On 17.09.20 20:14, Oliver Schwartz wrote:
On 17 Sep 2020, at 18:21, Ralf Ramsauer
<ralf.ramsa...@oth-regensburg.de
<mailto:ralf.ramsa...@oth-regensburg.de>> wrote:
On 9/17/20 8:26 AM, Jan Kiszka wrote:
Subject tag should be "arm64". And the patch should go over next first.
I can fix up both.
On 16.09.20 15:07, Oliver Schwartz wrote:
SMC/HVC calls may modify registers x0 to x3. To make sure the compiler
doesn't assume input registers to be constant, also mark these registers
as output when used as input.
Signed-off-by: Oliver Schwartz <oliver.schwa...@gmx.de
<mailto:oliver.schwa...@gmx.de>>
---
hypervisor/arch/arm64/include/asm/smc.h | 13 ++++++-------
include/arch/arm64/asm/jailhouse_hypercall.h | 20 +++++++++++---------
2 files changed, 17 insertions(+), 16 deletions(-)
diff --git a/hypervisor/arch/arm64/include/asm/smc.h
b/hypervisor/arch/arm64/include/asm/smc.h
index 1a5d5c8..c80fe15 100644
--- a/hypervisor/arch/arm64/include/asm/smc.h
+++ b/hypervisor/arch/arm64/include/asm/smc.h
@@ -28,8 +28,8 @@ static inline long smc_arg1(unsigned long id,
unsigned long par1)
register unsigned long __par1 asm("r1") = par1;
asm volatile ("smc #0\n\t"
- : "=r" (__id)
- : "r"(__id), "r"(__par1)
+ : "+r" (__id), "+r"(__par1)
+ :
: "memory", "x2", "x3");
For SMCCC, I'm considering to align fully with Linux, i.e. convert the
remaining register clobberings into early ones, but I also have no
strong argument for it.
Ralf, thoughts?
Just had a look at Linux's implementation. And now I recall why I didn't
c&p it from Linux: it's an unreadable macro hell [1] that would benefit
from being open-coded. But trying to follow Linux's __constraint macro,
shouldn't we protect r2 and r3 as well?
Linux would unroll __constraint_read_1 and __constraint_write_1, and
__constraint_read_1 would protect r2 and r3 as well. We protect x2 and
x3 via the clobber list -- is that enough?
I’m by no means an expert on arm assembler. However, I’m quite positive
that there are no physical registers r2 and r3 that can be protected in
arm64. These are just placeholders that are translated by the compiler
to x2 and x3. So protecting r2/r3 as dummy arguments or x2/x3 via
clobber list is in effect the same, just more explicit.
I've merged both patches to next now using your pattern. I applied some
massaging though. Please have a look at the result.
Some consolidation patch for the hypercalls will also follow.
Jan
--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux
--
You received this message because you are subscribed to the Google Groups
"Jailhouse" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to jailhouse-dev+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/jailhouse-dev/7fc7e7ce-4bb5-8a9d-7299-63d85db3d256%40siemens.com.
