| Message ID | 20170626131625.79059-1-luc.vanoostenryck@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Jun 26, 2017 at 03:16:25PM +0200, Luc Van Oostenryck wrote: > Here the entrypoint, declared as a 64 bit integer, is read from > a pointer to 64bit integer but the read is done via readl_relaxed() > which is for 32bit quantities. > > All the high bits will thus be lost which change the meaning > of the test against zero done later. > > Fix this by using readq_relaxed() instead as it should be for > 64bit quantities. > > Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> > > --- > > !!! Warning !!! > I don't have the HW to test this, nor have I read the ACPI specs. > This is only compile tested. This looks correct to me, but I'd like an Ack from Lorenzo before I take it because it's not clear to me what issues can arise if firmware only writes zero to the bottom 32 bits of the jump address and leaves the upper bits intact. If firmware does this, then this patch will result in a WARN, but I can't figure out what this part of the spec is trying to achieve. Will > --- > arch/arm64/kernel/acpi_parking_protocol.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm64/kernel/acpi_parking_protocol.c b/arch/arm64/kernel/acpi_parking_protocol.c > index 1f5655cd9..f35e80aad 100644 > --- a/arch/arm64/kernel/acpi_parking_protocol.c > +++ b/arch/arm64/kernel/acpi_parking_protocol.c > @@ -125,7 +125,7 @@ static void acpi_parking_protocol_cpu_postboot(void) > struct parking_protocol_mailbox __iomem *mailbox = cpu_entry->mailbox; > __le64 entry_point; > > - entry_point = readl_relaxed(&mailbox->entry_point); > + entry_point = readq_relaxed(&mailbox->entry_point); > /* > * Check if firmware has cleared the entry_point as expected > * by the protocol specification. > -- > 2.13.0 >
[resending as I bungled modifying the Cc list] On Mon, Jun 26, 2017 at 03:16:25PM +0200, Luc Van Oostenryck wrote: > Here the entrypoint, declared as a 64 bit integer, is read from > a pointer to 64bit integer but the read is done via readl_relaxed() > which is for 32bit quantities. > > All the high bits will thus be lost which change the meaning > of the test against zero done later. > > Fix this by using readq_relaxed() instead as it should be for > 64bit quantities. > > Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> This looks right per my reading of the spec. FWIW: Acked-by: Mark Rutland <mark.rutland@arm.com> It would be good if someone could test this on HW implementing the parking protocol. I've Cc'd those who've given Tested-by in the past. Thanks, Mark. > > --- > > !!! Warning !!! > I don't have the HW to test this, nor have I read the ACPI specs. > This is only compile tested. > --- > arch/arm64/kernel/acpi_parking_protocol.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/arm64/kernel/acpi_parking_protocol.c b/arch/arm64/kernel/acpi_parking_protocol.c > index 1f5655cd9..f35e80aad 100644 > --- a/arch/arm64/kernel/acpi_parking_protocol.c > +++ b/arch/arm64/kernel/acpi_parking_protocol.c > @@ -125,7 +125,7 @@ static void acpi_parking_protocol_cpu_postboot(void) > struct parking_protocol_mailbox __iomem *mailbox = cpu_entry->mailbox; > __le64 entry_point; > > - entry_point = readl_relaxed(&mailbox->entry_point); > + entry_point = readq_relaxed(&mailbox->entry_point); > /* > * Check if firmware has cleared the entry_point as expected > * by the protocol specification. > -- > 2.13.0 >
On Mon, Jun 26, 2017 at 03:16:25PM +0200, Luc Van Oostenryck wrote: > Here the entrypoint, declared as a 64 bit integer, is read from > a pointer to 64bit integer but the read is done via readl_relaxed() > which is for 32bit quantities. > > All the high bits will thus be lost which change the meaning > of the test against zero done later. > > Fix this by using readq_relaxed() instead as it should be for > 64bit quantities. > > Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> > > --- > > !!! Warning !!! > I don't have the HW to test this, nor have I read the ACPI specs. > This is only compile tested. > --- > arch/arm64/kernel/acpi_parking_protocol.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) Patch is correct but as Mark already mentioned I would like some testing done on systems that rely on the parking protocol (hopefully FW clears the entry point correctly but I want to understand if this patch would trigger regressions that would force us to add quirks around it - even if we know this patch does the right thing from a kernel perspective). Acked-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> > diff --git a/arch/arm64/kernel/acpi_parking_protocol.c b/arch/arm64/kernel/acpi_parking_protocol.c > index 1f5655cd9..f35e80aad 100644 > --- a/arch/arm64/kernel/acpi_parking_protocol.c > +++ b/arch/arm64/kernel/acpi_parking_protocol.c > @@ -125,7 +125,7 @@ static void acpi_parking_protocol_cpu_postboot(void) > struct parking_protocol_mailbox __iomem *mailbox = cpu_entry->mailbox; > __le64 entry_point; > > - entry_point = readl_relaxed(&mailbox->entry_point); > + entry_point = readq_relaxed(&mailbox->entry_point); > /* > * Check if firmware has cleared the entry_point as expected > * by the protocol specification. > -- > 2.13.0 >
It is good with X-Gene 1/2.
Tested-by: Khuong Dinh <kdinh@apm.com>
It is good with X-Gene 1/2.
Tested-by: Khuong Dinh <kdinh@apm.com>
diff --git a/arch/arm64/kernel/acpi_parking_protocol.c b/arch/arm64/kernel/acpi_parking_protocol.c index 1f5655cd9..f35e80aad 100644 --- a/arch/arm64/kernel/acpi_parking_protocol.c +++ b/arch/arm64/kernel/acpi_parking_protocol.c @@ -125,7 +125,7 @@ static void acpi_parking_protocol_cpu_postboot(void) struct parking_protocol_mailbox __iomem *mailbox = cpu_entry->mailbox; __le64 entry_point; - entry_point = readl_relaxed(&mailbox->entry_point); + entry_point = readq_relaxed(&mailbox->entry_point); /* * Check if firmware has cleared the entry_point as expected * by the protocol specification.
Here the entrypoint, declared as a 64 bit integer, is read from a pointer to 64bit integer but the read is done via readl_relaxed() which is for 32bit quantities. All the high bits will thus be lost which change the meaning of the test against zero done later. Fix this by using readq_relaxed() instead as it should be for 64bit quantities. Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com> --- !!! Warning !!! I don't have the HW to test this, nor have I read the ACPI specs. This is only compile tested. --- arch/arm64/kernel/acpi_parking_protocol.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)