Re: [PATCH 11/14] fsi: Improve master indexing

From
Joel Stanley <joel@jms.id.au>
Date
2023-08-09 11:55:34
Also in
lkml
Reply to
Joel Stanley, 2023-08-09
On Wed, 9 Aug 2023 at 07:08, Joel Stanley [off-list ref] wrote:
On Mon, 12 Jun 2023 at 19:57, Eddie James [off-list ref] wrote:
quoted
Master indexing is problematic if a hub is rescanned while the
root master is being rescanned. Move the IDA free below the device
unregistration, lock the scan mutex in the probe function, and
request a specific idx in the hub driver.
I've applied this series, but taking a closer look at this patch I
think it can be improved. If you resend, just send this patch.
On hardware, it did this at FSI scan time:

 WARNING: CPU: 0 PID: 761 at /lib/idr.c:525 ida_free+0x140/0x154
 ida_free called for id=1 which is not allocated.
 CPU: 0 PID: 761 Comm: openpower-proc- Not tainted 6.1.34-d42f59e #1
 Hardware name: Generic DT based system
  unwind_backtrace from show_stack+0x18/0x1c
  show_stack from dump_stack_lvl+0x24/0x2c
  dump_stack_lvl from __warn+0x74/0xf0
  __warn from warn_slowpath_fmt+0x9c/0xd8
  warn_slowpath_fmt from ida_free+0x140/0x154
  ida_free from fsi_master_register+0xd0/0xf0
  fsi_master_register from hub_master_probe+0x11c/0x358
  hub_master_probe from really_probe+0xd4/0x3f0
  really_probe from driver_probe_device+0x38/0xd0
  driver_probe_device from __device_attach_driver+0xc8/0x148
  __device_attach_driver from bus_for_each_drv+0x90/0xdc
  bus_for_each_drv from __device_attach+0x114/0x1a4
  __device_attach from bus_probe_device+0x8c/0x94
  bus_probe_device from device_add+0x3a8/0x7fc
  device_add from fsi_master_scan+0x4e0/0x950
  fsi_master_scan from fsi_master_rescan+0x38/0x88
  fsi_master_rescan from master_rescan_store+0x14/0x20
  master_rescan_store from kernfs_fop_write_iter+0x114/0x200
  kernfs_fop_write_iter from vfs_write+0x1d0/0x374
  vfs_write from ksys_write+0x78/0x100
  ksys_write from ret_fast_syscall+0x0/0x54
 Exception stack(0x9fc51fa8 to 0x9fc51ff0)
 1fa0:                   00000001 01a01c78 00000003 01a01c78 00000001 00000001
 1fc0: 00000001 01a01c78 00000001 00000004 7eeb4ab0 7eeb4b3c 7eeb4ab4 7eeb499c
 1fe0: 76985abc 7eeb4928 76848af8 766f176c

quoted
Signed-off-by: Eddie James <eajames@linux.ibm.com>
---
 drivers/fsi/fsi-core.c       | 41 ++++++++++++++++++++++--------------
 drivers/fsi/fsi-master-hub.c |  2 ++
 2 files changed, 27 insertions(+), 16 deletions(-)
diff --git a/drivers/fsi/fsi-core.c b/drivers/fsi/fsi-core.c
index ec4d02264391..503061a6740b 100644
--- a/drivers/fsi/fsi-core.c
+++ b/drivers/fsi/fsi-core.c
@@ -1327,46 +1327,55 @@ static struct class fsi_master_class = {
 int fsi_master_register(struct fsi_master *master)
 {
        int rc;
-       struct device_node *np;

        mutex_init(&master->scan_lock);
-       master->idx = ida_alloc(&master_ida, GFP_KERNEL);
+
+       if (master->idx) {
Why do we allocate a new idx if there's already one?
quoted
+               master->idx = ida_alloc_range(&master_ida, master->idx,
+                                             master->idx, GFP_KERNEL);
If we can't get one in the range we want, we ask for any? Should this
print a warning?
quoted
+               if (master->idx < 0)
+                       master->idx = ida_alloc(&master_ida, GFP_KERNEL);
+       } else {
If ixd was zero, we create one. This is the "normal" case?
quoted
+               master->idx = ida_alloc(&master_ida, GFP_KERNEL);
+       }
+
We check the same error condition again.
quoted
        if (master->idx < 0)
                return master->idx;
quoted

-       dev_set_name(&master->dev, "fsi%d", master->idx);
+       if (!dev_name(&master->dev))
+               dev_set_name(&master->dev, "fsi%d", master->idx);
+
        master->dev.class = &fsi_master_class;

+       mutex_lock(&master->scan_lock);
        rc = device_register(&master->dev);
        if (rc) {
                ida_free(&master_ida, master->idx);
-               return rc;
-       }
+       } else {
+               struct device_node *np = dev_of_node(&master->dev);
This change looks a bit different to the idx changes. What's happening here?
quoted

-       np = dev_of_node(&master->dev);
-       if (!of_property_read_bool(np, "no-scan-on-init")) {
-               mutex_lock(&master->scan_lock);
-               fsi_master_scan(master);
-               mutex_unlock(&master->scan_lock);
+               if (!of_property_read_bool(np, "no-scan-on-init"))
+                       fsi_master_scan(master);
        }

-       return 0;
+       mutex_unlock(&master->scan_lock);
+       return rc;
 }
 EXPORT_SYMBOL_GPL(fsi_master_register);

 void fsi_master_unregister(struct fsi_master *master)
 {
-       trace_fsi_master_unregister(master);
+       int idx = master->idx;

-       if (master->idx >= 0) {
-               ida_free(&master_ida, master->idx);
-               master->idx = -1;
-       }
+       trace_fsi_master_unregister(master);

        mutex_lock(&master->scan_lock);
        fsi_master_unscan(master);
+       master->n_links = 0;
        mutex_unlock(&master->scan_lock);
+
        device_unregister(&master->dev);
+       ida_free(&master_ida, idx);
 }
 EXPORT_SYMBOL_GPL(fsi_master_unregister);
diff --git a/drivers/fsi/fsi-master-hub.c b/drivers/fsi/fsi-master-hub.c
index 6d8b6e8854e5..36da643b3201 100644
--- a/drivers/fsi/fsi-master-hub.c
+++ b/drivers/fsi/fsi-master-hub.c
@@ -12,6 +12,7 @@
 #include <linux/slab.h>

 #include "fsi-master.h"
+#include "fsi-slave.h"

 #define FSI_ENGID_HUB_MASTER           0x1c
@@ -229,6 +230,7 @@ static int hub_master_probe(struct device *dev)
        hub->master.dev.release = hub_master_release;
        hub->master.dev.of_node = of_node_get(dev_of_node(dev));

+       hub->master.idx = fsi_dev->slave->link + 1;
        hub->master.n_links = links;
        hub->master.read = hub_master_read;
        hub->master.write = hub_master_write;
--
2.31.1
Thread (21 messages) 21 messages, 3 authors, last reply 2023-08-17
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help