Skip to contentSkip to conversation

scsi: megaraid_sas: Fix MEGASAS_IOC_FIRMWARE regression

PATCH Applied Arnd Bergmann 3 participants 3 messages Jan 4, 2021 – Jan 8, 2021

Subsystems (2)
SCSI Megaraid SCSI/SAS drivers
lore
Share & tools
lore

Arnd Bergmann

author
#0 lore raw

PATCH 1/1 scsi: megaraid_sas: Fix MEGASAS_IOC_FIRMWARE regression

From: Arnd Bergmann <[email protected]>

Phil Oester reported that a fix for a possible buffer overrun that I sent caused a regression that manifests in this output:

Event Message: A PCI parity error was detected on a component at bus 0 device 5 function 0.
Severity: Critical
Message ID: PCI1308

The original code tried to handle the sense data pointer differently when using 32-bit 64-bit DMA addressing, which would lead to a 32-bit dma_addr_t value of 0x11223344 to get stored

32-bit kernel:       44 33 22 11 ?? ?? ?? ??
64-bit LE kernel:    44 33 22 11 00 00 00 00
64-bit BE kernel:    00 00 00 00 44 33 22 11

or a 64-bit dma_addr_t value of 0x1122334455667788 to get stored as

32-bit kernel: 88 77 66 55 ?? ?? ?? ??
64-bit kernel: 88 77 66 55 44 33 22 11

In my patch, I tried to ensure that the same value is used on both 32-bit and 64-bit kernels, and picked what seemed to be the most sensible combination, storing 32-bit addresses in the first four bytes (as 32-bit kernels already did), and 64-bit addresses in eight consecutive bytes (as 64-bit kernels already did), but evidently this was incorrect.

Always storing the dma_addr_t pointer as 64-bit little-endian, i.e. initializing the second four bytes to zero in case of 32-bit addressing, apparently solved the problem for Phil, and is consistent with what all 64-bit little-endian machines did before.

I also checked in the history that in previous versions of the code, the pointer was always in the first four bytes without padding, and that previous attempts to fix 64-bit user space, big-endian architectures and 64-bit DMA were clearly flawed and seem to have introduced made this worse.

Reported-by Phil Oester <[email protected]>
Fixes 381d34e376e3 ("scsi: megaraid_sas: Check user-provided offsets")
Fixes 107a60dd71b5 ("scsi: megaraid_sas: Add support for 64bit consistent DMA")
Fixes 94cd65ddf4d7 ("[SCSI] megaraid_sas: addded support for big endian architecture")
Fixes 7b2519afa1ab ("[SCSI] megaraid_sas: fix 64 bit sense pointer truncation")
Signed-off-by Arnd Bergmann <[email protected]>
---
 drivers/scsi/megaraid/megaraid_sas_base.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/scsi/megaraid/megaraid_sas_base.c +2 -4
index 6e4bf05c6d77..3b574c453414 100644
@@ -8205,11 +8205,9 @@ megasas_mgmt_fw_ioctl(struct megasas_instance *instance,
 			goto out;
 		}
 
+		/* always store 64 bits regardless of addressing */
 		sense_ptr = (void *)cmd->frame + ioc->sense_off;
-		if (instance->consistent_mask_64bit)
-			put_unaligned_le64(sense_handle, sense_ptr);
-		else
-			put_unaligned_le32(sense_handle, sense_ptr);
+		put_unaligned_le64(sense_handle, sense_ptr);
 	}
 
 	/*
Signature · 1 line
2.29.2
#1 lore raw
Quoted 37 lines · Arnd Bergmann

Phil Oester reported that a fix for a possible buffer overrun that I sent caused a regression that manifests in this output:

Event Message: A PCI parity error was detected on a component at bus 0 device 5 function 0.
Severity: Critical
Message ID: PCI1308

The original code tried to handle the sense data pointer differently when using 32-bit 64-bit DMA addressing, which would lead to a 32-bit dma_addr_t value of 0x11223344 to get stored

32-bit kernel:       44 33 22 11 ?? ?? ?? ??
64-bit LE kernel:    44 33 22 11 00 00 00 00
64-bit BE kernel:    00 00 00 00 44 33 22 11

or a 64-bit dma_addr_t value of 0x1122334455667788 to get stored as

32-bit kernel: 88 77 66 55 ?? ?? ?? ??
64-bit kernel: 88 77 66 55 44 33 22 11

In my patch, I tried to ensure that the same value is used on both 32-bit and 64-bit kernels, and picked what seemed to be the most sensible combination, storing 32-bit addresses in the first four bytes (as 32-bit kernels already did), and 64-bit addresses in eight consecutive bytes (as 64-bit kernels already did), but evidently this was incorrect.

Always storing the dma_addr_t pointer as 64-bit little-endian, i.e. initializing the second four bytes to zero in case of 32-bit addressing, apparently solved the problem for Phil, and is consistent with what all 64-bit little-endian machines did before.

I also checked in the history that in previous versions of the code, the pointer was always in the first four bytes without padding, and that previous attempts to fix 64-bit user space, big-endian architectures and 64-bit DMA were clearly flawed and seem to have introduced made this worse.

Quoted 6 lines
Reported-by Phil Oester <[email protected]>
Fixes 381d34e376e3 ("scsi: megaraid_sas: Check user-provided offsets")
Fixes 107a60dd71b5 ("scsi: megaraid_sas: Add support for 64bit consistent DMA")
Fixes 94cd65ddf4d7 ("[SCSI] megaraid_sas: addded support for big endian architecture")
Fixes 7b2519afa1ab ("[SCSI] megaraid_sas: fix 64 bit sense pointer truncation")
Signed-off-by Arnd Bergmann <[email protected]>

This solves the issue on our Dell servers, thanks Arnd.

Phil

Applied
#2 lore raw
In reply to Arnd Bergmann
Quoted 8 lines · Arnd Bergmann

Phil Oester reported that a fix for a possible buffer overrun that I sent caused a regression that manifests in this output:

Event Message: A PCI parity error was detected on a component at bus 0 device 5 function 0.
Severity: Critical
Message ID: PCI1308

[...]

Applied to 5.11/scsi-fixes, thanks!

[1/1] scsi: megaraid_sas: Fix MEGASAS_IOC_FIRMWARE regression

      https://git.kernel.org/mkp/scsi/c/b112036535ed
Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering