Date: Fri, 3 May 2024 12:16:31 +0100
From: Usama Arif <usamaarif642@...il.com>
To: Yosry Ahmed <yosryahmed@...gle.com>
Cc: akpm@...ux-foundation.org, hannes@...xchg.org, nphamcs@...il.com,
 chengming.zhou@...ux.dev, linux-mm@...ck.org, linux-kernel@...r.kernel.org,
 kernel-team@...a.com
Subject: Re: [PATCH v2 1/1] selftests: cgroup: add tests to verify the zswap
 writeback path


On 03/05/2024 03:36, Yosry Ahmed wrote:
> On Thu, May 2, 2024 at 12:02 PM Usama Arif <usamaarif642@...il.com> wrote:
>> Initate writeback with the below steps and check using
>> memory.stat.zswpwb if zswap writeback occurred:
>> 1. Allocate memory.
>> 2. Reclaim memory equal to the amount that was allocated in step 1.
>>     This will move it into zswap.
>> 3. Save current zswap usage.
>> 4. Move the memory allocated in step 1 back in from zswap.
>> 5. Set zswap.max to half the amount that was recorded in step 3.
>> 6. Attempt to reclaim memory equal to the amount that was allocated,
>>     this will either trigger writeback if its enabled, or reclamation
>>     will fail if writeback is disabled as there isn't enough zswap
>>     space.
>>
>> Suggested-by: Nhat Pham <nphamcs@...il.com>
>> Signed-off-by: Usama Arif <usamaarif642@...il.com>
>> ---
>>   tools/testing/selftests/cgroup/test_zswap.c | 125 +++++++++++++++++++-
>>   1 file changed, 124 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/testing/selftests/cgroup/test_zswap.c b/tools/testing/selftests/cgroup/test_zswap.c
>> index f0e488ed90d8..cd864ab825d0 100644
>> --- a/tools/testing/selftests/cgroup/test_zswap.c
>> +++ b/tools/testing/selftests/cgroup/test_zswap.c
>> @@ -50,7 +50,7 @@ static int get_zswap_stored_pages(size_t *value)
>>          return read_int("/sys/kernel/debug/zswap/stored_pages", value);
>>   }
>>
>> -static int get_cg_wb_count(const char *cg)
>> +static long get_cg_wb_count(const char *cg)
>>   {
>>          return cg_read_key_long(cg, "memory.stat", "zswpwb");
>>   }
>> @@ -248,6 +248,127 @@ static int test_zswapin(const char *root)
>>          return ret;
>>   }
>>
>> +/*
>> + * Initate writeback with the following steps:
>> + * 1. Allocate memory.
>> + * 2. Reclaim memory equal to the amount that was allocated in step 1.
>> +      This will move it into zswap.
>> + * 3. Save current zswap usage.
>> + * 4. Move the memory allocated in step 1 back in from zswap.
>> + * 5. Set zswap.max to half the amount that was recorded in step 3.
>> + * 6. Attempt to reclaim memory equal to the amount that was allocated,
>> +      this will either trigger writeback if its enabled, or reclamation
>> +      will fail if writeback is disabled as there isn't enough zswap space.
>> + */
>> +static int initiate_writeback(const char *cgroup, void *arg)
>> +{
>> +       char *test_group = arg;
>> +       size_t memsize = MB(4);
>> +       int ret = -1;
>> +       bool wb_enabled = cg_read_long(test_group, "memory.zswap.writeback");
>> +       long zswap_usage;
> Nit: Reverse christmas tree (i.e. is usually more aesthetically
> pleasing, but it isn't consistently followed but up to you. You can
> separate the declaration and initialization of wb_enabled to do that
> if you choose to.
>
Thanks for the review. Will change it in next revision.


>> +
>> +       if (cg_write(test_group, "memory.max", "8M"))
>> +               return ret;
> Why do we need to set this?
Not needed, will remove.
>> +
>> +       /* Allocate random memory to enusre high zswap memory usage */
> typo: ensure
>
>> +       char *mem = (char *)malloc(memsize);
>> +
>> +       if (!mem)
>> +               return ret;
>> +       for (int i = 0; i < memsize; i++)
>> +               mem[i] = rand() % 128;
> I am curious, what compression ratio do you observe in practice with
> this? Is there a risk of pages becoming totally incompressible and
> skipping zswap? One way to guarantee the page compressibility is to
> add a bunch of zeros. I usually fill half of it with zeros and half of
> it with random data. Not sure if this is actually required though.
>
With the default zswap parameters, zswap.current is 3491645, so about 
1.2:1. I had tried a few different zswap parameters and compression was 
still taking place. I initially tried the method from allocate_bytes, 
but the compression was too high with all the zeros and zswap was a 
really small value. Will switch to your method in next revision.

>> +
>> +       /* Try and reclaim allocated memory */
>> +       if (cg_write(test_group, "memory.reclaim", "4M")) {
>> +               ksft_print_msg("Failed to reclaim all of the requested memory\n");
>> +               goto out;
>> +       }
>> +
>> +       zswap_usage = cg_read_long(test_group, "memory.zswap.current");
>> +
>> +       /* zswpin */
>> +       for (int i = 0; i < memsize; i++) {
>> +               if (mem[i] < 0 || mem[i] > 127) {
>> +                       ksft_print_msg("invalid memory\n");
>> +                       ret = -1;
>> +               }
>> +       }
> I understand this correctness check is not strictly needed, but it is
> very weak right now. There is a 50% chance that corruption will be
> missed because the range we are checking is half the possible values.
>
> If we want a reliable correctness check, perhaps we should fill the
> page with increasing known values instead. Then after zswpin, we can
> check that the data is exactly what we filled in the first place.
>
> Is there any value in using random data here? If there is, we can
> store a second copy of the array to compare against instead.
So my thought over here was not to do a correctness check for zswap, 
just to access memory to do zswpin. Switching to the method in the 
comment above, we can do a correctness check as well, so will add that 
in next revision.
>> +
>> +       if (cg_write_numeric(test_group, "memory.zswap.max", zswap_usage/2))
>> +               goto out;
>> +
>> +       /*
>> +        * If writeback is enabled, trying to reclaim memory now will trigger a
>> +        * writeback as zswap.max is half of what was needed when reclaim ran the first time.
>> +        * If writeback is disabled, memory reclaim will fail as zswap is limited and
>> +        * it can't writeback to swap.
>> +        */
>> +       ret = cg_write(test_group, "memory.reclaim", "4M");
>> +       if (!wb_enabled && ret)
> Should we assert that ret is -EBUSY if !wb_enabled instead? Right now
> any error code will pass. The test will also pass if reclaim succeeds
> while writeback is disabled, which is not correct behavior.
I believe its -EAGAIN, but yes will change it.
>> +               ret = 0;
>> +out:
>> +       free(mem);
>> +       return ret;
>> +}
>> +
>> +/* Test to verify the zswap writeback path */
>> +static int test_zswap_writeback(const char *root, bool wb)
>> +{
>> +       int ret = KSFT_FAIL;
>> +       char *test_group;
>> +       long zswpwb_before, zswpwb_after;
>> +
>> +       test_group = cg_name(root,
>> +               wb ? "zswap_writeback_enabled_test" : "zswap_writeback_disabled_test");
> Why do we need different cgroup names? The tests do not run in
> parallel, do they?
>
It makes the tests independent of each other (for e.g. if cg_destroy for 
any magical reason fails for one of the tests, the cgroup creation of 
the other test wont be affected). Plus, I found it easier for debugging 
to examine cgroups after the test was executed. But no strong 
preference, will change it to one name.

>> +       if (!test_group)
>> +               goto out;
>> +       if (cg_create(test_group))
>> +               goto out;
>> +       if (cg_write(test_group, "memory.zswap.writeback", wb ? "1" : "0"))
>> +               return ret;
>> +
>> +       zswpwb_before = get_cg_wb_count(test_group);
>> +       if (zswpwb_before < 0) {
> Should we assert zswpwb_before == 0 instead?
Sure, will do in next revision.
>> +               ksft_print_msg("failed to get zswpwb_before\n");
>> +               goto out;
>> +       }
>> +
>> +       if (cg_run(test_group, initiate_writeback, (void *) test_group))
> Nit: initiate_writeback() is not a very descriptive name because it
> may or may not trigger writeback.
>
>> +               goto out;
>> +
>> +       /* Verify that zswap writeback occurred only if writeback was enabled */
>> +       zswpwb_after = get_cg_wb_count(test_group);
>> +       if (wb) {
>> +               if (zswpwb_after <= zswpwb_before) {
> If we assert zswpwb_before == 0 above, this can be simplified. Also, I
> think a single condition is enough, the message contents can be
> inferred by which test failed.
>
>> +                       ksft_print_msg("writeback enabled and zswpwb_after <= zswpwb_before\n");
>> +                       goto out;
>> +               }
>> +       } else {
>> +               if (zswpwb_after != zswpwb_before) {
>> +                       ksft_print_msg("writeback disabled and zswpwb_after != zswpwb_before\n");
>> +                       goto out;
>> +               }
>> +       }
>> +
>> +       ret = KSFT_PASS;
>> +
>> +out:
>> +       cg_destroy(test_group);
>> +       free(test_group);
>> +       return ret;
>> +}
>> +
>> +static int test_zswap_writeback_enabled(const char *root)
>> +{
>> +       return test_zswap_writeback(root, true);
>> +}
>> +
>> +static int test_zswap_writeback_disabled(const char *root)
>> +{
>> +       return test_zswap_writeback(root, false);
>> +}
>> +
>>   /*
>>    * When trying to store a memcg page in zswap, if the memcg hits its memory
>>    * limit in zswap, writeback should affect only the zswapped pages of that
>> @@ -425,6 +546,8 @@ struct zswap_test {
>>          T(test_zswap_usage),
>>          T(test_swapin_nozswap),
>>          T(test_zswapin),
>> +       T(test_zswap_writeback_enabled),
>> +       T(test_zswap_writeback_disabled),
>>          T(test_no_kmem_bypass),
>>          T(test_no_invasive_cgroup_shrink),
>>   };
>> --
>> 2.43.0
>>

Powered by blists - more mailing lists