cxl/pci: Fix the incorrect check of pci_config_read/write*() returns
Dave Jiang
maintainercover cxl/pci: Fix the incorrect check of pci_config_read/write*() returns
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Dave Jiang (2):
cxl/pci: Fix the incorrect check of pci_read_config_word() return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accesses
drivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)
base-commit: e43ffb69e0438cddd72aaa30898b4dc446f664f8
Signature · 1 line
2.54.0
Dave Jiang
maintainerPATCH 1/2 cxl/pci: Fix the incorrect check of pci_read_config_word() return
pci_read_config_word() returns PCIBIOS_* status on error which are positive values. The check should be for non-zero values to indicate error. Fix cxl_set_mem_enable() to check for non-zero return value instead of negative value.
While fixing this, also convert the error to negative errno value when returning on error path.
--- drivers/cxl/core/pci.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/cxl/core/pci.c +4 -4
@@ -187,8 +187,8 @@ static int cxl_set_mem_enable(struct cxl_dev_state *cxlds, u16 val)int rc; rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , &ctrl); -if (rc <0 ) -return rc; +if (rc) +return pcibios_err_to_errno (rc);if ((ctrl &PCI_DVSEC_CXL_MEM_ENABLE ) == val)return 1 ;
@@ -196,8 +196,8 @@ static int cxl_set_mem_enable(struct cxl_dev_state *cxlds, u16 val) ctrl |= val; rc =pci_write_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , ctrl); -if (rc <0 ) -return rc; +if (rc) +return pcibios_err_to_errno (rc);return 0 ; }
Signature · 1 line
2.54.0
Re: [PATCH 1/2] cxl/pci: Fix the incorrect check of pci_read_config_word() return
Quoted 45 lines · Dave Jiang
pci_read_config_word()returns PCIBIOS_* status on error which are positive values. The check should be for non-zero values to indicate error. Fixcxl_set_mem_enable()to check for non-zero return value instead of negative value.While fixing this, also convert the error to negative errno value when returning on error path.
Fixes 34e37b4c432c ("cxl/port: Enable HDM Capability after validating DVSEC Ranges")Assisted-by Claude:claude-opus-4-8Signed-off-by Dave Jiang <[email protected]>--- drivers/cxl/core/pci.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)drivers/cxl/core/pci.c +4
-4@@ -187,8 +187,8 @@ static int cxl_set_mem_enable(struct cxl_dev_state *cxlds, u16 val)int rc; rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , &ctrl); -if (rc <0 ) -return rc; +if (rc) +return pcibios_err_to_errno (rc);if ((ctrl &PCI_DVSEC_CXL_MEM_ENABLE ) == val)return 1 ;@@ -196,8 +196,8 @@ static int cxl_set_mem_enable(struct cxl_dev_state *cxlds, u16 val) ctrl |= val; rc =pci_write_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , ctrl); -if (rc <0 ) -return rc; +if (rc) +return pcibios_err_to_errno (rc);return 0 ; }Signature · 1 line
2.54.0
Re: [PATCH 1/2] cxl/pci: Fix the incorrect check of pci_read_config_word() return
In reply to Dave JiangQuoted 7 lines · Dave Jiang
pci_read_config_word()returns PCIBIOS_* status on error which are positive values. The check should be for non-zero values to indicate error. Fixcxl_set_mem_enable()to check for non-zero return value instead of negative value.While fixing this, also convert the error to negative errno value when returning on error path.
Dave Jiang
maintainerPATCH 2/2 cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accesses
In reply to Dave JiangPCI config space accessors return positive PCIBIOS_* status codes on failure that are positive integers. Several DVSEC accesses in the CXL core propagated these raw values to callers that test for failure against less than 0. Thus silently misinterpret the return value as success.
cxl_dvsec_rr_decode() and cxl_dvsec_mem_range_valid() are the two that have callers make such assumptions so far in the cxl/core. Convert the postive error values to negative errno values so the checks are correct on error paths.
While the chances of a config access failure are low, fix for correctness and to avoid confusion in the future when more DVSEC accesses are added.
--- drivers/cxl/core/pci.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/cxl/core/pci.c +7 -7
@@ -89,7 +89,7 @@ static int cxl_dvsec_mem_range_valid(struct cxl_dev_state *cxlds, int id) d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (id), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); valid =FIELD_GET (PCI_DVSEC_CXL_MEM_INFO_VALID , temp);if (valid)
@@ -275,7 +275,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CAP , &cap);if (rc) -return rc; +return pcibios_err_to_errno (rc);if (!(cap &PCI_DVSEC_CXL_MEM_CAPABLE )) {dev_dbg (dev,"Not MEM Capable\n" );
@@ -299,7 +299,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds,*/ rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , &ctrl);if (rc) -return rc; +return pcibios_err_to_errno (rc); info->mem_enabled =FIELD_GET (PCI_DVSEC_CXL_MEM_ENABLE , ctrl);if (!info->mem_enabled)
@@ -316,14 +316,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size |= temp &PCI_DVSEC_CXL_MEM_SIZE_LOW ;if (!size) {
@@ -333,14 +333,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base |= temp &PCI_DVSEC_CXL_MEM_BASE_LOW ;
Signature · 1 line
2.54.0
Re: [PATCH 2/2] cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accesses
Quoted 90 lines · Dave Jiang
PCI config space accessors return positive PCIBIOS_* status codes on failure that are positive integers. Several DVSEC accesses in the CXL core propagated these raw values to callers that test for failure against less than 0. Thus silently misinterpret the return value as success.
cxl_dvsec_rr_decode()andcxl_dvsec_mem_range_valid()are the two that have callers make such assumptions so far in the cxl/core. Convert the postive error values to negative errno values so the checks are correct on error paths.While the chances of a config access failure are low, fix for correctness and to avoid confusion in the future when more DVSEC accesses are added.
Fixes 6612bd9ff0b1 ("PCI: Update CXL DVSEC definitions")Fixes ce17ad0d5498 ("cxl: Wait Memory_Info_Valid before access memory related info")Assisted-by Claude:claude-opus-4-8Signed-off-by Dave Jiang <[email protected]>--- drivers/cxl/core/pci.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)drivers/cxl/core/pci.c +7
-7@@ -89,7 +89,7 @@ static int cxl_dvsec_mem_range_valid(struct cxl_dev_state *cxlds, int id) d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (id), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); valid =FIELD_GET (PCI_DVSEC_CXL_MEM_INFO_VALID , temp);if (valid)@@ -275,7 +275,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CAP , &cap);if (rc) -return rc; +return pcibios_err_to_errno (rc);if (!(cap &PCI_DVSEC_CXL_MEM_CAPABLE )) {dev_dbg (dev,"Not MEM Capable\n" );@@ -299,7 +299,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds,*/ rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , &ctrl);if (rc) -return rc; +return pcibios_err_to_errno (rc); info->mem_enabled =FIELD_GET (PCI_DVSEC_CXL_MEM_ENABLE , ctrl);if (!info->mem_enabled)@@ -316,14 +316,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size |= temp &PCI_DVSEC_CXL_MEM_SIZE_LOW ;if (!size) {@@ -333,14 +333,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base |= temp &PCI_DVSEC_CXL_MEM_BASE_LOW ;Signature · 1 line
2.54.0
Alison Schofield
maintainerRe: [PATCH 2/2] cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accesses
In reply to Dave JiangQuoted 10 lines · Dave Jiang
PCI config space accessors return positive PCIBIOS_* status codes on failure that are positive integers. Several DVSEC accesses in the CXL core propagated these raw values to callers that test for failure against less than 0. Thus silently misinterpret the return value as success.
cxl_dvsec_rr_decode()andcxl_dvsec_mem_range_valid()are the two that have callers make such assumptions so far in the cxl/core. Convert the postive error values to negative errno values so the checks are correct on error paths.
Quoted 4 lines
While the chances of a config access failure are low, fix for correctness and to avoid confusion in the future when more DVSEC accesses are added.
Fixes 6612bd9ff0b1 ("PCI: Update CXL DVSEC definitions")
This first fixes tag is a macro rename. Look at this one:1acba6e9206c ("cxl/pci: Break out range register decoding from cxl_hdm_decode_init()")
Quoted 12 lines
Fixes ce17ad0d5498 ("cxl: Wait Memory_Info_Valid before access memory related info")Assisted-by Claude:claude-opus-4-8Signed-off-by Dave Jiang <[email protected]>--- drivers/cxl/core/pci.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)drivers/cxl/core/pci.c +0
-0
drivers/cxl/core/pci.c static int cxl_dvsec_mem_range_valid(struct cxl_dev_state *cxlds, int id) · line 95
Quoted 8 lines
drivers/cxl/core/pci.c +1
-1@@ -89,7 +89,7 @@ static int cxl_dvsec_mem_range_valid(struct cxl_dev_state *cxlds, int id) d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (id), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); valid =FIELD_GET (PCI_DVSEC_CXL_MEM_INFO_VALID , temp);if (valid)
Above converts one twin and not the other, cxl_dvsec_mem_range_active()
With that cxl_await_media_ready() calls both, _valid & _active, and returns different error codes depending on which one fails.
Actually, cxl_await_media_ready() has 3 return sites:
- the cap read (pci_read_config_word) returns raw PCIBIOS
- cxl_dvsec_mem_range_active() returns raw PCIBIOS
- cxl_dvsec_mem_range_valid() returns negative errno (w this patch)
My suggestion is that cxl_await_media_ready() be made consistent, those 3 sites return negative errno, in one fixup patch.
Quoted 55 lines
drivers/cxl/core/pci.c +6
-6@@ -275,7 +275,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CAP , &cap);if (rc) -return rc; +return pcibios_err_to_errno (rc);if (!(cap &PCI_DVSEC_CXL_MEM_CAPABLE )) {dev_dbg (dev,"Not MEM Capable\n" );@@ -299,7 +299,7 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds,*/ rc =pci_read_config_word (pdev, d +PCI_DVSEC_CXL_CTRL , &ctrl);if (rc) -return rc; +return pcibios_err_to_errno (rc); info->mem_enabled =FIELD_GET (PCI_DVSEC_CXL_MEM_ENABLE , ctrl);if (!info->mem_enabled)@@ -316,14 +316,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_SIZE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); size |= temp &PCI_DVSEC_CXL_MEM_SIZE_LOW ;if (!size) {@@ -333,14 +333,14 @@ int cxl_dvsec_rr_decode(struct cxl_dev_state *cxlds, rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_HIGH (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base = (u64 )temp <<32 ; rc =pci_read_config_dword ( pdev, d +PCI_DVSEC_CXL_RANGE_BASE_LOW (i), &temp);if (rc) -return rc; +return pcibios_err_to_errno (rc); base |= temp &PCI_DVSEC_CXL_MEM_BASE_LOW ;Signature · 1 line
2.54.0
Quoted 6 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Quoted 10 lines
Dave Jiang (2):
cxl/pci: Fix the incorrect check ofpci_read_config_word()return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accessesdrivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)base-commit:
e43ffb69e0438cddd72aaa30898b4dc446f664f8
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Hi Dave,
Thanks for this, sorry I miss the email somehow I'll look into details later, seems reasonable to me from a glimpse.
Best regards,
Richard Cheng.
Quoted 15 lines
Quoted 1 line
Reviewed-by Jonathan Cameron <[email protected]>Quoted 10 lines
Dave Jiang (2):
cxl/pci: Fix the incorrect check ofpci_read_config_word()return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accessesdrivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)base-commit:
e43ffb69e0438cddd72aaa30898b4dc446f664f8
Richard Cheng
Reviewed-byQuoted 10 lines · Richard Cheng
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Quoted 12 lines
Hi Dave,
Thanks for this, sorry I miss the email somehow I'll look into details later, seems reasonable to me from a glimpse.
Best regards,
Richard Cheng.Quoted 3 lines
Quoted 1 line
Reviewed-by Jonathan Cameron <[email protected]>
The series looks good to me. There is still some minor cleanup left — for example, cxl_dvsec_mem_range_active() still has the same problem. I plan to send follow-up patches for the remaining sites unless you prefer to fold them into this series.
Let me know what you think.
Quoted 13 lines
Quoted 12 lines
Quoted 10 lines
Dave Jiang (2):
cxl/pci: Fix the incorrect check ofpci_read_config_word()return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accessesdrivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)base-commit:
e43ffb69e0438cddd72aaa30898b4dc446f664f8
Dave Jiang
maintainerQuoted 25 lines · Richard Cheng
Quoted 10 lines · Richard Cheng
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Quoted 12 lines
Hi Dave,
Thanks for this, sorry I miss the email somehow I'll look into details later, seems reasonable to me from a glimpse.
Best regards,
Richard Cheng.Quoted 3 lines
Quoted 1 line
Reviewed-by Jonathan Cameron <[email protected]>
Quoted 5 lines
The series looks good to me. There is still some minor cleanup left — for example,
cxl_dvsec_mem_range_active()still has the same problem. I plan to send follow-up patches for the remaining sites unless you prefer to fold them into this series.
Let me know what you think.
Go ahead and send your follow on patches since you know which sites you want to fix. I'll create a branch that you can apply on top of with these two.
DJ
Quoted 16 lines
Reviewed-by Richard Cheng <[email protected]>Quoted 13 lines
Quoted 12 lines
Quoted 10 lines
Dave Jiang (2):
cxl/pci: Fix the incorrect check ofpci_read_config_word()return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accessesdrivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)base-commit:
e43ffb69e0438cddd72aaa30898b4dc446f664f8
Dave Jiang
maintainerQuoted 28 lines · Dave Jiang
Quoted 25 lines · Richard Cheng
Quoted 22 lines · Richard Cheng
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Hi Dave,
Thanks for this, sorry I miss the email somehow I'll look into details later, seems reasonable to me from a glimpse.
Best regards,
Richard Cheng.Quoted 3 lines
Quoted 1 line
Reviewed-by Jonathan Cameron <[email protected]>
Quoted 7 lines
Quoted 5 lines
The series looks good to me. There is still some minor cleanup left — for example,
cxl_dvsec_mem_range_active()still has the same problem. I plan to send follow-up patches for the remaining sites unless you prefer to fold them into this series.
Let me know what you think.Go ahead and send your follow on patches since you know which sites you want to fix. I'll create a branch that you can apply on top of with these two.
Here's the branch:
https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/log/?h=for-7.2/cxl-pci-fixes
Dave Jiang
maintainerQuoted 10 lines · Richard Cheng
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Wed, 3 Jun 2026 11:57:50 -0700 Dave Jiang <[email protected]> wrote:
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Quoted 5 lines
Hi Dave,
Thanks for this, sorry I miss the email somehow I'll look into details later, seems reasonable to me from a glimpse.
Thanks for the review. Totally my fault. I think my brain said cc Richard when I was sending, but somehow I missed it.
DJ
Quoted 19 lines
Best regards,
Richard Cheng.Quoted 15 lines
Quoted 1 line
Reviewed-by Jonathan Cameron <[email protected]>Quoted 10 lines
Dave Jiang (2):
cxl/pci: Fix the incorrect check ofpci_read_config_word()return cxl/pci: Convert PCIBIOS errors to errno on DVSEC config accessesdrivers/cxl/core/pci.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)base-commit:
e43ffb69e0438cddd72aaa30898b4dc446f664f8
Alison Schofield
maintainerQuoted 5 lines · Dave Jiang
The patches are inspired by comments from Richard Cheng WRT pci config accessors return PCIBIOS errors that are positive integers rather than negative errnos. I went and audited the cxl PCI code and found a few locations that need to be fixed. There are other locations that may trigger issues, but at this point we'll fix them as they happen.
Feels like 'fix them as they happen' if OK for unrelated call sites, but in Patch 2 partial fixups look incomplete. See Patch 2 reply.