Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
The dependency has been accepted, but this patch needs a respin
for the changed clocks.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
The dependency has been accepted, but this patch needs a respin
for the changed clocks.
Thank you for pointing this out. wrt resets the GIC has two signals
(which I learnt lately when the dependency path was accepted). Earlier
discussion in irc with Sudeep pointed out that there wouldn't be any
use case of having GIC resets in DTSI, so either we drop the resets
property in DT binding doc or correct it.
Let me know your thoughts on this and how we proceed further.
Cheers,
Prabhakar
The dependency has been accepted, but this patch needs a respin
for the changed clocks.
Thank you for pointing this out. wrt resets the GIC has two signals
(which I learnt lately when the dependency path was accepted). Earlier
discussion in irc with Sudeep pointed out that there wouldn't be any
use case of having GIC resets in DTSI, so either we drop the resets
property in DT binding doc or correct it.
Let me know your thoughts on this and how we proceed further.
DT Rule #1: DT describes hardware not software policy.
And a possible use case: the RT CPU core may want to reset the AP GIC.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
The dependency has been accepted, but this patch needs a respin
for the changed clocks.
Thank you for pointing this out. wrt resets the GIC has two signals
(which I learnt lately when the dependency path was accepted). Earlier
discussion in irc with Sudeep pointed out that there wouldn't be any
use case of having GIC resets in DTSI, so either we drop the resets
property in DT binding doc or correct it.
Let me know your thoughts on this and how we proceed further.
DT Rule #1: DT describes hardware not software policy.
Completely agreed, no disagreement 😄.
And a possible use case: the RT CPU core may want to reset the AP GIC.
I didn't want to add new bindings without details on the implementation
to avoid possible issues with backward compatibility as this was not
thought through completely and correctly before it was added.
OK, now let us discuss your use-case: *RT CPU wants to reset AP GIC*
1. Will it just reset AP GIC or will it request the AP reset as a whole ?
I am not sure if we can handle former, if you think otherwise what is
the reset notification mechanism ?
2. Will that bypass secure world/PSCI ? Again more details on this would
be helpful to visualise the entire use-case end-to-end better.
By GIC reset, I am assuming it will be complete GIC reset including it's
CPU interface.
I don't think we can reset GIC without actual CPU reset. Even if we get
some notification magically to the CPU that its GIC alone needs to be
reset, it needs to safely higher exceptions to get its GIC CPU interface
reprogrammed to correct (saved) values before OS can reprogram the NS
world values. All these seems overall complicated and may be unnecessary.
--
Regards,
Sudeep
The dependency has been accepted, but this patch needs a respin
for the changed clocks.
Thank you for pointing this out. wrt resets the GIC has two signals
(which I learnt lately when the dependency path was accepted). Earlier
discussion in irc with Sudeep pointed out that there wouldn't be any
use case of having GIC resets in DTSI, so either we drop the resets
property in DT binding doc or correct it.
Let me know your thoughts on this and how we proceed further.
DT Rule #1: DT describes hardware not software policy.
Completely agreed, no disagreement .
Good ;-)
quoted
And a possible use case: the RT CPU core may want to reset the AP GIC.
I didn't want to add new bindings without details on the implementation
to avoid possible issues with backward compatibility as this was not
thought through completely and correctly before it was added.
OK, now let us discuss your use-case: *RT CPU wants to reset AP GIC*
1. Will it just reset AP GIC or will it request the AP reset as a whole ?
I am not sure if we can handle former, if you think otherwise what is
the reset notification mechanism ?
2. Will that bypass secure world/PSCI ? Again more details on this would
be helpful to visualise the entire use-case end-to-end better.
By GIC reset, I am assuming it will be complete GIC reset including it's
CPU interface.
I don't think we can reset GIC without actual CPU reset. Even if we get
some notification magically to the CPU that its GIC alone needs to be
reset, it needs to safely higher exceptions to get its GIC CPU interface
reprogrammed to correct (saved) values before OS can reprogram the NS
world values. All these seems overall complicated and may be unnecessary.
Probably both. Might make sense to reset on wake-up, after having disabled
clocks and powered down the AP CPU, AP GIC, ...
If that bypasses PSCI: well, if the unsecure software can do it, it
means the hardware is not secure. Or at least Linux has to be trusted.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
On Tue, Jul 13, 2021 at 11:04:09AM +0200, Geert Uytterhoeven wrote:
Hi Sudeep,
On Tue, Jul 13, 2021 at 10:56 AM Sudeep Holla [off-list ref] wrote:
quoted
On Tue, Jul 13, 2021 at 10:30:36AM +0200, Geert Uytterhoeven wrote:
[...]
quoted
quoted
And a possible use case: the RT CPU core may want to reset the AP GIC.
I didn't want to add new bindings without details on the implementation
to avoid possible issues with backward compatibility as this was not
thought through completely and correctly before it was added.
OK, now let us discuss your use-case: *RT CPU wants to reset AP GIC*
1. Will it just reset AP GIC or will it request the AP reset as a whole ?
I am not sure if we can handle former, if you think otherwise what is
the reset notification mechanism ?
2. Will that bypass secure world/PSCI ? Again more details on this would
be helpful to visualise the entire use-case end-to-end better.
By GIC reset, I am assuming it will be complete GIC reset including it's
CPU interface.
I don't think we can reset GIC without actual CPU reset. Even if we get
some notification magically to the CPU that its GIC alone needs to be
reset, it needs to safely higher exceptions to get its GIC CPU interface
reprogrammed to correct (saved) values before OS can reprogram the NS
world values. All these seems overall complicated and may be unnecessary.
Probably both. Might make sense to reset on wake-up, after having disabled
clocks and powered down the AP CPU, AP GIC, ...
/me confused. If this is arm64 platform, then you have to use *PSCI* and
I expect the reset to be done as part of CPU wake-up in PSCI firmware.
If that bypasses PSCI: well, if the unsecure software can do it, it
means the hardware is not secure. Or at least Linux has to be trusted.
No, if the system has PSCI, then you simply can't bypass that for GIC
reset. Or at-least I am failing to understand the complete flow of that.
--
Regards,
Sudeep
Hi Sudeep,
On Tue, Jul 13, 2021 at 11:16 AM Sudeep Holla [off-list ref] wrote:
On Tue, Jul 13, 2021 at 11:04:09AM +0200, Geert Uytterhoeven wrote:
quoted
On Tue, Jul 13, 2021 at 10:56 AM Sudeep Holla [off-list ref] wrote:
quoted
On Tue, Jul 13, 2021 at 10:30:36AM +0200, Geert Uytterhoeven wrote:
[...]
quoted
quoted
quoted
And a possible use case: the RT CPU core may want to reset the AP GIC.
I didn't want to add new bindings without details on the implementation
to avoid possible issues with backward compatibility as this was not
thought through completely and correctly before it was added.
OK, now let us discuss your use-case: *RT CPU wants to reset AP GIC*
1. Will it just reset AP GIC or will it request the AP reset as a whole ?
I am not sure if we can handle former, if you think otherwise what is
the reset notification mechanism ?
2. Will that bypass secure world/PSCI ? Again more details on this would
be helpful to visualise the entire use-case end-to-end better.
By GIC reset, I am assuming it will be complete GIC reset including it's
CPU interface.
I don't think we can reset GIC without actual CPU reset. Even if we get
some notification magically to the CPU that its GIC alone needs to be
reset, it needs to safely higher exceptions to get its GIC CPU interface
reprogrammed to correct (saved) values before OS can reprogram the NS
world values. All these seems overall complicated and may be unnecessary.
Probably both. Might make sense to reset on wake-up, after having disabled
clocks and powered down the AP CPU, AP GIC, ...
/me confused. If this is arm64 platform, then you have to use *PSCI* and
I expect the reset to be done as part of CPU wake-up in PSCI firmware.
DT Rule #1: DT describes hardware not software policy.
The fact that _Linux_ must use PSCI is a (unfortunate) software policy.
What about other OSes, or bare-metal software?
quoted
If that bypasses PSCI: well, if the unsecure software can do it, it
means the hardware is not secure. Or at least Linux has to be trusted.
No, if the system has PSCI, then you simply can't bypass that for GIC
reset. Or at-least I am failing to understand the complete flow of that.
PSCI can only prevent other software from bypassing GIC reset if PSCI
programs the hardware to prevent access to the GIC reset (if possible
at all).
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
On Tue, Jul 13, 2021 at 11:24:24AM +0200, Geert Uytterhoeven wrote:
Hi Sudeep,
On Tue, Jul 13, 2021 at 11:16 AM Sudeep Holla [off-list ref] wrote:
quoted
On Tue, Jul 13, 2021 at 11:04:09AM +0200, Geert Uytterhoeven wrote:
[...]
quoted
quoted
Probably both. Might make sense to reset on wake-up, after having disabled
clocks and powered down the AP CPU, AP GIC, ...
/me confused. If this is arm64 platform, then you have to use *PSCI* and
I expect the reset to be done as part of CPU wake-up in PSCI firmware.
DT Rule #1: DT describes hardware not software policy.
As mentioned before I agree on that. But I assume you too agree that not
all bits and pieces of hardware are represented in DT. Only ones that are
essential for any software to understand the hardware and make it work.
So my opinion is that this GIC reset information is implicit like many
other hardware information.
The fact that _Linux_ must use PSCI is a (unfortunate) software policy.
What about other OSes, or bare-metal software?
Disagree. PSCI is OS agnostic and _Linux_ is not the sole user. Do you
have examples of other OS that deploy alternate to PSCI ? Or bare-metal
that uses DT ? Again if there is a use-case, you need to spell out details
on when this can be used and where it can't be of much use(of-course with
the mention of Linux). There are complaints that DT bindings are too Linux
specific, so please pull in the other OS folks or other users so that we
get details on use-cases. Based on your argument we should have loads of
other information in DT on CPU or CPU peripherals for example even if
PSCI hides them for OS. Do we really want to get down that path ?
I agree PSCI is software policy but definitely fortunate, helped prevent
lot of non-sense in OS 😉.
quoted
quoted
If that bypasses PSCI: well, if the unsecure software can do it, it
means the hardware is not secure. Or at least Linux has to be trusted.
No, if the system has PSCI, then you simply can't bypass that for GIC
reset. Or at-least I am failing to understand the complete flow of that.
PSCI can only prevent other software from bypassing GIC reset if PSCI
programs the hardware to prevent access to the GIC reset (if possible
at all).
It is not matter of preventing, it is matter of what we can support
architecturally and sanely. While I agree with non-Linux use-case, we
can't support this in Linux.
--
Regards,
Sudeep