EDAC/ghes: Clear scanned data on unload

Commit

  b972fdb ("EDAC/ghes: Fix NULL pointer dereference in ghes_edac_register()")

didn't clear all the information from the scanned system and, more
specifically, left ghes_hw.num_dimms to its previous value. On a
second load (CONFIG_DEBUG_TEST_DRIVER_REMOVE=y), the driver would use
the leftover num_dimms value which is not 0 and thus the 0 check in
enumerate_dimms() will get bypassed and it would go directly to the
pointer deref:

  d = &hw->dimms[hw->num_dimms];

which is, of course, NULL:

  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] PREEMPT SMP
  CPU: 7 PID: 1 Comm: swapper/0 Not tainted 5.9.0-rc4+ #7
  Hardware name: GIGABYTE MZ01-CE1-00/MZ01-CE1-00, BIOS F02 08/29/2018
  RIP: 0010:enumerate_dimms.cold+0x7b/0x375

Reset the whole ghes_hw on driver unregister so that no stale values are
used on a second system scan.

Fixes: b972fdb ("EDAC/ghes: Fix NULL pointer dereference in ghes_edac_register()")
Cc: Shiju Jose <shiju.jose@huawei.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200911164817.GA19320@zn.tnic

drivers/edac/ghes_edac.c

@@ -632,6 +632,7 @@ void ghes_edac_unregister(struct ghes *ghes)
 	mutex_lock(&ghes_reg_mutex);
 
 	system_scanned = false;
+	memset(&ghes_hw, 0, sizeof(struct ghes_hw_desc));
 
 	if (!refcount_dec_and_test(&ghes_refcount))
 		goto unlock;