Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Reply-To: crecklin@redhat.com
To:     "Matthew Wilcox (Oracle)" <willy@infradead.org>,
        LKML <linux-kernel@vger.kernel.org>
From:   Chris von Recklinghausen <crecklin@redhat.com>
Subject: tools/testing/radix-tree/idr-test gets a failed assertion on single
 cpu systems
Organization: Red Hat
Message-ID: <def0fd33-da11-6283-d98f-66eb9a8dd201@redhat.com>
Date:   Fri, 26 Mar 2021 14:23:39 -0400
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.8.0
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Precedence: bulk

Hi Matthew,

I made the observation that while tools/testing/radix-tree/idr-test runs 
and passes just fine on a system with more than one cpu, it gets an 
assertion failure when run on a single cpu system. My test system is 
Fedora 34 running on an x86_64 system. It can be easily reproduced by 
offlining all cpus but cpu0.

[root@hpe-ml110g7-01 linux]# tools/testing/radix-tree/idr-test
vvv Ignore these warnings
assertion failed at idr.c:250
assertion failed at idr.c:206
^^^ Warnings over
idr-test: idr-test.c:320: idr_find_test_1: Assertion `!(entry != 
xa_mk_value(id))' failed.
Aborted (core dumped)

I bisected the change to 5c089fd0c734 ("idr: Fix idr_get_next race with 
idr_remove").

Since idr_get_next can return NULL, I stuck a BUG_ON(!entry) just above 
the failing assert, and in this case idr_get_next is returning NULL.

Next, I stuck a BUG_ON in the place that idr_get_next_ul returns NULL 
and commented out the contents of idr_u32_test1 so we're not knowingly 
passing it bad values, and we seem to fail because the list has been 
gone through.

void *idr_get_next_ul(struct idr *idr, unsigned long *nextid)
{
     struct radix_tree_iter iter;
     void __rcu **slot;
     void *entry = NULL;
     unsigned long base = idr->idr_base;
     unsigned long id = *nextid;

     id = (id < base) ? 0 : id - base;
     radix_tree_for_each_slot(slot, &idr->idr_rt, &iter, id) {
         entry = rcu_dereference_raw(*slot);
         if (!entry)
             continue;
         if (!xa_is_internal(entry))
             break;
         if (slot != &idr->idr_rt.xa_head && !xa_is_retry(entry))
             break;
         slot = radix_tree_iter_retry(&iter);
     }
     if (!slot)
         return NULL; <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<

     *nextid = iter.index + base;
     return entry;
}
EXPORT_SYMBOL(idr_get_next_ul);

I'm not sure if this is a test issue or possibly an issue with user 
level RCU when there's only a single cpu in the system, but I figured it 
was worth bringing it to your attention. If there's anything I can do to 
help to further analyze this or try out a fix, I'm happy to help.

Thanks,

Chris von Recklinghausen

Red Hat