arc: add support for TIF_NOTIFY_SIGNAL
PATCH 1/1 arc: add support for TIF_NOTIFY_SIGNAL
Wire up TIF_NOTIFY_SIGNAL handling for arc.
---
5.11 has support queued up for TIF_NOTIFY_SIGNAL, see this posting for details:
https://lore.kernel.org/io-uring/[email protected]/
As part of that work, I'm adding TIF_NOTIFY_SIGNAL support to all archs, as that will enable a set of cleanups once all of them support it. I'm happy carrying this patch if need be, or it can be funelled through the arch tree. Let me know.
arch/arc/include/asm/thread_info.h | 4 +++- arch/arc/kernel/entry.S | 2 +- arch/arc/kernel/signal.c | 2 +- 3 files changed, 5 insertions(+), 3 deletions(-)
arch/arc/include/asm/thread_info.h +3 -1
@@ -79,6 +79,7 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define TIF_SIGPENDING 2/* signal pending */ #define TIF_NEED_RESCHED 3/* rescheduling necessary */ #define TIF_SYSCALL_AUDIT 4/* syscall auditing active */ +#define TIF_NOTIFY_SIGNAL 5/* signal notifications exist */ #define TIF_SYSCALL_TRACE 15/* syscall trace active */ /* true if poll_idle() is polling TIF_NEED_RESCHED */
@@ -89,11 +90,12 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define _TIF_SIGPENDING (1<<TIF_SIGPENDING)#define _TIF_NEED_RESCHED (1<<TIF_NEED_RESCHED)#define _TIF_SYSCALL_AUDIT (1<<TIF_SYSCALL_AUDIT) +#define _TIF_NOTIFY_SIGNAL (1<<TIF_NOTIFY_SIGNAL)#define _TIF_MEMDIE (1<<TIF_MEMDIE)/* work to do on interrupt/exception return */ #define _TIF_WORK_MASK (_TIF_NEED_RESCHED | _TIF_SIGPENDING | \ - _TIF_NOTIFY_RESUME) + _TIF_NOTIFY_RESUME | _TIF_NOTIFY_SIGNAL)/* * _TIF_ALLWORK_MASK includes SYSCALL_TRACE, but we don't need it.
arch/arc/kernel/entry.S +1 -1
@@ -307,7 +307,7 @@ resume_user_mode_begin: mov r0, sp ; pt_regs for arg to do_signal()/do_notify_resume() GET_CURR_THR_INFO_FLAGS r9 - bbit0 r9, TIF_SIGPENDING, .Lchk_notify_resume + and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL, .Lchk_notify_resume ; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in pt_reg since the "C" ABI (kernel code) will automatically
arch/arc/kernel/signal.c +1 -1
@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs); -if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */
Signature · 10 lines
2.29.0 -- Jens Axboe _______________________________________________ linux-snps-arc mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Quoted 3 lines · Jens Axboe
Wire up
TIF_NOTIFY_SIGNALhandling for arc.
Just to be clear, ARC and ARM seem to differ in 1 letter, they are in no way related :-)
Quoted 7 lines
Signed-off-by Jens Axboe<[email protected]>---
5.11 has support queued up for
TIF_NOTIFY_SIGNAL, see this posting for details:https://urldefense.com/v3/__https://lore.kernel.org/io-uring/[email protected]/__;!!A4F2R9G_pg!Oo4kwr6Yy8y5ByC8jIONjU28gLsqpyfzT10u5kfdsj2E-U2VIXQs7SvTmBRIkupZ$
TL'DR ignore this url rewrite.
Corporate mailer fudging the urls and I can't seem to find this msg in my mailing group subscription in Thunderbird (although it shows up in web achieve) hence need to reply from office account rather than the newsgroup account.
Quoted 4 lines
As part of that work, I'm adding
TIF_NOTIFY_SIGNALsupport to all archs, as that will enable a set of cleanups once all of them support it. I'm happy carrying this patch if need be, or it can be funelled through the arch tree. Let me know.
I'm fine either ways too, best to do whatever you are doing for other arches. Although this patch per se is broken, please see below.
arch/arc/kernel/entry.S resume_user_mode_begin: · line 310
Quoted 4 lines
GET_CURR_THR_INFO_FLAGSr9arch/arc/kernel/entry.S +1
-1- bbit0 r9, TIF_SIGPENDING, .Lchk_notify_resume + and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL, .Lchk_notify_resume
This is not correct, AND is a simple ALU instruction with format: AND dest, src2, src1
With dest 0, it won't update any register. With .f suffix it would set CC flag based on ALU operation which can subsequent used for a predicated instruction such as BZ.
So you need something like
and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL
bz .Lchk_notify_resume
Quoted 8 lines
; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in
pt_regsince the "C" ABI (kernel code) will automaticallyarch/arc/kernel/signal.c +0
-0
arch/arc/kernel/signal.c void do_signal(struct pt_regs *regs) · line 368
Quoted 8 lines
arch/arc/kernel/signal.c +1
-1@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs); -if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */
I've not seen your entire patchset, but it seems we are now hitting do_signal() for either of TIF_{SIGPENDING|NOTIFY_SIGNAL} but then only handling signal for TIF_SIGPENDING, so why even bother to come here. Do you plan to add additional arch handling later ?
Thx,
-Vineet
_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Quoted 8 lines · Vineet Gupta
Quoted 3 lines · Jens Axboe
Wire up
TIF_NOTIFY_SIGNALhandling for arc.Just to be clear, ARC and ARM seem to differ in 1 letter, they are in no way related :-)
Oops, fat fingered that one. Will update :-)
Quoted 21 lines
Quoted 4 lines
As part of that work, I'm adding
TIF_NOTIFY_SIGNALsupport to all archs, as that will enable a set of cleanups once all of them support it. I'm happy carrying this patch if need be, or it can be funelled through the arch tree. Let me know.I'm fine either ways too, best to do whatever you are doing for other arches. Although this patch per se is broken, please see below.
Review
Quoted 4 lines
GET_CURR_THR_INFO_FLAGSr9(quoted diff) +1
-1- bbit0 r9, TIF_SIGPENDING, .Lchk_notify_resume + and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL, .Lchk_notify_resumeThis is not correct, AND is a simple ALU instruction with format: AND dest, src2, src1
With dest 0, it won't update any register. With .f suffix it would set CC flag based on ALU operation which can subsequent used for a predicated instruction such as BZ.
Quoted 4 lines
So you need something like
and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL
bz .Lchk_notify_resume
Ah thanks, I'll make that change. Hard for me to test a lot of these, so I really appreciate someone knowledgable taking a look at it.
Quoted 11 lines
Quoted 11 lines
; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in
pt_regsince the "C" ABI (kernel code) will automaticallyarch/arc/kernel/signal.c +0
-0@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs);
Quoted 11 lines
arch/arc/kernel/signal.c void do_signal(struct pt_regs *regs) · line 368
Quoted 5 lines
arch/arc/kernel/signal.c +1
-1-if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */ I've not seen your entire patchset, but it seems we are now hitting
do_signal()for either of TIF_{SIGPENDING|NOTIFY_SIGNAL} but then only handling signal forTIF_SIGPENDING, so why even bother to come here. Do you plan to add additional arch handling later ?
Nope, that's all that's needed for each arch. The behavior you describe is how it works. It makes it so that TIF_SIGPENDING will do the signal delivery and syscall restart as it always has done, but TIF_NOTIFY_SIGNAL will only do the syscall restart. The latter is the intent, hence TIF_NOTIFY_SIGNAL provides a way to interrupt a process and have it process task_work without going through signal delivery like task_work with TWA_SIGNAL does today.
Updated version below:
commit 3c6239647d95d03d1436bc826a004791c3f04617
Author: Jens Axboe <[email protected]> Date: Mon Oct 12 07:15:37 2020 -0600
arc: add support for TIF_NOTIFY_SIGNAL
Wire up TIF_NOTIFY_SIGNAL handling for arc.
arch/arc/include/asm/thread_info.h +3 -1
@@ -79,6 +79,7 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define TIF_SIGPENDING 2/* signal pending */ #define TIF_NEED_RESCHED 3/* rescheduling necessary */ #define TIF_SYSCALL_AUDIT 4/* syscall auditing active */ +#define TIF_NOTIFY_SIGNAL 5/* signal notifications exist */ #define TIF_SYSCALL_TRACE 15/* syscall trace active */ /* true if poll_idle() is polling TIF_NEED_RESCHED */
@@ -89,11 +90,12 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define _TIF_SIGPENDING (1<<TIF_SIGPENDING)#define _TIF_NEED_RESCHED (1<<TIF_NEED_RESCHED)#define _TIF_SYSCALL_AUDIT (1<<TIF_SYSCALL_AUDIT) +#define _TIF_NOTIFY_SIGNAL (1<<TIF_NOTIFY_SIGNAL)#define _TIF_MEMDIE (1<<TIF_MEMDIE)/* work to do on interrupt/exception return */ #define _TIF_WORK_MASK (_TIF_NEED_RESCHED | _TIF_SIGPENDING | \ - _TIF_NOTIFY_RESUME) + _TIF_NOTIFY_RESUME | _TIF_NOTIFY_SIGNAL)/* * _TIF_ALLWORK_MASK includes SYSCALL_TRACE, but we don't need it.
arch/arc/kernel/entry.S +2 -1
@@ -307,7 +307,8 @@ resume_user_mode_begin: mov r0, sp ; pt_regs for arg to do_signal()/do_notify_resume() GET_CURR_THR_INFO_FLAGS r9 - bbit0 r9, TIF_SIGPENDING, .Lchk_notify_resume + and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL + bz .Lchk_notify_resume ; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in pt_reg since the "C" ABI (kernel code) will automatically
arch/arc/kernel/signal.c +1 -1
@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs); -if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */
Signature · 7 lines
Jens Axboe _______________________________________________ linux-snps-arc mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Vineet Gupta
Acked-byQuoted 3 lines · Jens Axboe
Ah thanks, I'll make that change. Hard for me to test a lot of these, so I really appreciate someone knowledgable taking a look at it.
Sure, glad to help, thx to you for writing the arch patches in first place. It takes a bit of daring to venture in unchartered waters ;-)
These days it is super easy to get your hands on a ARC cross toolchain. We don't have them shipping as regular distro packages just yet, but you can download a prebuilt @
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/releases/download/arc-2020.09-release/arc_gnu_2020.09_prebuilt_glibc_le_archs_linux_install.tar.gz
Or you can just build upstream gcc + binutils for ARC if you so prefer.
Quoted 21 lines
Quoted 11 lines
Quoted 11 lines
; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in
pt_regsince the "C" ABI (kernel code) will automaticallyarch/arc/kernel/signal.c +0
-0@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs);Quoted 10 lines
arch/arc/kernel/signal.c void do_signal(struct pt_regs *regs) · line 368
Quoted 5 lines
arch/arc/kernel/signal.c +1
-1-if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */ I've not seen your entire patchset, but it seems we are now hitting
do_signal()for either of TIF_{SIGPENDING|NOTIFY_SIGNAL} but then only handling signal forTIF_SIGPENDING, so why even bother to come here. Do you plan to add additional arch handling later ?
Quoted 7 lines
Nope, that's all that's needed for each arch. The behavior you describe is how it works. It makes it so that
TIF_SIGPENDINGwill do the signal delivery and syscall restart as it always has done, butTIF_NOTIFY_SIGNALwill only do the syscall restart. The latter is the intent, henceTIF_NOTIFY_SIGNALprovides a way to interrupt a process and have it processtask_workwithout going through signal delivery liketask_workwithTWA_SIGNALdoes today.
Nice, thx for explaining that.
Quoted 8 lines
Updated version below:
commit
3c6239647d95d03d1436bc826a004791c3f04617Author: Jens Axboe <[email protected]> Date: Mon Oct 12 07:15:37 2020 -0600
Quoted 6 lines
arc: add support for
TIF_NOTIFY_SIGNALWire up
TIF_NOTIFY_SIGNALhandling for arc.Signed-off-by Jens Axboe <[email protected]>
Quoted 47 lines
arch/arc/include/asm/thread_info.h +3
-1@@ -79,6 +79,7 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define TIF_SIGPENDING 2/* signal pending */ #define TIF_NEED_RESCHED 3/* rescheduling necessary */ #define TIF_SYSCALL_AUDIT 4/* syscall auditing active */ +#define TIF_NOTIFY_SIGNAL 5/* signal notifications exist */ #define TIF_SYSCALL_TRACE 15/* syscall trace active */ /* true if poll_idle() is polling TIF_NEED_RESCHED */ @@ -89,11 +90,12 @@ static inline __attribute_const__ struct thread_info *current_thread_info(void)#define _TIF_SIGPENDING (1<<TIF_SIGPENDING)#define _TIF_NEED_RESCHED (1<<TIF_NEED_RESCHED)#define _TIF_SYSCALL_AUDIT (1<<TIF_SYSCALL_AUDIT) +#define _TIF_NOTIFY_SIGNAL (1<<TIF_NOTIFY_SIGNAL)#define _TIF_MEMDIE (1<<TIF_MEMDIE)/* work to do on interrupt/exception return */ #define _TIF_WORK_MASK (_TIF_NEED_RESCHED | _TIF_SIGPENDING | \ - _TIF_NOTIFY_RESUME) + _TIF_NOTIFY_RESUME | _TIF_NOTIFY_SIGNAL)/* * _TIF_ALLWORK_MASK includes SYSCALL_TRACE, but we don't need it. arch/arc/kernel/entry.S +2
-1@@ -307,7 +307,8 @@ resume_user_mode_begin: mov r0, sp ; pt_regs for arg to do_signal()/do_notify_resume() GET_CURR_THR_INFO_FLAGS r9 - bbit0 r9, TIF_SIGPENDING, .Lchk_notify_resume + and.f 0, r9, TIF_SIGPENDING|TIF_NOTIFY_SIGNAL + bz .Lchk_notify_resume ; Normal Trap/IRQ entry only saves Scratch (caller-saved) regs ; in pt_reg since the "C" ABI (kernel code) will automaticallyarch/arc/kernel/signal.c +0
-0
arch/arc/kernel/signal.c void do_signal(struct pt_regs *regs) · line 368
Quoted 8 lines
arch/arc/kernel/signal.c +1
-1@@ -362,7 +362,7 @@ void do_signal(struct pt_regs *regs) restart_scall =in_syscall (regs) &&syscall_restartable (regs); -if (get_signal (&ksig)) { +if (test_thread_flag (TIF_SIGPENDING ) &&get_signal (&ksig)) {if (restart_scall) {arc_restart_syscall (&ksig.ka, regs);syscall_wont_restart (regs);/* No more restarts */
_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Quoted 7 lines · Vineet Gupta
Quoted 3 lines · Jens Axboe
Ah thanks, I'll make that change. Hard for me to test a lot of these, so I really appreciate someone knowledgable taking a look at it.
Sure, glad to help, thx to you for writing the arch patches in first place. It takes a bit of daring to venture in unchartered waters ;-)
Nice to do something different for a change! And another motivation is that I can't drop some old dead code before all archs support TIF_NOTIFY_SIGNAL.
Quoted 6 lines
These days it is super easy to get your hands on a ARC cross toolchain. We don't have them shipping as regular distro packages just yet, but you can download a prebuilt @
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/releases/download/arc-2020.09-release/arc_gnu_2020.09_prebuilt_glibc_le_archs_linux_install.tar.gzOr you can just build upstream gcc + binutils for ARC if you so prefer.
Thanks, that's very helpful. Unfortunately my distro didn't have gcc+binutils packaged for arc, like it did for a lot of the other targets.
Quoted 4 lines
Quoted 4 lines
commit
3c6239647d95d03d1436bc826a004791c3f04617Author: Jens Axboe <[email protected]> Date: Mon Oct 12 07:15:37 2020 -0600
Quoted 8 lines
Quoted 6 lines
arc: add support for
TIF_NOTIFY_SIGNALWire up
TIF_NOTIFY_SIGNALhandling for arc.Signed-off-by Jens Axboe <[email protected]>Acked-by Vineet Gupta <[email protected]>
Added, thanks!
Signature · 7 lines
Jens Axboe _______________________________________________ linux-snps-arc mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-snps-arc