PS8 does not use huge pages on Ubuntu 20.04 LTS with apparmor enabled

/etc/sysctl.conf: entries for 600MB shared memoryvm.nr_hugepages=300
kernel.shmall=153600
vm.hugetlb_shm_group=998
kernel.shmmax=629145600

/etc/mysql/.../...cnf: entries for 256MB innodb buffer
innodb_buffer_pool_size = 256m
large-pages = ON
innodb_buffer_pool_instances = 1

/etc/apparmor.d/local/usr.sbin.mysqld: entries for local path and hugepages
 ...
 /{,var/}run/systemd/notify w,
 capability sys_rawio,
 capability ipc_lock,
 capability sys_resource,
 capability dac_read_search,
 /proc/*/status r,
 /etc/ssl/openssl.cnf r

2021-10-26T14:56:34.284043+02:00 0 [Warning] [MY-012678] [InnoDB] Failed to attach shared memory segment, errno 13
2021-10-26T14:56:34.284133+02:00 0 [Warning] [MY-012679] [InnoDB] Using conventional memory pool
2021-10-26T14:56:34.292204+02:00 0 [Warning] [MY-012678] [InnoDB] Failed to attach shared memory segment, errno 13
2021-10-26T14:56:34.292301+02:00 0 [Warning] [MY-012679] [InnoDB] Using conventional memory pool

kern.log:
ct 26 15:05:09 db-zzz-115-u kernel: [ 264.933778] audit: type=1400 audit(1635253509.366:21): apparmor="ALLOWED" operation="file_mmap" info="Failed name lookup - disconnected path" error=-13 profile="/usr/sbin/mysqld" name="" pid=10731 comm="mysqld" requested_mask="wr" denied_mask="wr" fsuid=996 ouid=996
Oct 26 15:05:09 db-zzz-115-u kernel: [ 264.941470] audit: type=1400 audit(1635253509.374:22): apparmor="ALLOWED" operation="file_mmap" info="Failed name lookup - disconnected path" error=-13 profile="/usr/sbin/mysqld" name="" pid=10731 comm="mysqld" requested_mask="wr" denied_mask="wr" fsuid=996 ouid=996

98 /** Allocates large pages memory.
99 @param[in,out] n Number of bytes to allocate
100 @return allocated memory */
101 void *os_mem_alloc_large(ulint *n, bool populate) {
102 void *ptr;
103 ulint size;
104 #if defined HAVE_LINUX_LARGE_PAGES && defined UNIV_LINUX
105 int shmid;
106 struct shmid_ds buf;
107
108 if (!os_use_large_pages || !os_large_page_size) {
109 goto skip;
110 }
111
112 /* Align block size to os_large_page_size */
113 ut_ad(ut_is_2pow(os_large_page_size));
114 size = ut_2pow_round(*n + (os_large_page_size - 1), os_large_page_size);
115
116 shmid = shmget(IPC_PRIVATE, (size_t)size, SHM_HUGETLB | SHM_R | SHM_W);
117 if (shmid < 0) {
118 ib::warn(ER_IB_MSG_852)
119 << "Failed to allocate " << size << " bytes. errno " << errno;
120 ptr = nullptr;
121 } else {
122 ptr = shmat(shmid, nullptr, 0);
123 if (ptr == (void *)-1) {
124 ib::warn(ER_IB_MSG_853) << "Failed to attach shared memory segment,"
125 " errno "
126 << errno;
127 ptr = nullptr;
128 }
129
130 /* Remove the shared memory segment so that it will be
131 automatically freed after memory is detached or
132 process exits */
133 shmctl(shmid, IPC_RMID, &buf);
134 }
135
136 if (ptr) {
137 *n = size;
138 os_total_large_mem_allocated.fetch_add(size);
139
140 UNIV_MEM_ALLOC(ptr, size);
141 return (ptr);
142 }
143
144 ib::warn(ER_IB_MSG_854) << "Using conventional memory pool";

so shmget in line 116 delivers a shmid >= 0 as it outputs the error in line 124
but shmat does get a -1
What I stumble upon is line 133 as is executed if a shared mem segment was allocated, perhaps this is causing the empty name in the kern.log, but that's wild guessing from my side. (and might cause problems without apparmor if I'm guessing correctly here?)

From one side I guess I might have missed an apparmor setting needed for hugepages, but which?
On the other side apparmor logs ALLOWED in kern.log.