a previous mail.
>>>
>>> ok, thanks for pointing that out to me.
>>>>
>>>>> David points out that terminology is tricky here due to differing
>>>>> behaviors.
>>>>> And perhaps that is your point in asking for guidance text. It can be
>>>>> complicated
>>>>
>>>> Which means you need an explanation how to use this even more.
>>>>
>>>> And why is CONFIG_CRASH_MAX_MEMORY_RANGES even a Kconfig item and not
>>>> something you discover from the hardware?
>>>
>>> No, is the short answer.
>>>
>>>>
>>>> Your help text talks about System RAM entries in /proc/iomem which
>>>> means
>>>> that those entries are present somewhere in the kernel and you can
>>>> read
>>>> them out and do the proper calculations dynamically instead of
>>>> doing the
>>>> static CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES thing.
>>>
>>> The intent is to compute the max size buffer needed to contain a
>>> maximum populated elfcorehdr, which is primarily based on the number
>>> of CPUs and memory regions. Thus far I (and others involved) have
>>> not found a kernel method to determine the maximum number of memory
>>> regions possible (if you are aware of one, please let me know!).
>>> Thus CONFIG_CRASH_MAX_MEMORY_RANGES was born (rather borrowed from
>>> kexec-tools).
>>>
>>> So no dynamic computation is possible, yet.
>>>
>>>>
>>>>> , but it all comes down to System RAM entries.
>>>>>
>>>>> I could perhaps offer an overly simplified example such that for
>>>>> 1GiB block
>>>>> size, for example, the CRASH_MAX_MEMORY_RANGES of 32768 would
>>>>> allow for 32TiB
>>>>> of memory?
>>>>
>>>> Yes, and stick it somewhere in Documentation/admin-guide/kdump/ and
>>>> refer to it in that help text so that people can find it and read
>>>> how to
>>>> use your new option.
>>>>
>>> ok
>>>
>>>>> The kbuf.bufsz value is obtained via a call to
>>>>> prepare_elf_headers(); I can
>>>>> not initialize it at its declaration.
>>>>
>>>> Sorry, I meant this:
>>>>
>>>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>>>> index 8fc7d678ac72..ee6fd9f1b2b9 100644
>>>> --- a/arch/x86/kernel/crash.c
>>>> +++ b/arch/x86/kernel/crash.c
>>>> @@ -395,8 +395,9 @@ int crash_load_segments(struct kimage *image)
>>>> Â Â Â Â Â if (ret)
>>>> Â Â Â Â Â Â Â Â Â return ret;
>>>> Â -Â Â Â image->elf_headers = kbuf.buffer;
>>>> -Â Â Â image->elf_headers_sz = kbuf.bufsz;
>>>> +   image->elf_headers   = kbuf.buffer;
>>>> +   image->elf_headers_sz   = kbuf.bufsz;
>>>> +   kbuf.memsz       = kbuf.bufsz;
>>>> Â Â #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>>> Â Â Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>>>> @@ -407,9 +408,8 @@ int crash_load_segments(struct kimage *image)
>>>> Â Â Â Â Â image->elf_headers_sz = kbuf.memsz;
>>>> Â Â Â Â Â image->elfcorehdr_index = image->nr_segments;
>>>> Â Â Â Â Â image->elfcorehdr_index_valid = true;
>>>> -#else
>>>> -Â Â Â kbuf.memsz = kbuf.bufsz;
>>>> Â #endif
>>>> +
>>>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>>>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>>>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>>>>
>>> ok
>>>
>>>>> I'm at a loss as to what to do differently here. You've raised
>>>>> this issue
>>>>> before and I went back and looked at the suggestions then and I
>>>>> don't see
>>>>> how that applies to this situation. How is this situation
>>>>> different than the
>>>>> #ifdef CONFIG_KEXEC_FILE that immediately preceeds it?
>>>>
>>>> See the diff at the end. I'm not saying this is how you should do it
>>>> but it should give you a better idea. The logic being, the functions
>>>> in the .c file don't really need ifdeffery around them - you're adding
>>>> 1-2 functions and crash.c is not that big - so they can be built in
>>>> unconditionally. You'd need the ifdeffery *in the header only* when
>>>> crash.c is not being built.
>>> ok; I've overlooked that scenario.
>>>>
>>>> But I've done it with ifdeffery in the .c file now because yes, the
>>>> kexec code is a minefield of ifdeffery. Hell, there's ifdeffery
>>>> even in
>>>> the headers for structs. Ifdeffery you don't really need. Someone
>>>> should
>>>> clean that up and simplify this immensely.
>>>
>>> ok
>>>
>>>>
>>>>> Currently there is a concurrent effort for PPC support by Sourabh
>>>>> Jain, and in that effort arch_map_crash_pages() is using __va(paddr).
>>>>
>>>> Why?
>>>>
>>>>> I do not know the nuances between kmap_local_page() and __va() to
>>>>> answer the question.
>>>>
>>>> kmap_local_page() is a generic interface and it should work on any
>>>> arch.
>>>>
>>>> And it is documented even:
>>>>
>>>> $ git grep kmap_local_page Documentation/
>>>>
>>>>> If kmap_local_page() works for all archs, then I'm happy to drop
>>>>> these
>>>>> arch_ variants and use it directly.
>>>>
>>>> Yes, pls do.
>>>
>>> I'll check with Sourabh to see if PPC can work with kmap_local_page().
>> I think kmap_local_page do support on PowerPC. But can you explain
>> why we need this
>> function here, aren't the reserve memory already available to use?
>
> On x86, attempts to access the elfcorehdr without mapping it did not
> work (resulted
> in a fault).
>
> Let me know if using kmap_local_page() in place of __va() in
> arch_map_crash_pages().
> If it does, then I can eliminate arch_un/map_crash_pages() and use
> kmap_local_page()
> directly.
Hello Eric,
Atleast on ppc64 we have direct mapping available and hence just by
doing page shift
on physical address (__va) we can get valid virtual address on powerpc.
In short we don't
have to generate mapping again to access reserved region.
Regardless let's go with kdump_local_page API, it is supported on powerpc.
Thanks,
Sourabh Jain
On 10/11/22 23:55, Sourabh Jain wrote:
>
>>>>>> If kmap_local_page() works for all archs, then I'm happy to drop these
>>>>>> arch_ variants and use it directly.
>>>>>
>>>>> Yes, pls do.
>>>>
>>>> I'll check with Sourabh to see if PPC can work with kmap_local_page().
>>> I think kmap_local_page do support on PowerPC. But can you explain why we need this
>>> function here, aren't the reserve memory already available to use?
>>
>> On x86, attempts to access the elfcorehdr without mapping it did not work (resulted
>> in a fault).
>>
>> Let me know if using kmap_local_page() in place of __va() in arch_map_crash_pages().
>> If it does, then I can eliminate arch_un/map_crash_pages() and use kmap_local_page()
>> directly.
> Hello Eric,
>
> Atleast on ppc64 we have direct mapping available and hence just by doing page shift
> on physical address (__va) we can get valid virtual address on powerpc. In short we don't
> have to generate mapping again to access reserved region.
>
> Regardless let's go with kdump_local_page API, it is supported on powerpc.
>
> Thanks,
> Sourabh Jain
Ok, I will go that route.
Thanks!
eric
On 9/30/22 12:40, Borislav Petkov wrote:
> On Fri, Sep 30, 2022 at 12:11:26PM -0500, Eric DeVolder wrote:
>> There is of course a way to enumerate the memory regions in use on the
>> machine, that is not what this code needs. In order to compute the maximum
>> buffer size needed (this buffer size is computed once), the count of the
>> maximum number of memory regions possible (even if not currently in use) is
>> what is needed.
>
> Isn't that max number documented somewhere in memory hotplug docs?
>
> Because then you don't need that Kconfig item either. Imagine you're a
> distro kernel distributor and you want crash to work on all machines
> your kernel works.
>
> So you go and set that number to max. And that would be the 99% of the
> kernel configs out there.
>
> Which means, you can just set it to max without a Kconfig item.
>
>> Oh, that would be an error of haste on my part. This should be:
>> depends on CRASH_DUMP && MEMORY_HOTPLUG
>
> You need a Kconfig item which enables all this gunk as MEMORY_HOTPLUG is
> not a omnipresent feature. And that Kconfig item should depend on the
> other Kconfig items of the technology you need.
I once had CONFIG_CRASH_HOTPLUG, but you disagreed.
https://lore.kernel.org/lkml/Ylgot+LUDQl+G%[email protected]/
From there I simply went with
#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
which route do you prefer?
Thanks!
eric
>
>> Baoquan pointed me to:
>>
>> https://lore.kernel.org/lkml/[email protected]/T/
>
> In that thread says:
>
> "- arch_kexec_apply_relocations_add() is only overridden by x86 and s390.
> Retain the function prototype for those and move the weak
> implementation into the header as a static inline for other
> architectures."
>
> So yes, that's even better.
>
On Sat, Oct 08, 2022 at 10:35:14AM +0800, Baoquan He wrote:
> Memory hptlug is not limited by a certin or a max number of memory
> regions, but limited by how large of the linear mapping range which
> physical can be mapped into.
Memory hotplug is not limited by some abstract range but by the *actual*
possibility of how many DIMM slots on any motherboard can hotplug
memory. Certainly not 32K.
So you can choose a sane default which covers *all* actual systems out
there.
> For the Kconfig CRASH_MAX_MEMORY_RANGES Eric added, it's meaningful to
> me to set a fixed value which is enough in reality.
Yes, exactly.
> For extreme testing with special purpose, it could be broken easily,
> people need decide by self whether the CONFIG_CRASH_MAX_MEMORY_RANGES
> is enlarged or not.
I don't want for people to decide on one more thing where they have to
go and read a bunch of specs just to know what is a good value. So we
should set a sane, *practical* upper limit and simply go with it.
Everything else is testing stuff and if you test the kernel, then you
can change limits and values and so on as you want to.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/12/22 12:46, Borislav Petkov wrote:
> On Sat, Oct 08, 2022 at 10:35:14AM +0800, Baoquan He wrote:
>> Memory hptlug is not limited by a certin or a max number of memory
>> regions, but limited by how large of the linear mapping range which
>> physical can be mapped into.
>
> Memory hotplug is not limited by some abstract range but by the *actual*
> possibility of how many DIMM slots on any motherboard can hotplug
> memory. Certainly not 32K.
>
> So you can choose a sane default which covers *all* actual systems out
> there.
We run here QEMU with the ability for 1024 DIMM slots. A DIMM can be any
reasonable power of 2 size, and then that DIMM is further divided into memblocks,
typically 128MiB.
So, for example, 1TiB requires 1024 DIMMs of 1GiB each with 128MiB memblocks, that results
in 8K possible memory regions. So just going to 4TiB reaches 32K memory regions.
This I can attest for virtualized DIMMs, not sure about other memory hotplug technologies
like virtio-mem or DynamicMemory. But it seems reasonable that those technologies could
also easily reach into these number ranges.
Eric
>
>> For the Kconfig CRASH_MAX_MEMORY_RANGES Eric added, it's meaningful to
>> me to set a fixed value which is enough in reality.
>
> Yes, exactly.
>
>> For extreme testing with special purpose, it could be broken easily,
>> people need decide by self whether the CONFIG_CRASH_MAX_MEMORY_RANGES
>> is enlarged or not.
>
> I don't want for people to decide on one more thing where they have to
> go and read a bunch of specs just to know what is a good value. So we
> should set a sane, *practical* upper limit and simply go with it.
>
> Everything else is testing stuff and if you test the kernel, then you
> can change limits and values and so on as you want to.
>
> Thx.
>
On 10/12/22 15:19, Eric DeVolder wrote:
>
>
> On 10/12/22 12:46, Borislav Petkov wrote:
>> On Sat, Oct 08, 2022 at 10:35:14AM +0800, Baoquan He wrote:
>>> Memory hptlug is not limited by a certin or a max number of memory
>>> regions, but limited by how large of the linear mapping range which
>>> physical can be mapped into.
>>
>> Memory hotplug is not limited by some abstract range but by the *actual*
>> possibility of how many DIMM slots on any motherboard can hotplug
>> memory. Certainly not 32K.
>>
>> So you can choose a sane default which covers *all* actual systems out
>> there.
>
>
> We run here QEMU with the ability for 1024 DIMM slots. A DIMM can be any
> reasonable power of 2 size, and then that DIMM is further divided into memblocks,
> typically 128MiB.
>
> So, for example, 1TiB requires 1024 DIMMs of 1GiB each with 128MiB memblocks, that results
> in 8K possible memory regions. So just going to 4TiB reaches 32K memory regions.
>
> This I can attest for virtualized DIMMs, not sure about other memory hotplug technologies
> like virtio-mem or DynamicMemory. But it seems reasonable that those technologies could
> also easily reach into these number ranges.
>
> Eric
Oh, to be fair, if the above were fully populated, it would essentially coalescence
into a single reported region via crash_prepare_elf64_headers(). But in the sadistic
case, where every other memblock was offlined, that would result in the need to
report half of the memory regions via the elfcorehdr.
$0.02.
eric
>
>>
>>> For the Kconfig CRASH_MAX_MEMORY_RANGES Eric added, it's meaningful to
>>> me to set a fixed value which is enough in reality.
>>
>> Yes, exactly.
>>
>>> For extreme testing with special purpose, it could be broken easily,
>>> people need decide by self whether the CONFIG_CRASH_MAX_MEMORY_RANGES
>>> is enlarged or not.
>>
>> I don't want for people to decide on one more thing where they have to
>> go and read a bunch of specs just to know what is a good value. So we
>> should set a sane, *practical* upper limit and simply go with it.
>>
>> Everything else is testing stuff and if you test the kernel, then you
>> can change limits and values and so on as you want to.
>>
>> Thx.
>>
On Wed, Oct 12, 2022 at 03:19:19PM -0500, Eric DeVolder wrote:
> We run here QEMU with the ability for 1024 DIMM slots.
QEMU, haha.
What is the highest count of DIMM slots which are hotpluggable on a
real, *physical* system today? Are you saying you can have 1K DIMM slots
on a board?
I hardly doubt that.
> So, for example, 1TiB requires 1024 DIMMs of 1GiB each with 128MiB
> memblocks, that results in 8K possible memory regions. So just going
> to 4TiB reaches 32K memory regions.
Lemme see if I understand this correctly: when a system like that
crashes, you want to kdump *all* those 4TiB in a vmcore? How long would
that dump take to complete? A day?
IOW, how does a realistic use case of this look like - not a QEMU one?
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/12/22 at 10:41pm, Borislav Petkov wrote:
> On Wed, Oct 12, 2022 at 03:19:19PM -0500, Eric DeVolder wrote:
> > We run here QEMU with the ability for 1024 DIMM slots.
>
> QEMU, haha.
>
> What is the highest count of DIMM slots which are hotpluggable on a
> real, *physical* system today? Are you saying you can have 1K DIMM slots
> on a board?
The concern to range number mainly is on Virt guest systems. On
baremetal system, basically only very high end server support memory hotplug.
I ever visited customer's lab and saw one server, it owns 8 slots, on
each slot a box containing about 20 cpus and 2T memory at most can be
plugged in at one time. So people won't make too many slots for
hotplugging since it's too expensive.
I checked user space kexec code, the maximum memory range number is
honored to x86_64 because of a HPE SGI system. After that, nobody
complains about it. Please see below user space kexec-tools commit in
https://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
The memory ranges may be not all made by different DIMM slots, could be
firmware reservatoin, e.g efi/BIOS diggged out physical memory, or the
cpu logical address space is occupied by pci or other stuffs. I don't
have a HPE SGI system at hand to check.
commit 4a6d67d9e938a7accf128aff23f8ad4bda67f729
Author: Xunlei Pang <[email protected]>
Date: Thu Mar 23 19:16:59 2017 +0800
x86: Support large number of memory ranges
We got a problem on one SGI 64TB machine, the current kexec-tools
failed to work due to the insufficient ranges(MAX_MEMORY_RANGES)
allowed which is defined as 1024(less than the ranges on the machine).
The kcore header is insufficient due to the same reason as well.
To solve this, this patch simply doubles "MAX_MEMORY_RANGES" and
"KCORE_ELF_HEADERS_SIZE".
Signed-off-by: Xunlei Pang <[email protected]>
Tested-by: Frank Ramsay <[email protected]>
Signed-off-by: Simon Horman <[email protected]>
diff --git a/kexec/arch/i386/kexec-x86.h b/kexec/arch/i386/kexec-x86.h
index 33df3524f4e2..51855f8db762 100644
--- a/kexec/arch/i386/kexec-x86.h
+++ b/kexec/arch/i386/kexec-x86.h
@@ -1,7 +1,7 @@
#ifndef KEXEC_X86_H
#define KEXEC_X86_H
-#define MAX_MEMORY_RANGES 1024
+#define MAX_MEMORY_RANGES 2048
>
> I hardly doubt that.
The questioning is reasonable. 32K truly looks too much.
Now CONFIG_NR_CPUS has the maximum number as 8192. And user space
kexec-tools has maximum memory range number as 2048. We can take
the current 8192 + 2048 = 10K as default value conservatively. Or
take 8192 + 2048 * 2 = 12K which has two times of maximum memory range
bumber in kexec-tools. What do you think?
>
> > So, for example, 1TiB requires 1024 DIMMs of 1GiB each with 128MiB
> > memblocks, that results in 8K possible memory regions. So just going
> > to 4TiB reaches 32K memory regions.
>
> Lemme see if I understand this correctly: when a system like that
> crashes, you want to kdump *all* those 4TiB in a vmcore? How long would
> that dump take to complete? A day?
That is not a problem. The time of vmcore dumping mainly depends on the
actual memory size, not on memory range numbers. when dumping vmcore,
people use makedumpfile to filter zero page, free page, cache page, or user
date page according to configuration. If memory is huge, they can use
nr_cpus=x to enable multiple cpu to do multi-thread dumping. Kdump now
support more than 10 TB vmcore dumping.
On 08/10/22 01:03, Eric DeVolder wrote:
>
>
> On 9/19/22 02:06, Sourabh Jain wrote:
>>
>> On 10/09/22 02:35, Eric DeVolder wrote:
>>> For x86_64, when CPU or memory is hot un/plugged, the crash
>>> elfcorehdr, which describes the CPUs and memory in the system,
>>> must also be updated.
>>>
>>> When loading the crash kernel via kexec_load or kexec_file_load,
>>> the elfcorehdr is identified at run time in
>>> crash_core:handle_hotplug_event().
>>>
>>> To update the elfcorehdr for x86_64, a new elfcorehdr must be
>>> generated from the available CPUs and memory. The new elfcorehdr
>>> is prepared into a buffer, and then installed over the top of
>>> the existing elfcorehdr.
>>>
>>> In the patch 'kexec: exclude elfcorehdr from the segment digest'
>>> the need to update purgatory due to the change in elfcorehdr was
>>> eliminated. As a result, no changes to purgatory or boot_params
>>> (as the elfcorehdr= kernel command line parameter pointer
>>> remains unchanged and correct) are needed, just elfcorehdr.
>>>
>>> To accommodate a growing number of resources via hotplug, the
>>> elfcorehdr segment must be sufficiently large enough to accommodate
>>> changes, see the CRASH_MAX_MEMORY_RANGES configure item.
>>>
>>> With this change, crash hotplug for kexec_file_load syscall
>>> is supported. The kexec_load is also supported, but also
>>> requires a corresponding change to userspace kexec-tools.
>>>
>>> Signed-off-by: Eric DeVolder <[email protected]>
>>> Acked-by: Baoquan He <[email protected]>
>>> ---
>>>  arch/x86/Kconfig            | 11 ++++
>>> Â arch/x86/include/asm/kexec.h |Â 20 +++++++
>>>  arch/x86/kernel/crash.c     | 102
>>> +++++++++++++++++++++++++++++++++++
>>> Â 3 files changed, 133 insertions(+)
>>>
>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>> index f9920f1341c8..cdfc9b2fdf98 100644
>>> --- a/arch/x86/Kconfig
>>> +++ b/arch/x86/Kconfig
>>> @@ -2056,6 +2056,17 @@ config CRASH_DUMP
>>> Â Â Â Â Â Â Â (CONFIG_RELOCATABLE=y).
>>> Â Â Â Â Â Â Â For more details see Documentation/admin-guide/kdump/kdump.rst
>>> +config CRASH_MAX_MEMORY_RANGES
>>> +Â Â Â depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU ||
>>> MEMORY_HOTPLUG)
>>> +Â Â Â int
>>> +Â Â Â default 32768
>>> +Â Â Â help
>>> +Â Â Â Â Â For the kexec_file_load path, specify the maximum number of
>>> +Â Â Â Â Â memory regions, eg. as represented by the 'System RAM' entries
>>> +Â Â Â Â Â in /proc/iomem, that the elfcorehdr buffer/segment can
>>> accommodate.
>>> +Â Â Â Â Â This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>>> +Â Â Â Â Â size to determine the final buffer size.
>>> +
>>> Â config KEXEC_JUMP
>>> Â Â Â Â Â bool "kexec jump"
>>> Â Â Â Â Â depends on KEXEC && HIBERNATION
>>> diff --git a/arch/x86/include/asm/kexec.h
>>> b/arch/x86/include/asm/kexec.h
>>> index a3760ca796aa..432073385b2d 100644
>>> --- a/arch/x86/include/asm/kexec.h
>>> +++ b/arch/x86/include/asm/kexec.h
>>> @@ -212,6 +212,26 @@ typedef void crash_vmclear_fn(void);
>>> Â extern crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss;
>>> Â extern void kdump_nmi_shootdown_cpus(void);
>>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size);
>>> +#define arch_map_crash_pages arch_map_crash_pages
>>> +
>>> +void arch_unmap_crash_pages(void **ptr);
>>> +#define arch_unmap_crash_pages arch_unmap_crash_pages
>>> +
>>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>>> +Â Â Â Â Â Â Â unsigned int hp_action);
>>> +#define arch_crash_handle_hotplug_event
>>> arch_crash_handle_hotplug_event
>>> +
>>> +#ifdef CONFIG_HOTPLUG_CPU
>>> +static inline int crash_hotplug_cpu_support(void) { return 1; }
>>> +#define crash_hotplug_cpu_support crash_hotplug_cpu_support
>>> +#endif
>>> +
>>> +#ifdef CONFIG_MEMORY_HOTPLUG
>>> +static inline int crash_hotplug_memory_support(void) { return 1; }
>>> +#define crash_hotplug_memory_support crash_hotplug_memory_support
>>> +#endif
>>> +
>>> Â #endif /* __ASSEMBLY__ */
>>> Â #endif /* _ASM_X86_KEXEC_H */
>>> diff --git a/arch/x86/kernel/crash.c b/arch/x86/kernel/crash.c
>>> index 9ceb93c176a6..8fc7d678ac72 100644
>>> --- a/arch/x86/kernel/crash.c
>>> +++ b/arch/x86/kernel/crash.c
>>> @@ -25,6 +25,7 @@
>>> Â #include <linux/slab.h>
>>> Â #include <linux/vmalloc.h>
>>> Â #include <linux/memblock.h>
>>> +#include <linux/highmem.h>
>>> Â #include <asm/processor.h>
>>> Â #include <asm/hardirq.h>
>>> @@ -397,7 +398,18 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â image->elf_headers = kbuf.buffer;
>>> Â Â Â Â Â image->elf_headers_sz = kbuf.bufsz;
>>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>> +Â Â Â /* Ensure elfcorehdr segment large enough for hotplug changes */
>>> +Â Â Â kbuf.memsz =
>>> +Â Â Â Â Â Â Â (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
>>> +Â Â Â Â Â Â Â Â Â Â Â sizeof(Elf64_Phdr);
>>> +Â Â Â /* Mark as usable to crash kernel, else crash kernel fails on
>>> boot */
>>> +Â Â Â image->elf_headers_sz = kbuf.memsz;
>>> +Â Â Â image->elfcorehdr_index = image->nr_segments;
>>> +Â Â Â image->elfcorehdr_index_valid = true;
>>> +#else
>>> Â Â Â Â Â kbuf.memsz = kbuf.bufsz;
>>> +#endif
>>> Â Â Â Â Â kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
>>> Â Â Â Â Â kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
>>> Â Â Â Â Â ret = kexec_add_buffer(&kbuf);
>>> @@ -412,3 +424,93 @@ int crash_load_segments(struct kimage *image)
>>> Â Â Â Â Â return ret;
>>> Â }
>>> Â #endif /* CONFIG_KEXEC_FILE */
>>> +
>>> +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>>> +/*
>>> + * NOTE: The addresses and sizes passed to this routine have
>>> + * already been fully aligned on page boundaries. There is no
>>> + * need for massaging the address or size.
>>> + */
>>> +void *arch_map_crash_pages(unsigned long paddr, unsigned long size)
>>> +{
>>> +Â Â Â void *ptr = NULL;
>>> +
>>> +Â Â Â if (size > 0) {
>>> +Â Â Â Â Â Â Â struct page *page = pfn_to_page(paddr >> PAGE_SHIFT);
>>> +
>>> +Â Â Â Â Â Â Â ptr = kmap_local_page(page);
>>> +Â Â Â }
>>> +
>>> +Â Â Â return ptr;
>>> +}
>>> +
>>> +void arch_unmap_crash_pages(void **ptr)
>>> +{
>>> +Â Â Â if (ptr) {
>>> +Â Â Â Â Â Â Â if (*ptr)
>>> +Â Â Â Â Â Â Â Â Â Â Â kunmap_local(*ptr);
>>> +Â Â Â Â Â Â Â *ptr = NULL;
>>> +Â Â Â }
>>> +}
>>> +
>>> +/**
>>> + * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr
>>> changes
>>> + * @image: the active struct kimage
>>> + * @hp_action: the hot un/plug action being handled
>>> + *
>>> + * To accurately reflect hot un/plug changes, the new elfcorehdr
>>> + * is prepared in a kernel buffer, and then it is written on top
>>> + * of the existing/old elfcorehdr.
>>> + */
>>> +void arch_crash_handle_hotplug_event(struct kimage *image,
>>> +Â Â Â unsigned int hp_action)
>>> +{
>>> +Â Â Â struct kexec_segment *ksegment;
>>> +Â Â Â unsigned char *ptr = NULL;
>>> +Â Â Â unsigned long elfsz = 0;
>>> +Â Â Â void *elfbuf = NULL;
>>> +Â Â Â unsigned long mem, memsz;
>>> +
>>> +Â Â Â /*
>>> +Â Â Â Â * Elfcorehdr_index_valid checked in
>>> crash_core:handle_hotplug_event()
>>> +Â Â Â Â */
>>> +Â Â Â ksegment = &image->segment[image->elfcorehdr_index];
>>> +Â Â Â mem = ksegment->mem;
>>> +Â Â Â memsz = ksegment->memsz;
>>> +
>>> +Â Â Â /*
>>> +Â Â Â Â * Create the new elfcorehdr reflecting the changes to CPU and/or
>>> +Â Â Â Â * memory resources.
>>> +Â Â Â Â */
>>> +Â Â Â if (prepare_elf_headers(image, &elfbuf, &elfsz)) {
>>> +Â Â Â Â Â Â Â pr_err("crash hp: unable to prepare elfcore headers");
>>> +Â Â Â Â Â Â Â goto out;
>>
>> On PowerPC, while preparing the elf core header the memblock
>> structure is used to prepare program header for memory regions of
>> elfcorehdr. Since the above arch specific hotplug handler gets
>> invoked when memory is marked offline (MEM_OFFLINE) which is before
>> memblock structure gets updated so on PowerPC the above handler may
>> not work for memory hotplug case.
>>
>> Just wondering which data structure is used to get the list of memory
>> regions while preparing program header for memory regions of
>> elfcorehdr on other architectures?
>>
>> Thanks,
>> Sourabh Jain
>
> I think your request to report the memory block address in comments of
> patch 3/7 "crash: add generic infrastructure" cover this scenario now.
Yes, the asked changes will make easy for PowerPC to recreate elfcorehdr.
Thanks,
Sourabh Jain
On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
> The concern to range number mainly is on Virt guest systems.
And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
real machine?
> On baremetal system, basically only very high end server support
> memory hotplug. I ever visited customer's lab and saw one server,
> it owns 8 slots, on each slot a box containing about 20 cpus and 2T
> memory at most can be plugged in at one time. So people won't make too
> many slots for hotplugging since it's too expensive.
There you have it - the persuading argument.
> I checked user space kexec code, the maximum memory range number is
> honored to x86_64 because of a HPE SGI system. After that, nobody
> complains about it. Please see below user space kexec-tools commit in
> https://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
>
> The memory ranges may be not all made by different DIMM slots, could be
> firmware reservatoin, e.g efi/BIOS diggged out physical memory,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
I don't know what that means.
If it is firmware crap, you want to exclude that from kdump anyway.
> Now CONFIG_NR_CPUS has the maximum number as 8192. And user space
> kexec-tools has maximum memory range number as 2048. We can take
> the current 8192 + 2048 = 10K as default value conservatively. Or
> take 8192 + 2048 * 2 = 12K which has two times of maximum memory range
> bumber in kexec-tools. What do you think?
I still think that we should stick to reality and support what is
possible not what is potentially and theoretically there.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Wed, Oct 12, 2022 at 11:20:59AM -0500, Eric DeVolder wrote:
> I once had CONFIG_CRASH_HOTPLUG, but you disagreed.
>
> https://lore.kernel.org/lkml/Ylgot+LUDQl+G%[email protected]/
>
> From there I simply went with
>
> #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
>
> which route do you prefer?
If you do a single Kconfig item which depends on those two, it probably
is cleaner this way. And if the max memory ranges are hardcoded you
don't need the other prompt asking the user something she most likely
doesn't know how to answer properly.
That is, unless you wanna have that crash hotplug built in all the time.
Because CONFIG_HOTPLUG_CPU is pretty much always enabled so you might
just as well add the crash hotplug support unconditionally, without any
Kconfig ifdeffery whatsoever except CONFIG_MEMORY_HOTPLUG as that is
special and not present on the majority of hardware.
But on a plain simple laptop or workstation which has CPU hotplug, would
it make sense for the crash ranges to get updated too when CPUs are
offlined?
If so, I think you want this code present there too, without a Kconfig
item.
If this is server-only anyway, then a single Kconfig item sounds like
not such a bad idea.
I hope that makes some sense.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 26.10.22 16:48, Baoquan He wrote:
> On 10/25/22 at 12:31pm, Borislav Petkov wrote:
>> On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
>>> The concern to range number mainly is on Virt guest systems.
>>
>> And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
>> real machine?
IIRC, ACPI only allows for 256 slots. PPC dlpar might provide more.
>
> Well, currently, mem hotpug is an important feature on virt system to
> dynamically increase/shrink memory on the system. If only emulating real
> machine, it won't be different than bare metal system.
>
> IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
> 1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
> later, it will take best effort way to hot remove memory. Means if any
> memory block is occupied, it will be kept there. Finally we could only
> remove every second blocks, 4 blocks altogether. Then the left
> un-removed blocks will produce 4 separate memory regions. Like this, a
> virt guest could have many memory regions in kernel after memory
> being added/removed.
>
> If I am wrong, Please correct me, David.
Yes, virtio-mem (but also PPC dlpar) can result in many individual
memory blocks with holes in between after hotunplug. Hotplug OTOH,
usually tries to "plug" these holes and reduce the total number of
memory blocks. It might be rare that our range will be heavily
fragmented after unplug, but it's certainly possible.
[...]
>
> Yes, now assume we have a HPE SGI system and it has memory hotplug
> capacity. The system itself has already got memory regions more than
> 1024. Then when we hot add extra memory board, we want to include the
> newly added memory regions into elfcorehdr so that it will be dumped out
> in kdump kernel.
>
> That's why I earlier suggested 2048 for number of memory regions.
The more the better, unless "it hurts". Assuming a single memory block
is 128 MiB, that would be 256 GiB.
Usually, on big systems, the memory block size is 2 GiB. So 4 TiB.
--
Thanks,
David / dhildenb
On 10/25/22 at 12:31pm, Borislav Petkov wrote:
> On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
> > The concern to range number mainly is on Virt guest systems.
>
> And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
> real machine?
Well, currently, mem hotpug is an important feature on virt system to
dynamically increase/shrink memory on the system. If only emulating real
machine, it won't be different than bare metal system.
IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
later, it will take best effort way to hot remove memory. Means if any
memory block is occupied, it will be kept there. Finally we could only
remove every second blocks, 4 blocks altogether. Then the left
un-removed blocks will produce 4 separate memory regions. Like this, a
virt guest could have many memory regions in kernel after memory
being added/removed.
If I am wrong, Please correct me, David.
>
> > On baremetal system, basically only very high end server support
> > memory hotplug. I ever visited customer's lab and saw one server,
> > it owns 8 slots, on each slot a box containing about 20 cpus and 2T
> > memory at most can be plugged in at one time. So people won't make too
> > many slots for hotplugging since it's too expensive.
>
> There you have it - the persuading argument.
>
> > I checked user space kexec code, the maximum memory range number is
> > honored to x86_64 because of a HPE SGI system. After that, nobody
> > complains about it. Please see below user space kexec-tools commit in
> > https://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
> >
> > The memory ranges may be not all made by different DIMM slots, could be
> > firmware reservatoin, e.g efi/BIOS diggged out physical memory,
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> I don't know what that means.
>
> If it is firmware crap, you want to exclude that from kdump anyway.
Yes, now assume we have a HPE SGI system and it has memory hotplug
capacity. The system itself has already got memory regions more than
1024. Then when we hot add extra memory board, we want to include the
newly added memory regions into elfcorehdr so that it will be dumped out
in kdump kernel.
That's why I earlier suggested 2048 for number of memory regions.
>
> > Now CONFIG_NR_CPUS has the maximum number as 8192. And user space
> > kexec-tools has maximum memory range number as 2048. We can take
> > the current 8192 + 2048 = 10K as default value conservatively. Or
> > take 8192 + 2048 * 2 = 12K which has two times of maximum memory range
> > bumber in kexec-tools. What do you think?
>
> I still think that we should stick to reality and support what is
> possible not what is potentially and theoretically there.
Yes, agree. We should try to get a number which satisfies needs in
reality.
For Kconfig CRASH_MAX_MEMORY_RANGES in this patch, I have three items to
suggest:
1) the name is not good, it doesn't reflect the fact that it's the
number of program headers of elfcorehdr which includes the cpu note
numbers and memory region numers.
2) default cpu number, I suggest 512 or 1024. The biggest number I
ever saw in reality is 384. On virt system, it won't be too big. Below
is abstracted from arch/x86/Kconfig. A smaller one is also OK, we can
enlarge it when people really have a super machine and run into the
problem.
config NR_CPUS_DEFAULT
int
depends on X86_64
default 8192 if MAXSMP
default 64 if SMP
default 1 if !SMP
3) For memory regions, I would suggest 2048. Likewise, smaller value is
also fine, we can enlarge it when a real system run into this.
I made a draft here for reference, with my undertanding. Please feel
free to change it.
+config CRASH_ELF_CORE_PHDRS_NUM
+ depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
+ int
+ default 3072
+ help
+ For the kexec_file_load path, specify the default number of
+ phdr for the vmcore. E.g the memory regions represented by the
+ 'System RAM' entries in /proc/iomem, the cpu notes of each
+ present cpu stored in /sys/devices/system/cpu/cpuX/crash_notes.
Thanks
On 10/26/22 at 04:54pm, David Hildenbrand wrote:
> On 26.10.22 16:48, Baoquan He wrote:
> > On 10/25/22 at 12:31pm, Borislav Petkov wrote:
> > > On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
> > > > The concern to range number mainly is on Virt guest systems.
> > >
> > > And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
> > > real machine?
>
> IIRC, ACPI only allows for 256 slots. PPC dlpar might provide more.
>
> >
> > Well, currently, mem hotpug is an important feature on virt system to
> > dynamically increase/shrink memory on the system. If only emulating real
> > machine, it won't be different than bare metal system.
> >
> > IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
> > 1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
> > later, it will take best effort way to hot remove memory. Means if any
> > memory block is occupied, it will be kept there. Finally we could only
> > remove every second blocks, 4 blocks altogether. Then the left
> > un-removed blocks will produce 4 separate memory regions. Like this, a
> > virt guest could have many memory regions in kernel after memory
> > being added/removed.
> >
> > If I am wrong, Please correct me, David.
>
> Yes, virtio-mem (but also PPC dlpar) can result in many individual memory
> blocks with holes in between after hotunplug. Hotplug OTOH, usually tries to
> "plug" these holes and reduce the total number of memory blocks. It might be
> rare that our range will be heavily fragmented after unplug, but it's
> certainly possible.
>
> [...]
>
> >
> > Yes, now assume we have a HPE SGI system and it has memory hotplug
> > capacity. The system itself has already got memory regions more than
> > 1024. Then when we hot add extra memory board, we want to include the
> > newly added memory regions into elfcorehdr so that it will be dumped out
> > in kdump kernel.
> >
> > That's why I earlier suggested 2048 for number of memory regions.
>
> The more the better, unless "it hurts". Assuming a single memory block is
> 128 MiB, that would be 256 GiB.
>
> Usually, on big systems, the memory block size is 2 GiB. So 4 TiB.
Thanks a lot for these valuable inputs, David.
Hi Boris, Eric
So what's your suggested value for the Kconfig option?
1) cpu number, 1024?
2) memory regions, 2048?
About below draft, any comment? We can decide a value based on our
knowledge, can adjust later if any real system has more than the number.
+config CRASH_ELF_CORE_PHDRS_NUM
+ depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
+ int
+ default 3072
+ help
+ For the kexec_file_load path, specify the default number of
+ phdr for the vmcore. E.g the memory regions represented by the
+ 'System RAM' entries in /proc/iomem, the cpu notes of each
+ present cpu stored in /sys/devices/system/cpu/cpuX/crash_notes.
Thanks
Baoquan
On 10/27/22 08:52, Baoquan He wrote:
> On 10/26/22 at 04:54pm, David Hildenbrand wrote:
>> On 26.10.22 16:48, Baoquan He wrote:
>>> On 10/25/22 at 12:31pm, Borislav Petkov wrote:
>>>> On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
>>>>> The concern to range number mainly is on Virt guest systems.
>>>>
>>>> And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
>>>> real machine?
>>
>> IIRC, ACPI only allows for 256 slots. PPC dlpar might provide more.
>>
>>>
>>> Well, currently, mem hotpug is an important feature on virt system to
>>> dynamically increase/shrink memory on the system. If only emulating real
>>> machine, it won't be different than bare metal system.
>>>
>>> IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
>>> 1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
>>> later, it will take best effort way to hot remove memory. Means if any
>>> memory block is occupied, it will be kept there. Finally we could only
>>> remove every second blocks, 4 blocks altogether. Then the left
>>> un-removed blocks will produce 4 separate memory regions. Like this, a
>>> virt guest could have many memory regions in kernel after memory
>>> being added/removed.
>>>
>>> If I am wrong, Please correct me, David.
>>
>> Yes, virtio-mem (but also PPC dlpar) can result in many individual memory
>> blocks with holes in between after hotunplug. Hotplug OTOH, usually tries to
>> "plug" these holes and reduce the total number of memory blocks. It might be
>> rare that our range will be heavily fragmented after unplug, but it's
>> certainly possible.
>>
>> [...]
>>
>>>
>>> Yes, now assume we have a HPE SGI system and it has memory hotplug
>>> capacity. The system itself has already got memory regions more than
>>> 1024. Then when we hot add extra memory board, we want to include the
>>> newly added memory regions into elfcorehdr so that it will be dumped out
>>> in kdump kernel.
>>>
>>> That's why I earlier suggested 2048 for number of memory regions.
>>
>> The more the better, unless "it hurts". Assuming a single memory block is
>> 128 MiB, that would be 256 GiB.
>>
>> Usually, on big systems, the memory block size is 2 GiB. So 4 TiB.
>
> Thanks a lot for these valuable inputs, David.
>
> Hi Boris, Eric
>
> So what's your suggested value for the Kconfig option?
>
> 1) cpu number, 1024?
> 2) memory regions, 2048?
>
> About below draft, any comment? We can decide a value based on our
> knowledge, can adjust later if any real system has more than the number.
>
> +config CRASH_ELF_CORE_PHDRS_NUM
> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> + int
> + default 3072
> + help
> + For the kexec_file_load path, specify the default number of
> + phdr for the vmcore. E.g the memory regions represented by the
> + 'System RAM' entries in /proc/iomem, the cpu notes of each
> + present cpu stored in /sys/devices/system/cpu/cpuX/crash_notes.
>
> Thanks
> Baoquan
>
I prefer to keep CRASH_MAX_MEMORY_RANGES, as explained in my response to your message on October 26.
eric
On 10/26/22 09:48, Baoquan He wrote:
> On 10/25/22 at 12:31pm, Borislav Petkov wrote:
>> On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
>>> The concern to range number mainly is on Virt guest systems.
>>
>> And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
>> real machine?
>
> Well, currently, mem hotpug is an important feature on virt system to
> dynamically increase/shrink memory on the system. If only emulating real
> machine, it won't be different than bare metal system.
>
> IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
> 1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
> later, it will take best effort way to hot remove memory. Means if any
> memory block is occupied, it will be kept there. Finally we could only
> remove every second blocks, 4 blocks altogether. Then the left
> un-removed blocks will produce 4 separate memory regions. Like this, a
> virt guest could have many memory regions in kernel after memory
> being added/removed.
>
> If I am wrong, Please correct me, David.
>
>>
>>> On baremetal system, basically only very high end server support
>>> memory hotplug. I ever visited customer's lab and saw one server,
>>> it owns 8 slots, on each slot a box containing about 20 cpus and 2T
>>> memory at most can be plugged in at one time. So people won't make too
>>> many slots for hotplugging since it's too expensive.
>>
>> There you have it - the persuading argument.
So after re-reading the exchanges, many times, I think I realized that I have introduced confusion
by using "hotplug", and specifically "memory hotplug" and DIMMs in the same breath, and thus that
perhaps equated hotplug with ACPI DIMMs in these discussions.
Allow me to state that "hotplug" in this patch series refers to CPU and memory hot un/plug, and does
*not* explicitly refer to any particular underlying technology to generate CPU and/or memory hot
un/plug events.
For example, I have been using DIMMs as my example because that has been my test vehicle for
exercising this code; as such I think the discussion cornered itself into real world vs virt
discussion about DIMMs, with no end in sight. To be plain, this patch series does not intend to
convey or change anything specific about ACPI DIMMs.
In reality, when I state "hotplug" in these patches, I am talking generically and therefore
inclusive of any technology that can hot un/plug CPUs or memory. For memory specifically, this
includes ACPI DIMMs (whether baremetal or QEMU), ballooning, virtio-mem, PPC dlpar (per David),
Microsoft DynamicMemory, and the upcoming CXL.mem technology. Probably others that I am not aware.
Any of these technologies can add or remove memory from a bare metal and/or virtual system.
I apologize.
What is important is the number of memory regions (ie. /proc/iomem entries) that can be considered
to be the maximum. There is no kernel definition of such. The need to identify a maximum number is
so that the buffer containing the elfcorehdr can be sized and allocated at kdump load time, *once*.
This elfcorehdr buffer is then modified/updated repeatedly as hot un/plug events occur. It is *not*
re-allocated on each hot un/plug event; that is what the current solution does, sort of.
>>
>>> I checked user space kexec code, the maximum memory range number is
>>> honored to x86_64 because of a HPE SGI system. After that, nobody
>>> complains about it. Please see below user space kexec-tools commit in
>>> https://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git
>>>
>>> The memory ranges may be not all made by different DIMM slots, could be
>>> firmware reservatoin, e.g efi/BIOS diggged out physical memory,
>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> I don't know what that means.
>>
>> If it is firmware crap, you want to exclude that from kdump anyway.
>
> Yes, now assume we have a HPE SGI system and it has memory hotplug
> capacity. The system itself has already got memory regions more than
> 1024. Then when we hot add extra memory board, we want to include the
> newly added memory regions into elfcorehdr so that it will be dumped out
> in kdump kernel.
>
> That's why I earlier suggested 2048 for number of memory regions.
>
>>
>>> Now CONFIG_NR_CPUS has the maximum number as 8192. And user space
>>> kexec-tools has maximum memory range number as 2048. We can take
>>> the current 8192 + 2048 = 10K as default value conservatively. Or
>>> take 8192 + 2048 * 2 = 12K which has two times of maximum memory range
>>> bumber in kexec-tools. What do you think?
>>
>> I still think that we should stick to reality and support what is
>> possible not what is potentially and theoretically there.
>
> Yes, agree. We should try to get a number which satisfies needs in
> reality.
>
> For Kconfig CRASH_MAX_MEMORY_RANGES in this patch, I have three items to
> suggest:
>
> 1) the name is not good, it doesn't reflect the fact that it's the
> number of program headers of elfcorehdr which includes the cpu note
> numbers and memory region numers.
The total number of program headers is, generally speaking, the number of CPUs and the number of
memory regions (plug one for VMCOREINFO and maybe a few others). The NR_CPUS_DEFAULT conveys the
maximum number of CPUs possible, and likewise the CRASH_MAX_MEMORY_RANGES is intended to convey the
maximum number of memory regions possible.
It is not misnamed, imho; rather, I think due to the confusion I outline above, misunderstood.
>
> 2) default cpu number, I suggest 512 or 1024. The biggest number I
> ever saw in reality is 384. On virt system, it won't be too big. Below
> is abstracted from arch/x86/Kconfig. A smaller one is also OK, we can
> enlarge it when people really have a super machine and run into the
> problem.
>
> config NR_CPUS_DEFAULT
> int
> depends on X86_64
> default 8192 if MAXSMP
> default 64 if SMP
> default 1 if !SMP
I'm all for making this a sane number, I'm just not sure this patch series is the place to do this?
>
> 3) For memory regions, I would suggest 2048. Likewise, smaller value is
> also fine, we can enlarge it when a real system run into this.
As David points out, if the memblock size if 128MiB, then 2048 allows for 256GiB, which I believe to
be too low for a maximum default. I'd at least target 1TiB with 128MiB memblock size, which would
put the number at 8192.
Should the memblock size be 2GiB, for really big systems, then 8192 entries allow handling 16GiB.
With sizeof(elf64_phdr) of 64 bytes, that means the elfcorehdr buffer/memory segment is essentially
512KiB.
Be aware, in reality, that if the system was fully populated, it would not actually consume all 8192
phdrs. Rather /proc/iomem would essentially show a large contiguous address space which would
require just a single phdr. The reason to consider having the larger number of phdrs is so that if
the memory becomes fragmented due to hot un/plug events, then you need the phdrs to record the
sparse mapping. The sadistic case is that every other memblock is offlined/removed, not likely, but
possible.
>
> I made a draft here for reference, with my undertanding. Please feel
> free to change it.
>
> +config CRASH_ELF_CORE_PHDRS_NUM
> + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> + int
> + default 3072
> + help
> + For the kexec_file_load path, specify the default number of
> + phdr for the vmcore. E.g the memory regions represented by the
> + 'System RAM' entries in /proc/iomem, the cpu notes of each
> + present cpu stored in /sys/devices/system/cpu/cpuX/crash_notes.
>
> Thanks
>
I'd prefer keeping CRASH_MAX_MEMORY_RANGES as that allow the maximum phdr number value to be
reflective of CPUs and/or memory; not all systems support both CPU and memory hotplug. For example,
I have queued up this change to reflect this:
if (IS_ENABLED(CONFIG_HOTPLUG_CPU) || IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
/* Ensure elfcorehdr segment large enough for hotplug changes */
unsigned long pnum = 2; /* VMCOREINFO and kernel_map */
if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
pnum += CONFIG_NR_CPUS_DEFAULT;
if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
pnum += CONFIG_CRASH_MAX_MEMORY_RANGES;
if (pnum < (unsigned long)PN_XNUM) {
kbuf.memsz = pnum * sizeof(Elf64_Phdr);
kbuf.memsz += sizeof(Elf64_Ehdr);
image->elfcorehdr_index = image->nr_segments;
image->elfcorehdr_index_valid = true;
/* Mark as usable to crash kernel, else crash kernel fails on boot *
image->elf_headers_sz = kbuf.memsz;
} else {
pr_err("number of Phdrs %lu exceeds max %lu\n", pnum, (unsigned long
}
}
I hope this helps clarify my intentions with this patch series.
Thanks,
eric
On Thu, Oct 27, 2022 at 02:24:11PM -0500, Eric DeVolder wrote:
> Be aware, in reality, that if the system was fully populated, it would not
> actually consume all 8192 phdrs. Rather /proc/iomem would essentially show a
> large contiguous address space which would require just a single phdr.
Then that from below:
pnum += CONFIG_CRASH_MAX_MEMORY_RANGES;
which then would end up allocating 8192 would be a total waste.
So why don't you make that number dynamic then?
You start with something sensible:
total_num_pheaders = num_online_cpus() + "some number of regions" + "some few others"
I.e., a number which is a good compromise on the majority of machines.
Then, on hotplug events you count how many new regions are coming in
and when you reach the total_num_pheaders number, you double it (or some
other increase stragegy), reallocate the ELF header buffers etc needed
for kdump and you're good.
This way, you don't waste memory unnecessarily on the majority of
systems and those who need more, get to allocate more.
> I'd prefer keeping CRASH_MAX_MEMORY_RANGES as that allow the maximum phdr
> number value to be reflective of CPUs and/or memory; not all systems support
> both CPU and memory hotplug. For example, I have queued up this change to
> reflect this:
>
> if (IS_ENABLED(CONFIG_HOTPLUG_CPU) || IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
If you're going to keep CRASH_MAX_MEMORY_RANGES, then you can test only
that thing as it expresses the dependency on CONFIG_HOTPLUG_CPU and
CONFIG_MEMORY_HOTPLUG already.
If you end up making the number dynamic, then you could make that a
different Kconfig item which contains all that crash code as most of the
people won't need it anyway.
Hmm?
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/28/22 05:19, Borislav Petkov wrote:
> On Thu, Oct 27, 2022 at 02:24:11PM -0500, Eric DeVolder wrote:
>> Be aware, in reality, that if the system was fully populated, it would not
>> actually consume all 8192 phdrs. Rather /proc/iomem would essentially show a
>> large contiguous address space which would require just a single phdr.
>
> Then that from below:
>
> pnum += CONFIG_CRASH_MAX_MEMORY_RANGES;
>
> which then would end up allocating 8192 would be a total waste.
>
> So why don't you make that number dynamic then?
>
> You start with something sensible:
>
> total_num_pheaders = num_online_cpus() + "some number of regions" + "some few others"
>
> I.e., a number which is a good compromise on the majority of machines.
>
> Then, on hotplug events you count how many new regions are coming in
> and when you reach the total_num_pheaders number, you double it (or some
> other increase stragegy), reallocate the ELF header buffers etc needed
> for kdump and you're good.
>
> This way, you don't waste memory unnecessarily on the majority of
> systems and those who need more, get to allocate more.
This patch series sizes and allocates the memory buffer/segment for the elfcorehdr once, at kdump
load time.
In order to dynamically resize the elcorehdr memory buffer/segment, that causes the following ripple
effects:
- Permitting resizing of the elfcorehdr requires a means to "allocate" a new size buffer from
within the crash kernel reserved area. There is no allocator today; currently it is a kind of
one-pass placement process that happens at load time. The critical side effect of allocating a new
elfcorehdr buffer memory/segment is that it creates a new address for the elfcorehdr.
- The elfcorehdr is passed to the crash kernel via the elfcorehdr= kernel cmdline option. As such,
a dynamic change to the size of the elfcorehdr size necessarily invites a change of address of that
buffer, and therefore a change to rewrite the crash kernel cmdline to reflect the new elfcorehdr
buffer address.
- A change to the cmdline, also invites a possible change of address of the buffer containing the
cmdline, and thus a change to the x86 boot_params, which contains the cmdline pointer.
- A change to the cmdline and/or boot_params, which are *not* excluded from the hash/digest, means
that the purgatory hash/digest needs to be recomputed, and purgatory re-linked with the new
hash/digest and replaced.
A fair amount of work, but I have had this working in the past, around the v1 patch series
timeframe. However, it only worked for the kexec_file_load() syscall as all the needed pieces of
information were available; but for kexec_load(), it isn't possible to relink purgatory as by that
point purgatory is but a user-space binary blob.
It was feedback on the v1/v2 that pointed out that by excluding the elfcorehdr from the hash/digest,
the "change of address" problem with the elfcorehdr buffer/segment goes away, and, in turn, negates
the need to: introduce an allocator for the crash kernel reserved space, rewrite the crash kernel
cmdline with a new elfcorehdr, update boot_params with a new cmdline and re-link and replace
purgatory with the updated digest. And it enables this hotplug efforts to support kexec_load()
syscall as well.
So it is with this in mind that I suggest we stay with the statically sized elfcorehdr buffer.
If that can be agreed upon, then it is "just a matter" of picking a useful elfcorehdr size.
Currently that size is derived from the NR_DEFAULT_CPUS and CRASH_MAX_MEMORY_RANGES. So, there is
still the CRASH_MAX_MEMORY_RANGES knob to help a dial in size, should there be some issue with the
default value/size.
Or if there is desire to drop computing the size from NR_DEFAULT_CPUs and CRASH_MAX_MEMORY_RANGES
and simply go with CRASH_HOTPLUG_ELFCOREHDR_SZ which simply specifies the buffer size, then I'm also
good with that.
I still owe a much better explanation of how to size the elfcorehdr. I can use the comments and
ideas from the discussion to provide the necessary insight when choosing this value, whether that be
CRASH_MAX_MEMORY_RANGES or CRASH_HOTPLUG_ELFCOREHDR_SZ.
>
>> I'd prefer keeping CRASH_MAX_MEMORY_RANGES as that allow the maximum phdr
>> number value to be reflective of CPUs and/or memory; not all systems support
>> both CPU and memory hotplug. For example, I have queued up this change to
>> reflect this:
>>
>> if (IS_ENABLED(CONFIG_HOTPLUG_CPU) || IS_ENABLED(CONFIG_MEMORY_HOTPLUG)) {
>
> If you're going to keep CRASH_MAX_MEMORY_RANGES, then you can test only
> that thing as it expresses the dependency on CONFIG_HOTPLUG_CPU and
> CONFIG_MEMORY_HOTPLUG already.
>
> If you end up making the number dynamic, then you could make that a
> different Kconfig item which contains all that crash code as most of the
> people won't need it anyway.
It is my intention to correct the CRASH_MAX_MEMORY_RANGES (if we keep it) as such:
config CRASH_MAX_MEMORY_RANGES
depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
CRASH_MAX_MEMORY_RANGES should have never had CPU_HOTPLUG as a dependency; that was a cut-n-paste
error on my part.
>
> Hmm?
>
Thank you for the time and thought on this topic!
eric
On Fri, Oct 28, 2022 at 10:29:45AM -0500, Eric DeVolder wrote:
> So it is with this in mind that I suggest we stay with the statically sized elfcorehdr buffer.
>
> If that can be agreed upon, then it is "just a matter" of picking a useful
> elfcorehdr size. Currently that size is derived from the NR_DEFAULT_CPUS and
> CRASH_MAX_MEMORY_RANGES. So, there is still the CRASH_MAX_MEMORY_RANGES knob
> to help a dial in size, should there be some issue with the default
> value/size.
Let's see
kbuf.memsz =
(CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
sizeof(Elf64_Phdr);
which, IINM, is
(8192 + 32768) * 56
which is something like 2M.
(CONFIG_NR_CPUS_DEFAULT = 8192 - this is because of MAXSMP which gets
set on distro kernels)
Now, since userspace kexec tools uses 2048 for max memory ranges, that
size becomes smaller - around half a Mb. And since y'all wanna be on the
safe side, you can quadruple it and have
(8192 + 8192) * 56
which is still under a megabyte. And that's fine, I guess, on a big
server.
> Or if there is desire to drop computing the size from NR_DEFAULT_CPUs and
I think you should leave the dependency on the Kconfig size so that
smaller machines which are configured this way, don't end up wasting
unnecessary memory.
> It is my intention to correct the CRASH_MAX_MEMORY_RANGES (if we keep it) as such:
>
> config CRASH_MAX_MEMORY_RANGES
> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
Yes, but don't leave it to the user to decide what number to choose
- choose a high enough number, explain why you've chosen this with a
comment and that's it.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/28/22 12:06, Borislav Petkov wrote:
> On Fri, Oct 28, 2022 at 10:29:45AM -0500, Eric DeVolder wrote:
>> So it is with this in mind that I suggest we stay with the statically sized elfcorehdr buffer.
>>
>> If that can be agreed upon, then it is "just a matter" of picking a useful
>> elfcorehdr size. Currently that size is derived from the NR_DEFAULT_CPUS and
>> CRASH_MAX_MEMORY_RANGES. So, there is still the CRASH_MAX_MEMORY_RANGES knob
>> to help a dial in size, should there be some issue with the default
>> value/size.
>
> Let's see
>
> kbuf.memsz =
> (CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES) *
> sizeof(Elf64_Phdr);
>
> which, IINM, is
>
> (8192 + 32768) * 56
>
> which is something like 2M.
>
> (CONFIG_NR_CPUS_DEFAULT = 8192 - this is because of MAXSMP which gets
> set on distro kernels)
>
> Now, since userspace kexec tools uses 2048 for max memory ranges, that
> size becomes smaller - around half a Mb. And since y'all wanna be on the
> safe side, you can quadruple it and have
>
> (8192 + 8192) * 56
>
> which is still under a megabyte. And that's fine, I guess, on a big
> server.
Excellent, I'll set CRASH_MAX_MEMORY_RANGES to 8192! That seems a quite fair trade off of elfcorehdr
size vs system size (ie 1TiB w/ 128MiB memblock size).
>
>> Or if there is desire to drop computing the size from NR_DEFAULT_CPUs and
>
> I think you should leave the dependency on the Kconfig size so that
> smaller machines which are configured this way, don't end up wasting
> unnecessary memory.
Excellent, I'll leave the computation as NR_DEFAULT_CPUS + CRASH_MAX_MEMORY_RANGES.
>
>> It is my intention to correct the CRASH_MAX_MEMORY_RANGES (if we keep it) as such:
>>
>> config CRASH_MAX_MEMORY_RANGES
>> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
>
> Yes, but don't leave it to the user to decide what number to choose
> - choose a high enough number, explain why you've chosen this with a
> comment and that's it.
I currently have the Kconfig item as:
config CRASH_MAX_MEMORY_RANGES
depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
int
default 8192
help
For the kexec_file_load path, specify the maximum number of
memory regions, eg. as represented by the 'System RAM' entries
in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
This value is combined with NR_CPUS and multiplied by Elf64_Phdr
size to determine the final buffer size.
I'll work to provide information a better explanation as to the 8192 number.
Thank you!
eric
>
> Thx.
>
On 10/28/22 15:30, Borislav Petkov wrote:
> On Fri, Oct 28, 2022 at 02:26:58PM -0500, Eric DeVolder wrote:
>> config CRASH_MAX_MEMORY_RANGES
>> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
>> int
>> default 8192
>> help
>> For the kexec_file_load path, specify the maximum number of
>> memory regions, eg. as represented by the 'System RAM' entries
>> in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
>> This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>> size to determine the final buffer size.
>
> No, do this:
>
> config CRASH_MEMORY_HOTPLUG_SUPPORT
> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
> help
> Help text explaining what this feature is
>
> this thing will simply get enabled when the user enables MEMORY_HOTPLUG
> and CRASH_DUMP.
>
> and then you do in the code:
>
> /*
> * A comment explaining how the 8192 value has been selected.
> */
> #define CRASH_MAX_MEMORY_RANGES 8192
>
> Thx.
>
ok, will do!
thanks!
eric
On Fri, Oct 28, 2022 at 02:26:58PM -0500, Eric DeVolder wrote:
> config CRASH_MAX_MEMORY_RANGES
> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
> int
> default 8192
> help
> For the kexec_file_load path, specify the maximum number of
> memory regions, eg. as represented by the 'System RAM' entries
> in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
> This value is combined with NR_CPUS and multiplied by Elf64_Phdr
> size to determine the final buffer size.
No, do this:
config CRASH_MEMORY_HOTPLUG_SUPPORT
depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
help
Help text explaining what this feature is
this thing will simply get enabled when the user enables MEMORY_HOTPLUG
and CRASH_DUMP.
and then you do in the code:
/*
* A comment explaining how the 8192 value has been selected.
*/
#define CRASH_MAX_MEMORY_RANGES 8192
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/28/22 15:30, Borislav Petkov wrote:
> On Fri, Oct 28, 2022 at 02:26:58PM -0500, Eric DeVolder wrote:
>> config CRASH_MAX_MEMORY_RANGES
>> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
>> int
>> default 8192
>> help
>> For the kexec_file_load path, specify the maximum number of
>> memory regions, eg. as represented by the 'System RAM' entries
>> in /proc/iomem, that the elfcorehdr buffer/segment can accommodate.
>> This value is combined with NR_CPUS and multiplied by Elf64_Phdr
>> size to determine the final buffer size.
>
> No, do this:
>
> config CRASH_MEMORY_HOTPLUG_SUPPORT
> depends on CRASH_DUMP && KEXEC_FILE && MEMORY_HOTPLUG
> help
> Help text explaining what this feature is
>
> this thing will simply get enabled when the user enables MEMORY_HOTPLUG
> and CRASH_DUMP.
>
> and then you do in the code:
>
> /*
> * A comment explaining how the 8192 value has been selected.
> */
> #define CRASH_MAX_MEMORY_RANGES 8192
>
> Thx.
>
How is this comment?
/*
* For the kexec_file_load() syscall path, specify the maximum number of
* memory regions that the elfcorehdr buffer/segment can accommodate.
* These regions are obtained via walk_system_ram_res(); eg. the
* 'System RAM' entries in /proc/iomem.
* This value is combined with NR_CPUS and multiplied by sizeof(Elf64_Phdr)
* to determine the final elfcorehdr memory buffer/segment size.
* The value 8192, for example, covers a (sparsely populated) 1TiB system
* consisting of 128MiB memblock size, while resulting in an elfcorehdr
* memory buffer/segment size under 1MiB.
*/
#define CRASH_MAX_MEMORY_RANGES 8192
On Fri, Oct 28, 2022 at 04:22:54PM -0500, Eric DeVolder wrote:
> /*
> * For the kexec_file_load() syscall path, specify the maximum number of
> * memory regions that the elfcorehdr buffer/segment can accommodate.
> * These regions are obtained via walk_system_ram_res(); eg. the
> * 'System RAM' entries in /proc/iomem.
> * This value is combined with NR_CPUS and multiplied by sizeof(Elf64_Phdr)
NR_CPUS_DEFAULT
> * to determine the final elfcorehdr memory buffer/segment size.
> * The value 8192, for example, covers a (sparsely populated) 1TiB system
> * consisting of 128MiB memblock size, while resulting in an elfcorehdr
> * memory buffer/segment size under 1MiB.
... and it is a sane choice trying to accomodate both actual baremetal
and VM configurations."
Yeah, it's a good start.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On 10/27/22 at 02:28pm, Eric DeVolder wrote:
>
>
> On 10/27/22 08:52, Baoquan He wrote:
> > On 10/26/22 at 04:54pm, David Hildenbrand wrote:
> > > On 26.10.22 16:48, Baoquan He wrote:
> > > > On 10/25/22 at 12:31pm, Borislav Petkov wrote:
> > > > > On Thu, Oct 13, 2022 at 10:57:28AM +0800, Baoquan He wrote:
> > > > > > The concern to range number mainly is on Virt guest systems.
> > > > >
> > > > > And why would virt emulate 1K hotpluggable DIMM slots and not emulate a
> > > > > real machine?
> > >
> > > IIRC, ACPI only allows for 256 slots. PPC dlpar might provide more.
> > >
> > > >
> > > > Well, currently, mem hotpug is an important feature on virt system to
> > > > dynamically increase/shrink memory on the system. If only emulating real
> > > > machine, it won't be different than bare metal system.
> > > >
> > > > IIRC, the ballon driver or virtio-mem feature can add memory board, e.g
> > > > 1G, block size is 128M, 8 blocks added. When shrinking this 1G memory
> > > > later, it will take best effort way to hot remove memory. Means if any
> > > > memory block is occupied, it will be kept there. Finally we could only
> > > > remove every second blocks, 4 blocks altogether. Then the left
> > > > un-removed blocks will produce 4 separate memory regions. Like this, a
> > > > virt guest could have many memory regions in kernel after memory
> > > > being added/removed.
> > > >
> > > > If I am wrong, Please correct me, David.
> > >
> > > Yes, virtio-mem (but also PPC dlpar) can result in many individual memory
> > > blocks with holes in between after hotunplug. Hotplug OTOH, usually tries to
> > > "plug" these holes and reduce the total number of memory blocks. It might be
> > > rare that our range will be heavily fragmented after unplug, but it's
> > > certainly possible.
> > >
> > > [...]
> > >
> > > >
> > > > Yes, now assume we have a HPE SGI system and it has memory hotplug
> > > > capacity. The system itself has already got memory regions more than
> > > > 1024. Then when we hot add extra memory board, we want to include the
> > > > newly added memory regions into elfcorehdr so that it will be dumped out
> > > > in kdump kernel.
> > > >
> > > > That's why I earlier suggested 2048 for number of memory regions.
> > >
> > > The more the better, unless "it hurts". Assuming a single memory block is
> > > 128 MiB, that would be 256 GiB.
> > >
> > > Usually, on big systems, the memory block size is 2 GiB. So 4 TiB.
> >
> > Thanks a lot for these valuable inputs, David.
> >
> > Hi Boris, Eric
> >
> > So what's your suggested value for the Kconfig option?
> >
> > 1) cpu number, 1024?
> > 2) memory regions, 2048?
> >
> > About below draft, any comment? We can decide a value based on our
> > knowledge, can adjust later if any real system has more than the number.
> >
> > +config CRASH_ELF_CORE_PHDRS_NUM
> > + depends on CRASH_DUMP && KEXEC_FILE && (HOTPLUG_CPU || MEMORY_HOTPLUG)
> > + int
> > + default 3072
> > + help
> > + For the kexec_file_load path, specify the default number of
> > + phdr for the vmcore. E.g the memory regions represented by the
> > + 'System RAM' entries in /proc/iomem, the cpu notes of each
> > + present cpu stored in /sys/devices/system/cpu/cpuX/crash_notes.
> >
> > Thanks
> > Baoquan
> >
>
> I prefer to keep CRASH_MAX_MEMORY_RANGES, as explained in my response to your message on October 26.
> eric
Ah, sorry, I mixed it up with NR_CPUS. I went on an office outing
yesterday, glad to see you and Boris have made an agreement on the code
change and value. Thanks.
>