Skip to contentSkip to conversation

treewide: remove unreachable memblock_reserve() return value checks in early boot

PATCH In review Sang-Heon Jeon 2 participants 9 messages Jul 6 – Jul 7

Subsystems (6)
Memory management x86 PowerPC SuperH +2 more subsystems
lore
Share & tools
lore
#0 lore raw

v2 cover treewide: remove unreachable memblock_reserve() return value checks in early boot

memblock_reserve() can only return an error after memblock_allow_resize() has been called. Before that it either succeeds or panics, never returning an error.

Before memblock_allow_resize() is called, the return value checks of memblock_reserve() are unreachable and can be removed.

All the call sites are in different trees, so I split the change into one patch per arch/subsystem. The patches are independent and can be applied separately.

---
Changes from v1 [1]
- add reviewed-by tag
- rebased onto latest memblock for-next

[1] https://lore.kernel.org/all/[email protected]/
---
Hello Mike,

Like the previous memblock cleanup series, could you review and take this patch series to the memblock tree?

Thank you for taking valuable time.

Best Regards,
Sang-Heon Jeon
---

Sang-Heon Jeon (5):
x86/setup: remove unreachable memblock_reserve() return value check
sh: remove unreachable memblock_reserve() return value check powerpc/fadump: remove unreachable memblock_reserve() return value
checks
xtensa: remove unreachable memblock_reserve() return value checks Revert "tpm: do not ignore memblock_reserve return value"

 arch/powerpc/kernel/fadump.c   | 14 +++-----------
 arch/sh/kernel/machine_kexec.c |  7 +------
 arch/x86/kernel/setup.c        |  7 ++-----
 arch/xtensa/kernel/setup.c     | 13 +++++++------
 drivers/firmware/efi/tpm.c     |  7 +------
 5 files changed, 14 insertions(+), 34 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/5 x86/setup: remove unreachable memblock_reserve() return value check

trim_snb_memory() runs from early_reserve_memory() in setup_arch(), before e820__memblock_setup() calls memblock_allow_resize(). At that point memblock_reserve() either succeeds or panics, never returning an error.

Therefore the return value check is unreachable, so remove it and the related logic.

No functional change.

---
 arch/x86/kernel/setup.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)
arch/x86/kernel/setup.c +2 -5
index 46882ce79c3a..f9395bc632ca 100644
@@ -742,11 +742,8 @@ static void __init trim_snb_memory(void)
 	 * setup_arch(), so there is no need to reserve it here.
 	 */
 
-	for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
-		if (memblock_reserve(bad_pages[i], PAGE_SIZE))
-			printk(KERN_WARNING "failed to reserve 0x%08lx\n",
-			       bad_pages[i]);
-	}
+	for (i = 0; i < ARRAY_SIZE(bad_pages); i++)
+		memblock_reserve(bad_pages[i], PAGE_SIZE);
 }
 
 static void __init trim_bios_range(void)
Signature · 1 line
2.43.0
#2 lore raw

PATCH v2 2/5 sh: remove unreachable memblock_reserve() return value check

In reply to Sang-Heon Jeon

reserve_crashkernel() runs from early_reserve_mem() in paging_init(), before the same function calls memblock_allow_resize(). At that point memblock_reserve() either succeeds or panics, never returning an error.

Therefore the return value check is unreachable, so remove it and the related logic.

No functional change.

---
 arch/sh/kernel/machine_kexec.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
arch/sh/kernel/machine_kexec.c +1 -6
index 37073ca1e0ad..ba9c412f14d9 100644
@@ -165,12 +165,7 @@ void __init reserve_crashkernel(void)
 			goto disable;
 		}
 	} else {
-		ret = memblock_reserve(crashk_res.start, crash_size);
-		if (unlikely(ret < 0)) {
-			pr_err("crashkernel reservation failed - "
-			       "memory is in use\n");
-			goto disable;
-		}
+		memblock_reserve(crashk_res.start, crash_size);
 	}
 
 	crashk_res.end = crashk_res.start + crash_size - 1;
Signature · 1 line
2.43.0
#3 lore raw

PATCH v2 3/5 powerpc/fadump: remove unreachable memblock_reserve() return value checks

In reply to Sang-Heon Jeon

fadump_append_bootargs() and fadump_reserve_mem() both run from early_init_devtree(), before it calls memblock_allow_resize(). At that point memblock_reserve() either succeeds or panics, never returning an error.

Therefore the return value checks are unreachable, so remove them and the related logic.

No functional change.

---
 arch/powerpc/kernel/fadump.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)
arch/powerpc/kernel/fadump.c +3 -11
index 7f79c9aea4a9..1629a82b9528 100644
@@ -154,13 +154,8 @@ void __init fadump_append_bootargs(void)
 	if (!fw_dump.dump_active || !fw_dump.param_area_supported || !fw_dump.param_area)
 		return;
 
-	if (fw_dump.param_area < fw_dump.boot_mem_top) {
-		if (memblock_reserve(fw_dump.param_area, COMMAND_LINE_SIZE)) {
-			pr_warn("WARNING: Can't use additional parameters area!\n");
-			fw_dump.param_area = 0;
-			return;
-		}
-	}
+	if (fw_dump.param_area < fw_dump.boot_mem_top)
+		memblock_reserve(fw_dump.param_area, COMMAND_LINE_SIZE);
 
 	append_args = (char *)fw_dump.param_area;
 	len = strlen(boot_command_line);
@@ -632,10 +627,7 @@ int __init fadump_reserve_mem(void)
 		    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
 			goto error_out;
 
-		if (memblock_reserve(base, size)) {
-			pr_err("Failed to reserve memory!\n");
-			goto error_out;
-		}
+		memblock_reserve(base, size);
 
 		pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
 			(size >> 20), base, (memblock_phys_mem_size() >> 20));
Signature · 1 line
2.43.0
#4 lore raw

PATCH v2 4/5 xtensa: remove unreachable memblock_reserve() return value checks

In reply to Sang-Heon Jeon

mem_reserve() wraps memblock_reserve(), and its callers check the return value. xtensa never calls memblock_allow_resize(), so memblock_reserve() either succeeds or panics, never returning an error.

Therefore the return value checks are unreachable, so remove them and make mem_reserve() return void.

No functional change.

---
 arch/xtensa/kernel/setup.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/xtensa/kernel/setup.c +7 -6
index f72e280363be..c7367b81cc36 100644
@@ -268,10 +268,10 @@ void __init init_arch(bp_tag_t *bp_start)
  * Initialize system. Setup memory and reserve regions.
  */
 
-static inline int __init_memblock mem_reserve(unsigned long start,
-					      unsigned long end)
+static inline void __init_memblock mem_reserve(unsigned long start,
+					       unsigned long end)
 {
-	return memblock_reserve(start, end - start);
+	memblock_reserve(start, end - start);
 }
 
 void __init setup_arch(char **cmdline_p)
@@ -290,11 +290,12 @@ void __init setup_arch(char **cmdline_p)
 	/* Reserve some memory regions */
 
 #ifdef CONFIG_BLK_DEV_INITRD
-	if (initrd_start < initrd_end &&
-	    !mem_reserve(__pa(initrd_start), __pa(initrd_end)))
+	if (initrd_start < initrd_end) {
+		mem_reserve(__pa(initrd_start), __pa(initrd_end));
 		initrd_below_start_ok = 1;
-	else
+	} else {
 		initrd_start = 0;
+	}
 #endif
 
 	mem_reserve(__pa(_stext), __pa(_end));
Signature · 1 line
2.43.0
#5 lore raw

PATCH v2 5/5 Revert "tpm: do not ignore memblock_reserve return value"

In reply to Sang-Heon Jeon

This reverts commit c33c28f9f6e017702374e1ab907dfa5a63db5301.

efi_tpm_eventlog_init() runs from efi_config_parse_tables() during early EFI initialization, before memblock_allow_resize(). At that point memblock_reserve() either succeeds or panics, never returning an error.

Therefore the return value check is unreachable, so restore the original unconditional memblock_reserve().

No functional change.

---
 drivers/firmware/efi/tpm.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
drivers/firmware/efi/tpm.c +1 -6
index cdd431027065..288bcf7a5b99 100644
@@ -61,12 +61,7 @@ int __init efi_tpm_eventlog_init(void)
 	}
 
 	tbl_size = sizeof(*log_tbl) + log_tbl->size;
-	if (memblock_reserve(efi.tpm_log, tbl_size)) {
-		pr_err("TPM Event Log memblock reserve fails (0x%lx, 0x%x)\n",
-		       efi.tpm_log, tbl_size);
-		ret = -ENOMEM;
-		goto out;
-	}
+	memblock_reserve(efi.tpm_log, tbl_size);
 
 	if (efi.tpm_final_log == EFI_INVALID_TABLE_ADDR) {
 		pr_info("TPM Final Events table not present\n");
Signature · 1 line
2.43.0
#6 lore raw
In reply to Sang-Heon Jeon

Hi Sang-Heon,

Quoted 6 lines · Sang-Heon Jeon

memblock_reserve() can only return an error after memblock_allow_resize() has been called. Before that it either succeeds or panics, never returning an error.

Before memblock_allow_resize() is called, the return value checks of memblock_reserve() are unreachable and can be removed.

I'd rather keep these checks.

Removing them relies on internal details of memblock_reserve() implementation and the existing event sequence. If the code would move around relying on panic in memblock_reserve() may not be correct.

And the few bytes and cycles the change saves do not worth the churn.

Signature · 2 lines
Sincerely yours,
Mike.
#7 lore raw

Hi Mike,

Quoted 11 lines · Mike Rapoport

Hi Sang-Heon,

Quoted 6 lines · Sang-Heon Jeon

memblock_reserve() can only return an error after memblock_allow_resize() has been called. Before that it either succeeds or panics, never returning an error.

Before memblock_allow_resize() is called, the return value checks of memblock_reserve() are unreachable and can be removed.

Quoted 7 lines

I'd rather keep these checks.

Removing them relies on internal details of memblock_reserve() implementation and the existing event sequence. If the code would move around relying on panic in memblock_reserve() may not be correct.

And the few bytes and cycles the change saves do not worth the churn.

Makes sense to me.

But most early boot callers of memblock_reserve() don't check the return value, so I thought we already rely on its internal behavior anyway. So the few remaining checks just looked a bit inconsistent to me.

Would you still prefer to keep these checks? If so, I'm fine with dropping this patch series. It's not a big deal :)

Thank you for taking the time to review this series.

Quoted 3 lines

--
Sincerely yours,
Mike.

Best Regards,
Sang-Heon Jeon

#8 lore raw
Quoted 20 lines · Sang-Heon Jeon

Hi Mike,

Quoted 17 lines · Mike Rapoport

Hi Sang-Heon,

Quoted 6 lines · Sang-Heon Jeon

memblock_reserve() can only return an error after memblock_allow_resize() has been called. Before that it either succeeds or panics, never returning an error.

Before memblock_allow_resize() is called, the return value checks of memblock_reserve() are unreachable and can be removed.

I'd rather keep these checks.

Removing them relies on internal details of memblock_reserve() implementation and the existing event sequence. If the code would move around relying on panic in memblock_reserve() may not be correct.

Quoted 8 lines
Quoted 1 line

And the few bytes and cycles the change saves do not worth the churn.

Makes sense to me.

But most early boot callers of memblock_reserve() don't check the return value, so I thought we already rely on its internal behavior anyway. So the few remaining checks just looked a bit inconsistent to me.

In reality it's very unlikely for memblock_reserve() to fail, especially after resize is allowed.
And if it does fail, the system would trip on a memory error, usually sooner than later.

Quoted 2 lines

Would you still prefer to keep these checks? If so, I'm fine with dropping this patch series. It's not a big deal :)

Let's keep the checks as they are now.

Quoted 2 lines

Best Regards,
Sang-Heon Jeon

Signature · 2 lines
Sincerely yours,
Mike.