Author: mw
Date: Tue May 26 15:54:32 2020
New Revision: 361524
URL: https://svnweb.freebsd.org/changeset/base/361524
Log:
Fix assumptions about number of IO queues in the ENA
Make the ena_adapter::num_io_queues a number of effectively used IO
queues. While the ena_adapter::max_num_io_queues is an upper-bound
specified by the HW, the ena_adapter::num_io_queues may be lower than
that, depending on runtime system resources availability.
On reset, there are called ena_destroy_device() and then
ena_restore_device(). The latter calls, in turn, ena_enable_msix(),
which will attempt to re-acquire ena_adapter::max_num_io_queues of
MSIX vectors again.
Thus, the value of ena_adapter::num_io_queues may be different before
and after reset. For this reason, free the IO rings structures (drbr,
counters) in ena_destroy_device() and allocate again in
ena_restore_device().
Submitted by: Maciej Bielski <m...@semihalf.com>
Obtained from: Semihalf
Sponsored by: Amazon, Inc.
Modified:
head/sys/dev/ena/ena.c
Modified: head/sys/dev/ena/ena.c
==============================================================================
--- head/sys/dev/ena/ena.c Tue May 26 15:50:30 2020 (r361523)
+++ head/sys/dev/ena/ena.c Tue May 26 15:54:32 2020 (r361524)
@@ -1562,7 +1562,6 @@ ena_enable_msix(struct ena_adapter *adapter)
}
device_printf(dev, "Enable only %d MSI-x (out of %d), reduce "
"the number of queues\n", msix_vecs, msix_req);
- adapter->num_io_queues = msix_vecs - ENA_ADMIN_MSIX_VEC;
}
adapter->msix_vecs = msix_vecs;
@@ -3116,6 +3115,15 @@ ena_destroy_device(struct ena_adapter *adapter, bool g
ena_disable_msix(adapter);
+ /*
+ * IO rings resources should be freed because `ena_restore_device()`
+ * calls (not directly) `ena_enable_msix()`, which re-allocates MSIX
+ * vectors. The amount of MSIX vectors after destroy-restore may be
+ * different than before. Therefore, IO rings resources should be
+ * established from scratch each time.
+ */
+ ena_free_all_io_rings_resources(adapter);
+
ena_com_abort_admin_commands(ena_dev);
ena_com_wait_for_abort_completion(ena_dev);
@@ -3192,6 +3200,18 @@ ena_restore_device(struct ena_adapter *adapter)
goto err_device_destroy;
}
+ /*
+ * Effective value of used MSIX vectors should be the same as before
+ * `ena_destroy_device()`, if possible, or closest to it if less vectors
+ * are available.
+ */
+ if ((adapter->msix_vecs - ENA_ADMIN_MSIX_VEC) < adapter->num_io_queues)
+ adapter->num_io_queues =
+ adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
+
+ /* Re-initialize rings basic information */
+ ena_init_io_rings(adapter);
+
/* If the interface was up before the reset bring it up */
if (ENA_FLAG_ISSET(ENA_FLAG_DEV_UP_BEFORE_RESET, adapter)) {
rc = ena_up(adapter);
@@ -3379,7 +3399,6 @@ ena_attach(device_t pdev)
adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
adapter->max_num_io_queues = max_num_io_queues;
- adapter->num_io_queues = max_num_io_queues;
adapter->buf_ring_size = ENA_DEFAULT_BUF_RING_SIZE;
@@ -3400,15 +3419,26 @@ ena_attach(device_t pdev)
goto err_tx_tag_free;
}
- /* initialize rings basic information */
- ena_init_io_rings(adapter);
-
+ /*
+ * The amount of requested MSIX vectors is equal to
+ * adapter::max_num_io_queues (see `ena_enable_msix()`), plus a constant
+ * number of admin queue interrupts. The former is initially determined
+ * by HW capabilities (see `ena_calc_max_io_queue_num())` but may not be
+ * achieved if there are not enough system resources. By default, the
+ * number of effectively used IO queues is the same but later on it can
+ * be limited by the user using sysctl interface.
+ */
rc = ena_enable_msix_and_set_admin_interrupts(adapter);
if (unlikely(rc != 0)) {
device_printf(pdev,
"Failed to enable and set the admin interrupts\n");
goto err_io_free;
}
+ /* By default all of allocated MSIX vectors are actively used */
+ adapter->num_io_queues = adapter->msix_vecs - ENA_ADMIN_MSIX_VEC;
+
+ /* initialize rings basic information */
+ ena_init_io_rings(adapter);
/* setup network interface */
rc = ena_setup_ifnet(pdev, adapter, &get_feat_ctx);
_______________________________________________
svn-src-head@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
