Re: [PATCH v1] slab: support for compiler-assisted type-based slab cache partitioning
From: Marco Elver
Date: Thu Apr 02 2026 - 09:50:30 EST
On Thu, 2 Apr 2026 at 15:33, Dan Carpenter <error27@xxxxxxxxx> wrote:
>
> Hi Marco,
>
> kernel test robot noticed the following build warnings:
>
> https://git-scm.com/docs/git-format-patch#_base_tree_information]
>
> url: https://github.com/intel-lab-lkp/linux/commits/Marco-Elver/slab-support-for-compiler-assisted-type-based-slab-cache-partitioning/20260401-035608
> base: https://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm.git mm-everything
> patch link: https://lore.kernel.org/r/20260331111240.153913-1-elver%40google.com
> patch subject: [PATCH v1] slab: support for compiler-assisted type-based slab cache partitioning
> config: um-randconfig-r072-20260401
> compiler: gcc-14 (Debian 14.2.0-19) 14.2.0
> smatch: v0.5.0-9004-gb810ac53
>
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <lkp@xxxxxxxxx>
> | Reported-by: Dan Carpenter <error27@xxxxxxxxx>
> | Closes: https://lore.kernel.org/r/202604020400.jEq32K95-lkp@xxxxxxxxx/
>
> New smatch warnings:
> drivers/misc/lkdtm/heap.c:118 lkdtm_READ_AFTER_FREE() warn: potential pointer math issue ('base' is a 32 bit pointer)
> drivers/misc/lkdtm/heap.c:169 lkdtm_KFENCE_READ_AFTER_FREE() warn: potential pointer math issue ('base' is a 32 bit pointer)
How is this related to the patch I sent? Did the <linux/slab.h> change
force rechecking of all these files and it found latent issues?
> vim +118 drivers/misc/lkdtm/heap.c
>
> 73f62e60d80c2d drivers/misc/lkdtm/heap.c Kees Cook 2022-03-03 92 static void lkdtm_READ_AFTER_FREE(void)
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 93 {
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 94 int *base, *val, saw;
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 95 size_t len = 1024;
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 96 /*
> e12145cf1c3a80 drivers/misc/lkdtm/heap.c Kees Cook 2020-06-25 97 * The slub allocator will use the either the first word or
> e12145cf1c3a80 drivers/misc/lkdtm/heap.c Kees Cook 2020-06-25 98 * the middle of the allocation to store the free pointer,
> e12145cf1c3a80 drivers/misc/lkdtm/heap.c Kees Cook 2020-06-25 99 * depending on configurations. Store in the second word to
> e12145cf1c3a80 drivers/misc/lkdtm/heap.c Kees Cook 2020-06-25 100 * avoid running into the freelist.
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 101 */
> e12145cf1c3a80 drivers/misc/lkdtm/heap.c Kees Cook 2020-06-25 102 size_t offset = sizeof(*base);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 103
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 104 base = kmalloc(len, GFP_KERNEL);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 105 if (!base) {
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 106 pr_info("Unable to allocate base memory.\n");
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 107 return;
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 108 }
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 109
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 110 val = kmalloc(len, GFP_KERNEL);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 111 if (!val) {
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 112 pr_info("Unable to allocate val memory.\n");
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 113 kfree(base);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 114 return;
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 115 }
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 116
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 117 *val = 0x12345678;
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 @118 base[offset] = *val;
>
> This doesn't really matter, but the comment says we are writing to the
> second word. The base[] array holds 256 integers, so:
>
> base[sizeof(int)] = *val;
>
> would be the third word, right? A word is unsigned long, right? All
> of a sudden I am unsure. :P
???
So this is clearly a minor defect in this existing code (comment wrong
or code might want to match what the comment said), but "slab: support
for compiler-assisted type-based slab cache partitioning" didn't touch
that.
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 119 pr_info("Value in memory before free: %x\n", base[offset]);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 120
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 121 kfree(base);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 122
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 123 pr_info("Attempting bad read from freed memory\n");
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 124 saw = base[offset];
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 125 if (saw != *val) {
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 126 /* Good! Poisoning happened, so declare a win. */
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 127 pr_info("Memory correctly poisoned (%x)\n", saw);
> 5b777131bd8005 drivers/misc/lkdtm/heap.c Kees Cook 2021-06-23 128 } else {
> 5b777131bd8005 drivers/misc/lkdtm/heap.c Kees Cook 2021-06-23 129 pr_err("FAIL: Memory was not poisoned!\n");
> 5b777131bd8005 drivers/misc/lkdtm/heap.c Kees Cook 2021-06-23 130 pr_expected_config_param(CONFIG_INIT_ON_FREE_DEFAULT_ON, "init_on_free");
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 131 }
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 132
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 133 kfree(val);
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 134 }
> ffc514f3fcac4a drivers/misc/lkdtm_heap.c Kees Cook 2016-06-26 135
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 136 static void lkdtm_KFENCE_READ_AFTER_FREE(void)
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 137 {
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 138 int *base, val, saw;
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 139 unsigned long timeout, resched_after;
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 140 size_t len = 1024;
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 141 /*
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 142 * The slub allocator will use the either the first word or
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 143 * the middle of the allocation to store the free pointer,
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 144 * depending on configurations. Store in the second word to
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 145 * avoid running into the freelist.
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 146 */
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 147 size_t offset = sizeof(*base);
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 148
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 149 /*
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 150 * 100x the sample interval should be more than enough to ensure we get
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 151 * a KFENCE allocation eventually.
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 152 */
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 153 timeout = jiffies + msecs_to_jiffies(100 * kfence_sample_interval);
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 154 /*
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 155 * Especially for non-preemption kernels, ensure the allocation-gate
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 156 * timer can catch up: after @resched_after, every failed allocation
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 157 * attempt yields, to ensure the allocation-gate timer is scheduled.
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 158 */
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 159 resched_after = jiffies + msecs_to_jiffies(kfence_sample_interval);
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 160 do {
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 161 base = kmalloc(len, GFP_KERNEL);
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 162 if (!base) {
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 163 pr_err("FAIL: Unable to allocate kfence memory!\n");
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 164 return;
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 165 }
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 166
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 167 if (is_kfence_address(base)) {
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 168 val = 0x12345678;
> aabf7c37dfbce3 drivers/misc/lkdtm/heap.c Stephen Boyd 2023-11-29 @169 base[offset] = val;
> ^^^^^^^^^^^^^^^^^^
> Same here.
Unrelated to "slab: support for compiler-assisted type-based slab
cache partitioning".