[PATCH v2 0/4] enable bpf_prog_pack allocator for powerpc

[PATCH v2 0/4] enable bpf_prog_pack allocator for powerpc

[Hari Bathini]

Most BPF programs are small, but they consume a page each. For systems
with busy traffic and many BPF programs, this may also add significant
pressure on instruction TLB. High iTLB pressure usually slows down the
whole system causing visible performance degradation for production
workloads.

bpf_prog_pack, a customized allocator that packs multiple bpf programs
into preallocated memory chunks, was proposed [1] to address it. This
series extends this support on powerpc.

The first patch introduces patch_instructions() function to enable
patching more than one instruction at a time. This change showed
around 5X improvement in the time taken to run test_bpf test cases.
Patches 2 & 3 add the arch specific functions needed to support this
feature. Patch 4 enables the support for powerpc and ensures cleanup
is handled racefully. Tested the changes successfully.

[1] https://lore.kernel.org/bpf/20220204185742.271030-1-song@kernel.org/
[2] https://lore.kernel.org/all/20221110184303.393179-1-hbathini@linux.ibm.com/ 

Changes in v2:
* Introduced patch_instructions() to help with patching bpf programs.

Hari Bathini (4):
  powerpc/code-patching: introduce patch_instructions()
  powerpc/bpf: implement bpf_arch_text_copy
  powerpc/bpf: implement bpf_arch_text_invalidate for bpf_prog_pack
  powerpc/bpf: use bpf_jit_binary_pack_[alloc|finalize|free]

 arch/powerpc/include/asm/code-patching.h |   1 +
 arch/powerpc/lib/code-patching.c         | 151 ++++++++++++++++-------
 arch/powerpc/net/bpf_jit.h               |   7 +-
 arch/powerpc/net/bpf_jit_comp.c          | 142 ++++++++++++++++-----
 arch/powerpc/net/bpf_jit_comp32.c        |   4 +-
 arch/powerpc/net/bpf_jit_comp64.c        |   6 +-
 6 files changed, 226 insertions(+), 85 deletions(-)

-- 
2.39.2

[PATCH v2 1/4] powerpc/code-patching: introduce patch_instructions()

[Hari Bathini]

patch_instruction() entails setting up pte, patching the instruction,
clearing the pte and flushing the tlb. If multiple instructions need
to be patched, every instruction would have to go through the above
drill unnecessarily. Instead, introduce function patch_instructions()
that patches multiple instructions at one go while setting up the pte,
clearing the pte and flushing the tlb only once per page range of
instructions. Observed ~5X improvement in speed of execution using
patch_instructions() over patch_instructions(), when more instructions
are to be patched.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
---
 arch/powerpc/include/asm/code-patching.h |   1 +
 arch/powerpc/lib/code-patching.c         | 151 ++++++++++++++++-------
 2 files changed, 106 insertions(+), 46 deletions(-)

diff --git a/arch/powerpc/include/asm/code-patching.h b/arch/powerpc/include/asm/code-patching.h
index 3f881548fb61..059fc4fe700e 100644
--- a/arch/powerpc/include/asm/code-patching.h
+++ b/arch/powerpc/include/asm/code-patching.h
@@ -74,6 +74,7 @@ int create_cond_branch(ppc_inst_t *instr, const u32 *addr,
 int patch_branch(u32 *addr, unsigned long target, int flags);
 int patch_instruction(u32 *addr, ppc_inst_t instr);
 int raw_patch_instruction(u32 *addr, ppc_inst_t instr);
+int patch_instructions(u32 *addr, u32 *code, bool fill_inst, size_t len);
 
 static inline unsigned long patch_site_addr(s32 *site)
 {
diff --git a/arch/powerpc/lib/code-patching.c b/arch/powerpc/lib/code-patching.c
index b00112d7ad46..33857b9b53de 100644
--- a/arch/powerpc/lib/code-patching.c
+++ b/arch/powerpc/lib/code-patching.c
@@ -278,77 +278,117 @@ static void unmap_patch_area(unsigned long addr)
 	flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
 }
 
-static int __do_patch_instruction_mm(u32 *addr, ppc_inst_t instr)
+static int __do_patch_instructions_mm(u32 *addr, u32 *code, bool fill_inst, size_t len)
 {
-	int err;
-	u32 *patch_addr;
-	unsigned long text_poke_addr;
-	pte_t *pte;
-	unsigned long pfn = get_patch_pfn(addr);
-	struct mm_struct *patching_mm;
-	struct mm_struct *orig_mm;
+	struct mm_struct *patching_mm, *orig_mm;
+	unsigned long text_poke_addr, pfn;
+	u32 *patch_addr, *end, *pend;
+	ppc_inst_t instr;
 	spinlock_t *ptl;
+	int ilen, err;
+	pte_t *pte;
 
 	patching_mm = __this_cpu_read(cpu_patching_context.mm);
 	text_poke_addr = __this_cpu_read(cpu_patching_context.addr);
-	patch_addr = (u32 *)(text_poke_addr + offset_in_page(addr));
 
 	pte = get_locked_pte(patching_mm, text_poke_addr, &ptl);
 	if (!pte)
 		return -ENOMEM;
 
-	__set_pte_at(patching_mm, text_poke_addr, pte, pfn_pte(pfn, PAGE_KERNEL), 0);
+	end = (void *)addr + len;
+	do {
+		pfn = get_patch_pfn(addr);
+		__set_pte_at(patching_mm, text_poke_addr, pte, pfn_pte(pfn, PAGE_KERNEL), 0);
 
-	/* order PTE update before use, also serves as the hwsync */
-	asm volatile("ptesync": : :"memory");
-
-	/* order context switch after arbitrary prior code */
-	isync();
-
-	orig_mm = start_using_temp_mm(patching_mm);
-
-	err = __patch_instruction(addr, instr, patch_addr);
+		/* order PTE update before use, also serves as the hwsync */
+		asm volatile("ptesync": : :"memory");
 
-	/* hwsync performed by __patch_instruction (sync) if successful */
-	if (err)
-		mb();  /* sync */
+		/* order context switch after arbitrary prior code */
+		isync();
+
+		orig_mm = start_using_temp_mm(patching_mm);
+
+		patch_addr = (u32 *)(text_poke_addr + offset_in_page(addr));
+		pend = (void *)addr + PAGE_SIZE - offset_in_page(addr);
+		if (end < pend)
+			pend = end;
+
+		while (addr < pend) {
+			instr = ppc_inst_read(code);
+			ilen = ppc_inst_len(instr);
+			err = __patch_instruction(addr, instr, patch_addr);
+			/* hwsync performed by __patch_instruction (sync) if successful */
+			if (err) {
+				mb();  /* sync */
+				break;
+			}
+
+			patch_addr = (void *)patch_addr + ilen;
+			addr = (void *)addr + ilen;
+			if (!fill_inst)
+				code = (void *)code + ilen;
+		}
 
-	/* context synchronisation performed by __patch_instruction (isync or exception) */
-	stop_using_temp_mm(patching_mm, orig_mm);
+		/* context synchronisation performed by __patch_instruction (isync or exception) */
+		stop_using_temp_mm(patching_mm, orig_mm);
 
-	pte_clear(patching_mm, text_poke_addr, pte);
-	/*
-	 * ptesync to order PTE update before TLB invalidation done
-	 * by radix__local_flush_tlb_page_psize (in _tlbiel_va)
-	 */
-	local_flush_tlb_page_psize(patching_mm, text_poke_addr, mmu_virtual_psize);
+		pte_clear(patching_mm, text_poke_addr, pte);
+		/*
+		 * ptesync to order PTE update before TLB invalidation done
+		 * by radix__local_flush_tlb_page_psize (in _tlbiel_va)
+		 */
+		local_flush_tlb_page_psize(patching_mm, text_poke_addr, mmu_virtual_psize);
+		if (err)
+			break;
+	} while (addr < end);
 
 	pte_unmap_unlock(pte, ptl);
 
 	return err;
 }
 
-static int __do_patch_instruction(u32 *addr, ppc_inst_t instr)
+static int __do_patch_instructions(u32 *addr, u32 *code, bool fill_inst, size_t len)
 {
-	int err;
-	u32 *patch_addr;
-	unsigned long text_poke_addr;
+	unsigned long text_poke_addr, pfn;
+	u32 *patch_addr, *end, *pend;
+	ppc_inst_t instr;
+	int ilen, err;
 	pte_t *pte;
-	unsigned long pfn = get_patch_pfn(addr);
 
 	text_poke_addr = (unsigned long)__this_cpu_read(cpu_patching_context.addr) & PAGE_MASK;
-	patch_addr = (u32 *)(text_poke_addr + offset_in_page(addr));
-
 	pte = __this_cpu_read(cpu_patching_context.pte);
-	__set_pte_at(&init_mm, text_poke_addr, pte, pfn_pte(pfn, PAGE_KERNEL), 0);
-	/* See ptesync comment in radix__set_pte_at() */
-	if (radix_enabled())
-		asm volatile("ptesync": : :"memory");
 
-	err = __patch_instruction(addr, instr, patch_addr);
+	end = (void *)addr + len;
+	do {
+		pfn = get_patch_pfn(addr);
+		__set_pte_at(&init_mm, text_poke_addr, pte, pfn_pte(pfn, PAGE_KERNEL), 0);
+		/* See ptesync comment in radix__set_pte_at() */
+		if (radix_enabled())
+			asm volatile("ptesync": : :"memory");
+
+		patch_addr = (u32 *)(text_poke_addr + offset_in_page(addr));
+		pend = (void *)addr + PAGE_SIZE - offset_in_page(addr);
+		if (end < pend)
+			pend = end;
+
+		while (addr < pend) {
+			instr = ppc_inst_read(code);
+			ilen = ppc_inst_len(instr);
+			err = __patch_instruction(addr, instr, patch_addr);
+			if (err)
+				break;
+
+			patch_addr = (void *)patch_addr + ilen;
+			addr = (void *)addr + ilen;
+			if (!fill_inst)
+				code = (void *)code + ilen;
+		}
 
-	pte_clear(&init_mm, text_poke_addr, pte);
-	flush_tlb_kernel_range(text_poke_addr, text_poke_addr + PAGE_SIZE);
+		pte_clear(&init_mm, text_poke_addr, pte);
+		flush_tlb_kernel_range(text_poke_addr, text_poke_addr + PAGE_SIZE);
+		if (err)
+			break;
+	} while (addr < end);
 
 	return err;
 }
@@ -369,15 +409,34 @@ int patch_instruction(u32 *addr, ppc_inst_t instr)
 
 	local_irq_save(flags);
 	if (mm_patch_enabled())
-		err = __do_patch_instruction_mm(addr, instr);
+		err = __do_patch_instructions_mm(addr, (u32 *)&instr, false, ppc_inst_len(instr));
 	else
-		err = __do_patch_instruction(addr, instr);
+		err = __do_patch_instructions(addr, (u32 *)&instr, false, ppc_inst_len(instr));
 	local_irq_restore(flags);
 
 	return err;
 }
 NOKPROBE_SYMBOL(patch_instruction);
 
+/*
+ * Patch 'addr' with 'len' bytes of instructions from 'code'.
+ */
+int patch_instructions(u32 *addr, u32 *code, bool fill_inst, size_t len)
+{
+	unsigned long flags;
+	int err;
+
+	local_irq_save(flags);
+	if (mm_patch_enabled())
+		err = __do_patch_instructions_mm(addr, code, fill_inst, len);
+	else
+		err = __do_patch_instructions(addr, code, fill_inst, len);
+	local_irq_restore(flags);
+
+	return err;
+}
+NOKPROBE_SYMBOL(patch_instructions);
+
 int patch_branch(u32 *addr, unsigned long target, int flags)
 {
 	ppc_inst_t instr;
-- 
2.39.2

[PATCH v2 2/4] powerpc/bpf: implement bpf_arch_text_copy

[Hari Bathini]

bpf_arch_text_copy is used to dump JITed binary to RX page, allowing
multiple BPF programs to share the same page. Use the newly introduced
patch_instructions() to implement it. Around 5X improvement in speed
of execution observed, using the new patch_instructions() function
over patch_instruction(), while running the tests from test_bpf.ko.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
---
 arch/powerpc/net/bpf_jit_comp.c | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index e93aefcfb83f..0a70319116d1 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c
@@ -13,9 +13,12 @@
 #include <linux/netdevice.h>
 #include <linux/filter.h>
 #include <linux/if_vlan.h>
-#include <asm/kprobes.h>
+#include <linux/memory.h>
 #include <linux/bpf.h>
 
+#include <asm/kprobes.h>
+#include <asm/code-patching.h>
+
 #include "bpf_jit.h"
 
 static void bpf_jit_fill_ill_insns(void *area, unsigned int size)
@@ -272,3 +275,21 @@ int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, int pass, struct code
 	ctx->exentry_idx++;
 	return 0;
 }
+
+void *bpf_arch_text_copy(void *dst, void *src, size_t len)
+{
+	void *ret = ERR_PTR(-EINVAL);
+	int err;
+
+	if (WARN_ON_ONCE(core_kernel_text((unsigned long)dst)))
+		return ret;
+
+	ret = dst;
+	mutex_lock(&text_mutex);
+	err = patch_instructions(dst, src, false, len);
+	if (err)
+		ret = ERR_PTR(err);
+	mutex_unlock(&text_mutex);
+
+	return ret;
+}
-- 
2.39.2

[PATCH v2 3/4] powerpc/bpf: implement bpf_arch_text_invalidate for bpf_prog_pack

[Hari Bathini]

Implement bpf_arch_text_invalidate and use it to fill unused part of
the bpf_prog_pack with trap instructions when a BPF program is freed.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
---
 arch/powerpc/net/bpf_jit_comp.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index 0a70319116d1..d1794d9f0154 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c
@@ -293,3 +293,18 @@ void *bpf_arch_text_copy(void *dst, void *src, size_t len)
 
 	return ret;
 }
+
+int bpf_arch_text_invalidate(void *dst, size_t len)
+{
+	u32 inst = BREAKPOINT_INSTRUCTION;
+	int ret = -EINVAL;
+
+	if (WARN_ON_ONCE(core_kernel_text((unsigned long)dst)))
+		return ret;
+
+	mutex_lock(&text_mutex);
+	ret = patch_instructions(dst, &inst, true, len);
+	mutex_unlock(&text_mutex);
+
+	return ret;
+}
-- 
2.39.2

[PATCH v2 4/4] powerpc/bpf: use bpf_jit_binary_pack_[alloc|finalize|free]

[Hari Bathini]

Use bpf_jit_binary_pack_alloc in powerpc jit. The jit engine first
writes the program to the rw buffer. When the jit is done, the program
is copied to the final location with bpf_jit_binary_pack_finalize.
With multiple jit_subprogs, bpf_jit_free is called on some subprograms
that haven't got bpf_jit_binary_pack_finalize() yet. Implement custom
bpf_jit_free() like in commit 1d5f82d9dd47 ("bpf, x86: fix freeing of
not-finalized bpf_prog_pack") to call bpf_jit_binary_pack_finalize(),
if necessary. While here, correct the misnomer powerpc64_jit_data to
powerpc_jit_data as it is meant for both ppc32 and ppc64.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
---
 arch/powerpc/net/bpf_jit.h        |   7 +-
 arch/powerpc/net/bpf_jit_comp.c   | 104 +++++++++++++++++++++---------
 arch/powerpc/net/bpf_jit_comp32.c |   4 +-
 arch/powerpc/net/bpf_jit_comp64.c |   6 +-
 4 files changed, 83 insertions(+), 38 deletions(-)

diff --git a/arch/powerpc/net/bpf_jit.h b/arch/powerpc/net/bpf_jit.h
index d767e39d5645..a8b7480c4d43 100644
--- a/arch/powerpc/net/bpf_jit.h
+++ b/arch/powerpc/net/bpf_jit.h
@@ -168,15 +168,16 @@ static inline void bpf_clear_seen_register(struct codegen_context *ctx, int i)
 
 void bpf_jit_init_reg_mapping(struct codegen_context *ctx);
 int bpf_jit_emit_func_call_rel(u32 *image, struct codegen_context *ctx, u64 func);
-int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *ctx,
+int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, u32 *fimage, struct codegen_context *ctx,
 		       u32 *addrs, int pass, bool extra_pass);
 void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx);
 void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx);
 void bpf_jit_realloc_regs(struct codegen_context *ctx);
 int bpf_jit_emit_exit_insn(u32 *image, struct codegen_context *ctx, int tmp_reg, long exit_addr);
 
-int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, int pass, struct codegen_context *ctx,
-			  int insn_idx, int jmp_off, int dst_reg);
+int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, u32 *fimage, int pass,
+			  struct codegen_context *ctx, int insn_idx,
+			  int jmp_off, int dst_reg);
 
 #endif
 
diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index d1794d9f0154..ece75c829499 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c
@@ -42,10 +42,11 @@ int bpf_jit_emit_exit_insn(u32 *image, struct codegen_context *ctx, int tmp_reg,
 	return 0;
 }
 
-struct powerpc64_jit_data {
-	struct bpf_binary_header *header;
+struct powerpc_jit_data {
+	struct bpf_binary_header *hdr;
+	struct bpf_binary_header *fhdr;
 	u32 *addrs;
-	u8 *image;
+	u8 *fimage;
 	u32 proglen;
 	struct codegen_context ctx;
 };
@@ -62,15 +63,18 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 	u8 *image = NULL;
 	u32 *code_base;
 	u32 *addrs;
-	struct powerpc64_jit_data *jit_data;
+	struct powerpc_jit_data *jit_data;
 	struct codegen_context cgctx;
 	int pass;
 	int flen;
-	struct bpf_binary_header *bpf_hdr;
+	struct bpf_binary_header *fhdr = NULL;
+	struct bpf_binary_header *hdr = NULL;
 	struct bpf_prog *org_fp = fp;
 	struct bpf_prog *tmp_fp;
 	bool bpf_blinded = false;
 	bool extra_pass = false;
+	u8 *fimage = NULL;
+	u32 *fcode_base;
 	u32 extable_len;
 	u32 fixup_len;
 
@@ -100,9 +104,11 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 	addrs = jit_data->addrs;
 	if (addrs) {
 		cgctx = jit_data->ctx;
-		image = jit_data->image;
-		bpf_hdr = jit_data->header;
+		fimage = jit_data->fimage;
+		fhdr = jit_data->fhdr;
 		proglen = jit_data->proglen;
+		hdr = jit_data->hdr;
+		image = (void *)hdr + ((void *)fimage - (void *)fhdr);
 		extra_pass = true;
 		goto skip_init_ctx;
 	}
@@ -120,7 +126,7 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 	cgctx.stack_size = round_up(fp->aux->stack_depth, 16);
 
 	/* Scouting faux-generate pass 0 */
-	if (bpf_jit_build_body(fp, 0, &cgctx, addrs, 0, false)) {
+	if (bpf_jit_build_body(fp, NULL, NULL, &cgctx, addrs, 0, false)) {
 		/* We hit something illegal or unsupported. */
 		fp = org_fp;
 		goto out_addrs;
@@ -135,7 +141,7 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 	 */
 	if (cgctx.seen & SEEN_TAILCALL || !is_offset_in_branch_range((long)cgctx.idx * 4)) {
 		cgctx.idx = 0;
-		if (bpf_jit_build_body(fp, 0, &cgctx, addrs, 0, false)) {
+		if (bpf_jit_build_body(fp, NULL, NULL, &cgctx, addrs, 0, false)) {
 			fp = org_fp;
 			goto out_addrs;
 		}
@@ -157,17 +163,19 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 	proglen = cgctx.idx * 4;
 	alloclen = proglen + FUNCTION_DESCR_SIZE + fixup_len + extable_len;
 
-	bpf_hdr = bpf_jit_binary_alloc(alloclen, &image, 4, bpf_jit_fill_ill_insns);
-	if (!bpf_hdr) {
+	fhdr = bpf_jit_binary_pack_alloc(alloclen, &fimage, 4, &hdr, &image,
+					      bpf_jit_fill_ill_insns);
+	if (!fhdr) {
 		fp = org_fp;
 		goto out_addrs;
 	}
 
 	if (extable_len)
-		fp->aux->extable = (void *)image + FUNCTION_DESCR_SIZE + proglen + fixup_len;
+		fp->aux->extable = (void *)fimage + FUNCTION_DESCR_SIZE + proglen + fixup_len;
 
 skip_init_ctx:
 	code_base = (u32 *)(image + FUNCTION_DESCR_SIZE);
+	fcode_base = (u32 *)(fimage + FUNCTION_DESCR_SIZE);
 
 	/* Code generation passes 1-2 */
 	for (pass = 1; pass < 3; pass++) {
@@ -175,8 +183,9 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 		cgctx.idx = 0;
 		cgctx.alt_exit_addr = 0;
 		bpf_jit_build_prologue(code_base, &cgctx);
-		if (bpf_jit_build_body(fp, code_base, &cgctx, addrs, pass, extra_pass)) {
-			bpf_jit_binary_free(bpf_hdr);
+		if (bpf_jit_build_body(fp, code_base, fcode_base, &cgctx, addrs, pass, extra_pass)) {
+			bpf_arch_text_copy(&fhdr->size, &hdr->size, sizeof(hdr->size));
+			bpf_jit_binary_pack_free(fhdr, hdr);
 			fp = org_fp;
 			goto out_addrs;
 		}
@@ -192,21 +201,23 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 		 * Note that we output the base address of the code_base
 		 * rather than image, since opcodes are in code_base.
 		 */
-		bpf_jit_dump(flen, proglen, pass, code_base);
+		bpf_jit_dump(flen, proglen, pass, fcode_base);
 
 #ifdef CONFIG_PPC64_ELF_ABI_V1
 	/* Function descriptor nastiness: Address + TOC */
-	((u64 *)image)[0] = (u64)code_base;
+	((u64 *)image)[0] = (u64)fcode_base;
 	((u64 *)image)[1] = local_paca->kernel_toc;
 #endif
 
-	fp->bpf_func = (void *)image;
+	fp->bpf_func = (void *)fimage;
 	fp->jited = 1;
 	fp->jited_len = proglen + FUNCTION_DESCR_SIZE;
 
-	bpf_flush_icache(bpf_hdr, (u8 *)bpf_hdr + bpf_hdr->size);
 	if (!fp->is_func || extra_pass) {
-		bpf_jit_binary_lock_ro(bpf_hdr);
+		if (bpf_jit_binary_pack_finalize(fp, fhdr, hdr)) {
+			fp = org_fp;
+			goto out_addrs;
+		}
 		bpf_prog_fill_jited_linfo(fp, addrs);
 out_addrs:
 		kfree(addrs);
@@ -216,8 +227,9 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 		jit_data->addrs = addrs;
 		jit_data->ctx = cgctx;
 		jit_data->proglen = proglen;
-		jit_data->image = image;
-		jit_data->header = bpf_hdr;
+		jit_data->fimage = fimage;
+		jit_data->fhdr = fhdr;
+		jit_data->hdr = hdr;
 	}
 
 out:
@@ -231,12 +243,13 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
  * The caller should check for (BPF_MODE(code) == BPF_PROBE_MEM) before calling
  * this function, as this only applies to BPF_PROBE_MEM, for now.
  */
-int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, int pass, struct codegen_context *ctx,
-			  int insn_idx, int jmp_off, int dst_reg)
+int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, u32 *fimage, int pass,
+			  struct codegen_context *ctx, int insn_idx, int jmp_off,
+			  int dst_reg)
 {
 	off_t offset;
 	unsigned long pc;
-	struct exception_table_entry *ex;
+	struct exception_table_entry *ex, *ex_entry;
 	u32 *fixup;
 
 	/* Populate extable entries only in the last pass */
@@ -247,9 +260,16 @@ int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, int pass, struct code
 	    WARN_ON_ONCE(ctx->exentry_idx >= fp->aux->num_exentries))
 		return -EINVAL;
 
+	/*
+	 * Program is firt written to image before copying to the
+	 * final location (fimage). Accordingly, update in the image first.
+	 * As all offsets used are relative, copying as is to the
+	 * final location should be alright.
+	 */
 	pc = (unsigned long)&image[insn_idx];
+	ex = (void *)fp->aux->extable - (void *)fimage + (void *)image;
 
-	fixup = (void *)fp->aux->extable -
+	fixup = (void *)ex -
 		(fp->aux->num_exentries * BPF_FIXUP_LEN * 4) +
 		(ctx->exentry_idx * BPF_FIXUP_LEN * 4);
 
@@ -260,17 +280,17 @@ int bpf_add_extable_entry(struct bpf_prog *fp, u32 *image, int pass, struct code
 	fixup[BPF_FIXUP_LEN - 1] =
 		PPC_RAW_BRANCH((long)(pc + jmp_off) - (long)&fixup[BPF_FIXUP_LEN - 1]);
 
-	ex = &fp->aux->extable[ctx->exentry_idx];
+	ex_entry = &ex[ctx->exentry_idx];
 
-	offset = pc - (long)&ex->insn;
+	offset = pc - (long)&ex_entry->insn;
 	if (WARN_ON_ONCE(offset >= 0 || offset < INT_MIN))
 		return -ERANGE;
-	ex->insn = offset;
+	ex_entry->insn = offset;
 
-	offset = (long)fixup - (long)&ex->fixup;
+	offset = (long)fixup - (long)&ex_entry->fixup;
 	if (WARN_ON_ONCE(offset >= 0 || offset < INT_MIN))
 		return -ERANGE;
-	ex->fixup = offset;
+	ex_entry->fixup = offset;
 
 	ctx->exentry_idx++;
 	return 0;
@@ -308,3 +328,27 @@ int bpf_arch_text_invalidate(void *dst, size_t len)
 
 	return ret;
 }
+
+void bpf_jit_free(struct bpf_prog *fp)
+{
+	if (fp->jited) {
+		struct powerpc_jit_data *jit_data = fp->aux->jit_data;
+		struct bpf_binary_header *hdr;
+
+		/*
+		 * If we fail the final pass of JIT (from jit_subprogs),
+		 * the program may not be finalized yet. Call finalize here
+		 * before freeing it.
+		 */
+		if (jit_data) {
+			bpf_jit_binary_pack_finalize(fp, jit_data->fhdr, jit_data->hdr);
+			kvfree(jit_data->addrs);
+			kfree(jit_data);
+		}
+		hdr = bpf_jit_binary_pack_hdr(fp);
+		bpf_jit_binary_pack_free(hdr, NULL);
+		WARN_ON_ONCE(!bpf_prog_kallsyms_verify_off(fp));
+	}
+
+	bpf_prog_unlock_free(fp);
+}
diff --git a/arch/powerpc/net/bpf_jit_comp32.c b/arch/powerpc/net/bpf_jit_comp32.c
index 7f91ea064c08..fb2761b54d64 100644
--- a/arch/powerpc/net/bpf_jit_comp32.c
+++ b/arch/powerpc/net/bpf_jit_comp32.c
@@ -278,7 +278,7 @@ static int bpf_jit_emit_tail_call(u32 *image, struct codegen_context *ctx, u32 o
 }
 
 /* Assemble the body code between the prologue & epilogue */
-int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *ctx,
+int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, u32 *fimage, struct codegen_context *ctx,
 		       u32 *addrs, int pass, bool extra_pass)
 {
 	const struct bpf_insn *insn = fp->insnsi;
@@ -997,7 +997,7 @@ int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *
 					jmp_off += 4;
 				}
 
-				ret = bpf_add_extable_entry(fp, image, pass, ctx, insn_idx,
+				ret = bpf_add_extable_entry(fp, image, fimage, pass, ctx, insn_idx,
 							    jmp_off, dst_reg);
 				if (ret)
 					return ret;
diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
index 8dd3cabaa83a..37a8970a7065 100644
--- a/arch/powerpc/net/bpf_jit_comp64.c
+++ b/arch/powerpc/net/bpf_jit_comp64.c
@@ -343,7 +343,7 @@ asm (
 );
 
 /* Assemble the body code between the prologue & epilogue */
-int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *ctx,
+int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, u32 *fimage, struct codegen_context *ctx,
 		       u32 *addrs, int pass, bool extra_pass)
 {
 	enum stf_barrier_type stf_barrier = stf_barrier_type_get();
@@ -922,8 +922,8 @@ int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *
 				addrs[++i] = ctx->idx * 4;
 
 			if (BPF_MODE(code) == BPF_PROBE_MEM) {
-				ret = bpf_add_extable_entry(fp, image, pass, ctx, ctx->idx - 1,
-							    4, dst_reg);
+				ret = bpf_add_extable_entry(fp, image, fimage, pass, ctx,
+							    ctx->idx - 1, 4, dst_reg);
 				if (ret)
 					return ret;
 			}
-- 
2.39.2

[PATCH v2 0/4] enable bpf_prog_pack allocator for powerpc

[Hari Bathini]

Most BPF programs are small, but they consume a page each. For systems
with busy traffic and many BPF programs, this may also add significant
pressure on instruction TLB. High iTLB pressure usually slows down the
whole system causing visible performance degradation for production
workloads.

bpf_prog_pack, a customized allocator that packs multiple bpf programs
into preallocated memory chunks, was proposed [1] to address it. This
series extends this support on powerpc.

The first patch introduces patch_instructions() function to enable
patching more than one instruction at a time. This change showed
around 5X improvement in the time taken to run test_bpf test cases.
Patches 2 & 3 add the arch specific functions needed to support this
feature. Patch 4 enables the support for powerpc and ensures cleanup
is handled racefully. Tested the changes successfully.

[1] https://lore.kernel.org/bpf/20220204185742.271030-1-song@kernel.org/
[2] https://lore.kernel.org/all/20221110184303.393179-1-hbathini@linux.ibm.com/ 

Changes in v2:
* Introduced patch_instructions() to help with patching bpf programs.

Hari Bathini (4):
  powerpc/code-patching: introduce patch_instructions()
  powerpc/bpf: implement bpf_arch_text_copy
  powerpc/bpf: implement bpf_arch_text_invalidate for bpf_prog_pack
  powerpc/bpf: use bpf_jit_binary_pack_[alloc|finalize|free]

 arch/powerpc/include/asm/code-patching.h |   1 +
 arch/powerpc/lib/code-patching.c         | 151 ++++++++++++++++-------
 arch/powerpc/net/bpf_jit.h               |   7 +-
 arch/powerpc/net/bpf_jit_comp.c          | 142 ++++++++++++++++-----
 arch/powerpc/net/bpf_jit_comp32.c        |   4 +-
 arch/powerpc/net/bpf_jit_comp64.c        |   6 +-
 6 files changed, 226 insertions(+), 85 deletions(-)

-- 
2.39.2

[PATCH v2 3/4] powerpc/bpf: implement bpf_arch_text_invalidate for bpf_prog_pack

[Hari Bathini]

Implement bpf_arch_text_invalidate and use it to fill unused part of
the bpf_prog_pack with trap instructions when a BPF program is freed.

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
---
 arch/powerpc/net/bpf_jit_comp.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index 0a70319116d1..d1794d9f0154 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c
@@ -293,3 +293,18 @@ void *bpf_arch_text_copy(void *dst, void *src, size_t len)
 
 	return ret;
 }
+
+int bpf_arch_text_invalidate(void *dst, size_t len)
+{
+	u32 inst = BREAKPOINT_INSTRUCTION;
+	int ret = -EINVAL;
+
+	if (WARN_ON_ONCE(core_kernel_text((unsigned long)dst)))
+		return ret;
+
+	mutex_lock(&text_mutex);
+	ret = patch_instructions(dst, &inst, true, len);
+	mutex_unlock(&text_mutex);
+
+	return ret;
+}
-- 
2.39.2

Re: [PATCH v2 3/4] powerpc/bpf: implement bpf_arch_text_invalidate for bpf_prog_pack

[Song Liu]

On Thu, Mar 9, 2023 at 10:02 AM Hari Bathini <hbathini@linux.ibm.com> wrote:
>
> Implement bpf_arch_text_invalidate and use it to fill unused part of
> the bpf_prog_pack with trap instructions when a BPF program is freed.
>
> Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
> ---
>  arch/powerpc/net/bpf_jit_comp.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
>
> diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
> index 0a70319116d1..d1794d9f0154 100644
> --- a/arch/powerpc/net/bpf_jit_comp.c
> +++ b/arch/powerpc/net/bpf_jit_comp.c
> @@ -293,3 +293,18 @@ void *bpf_arch_text_copy(void *dst, void *src, size_t len)
>
>         return ret;
>  }
> +
> +int bpf_arch_text_invalidate(void *dst, size_t len)
> +{
> +       u32 inst = BREAKPOINT_INSTRUCTION;
> +       int ret = -EINVAL;

No need to set to -EINVAL here.

> +
> +       if (WARN_ON_ONCE(core_kernel_text((unsigned long)dst)))
> +               return ret;

Just return -EINVAL instead.

> +
> +       mutex_lock(&text_mutex);
> +       ret = patch_instructions(dst, &inst, true, len);
> +       mutex_unlock(&text_mutex);
> +
> +       return ret;
> +}
> --
> 2.39.2
>