sched_ext: Maintain an online cid mask in the scheduler arenagrafted

Schedulers on the default cid mapping treat [0, nr_online_cids) as the
online set and restart on hotplug. Schedulers that install their own mapping
with scx_bpf_cid_override() have no way to learn which cids are online: the
count no longer identifies members and the CPU-form cpumask is unusable from
cid programs. This is an obvious hole in the cid API.

Add scx_bpf_online_cmask(), a kernel-maintained cmask in the scheduler's
arena, allocated alongside the per-CPU scratch masks and populated after the
cid mapping is finalized and before ops.init(), for child schedulers too.
The pointer stays valid through ops.exit() with no reference to take. It is
the arena offset as a void pointer, the same form struct_ops arena arguments
arrive in. The verifier types the void return as a scalar for the program's
arena cast.

The mask follows the SCX hotplug notifications: seeded from cpu_active_mask
and updated before ops.cid_online/offline() runs, so it lags cpu_online_mask
only inside a hotplug transition. Updates walk the scheduler list under the
lock that also serializes unlinking. Reads are live, not atomic snapshots.
Root initialization excludes hotplug.

v2: Reworded the getter kerneldoc (Andrea Righi).

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>

3 files changed, 345 insertions, 0 deletions

diff --git a/arch/xtensa/boot/boot-redboot/Makefile b/arch/xtensa/boot/boot-redboot/Makefile
new file mode 100644
index 000000000..c0eef3f6f
--- /dev/null
+++ b/arch/xtensa/boot/boot-redboot/Makefile
@@ -0,0 +1,31 @@
+#
+# This file is subject to the terms and conditions of the GNU General Public
+# License.  See the file "COPYING" in the main directory of this archive
+# for more details.
+#
+
+OBJCOPY_ARGS := -O $(if $(CONFIG_CPU_BIG_ENDIAN),elf32-xtensa-be,elf32-xtensa-le)
+
+boot-y	:= bootstrap.o
+targets	+= $(boot-y)
+
+OBJS	:= $(addprefix $(obj)/,$(boot-y))
+LIBS	:= arch/xtensa/boot/lib/lib.a arch/xtensa/lib/lib.a
+
+$(obj)/zImage.o: $(obj)/../vmlinux.bin.gz $(OBJS)
+	$(Q)$(OBJCOPY) $(OBJCOPY_ARGS) -R .comment \
+		--add-section image=$< \
+		--set-section-flags image=contents,alloc,load,load,data \
+		$(OBJS) $@
+
+$(obj)/zImage.elf: $(obj)/zImage.o $(LIBS)
+	$(Q)$(LD) $(KBUILD_LDFLAGS) \
+		-T $(srctree)/$(obj)/boot.ld \
+		--build-id=none \
+		-o $@ $^
+
+$(obj)/../zImage.redboot: $(obj)/zImage.elf
+	$(Q)$(OBJCOPY) -S -O binary $< $@
+	$(Q)$(kecho) '  Kernel: $@ is ready'
+
+all zImage: $(obj)/../zImage.redboot
diff --git a/arch/xtensa/boot/boot-redboot/boot.ld b/arch/xtensa/boot/boot-redboot/boot.ld
new file mode 100644
index 000000000..b0b9e95b5
--- /dev/null
+++ b/arch/xtensa/boot/boot-redboot/boot.ld
@@ -0,0 +1,59 @@
+OUTPUT_ARCH(xtensa)
+
+SECTIONS
+{
+	.start 0xD1000000 : { *(.start) }
+
+	.text :
+	{
+		__reloc_start = . ;
+		_text_start = . ;
+		*(.literal .text.literal .text)
+		_text_end = . ;
+	}
+
+	.rodata ALIGN(0x04):
+	{
+		*(.rodata)
+		*(.rodata1)
+	}
+
+	.data ALIGN(0x04):
+	{
+		*(.data)
+		*(.data1)
+		*(.sdata)
+		*(.sdata2)
+		*(.got.plt)
+		*(.got)
+		*(.dynamic)
+	}
+
+	__reloc_end = . ;
+
+	. = ALIGN(0x10);
+	__image_load = . ;
+	.image 0xd0003000: AT(__image_load)
+	{
+		_image_start = .;
+		*(image)
+		. = (. + 3) & ~ 3;
+		_image_end = .	;
+	}
+
+
+	.bss ((LOADADDR(.image) + SIZEOF(.image) + 3) & ~ 3):
+	{
+		__bss_start = .;
+		*(.sbss)
+		*(.scommon)
+		*(.dynbss)
+		*(.bss)
+		__bss_end = .;
+	}
+	_end = .;
+	_param_start = .;
+
+
+	PROVIDE (end = .);
+}
diff --git a/arch/xtensa/boot/boot-redboot/bootstrap.S b/arch/xtensa/boot/boot-redboot/bootstrap.S
new file mode 100644
index 000000000..3ed94ad35
--- /dev/null
+++ b/arch/xtensa/boot/boot-redboot/bootstrap.S
@@ -0,0 +1,255 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#include <asm/core.h>
+#include <asm/regs.h>
+#include <asm/asmmacro.h>
+#include <asm/cacheasm.h>
+#include <asm/processor.h>
+	/*
+	 * RB-Data: RedBoot data/bss
+	 * P:	    Boot-Parameters
+	 * L:	    Kernel-Loader
+	 *
+	 * The Linux-Kernel image including the loader must be loaded
+	 * to a position so that the kernel and the boot parameters
+	 * can fit in the space before the load address.
+	 *  ______________________________________________________
+	 * |_RB-Data_|_P_|__________|_L_|___Linux-Kernel___|______|
+	 *                          ^
+	 *                          ^ Load address
+	 *  ______________________________________________________
+	 * |___Linux-Kernel___|_P_|_L_|___________________________|
+	 *
+	 * The loader copies the parameter to the position that will
+	 * be the end of the kernel and itself to the end of the
+	 * parameter list.
+	 */
+
+/* Make sure we have enough space for the 'uncompressor' */
+
+#define STACK_SIZE 32768
+#define HEAP_SIZE (131072*4)
+
+	# a2: Parameter list
+	# a3: Size of parameter list
+
+	.section .start, "ax"
+
+	.globl __start
+	/* this must be the first byte of the loader! */
+__start:
+	abi_entry(32)		# we do not intend to return
+	_call0	_start
+__start_a0:
+	.align 4
+
+	.section .text, "ax"
+	.literal_position
+	.begin literal_prefix .text
+
+	/* put literals in here! */
+
+	.globl _start
+_start:
+
+	/* 'reset' window registers */
+
+	movi	a4, 1
+	wsr	a4, ps
+	rsync
+#if XCHAL_HAVE_WINDOWED
+	rsr	a5, windowbase
+	ssl	a5
+	sll	a4, a4
+	wsr	a4, windowstart
+	rsync
+#endif
+	movi	a4, KERNEL_PS_WOE_MASK
+	wsr	a4, ps
+	rsync
+
+KABI_C0	mov	abi_saved0, abi_arg0
+
+	/* copy the loader to its address
+	 * Note: The loader itself is a very small piece, so we assume we
+	 *       don't partially overlap. We also assume (even more important)
+	 *	 that the kernel image is out of the way. Usually, when the
+	 *	 load address of this image is not at an arbitrary address,
+	 *	 but aligned to some 10K's we shouldn't overlap.
+	 */
+
+	/* Note: The assembler cannot relax "addi a0, a0, ..." to an
+	   l32r, so we load to a4 first. */
+
+	# addi	a4, a0, __start - __start_a0
+	# mov	a0, a4
+
+	movi	a4, __start
+	movi	a5, __start_a0
+	add	a4, a0, a4
+	sub	a0, a4, a5
+
+	movi	a4, __start
+	movi	a5, __reloc_end
+
+	# a0: address where this code has been loaded
+	# a4: compiled address of __start
+	# a5: compiled end address
+
+	mov.n	a7, a0
+	mov.n	a8, a4
+
+1:
+	l32i	a10, a7, 0
+	l32i	a11, a7, 4
+	s32i	a10, a8, 0
+	s32i	a11, a8, 4
+	l32i	a10, a7, 8
+	l32i	a11, a7, 12
+	s32i	a10, a8, 8
+	s32i	a11, a8, 12
+	addi	a8, a8, 16
+	addi	a7, a7, 16
+	blt	a8, a5, 1b
+
+
+	/* We have to flush and invalidate the caches here before we jump. */
+
+#if XCHAL_DCACHE_IS_WRITEBACK
+
+	___flush_dcache_all a5 a6
+
+#endif
+
+	___invalidate_icache_all a5 a6
+	isync
+
+	movi	a11, _reloc
+	jx	a11
+
+	.globl _reloc
+_reloc:
+
+	/* RedBoot is now at the end of the memory, so we don't have
+	 * to copy the parameter list. Keep the code around; in case
+	 * we need it again. */
+#if 0
+	# a0: load address
+	# a2: start address of parameter list
+	# a3: length of parameter list
+	# a4: __start
+
+	/* copy the parameter list out of the way */
+
+	movi	a6, _param_start
+	add	a3, a2, a3
+2:
+	l32i	a8, a2, 0
+	s32i	a8, a6, 0
+	addi	a2, a2, 4
+	addi	a6, a6, 4
+	blt	a2, a3, 2b
+#endif
+
+	/* clear BSS section */
+	movi	a6, __bss_start
+	movi	a7, __bss_end
+	movi.n	a5, 0
+3:
+	s32i	a5, a6, 0
+	addi	a6, a6, 4
+	blt	a6, a7, 3b
+
+	movi	a5, -16
+	movi	a1, _stack + STACK_SIZE
+	and	a1, a1, a5
+
+	/* Uncompress the kernel */
+
+	# a0: load address
+	# a2: boot parameter
+	# a4: __start
+
+	movi	a3, __image_load
+	sub	a4, a3, a4
+	add	abi_arg2, a0, a4
+
+	# a1  Stack
+	# a8(a4)  Load address of the image
+
+	movi	abi_arg0, _image_start
+	movi	abi_arg4, _image_end
+	movi	abi_arg1, 0x1000000
+	sub	abi_tmp0, abi_arg4, abi_arg0
+	movi	abi_arg3, complen
+	s32i	abi_tmp0, abi_arg3, 0
+
+	movi	a0, 0
+
+	# abi_arg0 destination
+	# abi_arg1 maximum size of destination
+	# abi_arg2 source
+	# abi_arg3 ptr to length
+
+	.extern gunzip
+	movi	abi_tmp0, gunzip
+	beqz	abi_tmp0, 1f
+
+	abi_callx	abi_tmp0
+
+	j	2f
+
+
+	# abi_arg0 destination start
+	# abi_arg1 maximum size of destination
+	# abi_arg2 source start
+	# abi_arg3 ptr to length
+	# abi_arg4 destination end
+
+1:
+        l32i    abi_tmp0, abi_arg2, 0
+        l32i    abi_tmp1, abi_arg2, 4
+        s32i    abi_tmp0, abi_arg0, 0
+        s32i    abi_tmp1, abi_arg0, 4
+        l32i    abi_tmp0, abi_arg2, 8
+        l32i    abi_tmp1, abi_arg2, 12
+        s32i    abi_tmp0, abi_arg0, 8
+        s32i    abi_tmp1, abi_arg0, 12
+        addi    abi_arg0, abi_arg0, 16
+        addi    abi_arg2, abi_arg2, 16
+        blt     abi_arg0, abi_arg4, 1b
+
+
+	/* jump to the kernel */
+2:
+#if XCHAL_DCACHE_IS_WRITEBACK
+
+	___flush_dcache_all a5 a6
+
+#endif
+
+	___invalidate_icache_all a5 a6
+
+	isync
+
+	# a2  Boot parameter list
+
+KABI_C0	mov	abi_arg0, abi_saved0
+	movi	a0, _image_start
+	jx	a0
+
+	.align 16
+	.data
+	.globl avail_ram
+avail_ram:
+	.long	_heap
+	.globl end_avail
+end_avail:
+	.long	_heap + HEAP_SIZE
+
+	.comm _stack, STACK_SIZE
+	.comm _heap, HEAP_SIZE
+
+	.globl end_avail
+	.comm complen, 4
+
+	.end	literal_prefix