Skip to contentSkip to conversation

ppc/fadump: collect dump if the collected size is lesser than reserved

PATCH Superseded Shivang Upadhyay 2 participants 4 messages Aug 26 – Aug 27 PowerPC

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 ppc/fadump: collect dump if the collected size is lesser than reserved

During Fadump in Qemu VM, when maxcpus value is set to more than current cpus, following failure is observed.

    [0.000000] rtas: Dump taken by platform is incomplete (-1)

This is because the CPU_STATE_DATA is allocated for maxcpus, while the data is only filled for current cpus. As per current implementation of Fadump, dumped_bytes and source_len for a region have to match, Which is failing for qemu's case, Even tough the dumped bytes are reported correctly for only the current cpus. After this /proc/vmcore generatition also fails.

Allowing dumped_bytes to be lesser than or equal to allocated length, for CPU_STATE_DATA Fadump region.

---
ChangeLog:

v2: Only allowing lesser size for CPU_STATE_DATE Fadump region.

v1: https://lore.kernel.org/qemu-devel/[email protected]/

---
 arch/powerpc/platforms/pseries/rtas-fadump.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)
arch/powerpc/platforms/pseries/rtas-fadump.c +15 -2
index 3bb4ac2ab6cc..ea1a0a8ac9eb 100644
@@ -459,7 +459,10 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
 	/* Check if the dump data is valid. */
 	for (int i = 0; i < be16_to_cpu(fdm_active->header.dump_num_sections); i++) {
 		int type = be16_to_cpu(fdm_active->rgn[i].source_data_type);
+		uint64_t bytes_dumped = be64_to_cpu(fdm_active->rgn[i].bytes_dumped);
+		uint64_t source_len = be64_to_cpu(fdm_active->rgn[i].source_len);
 		int rc = 0;
+		int region_collected;
 
 		switch (type) {
 		case RTAS_FADUMP_CPU_STATE_DATA:
@@ -469,8 +472,18 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
 				pr_err("Dump taken by platform is not valid (%d)\n", i);
 				rc = -EINVAL;
 			}
-			if (fdm_active->rgn[i].bytes_dumped != fdm_active->rgn[i].source_len) {
+			/*
+			 * Make sure that dump is collected for entire region.
+			 * CPU_STATE_DATA region is allowed to dump less than allocated space.
+			 */
+			region_collected = (bytes_dumped == source_len ||
+	                         (type == RTAS_FADUMP_CPU_STATE_DATA && bytes_dumped <= source_len));
+
+
+			if (! region_collected) {
 				pr_err("Dump taken by platform is incomplete (%d)\n", i);
+				pr_debug("type -> %d, bytes_dumped -> %llx, source_len -> %llx\n",
+				         type, bytes_dumped, source_len);
 				rc = -EINVAL;
 			}
 			if (rc) {
@@ -482,7 +495,7 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
 			}
 			break;
 		case RTAS_FADUMP_PARAM_AREA:
-			if (fdm_active->rgn[i].bytes_dumped != fdm_active->rgn[i].source_len ||
+			if (bytes_dumped != source_len ||
 			    fdm_active->rgn[i].error_flags != 0) {
 				pr_warn("Failed to process additional parameters! Proceeding anyway..\n");
 				fadump_conf->param_area = 0;
Signature · 1 line
2.54.0
#1 lore raw
Quoted 12 lines · Shivang Upadhyay

During Fadump in Qemu VM, when maxcpus value is set to more than current cpus, following failure is observed.

     [0.000000] rtas: Dump taken by platform is incomplete (-1)

This is because the CPU_STATE_DATA is allocated for maxcpus, while the data is only filled for current cpus. As per current implementation of Fadump, dumped_bytes and source_len for a region have to match, Which is failing for qemu's case, Even tough the dumped bytes are reported correctly for only the current cpus. After this /proc/vmcore generatition also fails.

Quoted 2 lines

Allowing dumped_bytes to be lesser than or equal to allocated length, for CPU_STATE_DATA Fadump region.

The overall logic looks good to me. And as per PAPR, firmware only promises to send CPU data of online CPUs, so it is OK to accept bytes dumped less than the source_len for CPU_STATE_DATA.

Quoted 29 lines

---
ChangeLog:

v2: Only allowing lesser size for CPU_STATE_DATE Fadump region.

v1: https://lore.kernel.org/qemu-devel/[email protected]/

---
  arch/powerpc/platforms/pseries/rtas-fadump.c | 17 +++++++++++++++--
  1 file changed, 15 insertions(+), 2 deletions(-)
arch/powerpc/platforms/pseries/rtas-fadump.c +3 -0
index 3bb4ac2ab6cc..ea1a0a8ac9eb 100644
@@ -459,7 +459,10 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
  	/* Check if the dump data is valid. */
  	for (int i = 0; i < be16_to_cpu(fdm_active->header.dump_num_sections); i++) {
  		int type = be16_to_cpu(fdm_active->rgn[i].source_data_type);
+		uint64_t bytes_dumped = be64_to_cpu(fdm_active->rgn[i].bytes_dumped);
+		uint64_t source_len = be64_to_cpu(fdm_active->rgn[i].source_len);
  		int rc = 0;
+		int region_collected;
 
  		switch (type) {
  		case RTAS_FADUMP_CPU_STATE_DATA:

arch/powerpc/platforms/pseries/rtas-fadump.c static int __init rtas_fadump_process(struct fw_dump *fadump_conf) · line 480

Quoted 10 lines
arch/powerpc/platforms/pseries/rtas-fadump.c +6 -1
@@ -469,8 +472,18 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
  				pr_err("Dump taken by platform is not valid (%d)\n", i);
  				rc = -EINVAL;
  			}
-			if (fdm_active->rgn[i].bytes_dumped != fdm_active->rgn[i].source_len) {
+			/*
+			 * Make sure that dump is collected for entire region.
+			 * CPU_STATE_DATA region is allowed to dump less than allocated space.
+			 */
+			region_collected = (bytes_dumped == source_len ||
+	                         (type == RTAS_FADUMP_CPU_STATE_DATA && bytes_dumped <= source_len));

Do we really need region_collected variable? Can't we manage with rc only?

Is bytes_dumped < source_len is good enough instead of <=. There are a couple of warnings/errors reported by the checkpatch script. Please address them in the next version.

- Sourabh Jain

Quoted 18 lines
arch/powerpc/platforms/pseries/rtas-fadump.c +6 -1
+
+
+			if (! region_collected) {
  				pr_err("Dump taken by platform is incomplete (%d)\n", i);
+				pr_debug("type -> %d, bytes_dumped -> %llx, source_len -> %llx\n",
+				         type, bytes_dumped, source_len);
  				rc = -EINVAL;
  			}
  			if (rc) {
@@ -482,7 +495,7 @@ static int __init rtas_fadump_process(struct fw_dump *fadump_conf)
  			}
  			break;
  		case RTAS_FADUMP_PARAM_AREA:
-			if (fdm_active->rgn[i].bytes_dumped != fdm_active->rgn[i].source_len ||
+			if (bytes_dumped != source_len ||
  			    fdm_active->rgn[i].error_flags != 0) {
  				pr_warn("Failed to process additional parameters! Proceeding anyway..\n");
  				fadump_conf->param_area = 0;
#2 lore raw
Quoted 6 lines · Sourabh Jain
Quoted 2 lines

bytes_dumped <= source_len));

Do we really need region_collected variable? Can't we manage with rc only?

Quoted 7 lines

Is bytes_dumped < source_len is good enough instead of <=. There are a
couple of warnings/errors reported by the checkpatch script. Please address them in the next version.

- Sourabh Jain

Hi Sourabh,

bytes_dumped < source_len should also be correct, but "<=" is more expressive. Why not leave such optimizations to compiler.

For the region_collected, RC should be used to manage the return code, so I as per me it is more appropriate to use new variable here. Or maybe, type the condition in the bracket itself(?).

Thanks for pointing out the checkpatch bugs, I'll send a revision soon.

~Shivang.

#3 lore raw
Quoted 12 lines · Shivang Upadhyay
Quoted 11 lines · Sourabh Jain
Quoted 1 line

bytes_dumped <= source_len));

Do we really need region_collected variable? Can't we manage with rc only?

Is bytes_dumped < source_len is good enough instead of <=. There are a
couple of warnings/errors reported by the checkpatch script. Please address them in the next version.

- Sourabh Jain

Quoted 8 lines

Hi Sourabh,

bytes_dumped < source_len should also be correct, but "<=" is more expressive. Why not leave such optimizations to compiler.

For the region_collected, RC should be used to manage the return code, so I as per me it is more appropriate to use new variable here. Or maybe, type the condition in the bracket itself(?).

Yes this also works.

- Sourabh Jain

Quoted 4 lines

Thanks for pointing out the checkpatch bugs, I'll send a revision soon.

~Shivang.