| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20211017161958.44351-18-kitakar@gmail.com>
Date: Mon, 18 Oct 2021 01:19:57 +0900
From: Tsuchiya Yuto <kitakar@...il.com>
To: unlisted-recipients:; (no To-header on input)
Cc: Hans de Goede <hdegoede@...hat.com>,
Patrik Gfeller <patrik.gfeller@...il.com>,
Alan <alan@...ux.intel.com>, Tsuchiya Yuto <kitakar@...il.com>,
Mauro Carvalho Chehab <mchehab@...nel.org>,
Sakari Ailus <sakari.ailus@...ux.intel.com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Hans Verkuil <hverkuil-cisco@...all.nl>,
Laurent Pinchart <laurent.pinchart@...asonboard.com>,
Arnd Bergmann <arnd@...db.de>,
Tomi Valkeinen <tomi.valkeinen@...asonboard.com>,
Yang Yingliang <yangyingliang@...wei.com>,
Aline Santana Cordeiro <alinesantanacordeiro@...il.com>,
Dinghao Liu <dinghao.liu@....edu.cn>,
linux-media@...r.kernel.org, linux-staging@...ts.linux.dev,
linux-kernel@...r.kernel.org
Subject: [PATCH 17/17] [NOT-FOR-MERGE] atomisp: Fix up the open v load race
From: Alan <alan@...ux.intel.com>
Applied this patch as-is for now to pass firmware load on probe.
I haven't looked into this and not written by me anyway.
Thus, NOT-FOR-MERGE. At least I can say this patch works just fine.
>8-----------------------------------------------------------------8<
From: Alan <alan@...ux.intel.com>
Patch from https://lore.kernel.org/linux-media/151001137594.77201.4306351721772580664.stgit@alans-desktop/raw
Here is the original commit message:
Subject: [PATCH 1/3] atomisp: Fix up the open v load race
From: Alan <alan@...ux.intel.com>
Date: Mon, 06 Nov 2017 23:36:36 +0000
This isn't the ideal final solution but it stops the main problem for now
where an open (often from udev) races the device initialization and we try
and load the firmware twice at the same time. This needless to say doesn't
usually end well.
Signed-off-by: Alan Cox <alan@...ux.intel.com>
This patch fixes the following "css init failed" error related to
firmware load. This issue often occurs when I insmod the module after
boot.
kern :err : [ 234.658869] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.755023] atomisp-isp2 0000:00:03.0: can't change power state from D0 to D3hot (config space inaccessible)
kern :err : [ 234.755053] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.755581] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.756146] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.757115] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.757582] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.758885] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.760317] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.760818] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.769102] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.769527] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.771353] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.771804] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.772299] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.774617] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.775090] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.775607] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.776230] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.779059] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
kern :err : [ 234.787726] atomisp-isp2 0000:00:03.0: can't change power state from D3cold to D0 (config space inaccessible)
kern :err : [ 234.788078] atomisp-isp2 0000:00:03.0: css init failed --- bad firmware?
Patch is adapted to v5.15-rc4.
Signed-off-by: Tsuchiya Yuto <kitakar@...il.com>
---
drivers/staging/media/atomisp/pci/atomisp_fops.c | 12 ++++++++++++
drivers/staging/media/atomisp/pci/atomisp_internal.h | 5 +++++
drivers/staging/media/atomisp/pci/atomisp_v4l2.c | 6 ++++++
3 files changed, 23 insertions(+)
diff --git a/drivers/staging/media/atomisp/pci/atomisp_fops.c b/drivers/staging/media/atomisp/pci/atomisp_fops.c
index f82bf082aa79..62535e0f4429 100644
--- a/drivers/staging/media/atomisp/pci/atomisp_fops.c
+++ b/drivers/staging/media/atomisp/pci/atomisp_fops.c
@@ -772,6 +772,18 @@ static int atomisp_open(struct file *file)
dev_dbg(isp->dev, "open device %s\n", vdev->name);
+ /* Ensure that if we are still loading we block. Once the loading
+ is over we can proceed. We can't blindly hold the lock until
+ that occurs as if the load fails we'll deadlock the unload */
+ rt_mutex_lock(&isp->loading);
+ /* Revisit this with a better check once the code structure is
+ cleaned up a bit more FIXME */
+ if (!isp->ready) {
+ rt_mutex_unlock(&isp->loading);
+ return -ENXIO;
+ }
+ rt_mutex_unlock(&isp->loading);
+
rt_mutex_lock(&isp->mutex);
acc_node = !strcmp(vdev->name, "ATOMISP ISP ACC");
diff --git a/drivers/staging/media/atomisp/pci/atomisp_internal.h b/drivers/staging/media/atomisp/pci/atomisp_internal.h
index c01db10bb735..356a7f1bb757 100644
--- a/drivers/staging/media/atomisp/pci/atomisp_internal.h
+++ b/drivers/staging/media/atomisp/pci/atomisp_internal.h
@@ -246,6 +246,11 @@ struct atomisp_device {
/* Purpose of mutex is to protect and serialize use of isp data
* structures and css API calls. */
struct rt_mutex mutex;
+ /* This mutex ensures that we don't allow an open to succeed while
+ * the initialization process is incomplete */
+ struct rt_mutex loading;
+ /* Set once the ISP is ready to allow opens */
+ bool ready;
/*
* Serialise streamoff: mutex is dropped during streamoff to
* cancel the watchdog queue. MUST be acquired BEFORE
diff --git a/drivers/staging/media/atomisp/pci/atomisp_v4l2.c b/drivers/staging/media/atomisp/pci/atomisp_v4l2.c
index 720963156d24..e2ccace77a73 100644
--- a/drivers/staging/media/atomisp/pci/atomisp_v4l2.c
+++ b/drivers/staging/media/atomisp/pci/atomisp_v4l2.c
@@ -1567,6 +1567,7 @@ static int atomisp_pci_probe(struct pci_dev *pdev, const struct pci_device_id *i
dev_dbg(&pdev->dev, "atomisp mmio base: %p\n", isp->base);
rt_mutex_init(&isp->mutex);
+ rt_mutex_init(&isp->loading);
mutex_init(&isp->streamoff_mutex);
spin_lock_init(&isp->lock);
@@ -1749,6 +1750,8 @@ static int atomisp_pci_probe(struct pci_dev *pdev, const struct pci_device_id *i
pci_write_config_dword(pdev, MRFLD_PCI_CSI_AFE_TRIM_CONTROL, csi_afe_trim);
}
+ rt_mutex_lock(&isp->loading);
+
err = atomisp_initialize_modules(isp);
if (err < 0) {
dev_err(&pdev->dev, "atomisp_initialize_modules (%d)\n", err);
@@ -1806,6 +1809,8 @@ static int atomisp_pci_probe(struct pci_dev *pdev, const struct pci_device_id *i
release_firmware(isp->firmware);
isp->firmware = NULL;
isp->css_env.isp_css_fw.data = NULL;
+ isp->ready = true;
+ rt_mutex_unlock(&isp->loading);
atomisp_drvfs_init(isp);
@@ -1825,6 +1830,7 @@ static int atomisp_pci_probe(struct pci_dev *pdev, const struct pci_device_id *i
register_entities_fail:
atomisp_uninitialize_modules(isp);
initialize_modules_fail:
+ rt_mutex_unlock(&isp->loading);
cpu_latency_qos_remove_request(&isp->pm_qos);
atomisp_msi_irq_uninit(isp);
pci_free_irq_vectors(pdev);
--
2.33.1
Powered by blists - more mailing lists