Re: [PATCH v13 2/8] PCI/DPC: Allow dpc_probe() even if firmware first mode is enabled
From: Kuppuswamy Sathyanarayanan
Date: Thu Jan 23 2020 - 13:18:29 EST
On 1/22/20 7:10 PM, Bjorn Helgaas wrote:
On Wed, Jan 22, 2020 at 04:42:48PM -0800, Kuppuswamy Sathyanarayanan wrote:
Hi Bjorn,
On 1/22/20 3:17 PM, Bjorn Helgaas wrote:
On Sat, Jan 18, 2020 at 08:00:31PM -0800, sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx wrote:
From: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
As per ACPI specification v6.3, sec 5.6.6, Error Disconnect Recover
(EDR) notification used by firmware to let OS know about the DPC event
and permit OS to perform error recovery when processing the EDR
notification.
I want to clear up some of this description because this is a very
confusing area and we should follow the spec carefully so we all
understand each other.
Also, as per PCI firmware specification r3.2 Downstream
Port Containment Related Enhancements ECN, sec 4.5.1, table 4-6, if DPC
is controlled by firmware (firmware first mode), it's responsible for
initializing Downstream Port Containment Extended Capability Structures
per firmware policy.
The above is actually not what the ECN says. What it does say is
this:
If control of this feature [DPC configuration] was requested and
denied, firmware is responsible for initializing Downstream Port
Containment Extended Capability Structures per firmware policy.
Neither the PCI Firmware Spec (r3.2) nor the ECN contains any
reference to "firmware first". As far as I can tell, "Firmware First"
is defined by the ACPI spec, and the FIRMWARE_FIRST bit in various
HEST entries (sec 18.3.2) is what tells us when a device is in
firmware-first mode.
That's a really long way of saying that from a spec point of view, DPC
being controlled by firmware does NOT imply anything about
firmware-first mode.
But current AER and DPC driver uses ACPI FIRMWARE_FIRST bit
(in pcie_aer_get_firmware_first()) to decide whether AER/DPC is
controlled by firmware or OS. That's why I used the term
"firmware first mode" interchangeably with a mode in which
DPC/AER is controlled by firmware.
Yes. I think the current use of pcie_aer_get_firmware_first() there
is probably not optimal and we should change that so it corresponds
better with the spec.
And, OS is permitted to read or write DPC Control
and Status registers of a port while processing an Error Disconnect
Recover (EDR) notification from firmware on that port.
Currently, if firmware controls DPC (firmware first mode), OS will not
create/enumerate DPC PCIe port services. But, if OS supports EDR
feature, then as mentioned in above spec references, it should permit
enumeration of DPC driver and also support handling ACPI EDR
notification. So as first step, allow dpc_probe() to continue even if
firmware first mode is enabled. Also add appropriate checks to ensure
device registers are not modified outside EDR notification window in
firmware first mode. This is a preparatory patch for adding EDR support.
Signed-off-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@xxxxxxxxxxxxxxx>
Acked-by: Keith Busch <keith.busch@xxxxxxxxx>
---
drivers/pci/pcie/dpc.c | 74 ++++++++++++++++++++++++++++++++++--------
1 file changed, 61 insertions(+), 13 deletions(-)
diff --git a/drivers/pci/pcie/dpc.c b/drivers/pci/pcie/dpc.c
index e06f42f58d3d..c583a90fa90d 100644
--- a/drivers/pci/pcie/dpc.c
+++ b/drivers/pci/pcie/dpc.c
@@ -22,6 +22,7 @@ struct dpc_dev {
u16 cap_pos;
bool rp_extensions;
u8 rp_log_size;
+ bool edr_enabled; /* EDR mode is supported */
This needs a better name, or perhaps it should be removed completely.
Any suggestions ? native_mode or firmware_dpc ?
EDR is not a mode that can be enabled or disabled. The _OSC "EDR
supported" bit tells the firmware whether the OS supports EDR
notification, but there's no corresponding bit that tells the OS
whether firmware will actually *send* those notifications.
I agree that EDR is not a mode. But with EDR support, DPC driver functions
can be triggered/executed in two contexts.
1. DPC is controlled by OS.
2. Via EDR notification if DPC is controlled by firmware.
Since we want to use same code for both scenarios, we need some method
to detect which context is currently in use.
I haven't gotten as far as looking at the actual EDR support yet, but
I wonder if this could be rearranged so that dpc.c contained two
things: (1) the existing DPC driver that claims DPC capability via
dpc_probe() and (2) a few functions exported for use by the EDR path.
Then maybe the EDR path could live in pci-acpi.c (or maybe a new edr.c
file) and would not depend on dpc_probe() actually *claiming* the
capability.
In EDR code path, we depend on common pcie_do_recovery() function in
pcie/err.c for doing the error recovery, and this function has
dependency (related to callbacks) on service (DPC/AER) from
which error recovery is triggered. So if we plan to separate EDR code
path from
DPC by exporting required functions, we might have to do similar thing
in err.c.
There will be also be some code duplication if we choose this path. So,
I am not
sure whether there is enough gain in taking this approach. Let me know
your comments.
Also, EDR is an ACPI concept, and dpc.c is a generic driver not
specific to ACPI, so we should try to avoid polluting it with
ACPI-isms.
};
static const char * const rp_pio_error_string[] = {
@@ -69,6 +70,14 @@ void pci_save_dpc_state(struct pci_dev *dev)
if (!dpc)
return;
+ /*
+ * If DPC is controlled by firmware then save/restore tasks are also
+ * controlled by firmware. So skip rest of the function if DPC is
+ * controlled by firmware.
+ */
+ if (dpc->edr_enabled)
+ return;
I think this should be something like:
if (!host->native_dpc)
return;
That's what the spec says: if firmware does not grant control of DPC
to the OS via _OSC, the OS may not read/write the DPC capability.
The usual situation would be that if firmware doesn't grant the OS
permission to use DPC, we wouldn't use the dpc.c driver at all.
But IIUC, the point of this EDR stuff is that we're adding a case
where the OS doesn't have permission to use DPC, but we *do* want to
use parts of dpc.c in limited cases while handling EDR notifications.
Yes, your assumption is correct.
I think you should probably take the DPC-related _OSC stuff from the
last patch ("PCI/ACPI: Enable EDR support") and move it before this
patch, so it *only* negotiates DPC ownership, per the ECN. That would
probably result in dpc.c being used only if _OSC grants DPC ownership
to the OS.
Patch titled ("PCI/ACPI: Enable EDR support") exposes EDR support to
firmware in _OSC negotiation. So you wanted me to move this patch to
the end of the series to make sure we don't expose EDR support until
we have necessary code changes in kernel.
It actually does *two* things: (1) adds OSC_PCI_EXPRESS_DPC_CONTROL
and native_dpc, and (2) adds OSC_PCI_EDR_SUPPORT and related code. I
think these should be split into two patches, and part 1 is what I
think could be done early before adding EDR support.
ok. makes sense.
+ * As per PCIe r5.0, sec 6.2.10, implementation note titled
+ * "Determination of DPC Control", to avoid conflicts over whether
+ * platform firmware or the operating system have control of DPC,
+ * it is recommended that platform firmware and operating systems
+ * always link the control of DPC to the control of Advanced Error
+ * Reporting.
+ *
+ * So use AER FF mode check API pcie_aer_get_firmware_first() to decide
+ * whether DPC is controlled by software or firmware.
AFAICT, this is not what the spec says. Firmware-first is not what
tells us whether DPC is controlled by firmware or the OS. For ACPI
systems, _OSC tells us whether the OS is allowed control of DPC.
But current DPC/AER driver use FIRMWARE_FIRST bit (in
pcie_aer_get_firmware_first()) determine firmware/OS ownership.
I think that's a problem and we should fix it. It's arguably a little
bit of feature creep to do that in this series, but it makes it very
confusing when the specs talk about ownership negotiated via _OSC and
the code talks about "firmware-first" and makes assumptions about
connections between the two.
Those things may very well *be* related in the firmware
implementation, but assuming those connections in Linux makes the code
hard to read and maintain. This series adds quite a bit of
ACPI-related stuff, and I think we should untangle confusions like
this before we add more.
Bjorn
--
Sathyanarayanan Kuppuswamy
Linux kernel developer