DMA Mapping Error Analysis - Revision history

Shuahkhan at 23:19, 13 September 2012

2012-09-13T23:19:44Z

Shuahkhan: /* dma_map_page() usages and error checking status */

2012-09-10T20:12:57Z

‎dma_map_page() usages and error checking status

Shuahkhan: /* dma_map_single() usages and error checking status */

2012-09-10T20:11:00Z

‎dma_map_single() usages and error checking status

Shuahkhan: /* dma_map_page() usages and error checking status */

2012-09-10T15:23:57Z

‎dma_map_page() usages and error checking status

Shuahkhan at 16:53, 7 September 2012

2012-09-07T16:53:22Z

Shuahkhan: /* DMA Mapping Error Check Analysis */

2012-09-07T15:52:16Z

‎DMA Mapping Error Check Analysis

Shuahkhan at 15:26, 7 September 2012

2012-09-07T15:26:13Z

Shuahkhan at 01:52, 7 September 2012

2012-09-07T01:52:48Z

Shuahkhan at 01:41, 7 September 2012

2012-09-07T01:41:14Z

Shuahkhan at 01:37, 7 September 2012

2012-09-07T01:37:38Z

@@ Line 255: / Line 255: @@
 |sound/sparc/dbri.c || 1 || Broken ||
 |-
-|drivers/scsi/advansys.c || 2 || Partially Broken ||
+|drivers/scsi/advansys.c || 2 || Partially Broken || driver is still not properly converted to the DMA API - please read FIXME in this file for details on what is broken and what needs to be fixed.
 |-
 |drivers/mmc/host/wbsd.c || 1 || Partially Broken || Checks returned address is aligned on 64k - however doesn't call dma_mapping_error()

@@ Line 511: / Line 511: @@
 |drivers/infiniband/hw/ipath/ipath_user_sdma.c || 3 || Unmap Broken ||
 |-
-|drivers/firewire/core-iso.c || 1 || Unmap Broken || On failure, the callers are responsible for cleanup by calling fw_iso_buffer_destroy() eventually. ioctl_create_iso_context() doesn't do this correctly if it's called multiple times.
+|drivers/firewire/core-iso.c || 1 || Unmap Broken || On failure, the callers are responsible for cleanup by calling fw_iso_buffer_destroy() eventually. ioctl_create_iso_context() doesn't do this correctly if it's called multiple times. https://lkml.org/lkml/2012/9/10/76
 |-
 |drivers/net/ethernet/sfc/rx.c || 1 || Unmap Broken ||
@@ Line 525: / Line 525: @@
 |drivers/infiniband/ulp/ipoib/ipoib_cm.c || 1 || Good ||
 |-
-|drivers/firewire/ohci.c || 1 || Good || ar_context_release() takes care of cleanup.
+|drivers/firewire/ohci.c || 1 || Good || ar_context_release() takes care of cleanup. https://lkml.org/lkml/2012/9/10/76
 |-
 |drivers/scsi/cxgbi/libcxgbi.c || 1 || Good ||

@@ Line 131: / Line 131: @@
 |drivers/usb/musb/musb_gadget.c || 1 || Broken ||
 |-
-|drivers/usb/core/usb.c || 1 || Cleanup || The code is #ifdef'ed out. It hasn't been used in many years; it should be removed.
+|drivers/usb/core/usb.c || 1 || Cleanup || The code is #ifdef'ed out. It hasn't been used in many years; it should be removed. https://lkml.org/lkml/2012/9/7/455
 |-
 |drivers/usb/gadget/atmel_usba_udc.c || 1 || Broken ||
@@ Line 449: / Line 449: @@
 |net/9p/trans_rdma.c || 2 || Good ||
 |}
 ===== dma_map_page() usages and error checking status =====

@@ Line 512: / Line 512: @@
 |drivers/infiniband/hw/ipath/ipath_user_sdma.c || 3 || Unmap Broken ||
 |-
-|drivers/firewire/core-iso.c || 1 || Unmap Broken ||
+|drivers/firewire/core-iso.c || 1 || Unmap Broken || On failure, the callers are responsible for cleanup by calling fw_iso_buffer_destroy() eventually. ioctl_create_iso_context() doesn't do this correctly if it's called multiple times.
-|-
-|drivers/firewire/ohci.c || 1 || Unmap Broken ||
 |-
 |drivers/net/ethernet/sfc/rx.c || 1 || Unmap Broken ||
@@ Line 527: / Line 525: @@
 |-
 |drivers/infiniband/ulp/ipoib/ipoib_cm.c || 1 || Good ||
 |-
 |drivers/scsi/cxgbi/libcxgbi.c || 1 || Good ||

@@ Line 131: / Line 131: @@
 |drivers/usb/musb/musb_gadget.c || 1 || Broken ||
 |-
-|drivers/usb/core/usb.c || 1 || Broken ||
+|drivers/usb/core/usb.c || 1 || Cleanup || The code is #ifdef'ed out. It hasn't been used in many years; it should be removed.
 |-
 |drivers/usb/gadget/atmel_usba_udc.c || 1 || Broken ||

@@ Line 1: / Line 1: @@
 ==== DMA Mapping Error Check Analysis ====
-I analyzed all calls to dma_map_single() and dma_map_page() in the kernel to see if dma mapping errors are checked by the callers after mapping routines return. The goal of this analysis is to find drivers that currently do not check dma mapping errors, and fix them. I did a grep for dma_map_single() and dma_map_page() and looked at the code that calls these routines. As a side note, this analysis stemmed from the discussion on my patch that disables swiotlb overflow as a first step towards removing the support all together.
+I analyzed all calls to dma_map_single() and dma_map_page() in the kernel, to see if callers check for mapping errors, before using the returned address. The goal of this analysis is to find drivers that currently do not check dma mapping errors, and fix them. I did a grep for dma_map_single() and dma_map_page() and looked at the code that calls these routines. As a side note, this analysis stemmed from the discussion on my patch that disables swiotlb overflow as a first step towards removing the support all together.
 '''linux-next August 6 2012''' is used as the base for this analysis.

@@ Line 56: / Line 56: @@
 * Circle back to where I started - remove swiotlb overflow buffer support to make it consistent with the rest of the iotlb implementations. Removing support will be done in phases starting with deprecation notice, before removing the feature. [V1 patch https://lkml.org/lkml/2012/8/24/426]
-In addition to the above, I am sharing the information with the community at large, so individual driver and module owners can take a look at this analysis and assess the severity of the reported problems. I propose that, we use this page to record comments and track progress made on fixing the problem drivers and modules. If you have comments, questions etc. please contact [[mailto:shuahkhan@gmail.com,shuah.khan@hp.com Shuah Khan]]
+In addition to the above, I am sharing the information with the community at large, so individual driver and module owners can take a look at this analysis and assess the severity of the reported problems. I propose that, we use this page to record comments and track progress made on fixing the problem drivers and modules. If you have comments, questions etc. please contact [[mailto:shuahkhan@gmail.com,shuah.khan@hp.com?cc=linux-kernel@vger.kernel.org Shuah Khan]]
 ==== Detailed Analysis ====

@@ Line 169: / Line 169: @@
 |drivers/net/ethernet/sun/sunbmac.c || 3 || Broken ||
 |-
-|drivers/net/ethernet/sun/niu.c || 1 || Broken || Fixed https://lkml.org/lkml/2012/7/20/444
+|drivers/net/ethernet/sun/niu.c || 1 || Broken || [Fixed https://lkml.org/lkml/2012/7/20/401]
 |-
 |drivers/net/ethernet/sun/sunhme.c || 5 || Broken ||

@@ Line 54: / Line 54: @@
 * Enhance dma-debug infrastructure to track dma map and unmap errors - I am working on this change. [V1 Patch https://lkml.org/lkml/2012/9/2/80]
 * Assess severity and fix drivers and modules as needed.
-* Circle back to where I started - remove swiotlb overflow buffer support to make it consistent with the rest of the iotlb implementations. Removing support will be done in phases starting with deprecation notice, before removing the feature. [V1 Patch https://lkml.org/lkml/2012/8/10/326]
+* Circle back to where I started - remove swiotlb overflow buffer support to make it consistent with the rest of the iotlb implementations. Removing support will be done in phases starting with deprecation notice, before removing the feature. [V1 patch https://lkml.org/lkml/2012/8/24/426]
 In addition to the above, I am sharing the information with the community at large, so individual driver and module owners can take a look at this analysis and assess the severity of the reported problems. I propose that, we use this page to record comments and track progress made on fixing the problem drivers and modules. If you have comments, questions etc. please contact [[mailto:shuahkhan@gmail.com,shuah.khan@hp.com Shuah Khan]]

@@ Line 21: / Line 21: @@
 # Checks dma mapping errors without unlikely()
 # Checks dma mapping errors with unlikely()
 I grouped the above three cases as good cases. Using unlikely() is icing on the cake, and something we don't need to be concerned about considering the other more serious problems in this area.
@@ Line 44: / Line 45: @@
 * Why don't we see failures related to missing dma mapping error checks?
 * Are they silent failures?
 I propose the following next steps to gather more information and understand the severity of the problem.
@@ Line 52: / Line 54: @@
 * Enhance dma-debug infrastructure to track dma map and unmap errors - I am working on this change. [V1 Patch https://lkml.org/lkml/2012/9/2/80]
 * Assess severity and fix drivers and modules as needed.
-* Circle back to where I started - remove swiotlb overflow buffer support to make it consistent with the rest of the iotlb implementations.
+* Circle back to where I started - remove swiotlb overflow buffer support to make it consistent with the rest of the iotlb implementations. Removing support will be done in phases starting with deprecation notice, before removing the feature. [V1 Patch https://lkml.org/lkml/2012/8/10/326]
 In addition to the above, I am sharing the information with the community at large, so individual driver and module owners can take a look at this analysis and assess the severity of the reported problems. I propose that, we use this page to record comments and track progress made on fixing the problem drivers and modules. If you have comments, questions etc. please contact [[mailto:shuahkhan@gmail.com,shuah.khan@hp.com Shuah Khan]]