module: avoid userspace pressure on unwanted allocations
Subsystems (3)
Luis Chamberlain
maintainerv3 cover module: avoid userspace pressure on unwanted allocations
This v3 series follows up on the second iteration of these patches [0]. This and other pending changes are avaiable on 20230413-module-alloc-opts branch [1] which is based on modules-next.
Changes on this v3:
o Catalin Marinas suggested we just use kmemleak_not_leak() for both ELF allocations even if its init stuff.
o Considerable amount of effort went into trying to see if there's relationship with CPU count and wasted virtual memory allocations. The new module debugfs counters helped with creating this evaluation. The result of that put me on a path to then add even more debugging facilities to rule out and identify the culprits. In the end I have patches now which can get this down to 0 bytes wasted. The patch in this series which helps reduce the allocations has a graph showing the findings of the relationship between wasted virtual memory allocations and CPU count all during boot. It is insanity that the graph has to go into gigabytes of wasted virtual memory all at boot.
o To help folks compare apples to apples I've put the stats debug patch *prior* to the one that helps with allocations. This way folks can see for themselves what the results look like.
o Enhanced the statistics a bit more and added an example with 255 CPUs.
o Went with atomic_long and casting for the debugs big counters.
o Rolled in the patch that moved a helper as David suggested.
o Minor fixes reported by 0-day
o Added tags for Reviews, etc.
[0] https://lkml.kernel.org/r/[email protected]
[1] https://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/linux.git/log/?h=20230413-module-alloc-opts
Luis Chamberlain (4):
module: fix kmemleak annotations for non init ELF sections
module: extract patient module check into helper
module: add debug stats to help identify memory pressure
module: avoid allocation if module is already present and ready
Documentation/core-api/kernel-api.rst | 22 +- kernel/module/Kconfig | 37 +++ kernel/module/Makefile | 1 + kernel/module/decompress.c | 4 + kernel/module/internal.h | 74 +++++ kernel/module/main.c | 194 ++++++++---- kernel/module/stats.c | 432 ++++++++++++++++++++++++++ kernel/module/tracking.c | 7 +- 8 files changed, 703 insertions(+), 68 deletions(-) create mode 100644 kernel/module/stats.c
Signature · 1 line
2.39.2
Luis Chamberlain
maintainerPATCH v3 1/4 module: fix kmemleak annotations for non init ELF sections
Commit ac3b43283923 ("module: replace module_layout with module_memory") reworked the way to handle memory allocations to make it clearer. But it lost in translation how we handled kmemleak_ignore() or kmemleak_not_leak() for different ELF sections.
Fix this and clarify the comments a bit more. Contrary to the old way of using kmemleak_ignore() for init.* ELF sections we stick now only to kmemleak_not_leak() as per suggestion by Catalin Marinas so to avoid any false positives and simplify the code.
--- kernel/module/main.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
kernel/module/main.c +10 -5
@@ -2231,13 +2231,18 @@ static int move_module(struct module *mod, struct load_info *info) } mod->mem[type].size =PAGE_ALIGN (mod->mem[type].size); ptr =module_memory_alloc (mod->mem[type].size, type); -/* -* The pointer to this block is stored in the module structure -* which is inside the block. Just mark it as not being a -* leak. +* The pointer to these blocks of memory are stored on the module +* structure and we keep that around so long as the module is +* around. We only free that memory when we unload the module. +* Just mark them as not being a leak then. The .init* ELF +* sections *do* get freed after boot so we *could* treat them +* slightly differently with kmemleak_ignore() and only grey +* them out as they work as typical memory allocations which +* *do* eventually get freed, but let's just keep things simple +* and avoid *any* false positives. */ -kmemleak_ignore (ptr); +kmemleak_not_leak (ptr);if (!ptr) { t = type; goto out_enomem;
Signature · 1 line
2.39.2
Luis Chamberlain
maintainerPATCH v3 2/4 module: extract patient module check into helper
In reply to Luis ChamberlainThe patient module check inside add_unformed_module() is large enough as we need it. It is a bit hard to read too, so just move it to a helper and do the inverse checks first to help shift the code and make it easier to read. The new helper then is module_patient_check_exists().
To make this work we need to mvoe the finished_loading() up, we do that without making any functional changes to that routine.
--- kernel/module/main.c | 112 +++++++++++++++++++++++-------------------- 1 file changed, 60 insertions(+), 52 deletions(-)
kernel/module/main.c +60 -52
@@ -2447,27 +2447,6 @@ static int post_relocation(struct module *mod, const struct load_info *info)return module_finalize (info->hdr, info->sechdrs, mod); } -/* Is this module of this name done loading? No locks held. */ -static bool finished_loading (const char *name) -{ -struct module *mod; -bool ret; - -/* -* The module_mutex should not be a heavily contended lock; -* if we get the occasional sleep here, we'll go an extra iteration -* in the wait_event_interruptible(), which is harmless. -*/ -sched_annotate_sleep (); -mutex_lock (&module_mutex); - mod =find_module_all (name,strlen (name), true); - ret = !mod || mod->state ==MODULE_STATE_LIVE - || mod->state ==MODULE_STATE_GOING ; -mutex_unlock (&module_mutex); - -return ret; -} -/* Call module constructors. */ static void do_mod_ctors (struct module *mod) {
@@ -2631,6 +2610,63 @@ static int may_init_module(void)return 0 ; } +/* Is this module of this name done loading? No locks held. */ +static bool finished_loading (const char *name) +{ +struct module *mod; +bool ret; + +/* +* The module_mutex should not be a heavily contended lock; +* if we get the occasional sleep here, we'll go an extra iteration +* in the wait_event_interruptible(), which is harmless. +*/ +sched_annotate_sleep (); +mutex_lock (&module_mutex); + mod =find_module_all (name,strlen (name), true); + ret = !mod || mod->state ==MODULE_STATE_LIVE + || mod->state ==MODULE_STATE_GOING ; +mutex_unlock (&module_mutex); + +return ret; +} + +/* Must be called with module_mutex held */ +static int module_patient_check_exists (const char *name) +{ +struct module *old; +int err =0 ; + + old =find_module_all (name,strlen (name), true); +if (old ==NULL ) +return 0 ; + +if (old->state ==MODULE_STATE_COMING || + old->state ==MODULE_STATE_UNFORMED ) { +/* Wait in case it fails to load. */ +mutex_unlock (&module_mutex); + err =wait_event_interruptible (module_wq, +finished_loading (name)); +mutex_lock (&module_mutex); +if (err) +return err; + +/* The module might have gone in the meantime. */ + old =find_module_all (name,strlen (name), true); + } + +/* +* We are here only when the same module was being loaded. Do +* not try to load it again right now. It prevents long delays +* caused by serialized module load failures. It might happen +* when more devices of the same type trigger load of +* a particular module. +*/ +if (old && old->state ==MODULE_STATE_LIVE ) +return -EEXIST ; +return -EBUSY ; +} +/* * We try to place it in the list now to make sure it's unique before * we dedicate too many resources. In particular, temporary percpu
@@ -2639,41 +2675,14 @@ static int may_init_module(void)static int add_unformed_module (struct module *mod) {int err; -struct module *old; mod->state =MODULE_STATE_UNFORMED ;mutex_lock (&module_mutex); - old =find_module_all (mod->name,strlen (mod->name), true); -if (old !=NULL ) { -if (old->state ==MODULE_STATE_COMING - || old->state ==MODULE_STATE_UNFORMED ) { -/* Wait in case it fails to load. */ -mutex_unlock (&module_mutex); - err =wait_event_interruptible (module_wq, -finished_loading (mod->name)); -if (err) - goto out_unlocked; - -/* The module might have gone in the meantime. */ -mutex_lock (&module_mutex); - old =find_module_all (mod->name,strlen (mod->name), - true); - } - -/* -* We are here only when the same module was being loaded. Do -* not try to load it again right now. It prevents long delays -* caused by serialized module load failures. It might happen -* when more devices of the same type trigger load of -* a particular module. -*/ -if (old && old->state ==MODULE_STATE_LIVE ) - err = -EEXIST ; -else - err = -EBUSY ; + err =module_patient_check_exists (mod->name); +if (err) goto out; - } +mod_update_bounds (mod);list_add_rcu (&mod->list, &modules);mod_tree_insert (mod);
@@ -2681,7 +2690,6 @@ static int add_unformed_module(struct module *mod) out:mutex_unlock (&module_mutex); -out_unlocked:return err; }
Signature · 1 line
2.39.2
Luis Chamberlain
maintainerPATCH v3 4/4 module: avoid allocation if module is already present and ready
In reply to Luis ChamberlainThe finit_module() system call can create unnecessary virtual memory pressure for duplicate modules. This is because load_module() can in the worse case allocate more than twice the size of a module in virtual memory. This saves at least a full size of the module in wasted vmalloc space memory by trying to avoid duplicates as soon as we can validate the module name in the read module structure.
This can only be an issue if a system is getting hammered with userspace loading modules. There are two ways to load modules typically on systems, one is the kernel moduile auto-loading (*request_module*() calls in-kernel) and the other is things like udev. The auto-loading is in-kernel, but that pings back to userspace to just call modprobe. We already have a way to restrict the amount of concurrent kernel auto-loads in a given time, however that still allows multiple requests for the same module to go through and force two threads in userspace racing to call modprobe for the same exact module. Even though libkmod which both modprobe and udev does check if a module is already loaded prior calling finit_module() races are still possible and this is clearly evident today when you have multiple CPUs.
To avoid memory pressure for such stupid cases put a stop gap for them. The *earliest* we can detect duplicates from the modules side of things is once we have blessed the module name, sadly after the first vmalloc allocation. We can check for the module being present *before* a secondary vmalloc() allocation.
There is a linear relationship between wasted virtual memory bytes and the number of CPU counts. The reason is that udev ends up racing to call tons of the same modules for each of the CPUs.
We can see the different linear relationships between wasted virtual memory and CPU count during after boot in the following graph:
+----------------------------------------------------------------------------+
14GB |-+ + + + + *+ +-|
| **** |
| *** |
| ** |
12GB |-+ ** +-|
| ** |
| ** |
| ** |
| ** |
10GB |-+ ** +-|
| ** |
| ** |
| ** |
8GB |-+ ** +-|
waste | ** ### |
| ** #### |
| ** ####### |
6GB |-+ **** #### +-|
| * #### |
| * #### |
| ***** #### |
4GB |-+ ** #### +-|
| ** #### |
| ** #### |
| ** #### |
2GB |-+ ** ##### +-|
| * #### |
| * #### Before ******* |
| **## + + + + After ####### |
+----------------------------------------------------------------------------+
0 50 100 150 200 250 300CPUs count
On the y-axis we can see gigabytes of wasted virtual memory during boot due to duplicate module requests which just end up failing. Trying to infer the slope this ends up being about ~463 MiB per CPU lost prior to this patch. After this patch we only loose about ~230 MiB per CPU, for a total savings of about ~233 MiB per CPU. This is all *just on bootup*!
On a 8vcpu 8 GiB RAM system using kdevops and testing against selftests kmod.sh -t 0008 I see a saving in the *highest* side of memory consumption of up to ~ 84 MiB with the Linux kernel selftests kmod test 0008. With the new stress-ng module test I see a 145 MiB difference in max memory consumption with 100 ops. The stress-ng module ops tests can be pretty pathalogical -- it is not realistic, however it was used to finally successfully reproduce issues which are only reported to happen on system with over 400 CPUs [0] by just usign 100 ops on a 8vcpu 8 GiB RAM system. Running out of virtual memory space is no surprise given the above graph, since at least on x86_64 we're capped at 128 MiB, eventually we'd hit a series of errors and once can use the above graph to guestimate when. This of course will vary depending on the features you have enabled. So for instance, enabling KASAN seems to make this much worse.
The results with kmod and stress-ng can be observed and visualized below. The time it takes to run the test is also not affected.
The kmod tests 0008:
The gnuplot is set to a range from 400000 KiB (390 Mib) - 580000 (566 Mib) given the tests peak around that range.
cat kmod.plot
set term dumb
set output fileout
set yrange [400000:580000]
plot filein with linespoints title "Memory usage (KiB)"
Before:
root@kmod ~ # /data/linux-next/tools/testing/selftests/kmod/kmod.sh -t 0008 root@kmod ~ # free -k -s 1 -c 40 | grep Mem | awk '{print $3}' > log-0008-before.txt ^C root@kmod ~ # sort -n -r log-0008-before.txt | head -1
528732
So ~516.33 MiB
After:
root@kmod ~ # /data/linux-next/tools/testing/selftests/kmod/kmod.sh -t 0008 root@kmod ~ # free -k -s 1 -c 40 | grep Mem | awk '{print $3}' > log-0008-after.txt ^C
root@kmod ~ # sort -n -r log-0008-after.txt | head -1 442516
So ~432.14 MiB
That's about 84 ~MiB in savings in the worst case. The graphs:
root@kmod ~ # gnuplot -e "filein='log-0008-before.txt'; fileout='graph-0008-before.txt'" kmod.plot root@kmod ~ # gnuplot -e "filein='log-0008-after.txt'; fileout='graph-0008-after.txt'" kmod.plot
root@kmod ~ # cat graph-0008-before.txt
580000 +-----------------------------------------------------------------+
| + + + + + + + |
560000 |-+ Memory usage (KiB) ***A***-|
| |
540000 |-+ +-|
| |
| *A *AA*AA*A*AA *A*AA A*A*A *AA*A*AA*A A |
520000 |-+A*A*AA *AA*A *A*AA*A*AA *A*A A *A+-|
|*A |
500000 |-+ +-|
| |
480000 |-+ +-|
| |
460000 |-+ +-|
| |
| |
440000 |-+ +-|
| |
420000 |-+ +-|
| + + + + + + + | 400000 +-----------------------------------------------------------------+
0 5 10 15 20 25 30 35 40
root@kmod ~ # cat graph-0008-after.txt
580000 +-----------------------------------------------------------------+
| + + + + + + + |
560000 |-+ Memory usage (KiB) ***A***-|
| |
540000 |-+ +-|
| |
| |
520000 |-+ +-|
| |
500000 |-+ +-|
| |
480000 |-+ +-|
| |
460000 |-+ +-|
| |
| *A *A*A |
440000 |-+A*A*AA*A A A*A*AA A*A*AA*A*AA*A*AA*A*AA*AA*A*AA*A*AA-|
|*A *A*AA*A |
420000 |-+ +-|
| + + + + + + + | 400000 +-----------------------------------------------------------------+
0 5 10 15 20 25 30 35 40
The stress-ng module tests:
This is used to run the test to try to reproduce the vmap issues reported by David:
echo 0 > /proc/sys/vm/oom_dump_tasks ./stress-ng --module 100 --module-name xfs
Prior to this commit:
root@kmod ~ # free -k -s 1 -c 40 | grep Mem | awk '{print $3}' > baseline-stress-ng.txt root@kmod ~ # sort -n -r baseline-stress-ng.txt | head -1
5046456
After this commit:
root@kmod ~ # free -k -s 1 -c 40 | grep Mem | awk '{print $3}' > after-stress-ng.txt root@kmod ~ # sort -n -r after-stress-ng.txt | head -1
4896972
5046456 - 4896972
149484
149484/1024
145.98046875000000000000
So this commit using stress-ng reveals saving about 145 MiB in memory using 100 ops from stress-ng which reproduced the vmap issue reported.
cat kmod.plot
set term dumb
set output fileout
set yrange [4700000:5070000]
plot filein with linespoints title "Memory usage (KiB)"
root@kmod ~ # gnuplot -e "filein='baseline-stress-ng.txt'; fileout='graph-stress-ng-before.txt'" kmod-simple-stress-ng.plot root@kmod ~ # gnuplot -e "filein='after-stress-ng.txt'; fileout='graph-stress-ng-after.txt'" kmod-simple-stress-ng.plot
root@kmod ~ # cat graph-stress-ng-before.txt
+---------------------------------------------------------------+
5.05e+06 |-+ + A + + + + + + +-|
| * Memory usage (KiB) ***A*** |
| * A |
5e+06 |-+ ** ** +-|
| ** * * A |
4.95e+06 |-+ * * A * A* +-|
| * * A A * * * * A |
| * * * * * * *A * * * A * |
4.9e+06 |-+ * * * A*A * A*AA*A A *A **A **A*A *+-|
| A A*A A * A * * A A * A * ** |
| * ** ** * * * * * * * |
4.85e+06 |-+ A A A ** * * ** *-|
| * * * * ** * |
| * A * * * * |
4.8e+06 |-+ * * * A A-|
| * * * |
4.75e+06 |-+ * * * +-|
| * ** |
| * + + + + + + ** + | 4.7e+06 +---------------------------------------------------------------+
0 5 10 15 20 25 30 35 40
root@kmod ~ # cat graph-stress-ng-after.txt
+---------------------------------------------------------------+
5.05e+06 |-+ + + + + + + + +-|
| Memory usage (KiB) ***A*** |
| |
5e+06 |-+ +-|
| |
4.95e+06 |-+ +-|
| |
| |
4.9e+06 |-+ *AA +-|
| A*AA*A*A A A*AA*AA*A*AA*A A A A*A *AA*A*A A A*AA*AA |
| * * ** * * * ** * *** * |
4.85e+06 |-+* *** * * * * *** A * * +-|
| * A * * ** * * A * * |
| * * * * ** * * |
4.8e+06 |-+* * * A * * * +-|
| * * * A * * |
4.75e+06 |-* * * * * +-|
| * * * * * |
| * + * *+ + + + + * *+ | 4.7e+06 +---------------------------------------------------------------+
0 5 10 15 20 25 30 35 40
[0] https://lkml.kernel.org/r/[email protected]
--- kernel/module/main.c | 6 +++++- kernel/module/stats.c | 14 +++++++------- 2 files changed, 12 insertions(+), 8 deletions(-)
kernel/module/main.c +5 -1
@@ -2815,7 +2815,11 @@ static int early_mod_check(struct load_info *info, int flags)if (err)return err; -return 0 ; +mutex_lock (&module_mutex); + err =module_patient_check_exists (info->mod->name,FAIL_DUP_MOD_BECOMING ); +mutex_unlock (&module_mutex); + +return err; }/*
kernel/module/stats.c +7 -7
@@ -87,7 +87,7 @@ extern struct dentry *mod_debugfs_root;* calls: * * a) FAIL_DUP_MOD_BECOMING: at the end of early_mod_check() before -* layout_and_allocate(). This does not yet happen. +* layout_and_allocate(). * - with module decompression: 2 virtual memory allocation calls * - without module decompression: 1 virtual memory allocation calls * b) FAIL_DUP_MOD_LOAD: after layout_and_allocate() on add_unformed_module()
@@ -130,15 +130,15 @@ static LIST_HEAD(dup_failed_modules);* * invalid_becoming_bytes: total number of bytes wasted due to * allocations used to read the kernel module userspace wants us to read * before we promote it to be processed to be added to our @modules linked -* list. These failures could in theory happen in if we had a check in between -* between a successful kernel_read_file_from_fd() call and right before -* we allocate the our private memory for the module which would be kept if -* the module is successfully loaded. The most common reason for this failure +* list. These failures could can happen in between a successful +* kernel_read_file_from_fd() call and right before we allocate the our +* private memory for the module which would be kept if the module is +* successfully loaded. The most common reason for this failure * is when userspace is racing to load a module which it does not yet see * loaded. The first module to succeed in add_unformed_module() will add a * module to our &modules list and subsequent loads of modules with the -* same name will error out at the end of early_mod_check(). A check -* for module_patient_check_exists() at the end of early_mod_check() could be +* same name will error out at the end of early_mod_check(). The check +* for module_patient_check_exists() at the end of early_mod_check() was * added to prevent duplicate allocations on layout_and_allocate() for * modules already being processed. These duplicate failed modules are * non-fatal, however they typically are indicative of userspace not seeing
Signature · 1 line
2.39.2
Luis Chamberlain
maintainerPATCH v3 3/4 module: add debug stats to help identify memory pressure
In reply to Luis ChamberlainLoading modules with finit_module() can end up using vmalloc(), vmap() and vmalloc() again, for a total of up to 3 separate allocations in the worst case for a single module. We always kernel_read*() the module, that's a vmalloc(). Then vmap() is used for the module decompression, and if so the last read buffer is freed as we use the now decompressed module buffer to stuff data into our copy module. The last allocation is specific to each architectures but pretty much that's generally a series of vmalloc() calls or a variation of vmalloc to handle ELF sections with special permissions.
Evaluation with new stress-ng module support [1] with just 100 ops is proving that you can end up using GiBs of data easily even with all care we have in the kernel and userspace today in trying to not load modules which are already loaded. 100 ops seems to resemble the sort of pressure a system with about 400 CPUs can create on module loading. Although issues relating to duplicate module requests due to each CPU inucurring a new module reuest is silly and some of these are being fixed, we currently lack proper tooling to help diagnose easily what happened, when it happened and who likely is to blame -- userspace or kernel module autoloading.
Provide an initial set of stats which use debugfs to let us easily scrape post-boot information about failed loads. This sort of information can be used on production worklaods to try to optimize *avoiding* redundant memory pressure using finit_module().
There's a few examples that can be provided:
A 255 vCPU system without the next patch in this series applied:
Startup finished in 19.143s (kernel) + 7.078s (userspace) = 26.221s graphical.target reached after 6.988s in userspace
And 13.58 GiB of virtual memory space lost due to failed module loading:
root@big ~ # cat /sys/kernel/debug/modules/stats
Mods ever loaded 67
Mods failed on kread 0
Mods failed on decompress 0
Mods failed on becoming 0
Mods failed on load 1411
Total module size 11464704
Total mod text size 4194304
Failed kread bytes 0
Failed decompress bytes 0
Failed becoming bytes 0
Failed kmod bytes 14588526272
Virtual mem wasted bytes 14588526272
Average mod size 171115
Average mod text size 62602
Average fail load bytes 10339140Duplicate failed modules:
module-name How-many-times Reason
kvm_intel 249 Load
kvm 249 Load
irqbypass 8 Load
crct10dif_pclmul 128 Load
ghash_clmulni_intel 27 Load
sha512_ssse3 50 Load
sha512_generic 200 Load
aesni_intel 249 Load
crypto_simd 41 Load
cryptd 131 Load
evdev 2 Load
serio_raw 1 Load
virtio_pci 3 Load
nvme 3 Load
nvme_core 3 Load
virtio_pci_legacy_dev 3 Load
virtio_pci_modern_dev 3 Load
t10_pi 3 Load
virtio 3 Load
crc32_pclmul 6 Load
crc64_rocksoft 3 Load
crc32c_intel 40 Load
virtio_ring 3 Load
crc64 3 Load
The following screen shot, of a simple 8vcpu 8 GiB KVM guest with the next patch in this series applied, shows 226.53 MiB are wasted in virtual memory allocations which due to duplicate module requests during boot. It also shows an average module memory size of 167.10 KiB and an an average module .text + .init.text size of 61.13 KiB. The end shows all modules which were detected as duplicate requests and whether or not they failed early after just the first kernel_read*() call or late after we've already allocated the private space for the module in layout_and_allocate(). A system with module decompression would reveal more wasted virtual memory space.
We should put effort now into identifying the source of these duplicate module requests and trimming these down as much possible. Larger systems will obviously show much more wasted virtual memory allocations.
root@kmod ~ # cat /sys/kernel/debug/modules/stats
Mods ever loaded 67
Mods failed on kread 0
Mods failed on decompress 0
Mods failed on becoming 83
Mods failed on load 16
Total module size 11464704
Total mod text size 4194304
Failed kread bytes 0
Failed decompress bytes 0
Failed becoming bytes 228959096
Failed kmod bytes 8578080
Virtual mem wasted bytes 237537176
Average mod size 171115
Average mod text size 62602
Avg fail becoming bytes 2758544
Average fail load bytes 536130Duplicate failed modules:
module-name How-many-times Reason
kvm_intel 7 Becoming
kvm 7 Becoming
irqbypass 6 Becoming & Load
crct10dif_pclmul 7 Becoming & Load
ghash_clmulni_intel 7 Becoming & Load
sha512_ssse3 6 Becoming & Load
sha512_generic 7 Becoming & Load
aesni_intel 7 Becoming
crypto_simd 7 Becoming & Load
cryptd 3 Becoming & Load
evdev 1 Becoming
serio_raw 1 Becoming
nvme 3 Becoming
nvme_core 3 Becoming
t10_pi 3 Becoming
virtio_pci 3 Becoming
crc32_pclmul 6 Becoming & Load
crc64_rocksoft 3 Becoming
crc32c_intel 3 Becoming
virtio_pci_modern_dev 2 Becoming
virtio_pci_legacy_dev 1 Becoming
crc64 2 Becoming
virtio 2 Becoming
virtio_ring 2 Becoming
[0] https://github.com/ColinIanKing/stress-ng.git
[1] echo 0 > /proc/sys/vm/oom_dump_tasks
./stress-ng --module 100 --module-name xfs
--- Documentation/core-api/kernel-api.rst | 22 +- kernel/module/Kconfig | 37 +++ kernel/module/Makefile | 1 + kernel/module/decompress.c | 4 + kernel/module/internal.h | 74 +++++ kernel/module/main.c | 65 +++- kernel/module/stats.c | 432 ++++++++++++++++++++++++++ kernel/module/tracking.c | 7 +- 8 files changed, 630 insertions(+), 12 deletions(-) create mode 100644 kernel/module/stats.c
Documentation/core-api/kernel-api.rst +20 -2
@@ -220,12 +220,30 @@ relay interface Module Support ============== -Module Loading --------------- +Kernel module auto-loading +-------------------------- .. kernel-doc:: kernel/module/kmod.c :export: +Module debugging +---------------- + +.. kernel-doc:: kernel/module/stats.c + :doc: module debugging statistics overview + +dup_failed_modules - tracks duplicate failed modules +**************************************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: dup_failed_modules - tracks duplicate failed modules + +module statistics debugfs counters +********************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: module statistics debugfs counters + Inter Module support --------------------
kernel/module/Kconfig +37 -0
@@ -22,6 +22,43 @@ menuconfig MODULES if MODULES +config MODULE_DEBUG + bool "Module debugging" + depends on DEBUG_FS + help + Allows you to enable / disable features which can help you debug + modules. You don't need these options in produciton systems. You can + and probably should enable this prior to making your kernel + produciton ready though. + +if MODULE_DEBUG + +config MODULE_STATS + bool "Module statistics" + depends on DEBUG_FS + help + This option allows you to maintain a record of module statistics. + For example each all modules size, average size, text size, and + failed modules and the size for each of those. For failed + modules we keep track of module which failed due to either the + existing module taking too long to load or that module already + was loaded. + + You should enable this if you are debugging production loads + and want to see if userspace or the kernel is doing stupid things + with loading modules when it shouldn't or if you want to help + optimize userspace / kernel space module autoloading schemes. + You might want to do this because failed modules tend to use + use up significan amount of memory, and so you'd be doing everyone + a favor in avoiding these failure proactively. + + This functionality is also useful for those experimenting with + module .text ELF section optimization. + + If unsure, say N. + +endif # MODULE_DEBUG + config MODULE_FORCE_LOAD bool "Forced module loading" default n
kernel/module/Makefile +1 -0
@@ -21,3 +21,4 @@ obj-$(CONFIG_SYSFS) += sysfs.o obj-$(CONFIG_KGDB_KDB) += kdb.o obj-$(CONFIG_MODVERSIONS) += version.o obj-$(CONFIG_MODULE_UNLOAD_TAINT_TRACKING) += tracking.o +obj-$(CONFIG_MODULE_STATS) += stats.o
kernel/module/decompress.c +4 -0
@@ -297,6 +297,10 @@ int module_decompress(struct load_info *info, const void *buf, size_t size)ssize_t data_size;int error; +#if defined(CONFIG_MODULE_STATS ) + info->compressed_len = size; +#endif +/* * Start with number of pages twice as big as needed for * compressed data.
kernel/module/internal.h +74 -0
86 lines not on this page · raw message
kernel/module/main.c +60 -5
165 lines not on this page · raw message
kernel/module/stats.c new file +432 -0
432 lines not on this page · raw message
kernel/module/tracking.c +2 -5
@@ -15,6 +15,7 @@#include "internal.h" static LIST_HEAD (unloaded_tainted_modules); +extern struct dentry *mod_debugfs_root;int try_add_tainted_module (struct module *mod) {
@@ -120,12 +121,8 @@ static const struct file_operations unloaded_tainted_modules_fops = {static int __init unloaded_tainted_modules_init (void ) { -struct dentry *dir; - - dir =debugfs_create_dir ("modules" ,NULL ); -debugfs_create_file ("unloaded_tainted" ,0444 , dir,NULL , +debugfs_create_file ("unloaded_tainted" ,0444 , mod_debugfs_root,NULL , &unloaded_tainted_modules_fops); -return 0 ; }module_init (unloaded_tainted_modules_init);
Signature · 1 line
2.39.2
Re: [PATCH v3 1/4] module: fix kmemleak annotations for non init ELF sections
In reply to Luis ChamberlainQuoted 10 lines · Luis Chamberlain
Commit
ac3b43283923("module: replacemodule_layoutwithmodule_memory") reworked the way to handle memory allocations to make it clearer. But it lost in translation how we handledkmemleak_ignore()orkmemleak_not_leak()for different ELF sections.Fix this and clarify the comments a bit more. Contrary to the old way of using
kmemleak_ignore()for init.* ELF sections we stick now only tokmemleak_not_leak()as per suggestion by Catalin Marinas so to avoid any false positives and simplify the code.
Quoted 5 lines
Fixes ac3b43283923 ("module: replace module_layout with module_memory")Reported-by Jim Cromie <[email protected]>Acked-by Song Liu <[email protected]>Suggested-by Catalin Marinas <[email protected]>Signed-off-by Luis Chamberlain <[email protected]>
Petr Pavlu
maintainerRe: [PATCH v3 3/4] module: add debug stats to help identify memory pressure
In reply to Luis ChamberlainQuoted 20 lines · Luis Chamberlain
Loading modules with
finit_module()can end up usingvmalloc(),vmap()andvmalloc()again, for a total of up to 3 separate allocations in the worst case for a single module. We always kernel_read*() the module, that's avmalloc(). Thenvmap()is used for the module decompression, and if so the last read buffer is freed as we use the now decompressed module buffer to stuff data into our copy module. The last allocation is specific to each architectures but pretty much that's generally a series ofvmalloc()calls or a variation of vmalloc to handle ELF sections with special permissions.Evaluation with new stress-ng module support [1] with just 100 ops is proving that you can end up using GiBs of data easily even with all care we have in the kernel and userspace today in trying to not load modules which are already loaded. 100 ops seems to resemble the sort of pressure a system with about 400 CPUs can create on module loading. Although issues relating to duplicate module requests due to each CPU inucurring a new module reuest is silly and some of these are being fixed, we currently lack proper tooling to help diagnose easily what happened, when it happened and who likely is to blame -- userspace or kernel module autoloading.
Quoted 4 lines
Provide an initial set of stats which use debugfs to let us easily scrape post-boot information about failed loads. This sort of information can be used on production worklaods to try to optimize *avoiding* redundant memory pressure using
finit_module().
This looks useful, thanks for it. Some comments below.
Quoted 172 lines
There's a few examples that can be provided:
A 255 vCPU system without the next patch in this series applied:
Startup finished in 19.143s (kernel) + 7.078s (userspace) = 26.221s graphical.target reached after 6.988s in userspace
And 13.58 GiB of virtual memory space lost due to failed module loading:
root@big ~ # cat
/sys/kernel/debug/modules/statsMods ever loaded 67 Mods failed on kread 0 Mods failed on decompress 0 Mods failed on becoming 0 Mods failed on load 1411 Total module size 11464704 Total mod text size 4194304 Failed kread bytes 0 Failed decompress bytes 0 Failed becoming bytes 0 Failed kmod bytes 14588526272 Virtual mem wasted bytes 14588526272 Average mod size 171115 Average mod text size 62602 Average fail load bytes 10339140Duplicate failed modules:
module-name How-many-times Reason
kvm_intel249 Load
kvm 249 Load
irqbypass 8 Load
crct10dif_pclmul128 Load
ghash_clmulni_intel27 Load
sha512_ssse350 Load
sha512_generic200 Load
aesni_intel249 Load
crypto_simd41 Load
cryptd 131 Load
evdev 2 Load
serio_raw1 Load
virtio_pci3 Load
nvme 3 Load
nvme_core3 Load
virtio_pci_legacy_dev3 Load
virtio_pci_modern_dev3 Load
t10_pi3 Load
virtio 3 Load
crc32_pclmul6 Load
crc64_rocksoft3 Load
crc32c_intel40 Load
virtio_ring3 Load
crc64 3 LoadThe following screen shot, of a simple 8vcpu 8 GiB KVM guest with the next patch in this series applied, shows 226.53 MiB are wasted in virtual memory allocations which due to duplicate module requests during boot. It also shows an average module memory size of 167.10 KiB and an an average module .text + .init.text size of 61.13 KiB. The end shows all modules which were detected as duplicate requests and whether or not they failed early after just the first kernel_read*() call or late after we've already allocated the private space for the module in
layout_and_allocate(). A system with module decompression would reveal more wasted virtual memory space.We should put effort now into identifying the source of these duplicate module requests and trimming these down as much possible. Larger systems will obviously show much more wasted virtual memory allocations.
root@kmod ~ # cat
/sys/kernel/debug/modules/statsMods ever loaded 67 Mods failed on kread 0 Mods failed on decompress 0 Mods failed on becoming 83 Mods failed on load 16 Total module size 11464704 Total mod text size 4194304 Failed kread bytes 0 Failed decompress bytes 0 Failed becoming bytes 228959096 Failed kmod bytes 8578080 Virtual mem wasted bytes 237537176 Average mod size 171115 Average mod text size 62602 Avg fail becoming bytes 2758544 Average fail load bytes 536130Duplicate failed modules:
module-name How-many-times Reason
kvm_intel7 Becoming
kvm 7 Becoming
irqbypass 6 Becoming & Load
crct10dif_pclmul7 Becoming & Load
ghash_clmulni_intel7 Becoming & Load
sha512_ssse36 Becoming & Load
sha512_generic7 Becoming & Load
aesni_intel7 Becoming
crypto_simd7 Becoming & Load
cryptd 3 Becoming & Load
evdev 1 Becoming
serio_raw1 Becoming
nvme 3 Becoming
nvme_core3 Becoming
t10_pi3 Becoming
virtio_pci3 Becoming
crc32_pclmul6 Becoming & Load
crc64_rocksoft3 Becoming
crc32c_intel3 Becoming
virtio_pci_modern_dev2 Becoming
virtio_pci_legacy_dev1 Becoming
crc64 2 Becoming
virtio 2 Becoming
virtio_ring2 Becoming[0] https://github.com/ColinIanKing/stress-ng.git
[1] echo 0 >/proc/sys/vm/oom_dump_tasks./stress-ng --module 100 --module-name xfsSigned-off-by Luis Chamberlain <[email protected]>--- Documentation/core-api/kernel-api.rst | 22 +- kernel/module/Kconfig | 37 +++ kernel/module/Makefile | 1 + kernel/module/decompress.c | 4 + kernel/module/internal.h | 74 +++++ kernel/module/main.c | 65 +++- kernel/module/stats.c | 432 ++++++++++++++++++++++++++ kernel/module/tracking.c | 7 +- 8 files changed, 630 insertions(+), 12 deletions(-) create mode 100644 kernel/module/stats.cDocumentation/core-api/kernel-api.rst +20
-2@@ -220,12 +220,30 @@ relay interface Module Support ============== -Module Loading --------------- +Kernel module auto-loading +-------------------------- .. kernel-doc:: kernel/module/kmod.c :export: +Module debugging +---------------- + +.. kernel-doc:: kernel/module/stats.c + :doc: module debugging statistics overview + +dup_failed_modules - tracks duplicate failed modules +**************************************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: dup_failed_modules - tracks duplicate failed modules + +module statistics debugfs counters +********************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: module statistics debugfs counters + Inter Module support --------------------kernel/module/Kconfig +0
-0@@ -22,6 +22,43 @@ menuconfig MODULES
kernel/module/Kconfig menuconfig MODULES · line 32
Quoted 10 lines
kernel/module/Kconfig +8
-0if MODULES +config MODULE_DEBUG + bool "Module debugging" + depends on DEBUG_FS + help + Allows you to enable / disable features which can help you debug + modules. You don't need these options in produciton systems. You can + and probably should enable this prior to making your kernel + produciton ready though.
2x typo: produciton -> production.
The last sentence could be misinterpreted to mean that you should enable this to make your kernel production ready. Not sure, maybe I would just drop this sentence.
Note that there are plenty of other typos in the added comments and documentation. Please review them with a spell checker.
kernel/module/Kconfig menuconfig MODULES · line 42
Quoted 10 lines
kernel/module/Kconfig +10
-0+ +if MODULE_DEBUG + +config MODULE_STATS + bool "Module statistics" + depends on DEBUG_FS + help + This option allows you to maintain a record of module statistics. + For example each all modules size, average size, text size, and + failed modules and the size for each of those. For failed
This sentence doesn't quite make sense. I guess it should say something as:
For example, size of all modules, average size, text size, a list of failed modules and the size for each of those.
kernel/module/Kconfig menuconfig MODULES · line 52
Quoted 10 lines
kernel/module/Kconfig +10
-0+ modules we keep track of module which failed due to either the + existing module taking too long to load or that module already + was loaded. + + You should enable this if you are debugging production loads + and want to see if userspace or the kernel is doing stupid things + with loading modules when it shouldn't or if you want to help + optimize userspace / kernel space module autoloading schemes. + You might want to do this because failed modules tend to use + use up significan amount of memory, and so you'd be doing everyone
Word 'use' is repeated twice.
Quoted 152 lines
kernel/module/Kconfig +9
-0+ a favor in avoiding these failure proactively. + + This functionality is also useful for those experimenting with + module .text ELF section optimization. + + If unsure, say N. + +endif # MODULE_DEBUG + config MODULE_FORCE_LOAD bool "Forced module loading" default nkernel/module/Makefile +1
-0@@ -21,3 +21,4 @@ obj-$(CONFIG_SYSFS) += sysfs.o obj-$(CONFIG_KGDB_KDB) += kdb.o obj-$(CONFIG_MODVERSIONS) += version.o obj-$(CONFIG_MODULE_UNLOAD_TAINT_TRACKING) += tracking.o +obj-$(CONFIG_MODULE_STATS) += stats.okernel/module/decompress.c +4
-0@@ -297,6 +297,10 @@ int module_decompress(struct load_info *info, const void *buf, size_t size)ssize_t data_size;int error; +#if defined(CONFIG_MODULE_STATS ) + info->compressed_len = size; +#endif +/* * Start with number of pages twice as big as needed for * compressed data. kernel/module/internal.h +74
-0@@ -59,6 +59,9 @@ struct load_info {unsigned long mod_kallsyms_init_off;#endif #ifdef CONFIG_MODULE_DECOMPRESS +#ifdef CONFIG_MODULE_STATS +unsigned long compressed_len; +#endif struct page **pages;unsigned int max_pages;unsigned int used_pages;@@ -143,6 +146,77 @@ static inline bool set_livepatch_module(struct module *mod)#endif } +/** +* enum fail_dup_mod_reason - state at which a duplicate module was detected +* +* @FAIL_DUP_MOD_BECOMING: the module is read properly, passes all checks but +* we've determined that another module with the same name is already loaded +* or being processed on our &modules list. This happens on early_mod_check() +* right before layout_and_allocate(). The kernel would have already +* vmalloc()'d space for the entire module through finit_module(). If +* decompression was used two vmap() spaces were used. These failures can +* happen when userspace has not seen the module present on the kernel and +* tries to load the module multiple times at same time. +* @FAIL_DUP_MOD_LOAD: the module has been read properly, passes all validation +* checks and the kernel determines that the module was unique and because +* of this allocated yet another private kernel copy of the module space in +* layout_and_allocate() but after this determined in add_unformed_module() +* that another module with the same name is already loaded or being processed. +* These failures should be mitigated as much as possible and are indicative +* of really fast races in loading modules. Without module decompression +* they waste twice as much vmap space. With module decompression three +* times the module's size vmap space is wasted. +*/ +enum fail_dup_mod_reason { +FAIL_DUP_MOD_BECOMING =0 , +FAIL_DUP_MOD_LOAD , +}; + +#ifdef CONFIG_MODULE_STATS + +#define mod_stat_add_long (count, var) atomic_long_add(count, var) +#define mod_stat_inc (name) atomic_inc(name) + +extern atomic_long_t total_mod_size; +extern atomic_long_t total_text_size; +extern atomic_long_t invalid_kread_bytes; +extern atomic_long_t invalid_decompress_bytes; + +extern atomic_t modcount; +extern atomic_t failed_kreads; +extern atomic_t failed_decompress; +struct mod_fail_load { +struct list_head list; +char name[MODULE_NAME_LEN ]; +atomic_long_t count; +unsigned long dup_fail_mask; +}; + +int try_add_failed_module (const char *name,size_t len,enum fail_dup_mod_reason reason); +void mod_stat_bump_invalid (struct load_info *info,int flags); +void mod_stat_bump_becoming (struct load_info *info,int flags); + +#else + +#define mod_stat_add_long (name, var) +#define mod_stat_inc (name) + +static inline int try_add_failed_module (const char *name,size_t len, +enum fail_dup_mod_reason reason) +{ +return 0 ; +} + +static inline void mod_stat_bump_invalid (struct load_info *info,int flags) +{ +} + +static inline void mod_stat_bump_becoming (struct load_info *info,int flags) +{ +} + +#endif /* CONFIG_MODULE_STATS */ + #ifdef CONFIG_MODULE_UNLOAD_TAINT_TRACKING struct mod_unload_taint { struct list_head list;kernel/module/main.c +3
-0@@ -56,6 +56,7 @@#include <linux/dynamic_debug.h> #include <linux/audit.h> #include <linux/cfi.h> +#include <linux/debugfs.h> #include <uapi/linux/module.h> #include "internal.h" @@ -87,6 +88,8 @@ struct symsearch {enum mod_license license; }; +struct dentry *mod_debugfs_root; +/* * Bounds of module memory, for speeding up __module_address. * Protected by module_mutex. @@ -2500,6 +2503,18 @@ static noinline int do_init_module(struct module *mod) {int ret =0 ;
kernel/module/main.c static noinline int do_init_module(struct module *mod) · line 2514
Quoted 10 lines
kernel/module/main.c +9
-0struct mod_initfree *freeinit; +#if defined(CONFIG_MODULE_STATS ) +unsigned int text_size =0 , total_size =0 ; + +for_each_mod_mem_type (type) { +const struct module_memory *mod_mem = &mod->mem[type]; +if (mod_mem->size) { + total_size += mod_mem->size; +if (type ==MOD_TEXT || type ==MOD_INIT_TEXT ) + text_size += mod->mem[type].size;
'text_size += mod_mem->size;' would be simpler.
Quoted 209 lines
kernel/module/main.c +48
-5+ } + } +#endif freeinit =kmalloc (sizeof (*freeinit),GFP_KERNEL );if (!freeinit) {@@ -2561,6 +2576,7 @@ static noinline int do_init_module(struct module *mod) mod->mem[type].base =NULL ; mod->mem[type].size =0 ; } +#ifdef CONFIG_DEBUG_INFO_BTF_MODULES /* .BTF is not SHF_ALLOC and will get removed, so sanitize pointer */ mod->btf_data =NULL ;@@ -2584,6 +2600,11 @@ static noinline int do_init_module(struct module *mod)mutex_unlock (&module_mutex);wake_up_all (&module_wq); +mod_stat_add_long (text_size, &total_text_size); +mod_stat_add_long (total_size, &total_mod_size); + +mod_stat_inc (&modcount); +return 0 ; fail_free_freeinit:@@ -2599,6 +2620,7 @@ static noinline int do_init_module(struct module *mod)ftrace_release_mod (mod);free_module (mod);wake_up_all (&module_wq); +return ret; }@@ -2632,7 +2654,8 @@ static bool finished_loading(const char *name) }/* Must be called with module_mutex held */ -static int module_patient_check_exists (const char *name) +static int module_patient_check_exists (const char *name, +enum fail_dup_mod_reason reason) {struct module *old;int err =0 ;@@ -2655,6 +2678,9 @@ static int module_patient_check_exists(const char *name) old =find_module_all (name,strlen (name), true); } +if (try_add_failed_module (name,strlen (name), reason)) +pr_warn ("Could not add fail-tracking for module: %s\n" , name); +/* * We are here only when the same module was being loaded. Do * not try to load it again right now. It prevents long delays @@ -2679,7 +2705,7 @@ static int add_unformed_module(struct module *mod) mod->state =MODULE_STATE_UNFORMED ;mutex_lock (&module_mutex); - err =module_patient_check_exists (mod->name); + err =module_patient_check_exists (mod->name,FAIL_DUP_MOD_LOAD );if (err) goto out;@@ -2800,6 +2826,7 @@ static int load_module(struct load_info *info, const char __user *uargs, int flags) {struct module *mod; +bool module_allocated = false;long err =0 ;char *after_dashes;@@ -2839,6 +2866,8 @@ static int load_module(struct load_info *info, const char __user *uargs, goto free_copy; } + module_allocated = true; +audit_log_kern_module (mod->name);/* Reserve our place in the list. */ @@ -2983,6 +3012,7 @@ static int load_module(struct load_info *info, const char __user *uargs,synchronize_rcu ();mutex_unlock (&module_mutex); free_module: +mod_stat_bump_invalid (info, flags);/* Free lock-classes; relies on the preceding sync_rcu() */ for_class_mod_mem_type(type, core_data) { lockdep_free_key_range(mod->mem[type].base,@@ -2991,6 +3021,13 @@ static int load_module(struct load_info *info, const char __user *uargs,module_deallocate (mod, info); free_copy: +/* +* The info->len is always set. We distinguish between +* failures once the proper module was allocated and +* before that. +*/ +if (!module_allocated) +mod_stat_bump_becoming (info, flags);free_copy (info, flags);return err; }@@ -3009,8 +3046,11 @@ SYSCALL_DEFINE3(init_module, void __user *, umod, umod, len, uargs); err =copy_module_from_user (umod, len, &info); -if (err) +if (err) { +mod_stat_inc (&failed_kreads); +mod_stat_add_long (len, &invalid_kread_bytes);return err; + }return load_module (&info, uargs,0 ); }@@ -3035,14 +3075,20 @@ SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags) len =kernel_read_file_from_fd (fd,0 , &buf,INT_MAX ,NULL ,READING_MODULE ); -if (len <0 ) +if (len <0 ) { +mod_stat_inc (&failed_kreads); +mod_stat_add_long (len, &invalid_kread_bytes);return len; + }if (flags &MODULE_INIT_COMPRESSED_FILE ) { err =module_decompress (&info, buf, len);vfree (buf);/* compressed data is no longer needed */ -if (err) +if (err) { +mod_stat_inc (&failed_decompress); +mod_stat_add_long (len, &invalid_decompress_bytes);return err; + } }else { info.hdr = buf; info.len = len;@@ -3216,3 +3262,12 @@ void print_modules(void) last_unloaded_module.taints);pr_cont ("\n" ); } + +#ifdef CONFIG_MODULE_DEBUG +static int module_debugfs_init (void ) +{ + mod_debugfs_root =debugfs_create_dir ("modules" ,NULL ); +return 0 ; +} +module_init (module_debugfs_init); +#endif kernel/module/stats.c new file +52
-0@@ -0,0 +1,432 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* Debugging module statistics. +* +* Copyright (C) 2023 Luis Chamberlain <[email protected]> +*/ + +#include <linux/module.h> +#include <linux/string.h> +#include <linux/printk.h> +#include <linux/slab.h> +#include <linux/list.h> +#include <linux/debugfs.h> +#include <linux/rculist.h> +#include <linux/math.h> + +#include "internal.h" + +/** +* DOC: module debugging statistics overview +* +* Enabling CONFIG_MODULE_STATS enables module debugging statistics which +* are useful to monitor and root cause memory pressure issues with module +* loading. These statistics are useful to allow us to improve production +* workloads. +* +* The current module debugging statistics supported help keep track of module +* loading failures to enable improvements either for kernel module +* auto-loading usage (request_module()) or interactions with userspace. +* Statistics are provided to track of all possible failures in the +* finit_module() path and memory wasted in this process space. Each of the +* failure counters are associated to a type of module loading failure which +* is known to incur a certain amount of memory allocation loss. In the worst +* case loading a module will fail after a 3 step memory allocation process: +* +* a) memory allocated with kernel_read_file_from_fd() +* b) module decompression processes the file read from +* kernel_read_file_from_fd(), and vmap() is used to map +* the decompressed module to a new local buffer which represents +* a copy of the decompressed module passed from userspace. The buffer +* from kernel_read_file_from_fd() is freed right away. +* c) layout_and_allocate() allocates space for the final resting +* place where we would keep the module if it were to be processed +* successfully. +* +* If a failure occurs after these three different allocations only one +* counters will be incremetned with the summation of the lost bytes incurred +* during this failure. Likewise, if a module loading failed only after step b) +* a separate counter is used and incremented for the bytes lost during both +* of those allocations. +* +* Virtual memory space can be limited, for example on x86 virtual memory size
kernel/module/stats.c line 62
Quoted 10 lines
kernel/module/stats.c new file +10
-0+* defaults to 128 MiB. We should strive to limit and avoid wasting virtual +* memory allocations when possible. These module dubugging statistics help +* to evaluate how much memory is being wasted on bootup due to module loading +* failures. +* +* All counters are designed to be incremental. Atomic counters are used so to +* remain simple and avoid delays and deadlocks. +*/ + +extern struct dentry *mod_debugfs_root;
Files kernel/module/stats.c and kernel/module/tracking.c both add this extern declaration. Can it be moved to kernel/module/internal.h?
Quoted 156 lines
kernel/module/stats.c new file +156
-0+ +/** +* DOC: dup_failed_modules - tracks duplicate failed modules +* +* Linked list of modules which failed to be loaded because an already existing +* module with the same name was already being processed or already loaded. +* The finit_module() system call incurs heavy virtual memory allocations. In +* the worst case an finit_module() system call can end up allocating virtual +* memory 3 times: +* +* 1) kernel_read_file_from_fd() call uses vmalloc() +* 2) optional module decompression uses vmap() +* 3) layout_and allocate() can use vzalloc() or an arch specific variation of +* vmalloc to deal with ELF sections requiring special permissions +* +* In practice on a typical boot today most finit_module() calls fail due to +* the module with the same name already being loaded or about to be processed. +* All virtual memory allocated to these failed modules will be lost with +* no functional use. +* +* To help with this the dup_failed_modules allows us to track modules which +* failed to load due to the fact that a module already was loaded or being +* processed already. There are only two points at which we can fail such +* calls, we list them below along with the number of virtual memory allocation +* calls: +* +* a) FAIL_DUP_MOD_BECOMING: at the end of early_mod_check() before +* layout_and_allocate(). This does not yet happen. +* - with module decompression: 2 virtual memory allocation calls +* - without module decompression: 1 virtual memory allocation calls +* b) FAIL_DUP_MOD_LOAD: after layout_and_allocate() on add_unformed_module() +* - with module decompression 3 virtual memory allocation calls +* - without module decompression 2 virtual memory allocation calls +* +* We should strive to get this list to be as small as possible. If this list +* is not empty it is a reflection of possible work or optimizations possible +* either in-kernel or in userspace. +*/ +static LIST_HEAD (dup_failed_modules); + +/** +* DOC: module statistics debugfs counters +* +* The total amount of wasted virtual memory allocation space during module +* loading can be computed by adding the total from the summation: +* +* * @invalid_kread_bytes + +* @invalid_decompress_bytes + +* @invalid_becoming_bytes + +* @invalid_mod_bytes +* +* The following debugfs counters are available to inspect module loading +* failures: +* +* * total_mod_size: total bytes ever used by all modules we've dealt with on +* this system +* * total_text_size: total bytes of the .text and .init.text ELF section +* sizes we've dealt with on this system +* * invalid_kread_bytes: bytes wasted in failures which happen due to +* memory allocations with the initial kernel_read_file_from_fd(). +* kernel_read_file_from_fd() uses vmalloc() and so these are wasted +* vmalloc() memory allocations. These should typically not happen unless +* your system is under memory pressure. +* * invalid_decompress_bytes: number of bytes wasted due to +* memory allocations in the module decompression path that use vmap(). +* These typically should not happen unless your system is under memory +* presssure. +* * invalid_becoming_bytes: total number of bytes wasted due to +* allocations used to read the kernel module userspace wants us to read +* before we promote it to be processed to be added to our @modules linked +* list. These failures could in theory happen in if we had a check in between +* between a successful kernel_read_file_from_fd() call and right before +* we allocate the our private memory for the module which would be kept if +* the module is successfully loaded. The most common reason for this failure +* is when userspace is racing to load a module which it does not yet see +* loaded. The first module to succeed in add_unformed_module() will add a +* module to our &modules list and subsequent loads of modules with the +* same name will error out at the end of early_mod_check(). A check +* for module_patient_check_exists() at the end of early_mod_check() could be +* added to prevent duplicate allocations on layout_and_allocate() for +* modules already being processed. These duplicate failed modules are +* non-fatal, however they typically are indicative of userspace not seeing +* a module in userspace loaded yet and unecessarily trying to load a +* module before the kernel even has a chance to begin to process prior +* requests. Although duplicate failures can be non-fatal, we should try to +* reduce vmalloc() pressure proactively, so ideally after boot this will +* be close to as 0 as possible. If module decompression was used we also +* add to this counter the cost of the initial kernel_read_file_from_fd() +* of the compressed module. If module decompression was not used the +* value represents the total wasted allocations in kernel_read_file_from_fd() +* calls for these type of failures. These failures can occur because: +* +* * module_sig_check() - module signature checks +* * elf_validity_cache_copy() - some ELF validation issue +* * early_mod_check(): +* +* * blacklisting +* * failed to rewrite section headers +* * version magic +* * live patch requirements didn't check out +* * the module was detected as being already present +* +* * invalid_mod_bytes: these are the total number of bytes lost due to +* failures after we did all the sanity checks of the module which userspace +* passed to us and after our first check that the module is unique. A +* module can still fail to load if we detect the module is loaded after we +* allocate space for it with layout_and_allocate(), we do this check right +* before processing the module as live and run its initialiation routines. +* Note that you have a failure of this type it also means the respective +* kernel_read_file_from_fd() memory space was also wasted, and so we +* increment this counter with twice the size of the module. Additionally +* if you used module decompression the size of the compressed module is +* also added to this counter. +* +* * modcount: how many modules we've loaded in our kernel life time +* * failed_kreads: how many modules failed due to failed kernel_read_file_from_fd() +* * failed_decompress: how many failed module decompression attempts we've had. +* These really should not happen unless your compression / decompression +* might be broken. +* * failed_becoming: how many modules failed after we kernel_read_file_from_fd() +* it and before we allocate memory for it with layout_and_allocate(). This +* counter is never incremented if you manage to validate the module and +* call layout_and_allocate() for it. +* * failed_load_modules: how many modules failed once we've allocated our +* private space for our module using layout_and_allocate(). These failures +* should hopefully mostly be dealt with already. Races in theory could +* still exist here, but it would just mean the kernel had started processing +* two threads concurrently up to early_mod_check() and then one just one +* thread won. These failures are good signs the kernel or userspace is +* doing something seriously stupid or that could be improved. We should +* strive to fix these, but it is perhaps not easy to fix them. +* A recent example are the modules requests incurred for frequency modules, +* a separate module request was being issued for each CPU on a system. +*/ + +atomic_long_t total_mod_size; +atomic_long_t total_text_size; +atomic_long_t invalid_kread_bytes; +atomic_long_t invalid_decompress_bytes; +static atomic_long_t invalid_becoming_bytes; +static atomic_long_t invalid_mod_bytes; +atomic_t modcount; +atomic_t failed_kreads; +atomic_t failed_decompress; +static atomic_t failed_becoming; +static atomic_t failed_load_modules; + +static const char *mod_fail_to_str (struct mod_fail_load *mod_fail) +{ +if (test_bit (FAIL_DUP_MOD_BECOMING , &mod_fail->dup_fail_mask) && +test_bit (FAIL_DUP_MOD_LOAD , &mod_fail->dup_fail_mask)) +return "Becoming & Load" ; +if (test_bit (FAIL_DUP_MOD_BECOMING , &mod_fail->dup_fail_mask)) +return "Becoming" ; +if (test_bit (FAIL_DUP_MOD_LOAD , &mod_fail->dup_fail_mask)) +return "Load" ;
kernel/module/stats.c line 228
Quoted 10 lines
kernel/module/stats.c new file +10
-0+return "Bug-on-stats" ; +} + +void mod_stat_bump_invalid (struct load_info *info,int flags) +{ +atomic_long_add (info->len *2 , &invalid_mod_bytes); +atomic_inc (&failed_load_modules); +#if defined(CONFIG_MODULE_DECOMPRESS ) +if (flags &MODULE_INIT_COMPRESSED_FILE ) +atomic_long_add (info->compressed_len, &invalid_mod_byte);
Variable invalid_mod_byte is not declared, should be invalid_mod_bytes.
Quoted 4 lines
kernel/module/stats.c new file +4
-0+#endif +} + +void mod_stat_bump_becoming (struct load_info *info,int flags)
kernel/module/stats.c line 242
Quoted 10 lines
kernel/module/stats.c new file +10
-0+{ +atomic_inc (&failed_becoming); +atomic_long_add (info->len, &invalid_becoming_bytes); +#if defined(CONFIG_MODULE_DECOMPRESS ) +if (flags &MODULE_INIT_COMPRESSED_FILE ) +atomic_long_add (info->compressed_len, &invalid_becoming_bytes); +#endif +} + +int try_add_failed_module (const char *name,size_t len,enum fail_dup_mod_reason reason)
Function try_add_failed_module() is only called from module_patient_check_exists() which always passes in a NUL-terminated string. The len parameter could be then dropped and the comparison in try_add_failed_module() could simply use strcmp().
Quoted 12 lines
kernel/module/stats.c new file +12
-0+{ +struct mod_fail_load *mod_fail; + +list_for_each_entry_rcu (mod_fail, &dup_failed_modules, list, +lockdep_is_held (&module_mutex)) { +if (strlen (mod_fail->name) == len && !memcmp (mod_fail->name, name, len)) { +atomic_long_inc (&mod_fail->count); +__set_bit (reason, &mod_fail->dup_fail_mask); + goto out; + } + } +
kernel/module/stats.c line 264
Quoted 10 lines
kernel/module/stats.c new file +10
-0+ mod_fail =kzalloc (sizeof (*mod_fail),GFP_KERNEL ); +if (!mod_fail) +return -ENOMEM ; +memcpy (mod_fail->name, name, len); +__set_bit (reason, &mod_fail->dup_fail_mask); +atomic_long_inc (&mod_fail->count); +list_add_rcu (&mod_fail->list, &dup_failed_modules); +out: +return 0 ; +}
Indentation in try_add_failed_module() uses spaces instead of tabs in a few places.
Quoted 27 lines
kernel/module/stats.c new file +27
-0+ +/* +* At 64 bytes per module and assuming a 1024 bytes preamble we can fit the +* 112 module prints within 8k. +* +* 1024 + (64*112) = 8k +*/ +#define MAX_PREAMBLE 1024 +#define MAX_FAILED_MOD_PRINT 112 +#define MAX_BYTES_PER_MOD 64 +static ssize_t read_file_mod_stats (struct file *file,char __user *user_buf, +size_t count,loff_t *ppos) +{ +struct mod_fail_load *mod_fail; +unsigned int len, size, count_failed =0 ; +char *buf; +u32 live_mod_count, fkreads, fdecompress, fbecoming, floads; +u64 total_size, text_size, ikread_bytes, ibecoming_bytes, idecompress_bytes, imod_bytes, + total_virtual_lost; + + live_mod_count =atomic_read (&modcount); + fkreads =atomic_read (&failed_kreads); + fdecompress =atomic_read (&failed_decompress); + fbecoming =atomic_read (&failed_becoming); + floads =atomic_read (&failed_load_modules); + + total_size =atomic64_read (&total_mod_size);
kernel/module/stats.c line 301
Quoted 10 lines
kernel/module/stats.c new file +10
-0+ text_size =atomic64_read (&total_text_size); + ikread_bytes =atomic64_read (&invalid_kread_bytes); + idecompress_bytes =atomic64_read (&invalid_decompress_bytes); + ibecoming_bytes =atomic64_read (&invalid_becoming_bytes); + imod_bytes =atomic64_read (&invalid_mod_bytes); + + total_virtual_lost = ikread_bytes + idecompress_bytes + ibecoming_bytes + imod_bytes; + + size =MAX_PREAMBLE +min ((unsigned int )(floads + fbecoming) *MAX_BYTES_PER_MOD , + (unsigned int )MAX_FAILED_MOD_PRINT *MAX_BYTES_PER_MOD );
Using
'size = MAX_PREAMBLE + min((unsigned int)(floads + fbecoming), (unsigned int)MAX_FAILED_MOD_PRINT) * MAX_BYTES_PER_MOD;' is a bit simpler and avoids any theoretical overflow of
'(floads + fbecoming) * MAX_BYTES_PER_MOD'.
Quoted 71 lines
kernel/module/stats.c new file +71
-0+ buf =kzalloc (size,GFP_KERNEL ); +if (buf ==NULL ) +return -ENOMEM ; + +/* The beginning of our debug preamble */ + len =scnprintf (buf +0 , size - len,"%25s\t%u\n" ,"Mods ever loaded" , live_mod_count); + + len +=scnprintf (buf + len, size - len,"%25s\t%u\n" ,"Mods failed on kread" , fkreads); + + len +=scnprintf (buf + len, size - len,"%25s\t%u\n" ,"Mods failed on decompress" , + fdecompress); + len +=scnprintf (buf + len, size - len,"%25s\t%u\n" ,"Mods failed on becoming" , fbecoming); + + len +=scnprintf (buf + len, size - len,"%25s\t%u\n" ,"Mods failed on load" , floads); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Total module size" , total_size); + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Total mod text size" , text_size); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Failed kread bytes" , ikread_bytes); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Failed decompress bytes" , + idecompress_bytes); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Failed becoming bytes" , ibecoming_bytes); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Failed kmod bytes" , imod_bytes); + + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Virtual mem wasted bytes" , total_virtual_lost); + +if (live_mod_count && total_size) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Average mod size" , +DIV_ROUND_UP (total_size, live_mod_count)); + } + +if (live_mod_count && text_size) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Average mod text size" , +DIV_ROUND_UP (text_size, live_mod_count)); + } + +/* +* We use WARN_ON_ONCE() for the counters to ensure we always have parity +* for keeping tabs on a type of failure with one type of byte counter. +* The counters for imod_bytes does not increase for fkreads failures +* for example, and so on. +*/ + +WARN_ON_ONCE (ikread_bytes && !fkreads); +if (fkreads && ikread_bytes) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Avg fail kread bytes" , +DIV_ROUND_UP (ikread_bytes, fkreads)); + } + +WARN_ON_ONCE (ibecoming_bytes && !fbecoming); +if (fbecoming && ibecoming_bytes) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Avg fail becoming bytes" , +DIV_ROUND_UP (ibecoming_bytes, fbecoming)); + } + +WARN_ON_ONCE (idecompress_bytes && !fdecompress); +if (fdecompress && idecompress_bytes) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Avg fail decomp bytes" , +DIV_ROUND_UP (idecompress_bytes, fdecompress)); + } + +WARN_ON_ONCE (imod_bytes && !floads); +if (floads && imod_bytes) { + len +=scnprintf (buf + len, size - len,"%25s\t%llu\n" ,"Average fail load bytes" , +DIV_ROUND_UP (imod_bytes, floads)); + } + +/* End of our debug preamble header. */
kernel/module/stats.c line 382
Quoted 10 lines
kernel/module/stats.c new file +10
-0+ +/* Catch when we've gone beyond our expected preamble */ +WARN_ON_ONCE (len >=MAX_PREAMBLE ); + +if (list_empty (&dup_failed_modules)) + goto out; + + len +=scnprintf (buf + len, size - len,"Duplicate failed modules:\n" ); + len +=scnprintf (buf + len, size - len,"%25s\t%15s\t%25s\n" , +"module-name" ,"How-many-times" ,"Reason" );
"module-name" -> "Module-name"
Quoted 40 lines
kernel/module/stats.c new file +40
-0+mutex_lock (&module_mutex); + + +list_for_each_entry_rcu (mod_fail, &dup_failed_modules, list) { +if (WARN_ON_ONCE (++count_failed >=MAX_FAILED_MOD_PRINT )) + goto out_unlock; + len +=scnprintf (buf + len, size - len,"%25s\t%15llu\t%25s\n" , mod_fail->name, +atomic64_read (&mod_fail->count),mod_fail_to_str (mod_fail)); + } +out_unlock: +mutex_unlock (&module_mutex); +out: +kfree (buf); +return simple_read_from_buffer (user_buf, count, ppos, buf, len); +} +#undef MAX_PREAMBLE +#undef MAX_FAILED_MOD_PRINT +#undef MAX_BYTES_PER_MOD + +static const struct file_operations fops_mod_stats = { + .read = read_file_mod_stats, + .open = simple_open, + .owner =THIS_MODULE , + .llseek = default_llseek, +}; + +#define mod_debug_add_ulong (name) debugfs_create_ulong(#name, 0400, mod_debugfs_root, (unsigned long *) &name.counter) +#define mod_debug_add_atomic (name) debugfs_create_atomic_t(#name, 0400, mod_debugfs_root, &name) +static int __init module_stats_init (void ) +{ +mod_debug_add_ulong (total_mod_size); +mod_debug_add_ulong (total_text_size); +mod_debug_add_ulong (invalid_kread_bytes); +mod_debug_add_ulong (invalid_decompress_bytes); +mod_debug_add_ulong (invalid_becoming_bytes); +mod_debug_add_ulong (invalid_mod_bytes); + +mod_debug_add_atomic (modcount); +mod_debug_add_atomic (failed_kreads); +mod_debug_add_atomic (failed_decompress);
kernel/module/stats.c line 432
Quoted 10 lines
kernel/module/stats.c new file +10
-0+mod_debug_add_atomic (failed_becoming); +mod_debug_add_atomic (failed_load_modules); + +debugfs_create_file ("stats" ,0400 , mod_debugfs_root, mod_debugfs_root, &fops_mod_stats); + +return 0 ; +} +#undef mod_debug_add_ulong +#undef mod_debug_add_atomic +module_init (module_stats_init);
Function module_stats_init() requires mod_debugfs_root being initialized which is done in module_debugfs_init(). Both functions are recorded to be called via module_init(). Just to make sure, is their ordering guaranteed in some way?
Quoted 13 lines
kernel/module/tracking.c +1
-0@@ -15,6 +15,7 @@#include "internal.h" static LIST_HEAD (unloaded_tainted_modules); +extern struct dentry *mod_debugfs_root;int try_add_tainted_module (struct module *mod) {
kernel/module/tracking.c static const struct file_operations unloaded_tainted_modules_fops = { · line 125
Quoted 9 lines
kernel/module/tracking.c +1
-4@@ -120,12 +121,8 @@ static const struct file_operations unloaded_tainted_modules_fops = {static int __init unloaded_tainted_modules_init (void ) { -struct dentry *dir; - - dir =debugfs_create_dir ("modules" ,NULL ); -debugfs_create_file ("unloaded_tainted" ,0444 , dir,NULL , +debugfs_create_file ("unloaded_tainted" ,0444 , mod_debugfs_root,NULL , &unloaded_tainted_modules_fops);
mod_debugfs_root is initialized in module_debugfs_init() only if CONFIG_MODULE_DEBUG is set. However, my reading is that feature CONFIG_MODULE_UNLOAD_TAINT_TRACKING is orthogonal to it and doesn't require CONFIG_MODULE_DEBUG, so it looks this change breaks this tracking?
kernel/module/tracking.c static const struct file_operations unloaded_tainted_modules_fops = { · line 128
Quoted 4 lines
kernel/module/tracking.c +0
-1-return 0 ; }module_init (unloaded_tainted_modules_init);
Cheers,
Petr
Luis Chamberlain
maintainerRe: [PATCH v3 3/4] module: add debug stats to help identify memory pressure
Quoted 1 line · Petr Pavlu
On 4/14/23 07:08, Luis Chamberlain wrote:
<-- Petr's spell checking -->
Quoted 2 lines
Note that there are plenty of other typos in the added comments and documentation. Please review them with a spell checker.
Yes I am terrible at that, I've now integrated a spell checker into my workflow. Fixed all these, thanks.
Quoted 7 lines
Quoted 7 lines
kernel/module/main.c +3
-0@@ -2500,6 +2503,18 @@ static noinline int do_init_module(struct module *mod) {int ret =0 ;struct mod_initfree *freeinit; +#if defined(CONFIG_MODULE_STATS ) +unsigned int text_size =0 , total_size =0 ; +
Quoted 8 lines
kernel/module/main.c static noinline int do_init_module(struct module *mod) · line 2514
Quoted 6 lines
kernel/module/main.c +6
-0+for_each_mod_mem_type (type) { +const struct module_memory *mod_mem = &mod->mem[type]; +if (mod_mem->size) { + total_size += mod_mem->size; +if (type ==MOD_TEXT || type ==MOD_INIT_TEXT ) + text_size += mod->mem[type].size;'
text_size+=mod_mem->size;' would be simpler.
Sure.
Quoted 4 lines
Review static noinline int do_init_module(struct module *mod) · line 2515
Quoted 1 line
(quoted diff) +1
-0+extern struct dentry *mod_debugfs_root;Files
kernel/module/stats.candkernel/module/tracking.cboth add this extern declaration. Can it be moved tokernel/module/internal.h?
Sure.
Quoted 5 lines
kernel/module/stats.c line 228
Quoted 3 lines
kernel/module/stats.c new file +3
-0+#if defined(CONFIG_MODULE_DECOMPRESS ) +if (flags &MODULE_INIT_COMPRESSED_FILE ) +atomic_long_add (info->compressed_len, &invalid_mod_byte);Variable
invalid_mod_byteis not declared, should beinvalid_mod_bytes.
Arnd already sent a fix for that, thanks.
Quoted 6 lines
kernel/module/stats.c line 242
Quoted 1 line
kernel/module/stats.c new file +1
-0+int try_add_failed_module (const char *name,size_t len,enum fail_dup_mod_reason reason)Function
try_add_failed_module()is only called frommodule_patient_check_exists()which always passes in a NUL-terminated string. The len parameter could be then dropped and the comparison intry_add_failed_module()could simply usestrcmp().
Sure, did that.
Quoted 2 lines
Indentation in
try_add_failed_module()uses spaces instead of tabs in a few places.
Fixed.
Quoted 7 lines
kernel/module/stats.c line 301
Quoted 2 lines
kernel/module/stats.c new file +2
-0+ size =MAX_PREAMBLE +min ((unsigned int )(floads + fbecoming) *MAX_BYTES_PER_MOD , + (unsigned int )MAX_FAILED_MOD_PRINT *MAX_BYTES_PER_MOD );Using
'size =MAX_PREAMBLE+min((unsigned int)(floads + fbecoming), (unsigned int)MAX_FAILED_MOD_PRINT)*MAX_BYTES_PER_MOD;' is a bit simpler and avoids any theoretical overflow of
'(floads + fbecoming) *MAX_BYTES_PER_MOD'.
Sure.
Quoted 4 lines
kernel/module/stats.c line 382
Quoted 2 lines
kernel/module/stats.c new file +2
-0+ len +=scnprintf (buf + len, size - len,"%25s\t%15s\t%25s\n" , +"module-name" ,"How-many-times" ,"Reason" );"module-name" -> "Module-name"
OK sure.
Quoted 3 lines
Function
module_stats_init()requiresmod_debugfs_rootbeing initialized which is done inmodule_debugfs_init(). Both functions are recorded to be called viamodule_init(). Just to make sure, is their ordering guaranteed in some way?
Link order takes care of that and main.o goes first.
Quoted 4 lines
mod_debugfs_rootis initialized inmodule_debugfs_init()only ifCONFIG_MODULE_DEBUGis set. However, my reading is that featureCONFIG_MODULE_UNLOAD_TAINT_TRACKINGis orthogonal to it and doesn't requireCONFIG_MODULE_DEBUG, so it looks this change breaks this tracking?
Ah yes We need a bool CONFIG_MODULE_DEBUGFS which is selected by those that need it. Added.
Luis
Luis Chamberlain
maintainerPATCH v4 1/1 module: add debug stats to help identify memory pressure
In reply to Luis ChamberlainLoading modules with finit_module() can end up using vmalloc(), vmap() and vmalloc() again, for a total of up to 3 separate allocations in the worst case for a single module. We always kernel_read*() the module, that's a vmalloc(). Then vmap() is used for the module decompression, and if so the last read buffer is freed as we use the now decompressed module buffer to stuff data into our copy module. The last allocation is specific to each architectures but pretty much that's generally a series of vmalloc() calls or a variation of vmalloc to handle ELF sections with special permissions.
Evaluation with new stress-ng module support [1] with just 100 ops is proving that you can end up using GiBs of data easily even with all care we have in the kernel and userspace today in trying to not load modules which are already loaded. 100 ops seems to resemble the sort of pressure a system with about 400 CPUs can create on module loading. Although issues relating to duplicate module requests due to each CPU inucurring a new module reuest is silly and some of these are being fixed, we currently lack proper tooling to help diagnose easily what happened, when it happened and who likely is to blame -- userspace or kernel module autoloading.
Provide an initial set of stats which use debugfs to let us easily scrape post-boot information about failed loads. This sort of information can be used on production worklaods to try to optimize *avoiding* redundant memory pressure using finit_module().
There's a few examples that can be provided:
A 255 vCPU system without the next patch in this series applied:
Startup finished in 19.143s (kernel) + 7.078s (userspace) = 26.221s graphical.target reached after 6.988s in userspace
And 13.58 GiB of virtual memory space lost due to failed module loading:
root@big ~ # cat /sys/kernel/debug/modules/stats
Mods ever loaded 67
Mods failed on kread 0
Mods failed on decompress 0
Mods failed on becoming 0
Mods failed on load 1411
Total module size 11464704
Total mod text size 4194304
Failed kread bytes 0
Failed decompress bytes 0
Failed becoming bytes 0
Failed kmod bytes 14588526272
Virtual mem wasted bytes 14588526272
Average mod size 171115
Average mod text size 62602
Average fail load bytes 10339140Duplicate failed modules:
module-name How-many-times Reason
kvm_intel 249 Load
kvm 249 Load
irqbypass 8 Load
crct10dif_pclmul 128 Load
ghash_clmulni_intel 27 Load
sha512_ssse3 50 Load
sha512_generic 200 Load
aesni_intel 249 Load
crypto_simd 41 Load
cryptd 131 Load
evdev 2 Load
serio_raw 1 Load
virtio_pci 3 Load
nvme 3 Load
nvme_core 3 Load
virtio_pci_legacy_dev 3 Load
virtio_pci_modern_dev 3 Load
t10_pi 3 Load
virtio 3 Load
crc32_pclmul 6 Load
crc64_rocksoft 3 Load
crc32c_intel 40 Load
virtio_ring 3 Load
crc64 3 Load
The following screen shot, of a simple 8vcpu 8 GiB KVM guest with the next patch in this series applied, shows 226.53 MiB are wasted in virtual memory allocations which due to duplicate module requests during boot. It also shows an average module memory size of 167.10 KiB and an an average module .text + .init.text size of 61.13 KiB. The end shows all modules which were detected as duplicate requests and whether or not they failed early after just the first kernel_read*() call or late after we've already allocated the private space for the module in layout_and_allocate(). A system with module decompression would reveal more wasted virtual memory space.
We should put effort now into identifying the source of these duplicate module requests and trimming these down as much possible. Larger systems will obviously show much more wasted virtual memory allocations.
root@kmod ~ # cat /sys/kernel/debug/modules/stats
Mods ever loaded 67
Mods failed on kread 0
Mods failed on decompress 0
Mods failed on becoming 83
Mods failed on load 16
Total module size 11464704
Total mod text size 4194304
Failed kread bytes 0
Failed decompress bytes 0
Failed becoming bytes 228959096
Failed kmod bytes 8578080
Virtual mem wasted bytes 237537176
Average mod size 171115
Average mod text size 62602
Avg fail becoming bytes 2758544
Average fail load bytes 536130Duplicate failed modules:
module-name How-many-times Reason
kvm_intel 7 Becoming
kvm 7 Becoming
irqbypass 6 Becoming & Load
crct10dif_pclmul 7 Becoming & Load
ghash_clmulni_intel 7 Becoming & Load
sha512_ssse3 6 Becoming & Load
sha512_generic 7 Becoming & Load
aesni_intel 7 Becoming
crypto_simd 7 Becoming & Load
cryptd 3 Becoming & Load
evdev 1 Becoming
serio_raw 1 Becoming
nvme 3 Becoming
nvme_core 3 Becoming
t10_pi 3 Becoming
virtio_pci 3 Becoming
crc32_pclmul 6 Becoming & Load
crc64_rocksoft 3 Becoming
crc32c_intel 3 Becoming
virtio_pci_modern_dev 2 Becoming
virtio_pci_legacy_dev 1 Becoming
crc64 2 Becoming
virtio 2 Becoming
virtio_ring 2 Becoming
[0] https://github.com/ColinIanKing/stress-ng.git
[1] echo 0 > /proc/sys/vm/oom_dump_tasks
./stress-ng --module 100 --module-name xfs
---
Only sending a v4 for this patch. All fixes are suggested by Petr. Changes on this v4 for this patch:
o simplify try_add_failed_module() with strcmp
o fix identation on try_add_failed_module()
o simplfy max bytes for debugfs
o Module-name for debug print
o add MODULE_DEBUGFS and make stats / taint tracking select it
o use extern for debugfs root
I'd hope we can move on with fixes using modules-next now.
Documentation/core-api/kernel-api.rst | 22 +- kernel/module/Kconfig | 41 ++- kernel/module/Makefile | 1 + kernel/module/decompress.c | 4 + kernel/module/internal.h | 78 +++++ kernel/module/main.c | 65 +++- kernel/module/stats.c | 430 ++++++++++++++++++++++++++ kernel/module/tracking.c | 7 +- 8 files changed, 635 insertions(+), 13 deletions(-) create mode 100644 kernel/module/stats.c
Documentation/core-api/kernel-api.rst +20 -2
@@ -220,12 +220,30 @@ relay interface Module Support ============== -Module Loading --------------- +Kernel module auto-loading +-------------------------- .. kernel-doc:: kernel/module/kmod.c :export: +Module debugging +---------------- + +.. kernel-doc:: kernel/module/stats.c + :doc: module debugging statistics overview + +dup_failed_modules - tracks duplicate failed modules +**************************************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: dup_failed_modules - tracks duplicate failed modules + +module statistics debugfs counters +********************************** + +.. kernel-doc:: kernel/module/stats.c + :doc: module statistics debugfs counters + Inter Module support --------------------
kernel/module/Kconfig +40 -1
53 lines not on this page · raw message
kernel/module/Makefile +1 -0
@@ -21,3 +21,4 @@ obj-$(CONFIG_SYSFS) += sysfs.o obj-$(CONFIG_KGDB_KDB) += kdb.o obj-$(CONFIG_MODVERSIONS) += version.o obj-$(CONFIG_MODULE_UNLOAD_TAINT_TRACKING) += tracking.o +obj-$(CONFIG_MODULE_STATS) += stats.o
kernel/module/decompress.c +4 -0
@@ -297,6 +297,10 @@ int module_decompress(struct load_info *info, const void *buf, size_t size)ssize_t data_size;int error; +#if defined(CONFIG_MODULE_STATS ) + info->compressed_len = size; +#endif +/* * Start with number of pages twice as big as needed for * compressed data.
kernel/module/internal.h +78 -0
90 lines not on this page · raw message
kernel/module/main.c +60 -5
159 lines not on this page · raw message
kernel/module/stats.c new file +430 -0
430 lines not on this page · raw message
kernel/module/tracking.c +2 -5
@@ -15,6 +15,7 @@#include "internal.h" static LIST_HEAD (unloaded_tainted_modules); +extern struct dentry *mod_debugfs_root;int try_add_tainted_module (struct module *mod) {
@@ -120,12 +121,8 @@ static const struct file_operations unloaded_tainted_modules_fops = {static int __init unloaded_tainted_modules_init (void ) { -struct dentry *dir; - - dir =debugfs_create_dir ("modules" ,NULL ); -debugfs_create_file ("unloaded_tainted" ,0444 , dir,NULL , +debugfs_create_file ("unloaded_tainted" ,0444 , mod_debugfs_root,NULL , &unloaded_tainted_modules_fops); -return 0 ; }module_init (unloaded_tainted_modules_init);
Signature · 1 line
2.39.2