ring-buffer: Do not check si_mem_available() before deferred init completes
Subsystems (4)
PATCH v5 1/1 ring-buffer: Do not check si_mem_available() before deferred init completes
With CONFIG_DEFERRED_STRUCT_PAGE_INIT=y, only about one memory section per node is initialized early in boot (16 MiB on powerpc). The rest is initialized on demand by deferred_grow_zone() when an allocation would otherwise fail, or later by page_alloc_init_late().
Until then NR_FREE_PAGES can be close to zero, because early allocations (SLUB, vmalloc, ftrace records) use up the first section. si_mem_available() is derived from NR_FREE_PAGES, so it under-reports.
__rb_allocate_pages() returns -ENOMEM when si_mem_available() is below the requested size. This check wrongly fails the buffers allocated for the boot CPU in early_trace_init() and for secondary CPUs in smp_init(), both of which run before page_alloc_init_late(). The resulting boot log shows:
ERROR: Failed to allocate trace buffer
ERROR: tracer: failed to allocate ring buffer!
Skip the check until SYSTEM_FREEING_INITMEM on CONFIG_DEFERRED_STRUCT_PAGE_INIT kernels. The allocation itself still fails if memory really is short.
This is the second half of the fix for the early boot crash reported on ppc64le KVM guests. Commit b3b974008388 ("tracing: Do not initialize tracefs work if tracing is disabled") stopped the NULL pointer dereference in __find_event_file() that followed the failed allocation. This patch fixes the allocation failure itself.
---
Changes since v4:
- v4: https://lore.kernel.org/all/[email protected]/
- Rewrote commit message for clarity and conciseness (Vincent, Steven).
- Structure the deferred init check using if/else with unlikely() (Steven).
- Added Vincent's R-b.
Changes since v3:
- v3: https://lore.kernel.org/all/[email protected]
- Replace #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT / #else / #endif with IS_ENABLED() + i = LONG_MAX to avoid duplicating the threshold check. (Vincent Donnefort)
- Trim the in-code comment to one sentence. (Vincent Donnefort)
Changes since v2:
- v2: https://lore.kernel.org/all/[email protected]
- Rebase on top of upstream ring_buffer.c changes from Vincent Donnefort; update threshold to i < (nr_pages << subbuf_order). (Michal Suchánek)
Changes since v1:
- v1: https://lore.kernel.org/all/[email protected]
- Extend the guard to cover SYSTEM_SCHEDULING as well: smp_init() calls trace_rb_cpu_prepare() -> __rb_allocate_pages() before page_alloc_init_late() completes, so the false negative window spans both SYSTEM_BOOTING and SYSTEM_SCHEDULING. Use system_state >= SYSTEM_FREEING_INITMEM as the correct boundary. (Sashiko, Vincent Donnefort)
- Wrap in #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT to preserve the original check unchanged on non-deferred kernels. (Vincent Donnefort)
- Drop Tested-by: guard condition changed, requesting fresh test.
kernel/trace/ring_buffer.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
kernel/trace/ring_buffer.c +10 -1
@@ -2468,8 +2468,17 @@ static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,* memory. It may not be accurate. But we don't care, we just want * to prevent doing any allocation when it is obvious that it is * not going to succeed. +* +* With CONFIG_DEFERRED_STRUCT_PAGE_INIT, si_mem_available() +* will reflect the actual available memory only after +* SYSTEM_FREEING_INITMEM. */ - i =si_mem_available (); +if (IS_ENABLED (CONFIG_DEFERRED_STRUCT_PAGE_INIT ) && +unlikely (system_state <SYSTEM_FREEING_INITMEM )) + i =LONG_MAX ; +else + i =si_mem_available (); +if (i < (nr_pages << subbuf_order))return -ENOMEM ;
base-commit: 20eef29b7c8465ac98d7dd6d6f2ac07a8d8056ca
Signature · 1 line
2.54.0 (Apple Git-157)