Hi,
I modified eal_legacy_hugepage_init and problem is solved.
Should the correction be added to dpdk upstream?
diff --git a/package/dpdk/dpdk-20.11/lib/librte_eal/linux/eal_memory.c
b/package/dpdk/dpdk-20.11/lib/librte_eal/linux/eal_memory.c
index 03a4f2dd..89a13e91 100644
--- a/package/dpdk/dpdk-20.11/lib/librte_eal/linux/eal_memory.c
+++ b/package/dpdk/dpdk-20.11/lib/librte_eal/linux/eal_memory.c
@@ -1458,6 +1458,7 @@ eal_legacy_hugepage_init(void)
mem_sz = msl->len;
munmap(msl->base_va, mem_sz);
msl->base_va = NULL;
+ msl->len = 0;
msl->heap = 0;
/* destroy backing fbarray */
Best regards
Yan Xiaoping
From: Yan, Xiaoping (NSB - CN/Hangzhou)
Sent: 2021年8月4日 15:14
To: 'users@dpdk.org' <users@dpdk.org>
Subject: dpdk-pdump prints "EAL: Error: Invalid memory"
Hi,
After updating dpdk version from 19.11 to 20.11
dpdk-pdump prints such error:
EAL: Error: Invalid memory
Port 7 MAC: 02 70 63 61 70 03
core (2), capture for (1) tuples
- port 0 device ((null)) queue 65535
^C
--legacy-mem is used for both primary primary and dpdk-pdump.
With some debug, I find that mlx5_mem_is_rte incorrectly consider this address
from os memory ((addr=0x4482b80)) as rte address, so mlx5_free
calls rte_free() to free it and caused error.
And this seems to because len of some unused memsegs is not set to 0 (so
rte_mem_virt2memseg_list(0x4482b80) returns a memseg).
Here is memsegs:
(gdb) p mcfg->memsegs
$3 = {{{base_va = 0x2aac0000000, addr_64 = 2932388921344}, page_sz =
1073741824, socket_id = 0,
version = 0, len = 34359738368, external = 0, heap = 1, memseg_arr = {
name = "memseg-1048576k-0-0", '\000' <repeats 44 times>, count = 5, len =
32,
elt_sz = 48, data = 0x2aaa302e000, rwlock = {cnt = 0}}}, {{base_va = 0x0,
addr_64 = 0},
page_sz = 1073741824, socket_id = 0, version = 0, len = 34359738368,
external = 0,
heap = 0, memseg_arr = {name = '\000' <repeats 63 times>, count = 0, len =
0, elt_sz = 0,
data = 0x0, rwlock = {cnt = 0}}}, {{base_va = 0x0, addr_64 = 0}, page_sz
= 1073741824,
socket_id = 1, version = 0, len = 34359738368, external = 0, heap = 0,
memseg_arr = {
name = '\000' <repeats 63 times>, count = 0, len = 0, elt_sz = 0, data =
0x0, rwlock = {
cnt = 0}}}, {{base_va = 0x0, addr_64 = 0}, page_sz = 1073741824,
socket_id = 1,
version = 0, len = 34359738368, external = 0, heap = 0, memseg_arr = {
name = '\000' <repeats 63 times>, count = 0, len = 0, elt_sz = 0, data =
0x0, rwlock = {
cnt = 0}}}, {{base_va = 0x0, addr_64 = 0}, page_sz = 0, socket_id = 0,
version = 0,
len = 0, external = 0, heap = 0, memseg_arr = {name = '\000' <repeats 63
times>, count = 0,
len = 0, elt_sz = 0, data = 0x0, rwlock = {cnt = 0}}} <repeats 124 times>}
Here is the stack trace
(gdb) bt
#0 mlx5_free (addr=0x4482b80) at
../dpdk-20.11/drivers/common/mlx5/mlx5_malloc.c:260
#1 0x0000000000706f5c in mlx5_mp_req_verbs_cmd_fd
(mp_id=mp_id@entry=0x7ffcdb6d9e50)
at ../dpdk-20.11/drivers/common/mlx5/mlx5_common_mp.c:140
#2 0x000000000050496f in mlx5_dev_spawn (config=0x7ffcdb6d9d70,
spawn=0x2ab753799c0,
dpdk_dev=0x4491400) at ../dpdk-20.11/drivers/net/mlx5/linux/mlx5_os.c:774
#3 mlx5_os_pci_probe (pci_drv=<optimized out>, pci_dev=<optimized out>)
at ../dpdk-20.11/drivers/net/mlx5/linux/mlx5_os.c:2154
#4 0x0000000000708b5a in drivers_probe (user_classes=1, pci_dev=0x44913f0,
pci_drv=0xe01800 <mlx5_pci_driver>, dev=0x2ab75379a80)
at ../dpdk-20.11/drivers/common/mlx5/mlx5_common_pci.c:246
#5 mlx5_common_pci_probe (pci_drv=0xe01800 <mlx5_pci_driver>,
pci_dev=0x44913f0)
at ../dpdk-20.11/drivers/common/mlx5/mlx5_common_pci.c:308
#6 0x00000000004268f9 in rte_pci_probe_one_driver (dev=0x44913f0,
dr=0xe01800 <mlx5_pci_driver>) at
../dpdk-20.11/drivers/bus/pci/pci_common.c:243
#7 pci_probe_all_drivers (dev=0x44913f0) at
../dpdk-20.11/drivers/bus/pci/pci_common.c:318
#8 pci_probe () at ../dpdk-20.11/drivers/bus/pci/pci_common.c:345
#9 0x00000000006bc4d3 in rte_bus_probe ()
at ../dpdk-20.11/lib/librte_eal/common/eal_common_bus.c:72
#10 0x0000000000422304 in rte_eal_init (argc=argc@entry=16,
argv=argv@entry=0x7ffcdb6da530)
at ../dpdk-20.11/lib/librte_eal/linux/eal.c:1210
#11 0x000000000056fee9 in main (argc=16, argv=0x7ffcdb6da748)
at ../dpdk-20.11/app/pdump/main.c:1118
(gdb) c
Continuing.
Thread 1 "dpdk-pdump" hit Breakpoint 5, rte_free (addr=0x4482b80)
It seems to me that below code piece from eal_legacy_hugepage_init should also
set len to 0?
/* we're not going to allocate more pages, so release VA space for
* unused memseg lists
*/
for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
struct rte_memseg_list *msl = &mcfg->memsegs[i];
size_t mem_sz;
/* skip inactive lists */
if (msl->base_va == NULL)
continue;
/* skip lists where there is at least one page
allocated */
if (msl->memseg_arr.count > 0)
continue;
/* this is an unused list, deallocate it */
mem_sz = msl->len;
munmap(msl->base_va, mem_sz);
msl->base_va = NULL;
// here, we should add msl->len = 0; ?
msl->heap = 0;
/* destroy backing fbarray */
rte_fbarray_destroy(&msl->memseg_arr);
}
Any comment?
Thank you.
Best regards
Yan Xiaoping
