[PATCH V2] cxl: Fix memory page not handled

The in-kernel 'library' API can be called by drivers to help
interaction with an IBM XSL on a POWER9 system.

The cxllib_handle_fault() API is used to handle memory fault. All memory
pages of the specified buffer have to be handled but under certain
conditions,the last page may not be touched, and the address the
adapter is trying to access is never sent to the kernel for resolution.

This patch reworks start address of the loop with an address aligned on
the page size. In this context, the last page is not missed.

Signed-off-by: Christophe Lombard <redacted>

Fixes: 3ced8d730063 ("cxl: Export library to support IBM XSL");

---
Changelog[v2]
 - Rebase to latest upstream.
 - Change the start address of the loop.
 - Rewrite the commit message.
---
 drivers/misc/cxl/cxllib.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/drivers/misc/cxl/cxllib.c b/drivers/misc/cxl/cxllib.c
index 5dba23c..ff7492d 100644
--- a/drivers/misc/cxl/cxllib.c
+++ b/drivers/misc/cxl/cxllib.c

@@ -219,8 +219,17 @@ int cxllib_handle_fault(struct mm_struct *mm, u64 addr, u64 size, u64 flags)
 
 	down_read(&mm->mmap_sem);
 
-	for (dar = addr; dar < addr + size; dar += page_size) {
-		if (!vma || dar < vma->vm_start || dar > vma->vm_end) {
+	vma = find_vma(mm, addr);
+	if (!vma) {
+		pr_err("Can't find vma for addr %016llx\n", addr);
+		rc = -EFAULT;
+		goto out;
+	}
+	/* get the size of the pages allocated */
+	page_size = vma_kernel_pagesize(vma);
+
+	for (dar = (addr & ~(page_size - 1)); dar < (addr + size); dar += page_size) {
+		if (dar < vma->vm_start || dar > vma->vm_end) {
 			vma = find_vma(mm, addr);
 			if (!vma) {
 				pr_err("Can't find vma for addr %016llx\n", addr);

-- 
2.7.4

Le 25/09/2017 à 10:58, Christophe Lombard a écrit :

The in-kernel 'library' API can be called by drivers to help
interaction with an IBM XSL on a POWER9 system.

The cxllib_handle_fault() API is used to handle memory fault. All memory
pages of the specified buffer have to be handled but under certain
conditions,the last page may not be touched, and the address the
adapter is trying to access is never sent to the kernel for resolution.

This patch reworks start address of the loop with an address aligned on
the page size. In this context, the last page is not missed.

Signed-off-by: Christophe Lombard <redacted>

Fixes: 3ced8d730063 ("cxl: Export library to support IBM XSL");

---

Thanks,

Acked-by: Frederic Barrat <redacted>

On 25/09/17 18:58, Christophe Lombard wrote:

The in-kernel 'library' API can be called by drivers to help
interaction with an IBM XSL on a POWER9 system.

The cxllib_handle_fault() API is used to handle memory fault. All memory
pages of the specified buffer have to be handled but under certain
conditions,the last page may not be touched, and the address the
adapter is trying to access is never sent to the kernel for resolution.

This patch reworks start address of the loop with an address aligned on
the page size. In this context, the last page is not missed.

Signed-off-by: Christophe Lombard <redacted>

Fixes: 3ced8d730063 ("cxl: Export library to support IBM XSL");

The duplication of the find_vma() depresses me slightly but I suppose 
there's not much that can be done there :)

Acked-by: Andrew Donnellan <redacted>

-- 
Andrew Donnellan              OzLabs, ADL Canberra
andrew.donnellan@au1.ibm.com  IBM Australia Limited

Hi Christophe,

A minor nitpick

Christophe Lombard [off-list ref] writes:

+	for (dar = (addr & ~(page_size - 1)); dar < (addr + size); dar += page_size) {
+		if (dar < vma->vm_start || dar > vma->vm_end) {

Code comment in mm_types.h for vm_end says it "The first byte after our
end address within vm_mm". So this condition should be
(dar < vma->vm_start || dar >= vma->vm_end)

~ Vaibhav

On 26/09/17 11:44, Vaibhav Jain wrote:

Hi Christophe,

A minor nitpick

Christophe Lombard [off-list ref] writes:

quoted

+	for (dar = (addr & ~(page_size - 1)); dar < (addr + size); dar += page_size) {
+		if (dar < vma->vm_start || dar > vma->vm_end) {

Code comment in mm_types.h for vm_end says it "The first byte after our
end address within vm_mm". So this condition should be
(dar < vma->vm_start || dar >= vma->vm_end)

Hmm, I *think* you're right... a quick grep through the tree seems to 
indicate that I'm not the only one to have made that mistake...

Christophe if you're happy to do a v3 with this fixed as well that would 
be great.

-- 
Andrew Donnellan              OzLabs, ADL Canberra
andrew.donnellan@au1.ibm.com  IBM Australia Limited

Le 26/09/2017 à 03:44, Vaibhav Jain a écrit :

Hi Christophe,

A minor nitpick

Christophe Lombard [off-list ref] writes:

quoted

+	for (dar = (addr & ~(page_size - 1)); dar < (addr + size); dar += page_size) {
+		if (dar < vma->vm_start || dar > vma->vm_end) {

Code comment in mm_types.h for vm_end says it "The first byte after our
end address within vm_mm". So this condition should be
(dar < vma->vm_start || dar >= vma->vm_end)

~ Vaibhav

You are right. Thanks for the review