Our Marvin testing has bisected down a regression in reaim-io-disk workload on
ext4 and xfs filesystems to:

5a9d041ba2f6da ("block: move io_context creation into where it's needed")

The differences look like:

Comparison
==========
                                  initial                initial               
 last                  penup                   last                  penup     
            first
                               good-v5.16              bad-v5.17        
bad-f46b81c5           bad-eca5892a          good-48b5c1fb         
good-25c4b5e0           bad-5a9d041b
Hmean     disk-1       1325.67 (   0.00%)     1308.33 *  -1.31%*     1300.96 *
-1.86%*     1304.35 (  -1.61%)     1315.21 (  -0.79%)     1305.48 *  -1.52%*   
 1314.06 (  -0.88%)
Hmean     disk-41     56138.75 (   0.00%)    54280.67 *  -3.31%*    53711.79 *
-4.32%*    53829.32 *  -4.11%*    55959.96 (  -0.32%)    55555.56 (  -1.04%)   
54089.71 *  -3.65%*
Hmean     disk-81     91284.75 (   0.00%)    86600.14 *  -5.13%*    84698.50 *
-7.22%*    84639.50 *  -7.28%*    91628.96 (   0.38%)    89469.81 (  -1.99%)   
84905.66 *  -6.99%*
Hmean     disk-121   111486.49 (   0.00%)   103242.32 *  -7.39%*   102513.41 *
-8.05%*   102340.00 *  -8.20%*   113543.95 (   1.85%)   112873.13 (   1.24%)  
102600.34 *  -7.97%*
Hmean     disk-161   125748.50 (   0.00%)   125683.06 (  -0.05%)   126439.79 ( 
0.55%)   125129.53 (  -0.49%)   125552.38 (  -0.16%)   125683.06 (  -0.05%)  
125978.09 (   0.18%)
Hmean     disk-201   141416.51 (   0.00%)   130097.09 *  -8.00%*   127943.99 *
-9.53%*   128107.08 *  -9.41%*   143162.39 *   1.23%*   141715.63 (   0.21%)  
128270.58 *  -9.30%*
Hmean     disk-241   154851.14 (   0.00%)   137583.26 * -11.15%*   137740.52
*-11.05%*   137374.12 * -11.29%*   154652.41 (  -0.13%)   154553.23 (  -0.19%) 
 139065.21 * -10.19%*
Hmean     disk-281   172427.90 (   0.00%)   154820.94 * -10.21%*   154650.53
*-10.31%*   155077.26 * -10.06%*   171690.43 (  -0.43%)   173922.01 (   0.87%) 
 154480.49 * -10.41%*