提交 7682a4c6 编写于 作者: D Dean Nelson 提交者: Tony Luck

[IA64-SGI] Silent data corruption caused by XPC V2.

Jack Steiner identified a problem where XPC can cause a silent
data corruption.  On module load, the placement may cause the
xpc_remote_copy_buffer to span two physical pages.  DMA transfers are
done to the start virtual address translated to physical.

This patch changes the buffer from a statically allocated buffer to a
kmalloc'd buffer.  Dean Nelson reviewed this before posting.  I have
tested it in the configuration that was showing the memory corruption
and verified it works.  I also added a BUG_ON statement to help catch
this if a similar situation is encountered.
Signed-off-by: NRobin Holt <holt@sgi.com>
Signed-off-by: NDean Nelson <dcn@sgi.com>
Signed-off-by: NJack Steiner <steiner@sgi.com>
Signed-off-by: NTony Luck <tony.luck@intel.com>
上级 9f737633
......@@ -279,8 +279,8 @@ xpc_pull_remote_cachelines(struct xpc_partition *part, void *dst,
return part->reason;
}
bte_ret = xp_bte_copy((u64) src, (u64) ia64_tpa((u64) dst),
(u64) cnt, (BTE_NORMAL | BTE_WACQUIRE), NULL);
bte_ret = xp_bte_copy((u64) src, (u64) dst, (u64) cnt,
(BTE_NORMAL | BTE_WACQUIRE), NULL);
if (bte_ret == BTE_SUCCESS) {
return xpcSuccess;
}
......
......@@ -1052,6 +1052,8 @@ xpc_do_exit(enum xpc_retval reason)
if (xpc_sysctl) {
unregister_sysctl_table(xpc_sysctl);
}
kfree(xpc_remote_copy_buffer_base);
}
......@@ -1212,24 +1214,20 @@ xpc_init(void)
partid_t partid;
struct xpc_partition *part;
pid_t pid;
size_t buf_size;
if (!ia64_platform_is("sn2")) {
return -ENODEV;
}
/*
* xpc_remote_copy_buffer is used as a temporary buffer for bte_copy'ng
* various portions of a partition's reserved page. Its size is based
* on the size of the reserved page header and part_nasids mask. So we
* need to ensure that the other items will fit as well.
*/
if (XPC_RP_VARS_SIZE > XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES) {
dev_err(xpc_part, "xpc_remote_copy_buffer is not big enough\n");
return -EPERM;
}
DBUG_ON((u64) xpc_remote_copy_buffer !=
L1_CACHE_ALIGN((u64) xpc_remote_copy_buffer));
buf_size = max(XPC_RP_VARS_SIZE,
XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES);
xpc_remote_copy_buffer = xpc_kmalloc_cacheline_aligned(buf_size,
GFP_KERNEL, &xpc_remote_copy_buffer_base);
if (xpc_remote_copy_buffer == NULL)
return -ENOMEM;
snprintf(xpc_part->bus_id, BUS_ID_SIZE, "part");
snprintf(xpc_chan->bus_id, BUS_ID_SIZE, "chan");
......@@ -1293,6 +1291,8 @@ xpc_init(void)
if (xpc_sysctl) {
unregister_sysctl_table(xpc_sysctl);
}
kfree(xpc_remote_copy_buffer_base);
return -EBUSY;
}
......@@ -1311,6 +1311,8 @@ xpc_init(void)
if (xpc_sysctl) {
unregister_sysctl_table(xpc_sysctl);
}
kfree(xpc_remote_copy_buffer_base);
return -EBUSY;
}
......@@ -1362,6 +1364,8 @@ xpc_init(void)
if (xpc_sysctl) {
unregister_sysctl_table(xpc_sysctl);
}
kfree(xpc_remote_copy_buffer_base);
return -EBUSY;
}
......
......@@ -71,19 +71,15 @@ struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
* Generic buffer used to store a local copy of portions of a remote
* partition's reserved page (either its header and part_nasids mask,
* or its vars).
*
* xpc_discovery runs only once and is a seperate thread that is
* very likely going to be processing in parallel with receiving
* interrupts.
*/
char ____cacheline_aligned xpc_remote_copy_buffer[XPC_RP_HEADER_SIZE +
XP_NASID_MASK_BYTES];
char *xpc_remote_copy_buffer;
void *xpc_remote_copy_buffer_base;
/*
* Guarantee that the kmalloc'd memory is cacheline aligned.
*/
static void *
void *
xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
{
/* see if kmalloc will give us cachline aligned memory by default */
......@@ -148,7 +144,7 @@ xpc_get_rsvd_page_pa(int nasid)
}
}
bte_res = xp_bte_copy(rp_pa, ia64_tpa(buf), buf_len,
bte_res = xp_bte_copy(rp_pa, buf, buf_len,
(BTE_NOTIFY | BTE_WACQUIRE), NULL);
if (bte_res != BTE_SUCCESS) {
dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
......@@ -447,7 +443,7 @@ xpc_check_remote_hb(void)
/* pull the remote_hb cache line */
bres = xp_bte_copy(part->remote_vars_pa,
ia64_tpa((u64) remote_vars),
(u64) remote_vars,
XPC_RP_VARS_SIZE,
(BTE_NOTIFY | BTE_WACQUIRE), NULL);
if (bres != BTE_SUCCESS) {
......@@ -498,8 +494,7 @@ xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
/* pull over the reserved page header and part_nasids mask */
bres = xp_bte_copy(*remote_rp_pa, ia64_tpa((u64) remote_rp),
bres = xp_bte_copy(*remote_rp_pa, (u64) remote_rp,
XPC_RP_HEADER_SIZE + xp_nasid_mask_bytes,
(BTE_NOTIFY | BTE_WACQUIRE), NULL);
if (bres != BTE_SUCCESS) {
......@@ -554,11 +549,8 @@ xpc_get_remote_vars(u64 remote_vars_pa, struct xpc_vars *remote_vars)
return xpcVarsNotSet;
}
/* pull over the cross partition variables */
bres = xp_bte_copy(remote_vars_pa, ia64_tpa((u64) remote_vars),
XPC_RP_VARS_SIZE,
bres = xp_bte_copy(remote_vars_pa, (u64) remote_vars, XPC_RP_VARS_SIZE,
(BTE_NOTIFY | BTE_WACQUIRE), NULL);
if (bres != BTE_SUCCESS) {
return xpc_map_bte_errors(bres);
......@@ -1239,7 +1231,7 @@ xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
part_nasid_pa = (u64) XPC_RP_PART_NASIDS(part->remote_rp_pa);
bte_res = xp_bte_copy(part_nasid_pa, ia64_tpa((u64) nasid_mask),
bte_res = xp_bte_copy(part_nasid_pa, (u64) nasid_mask,
xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE), NULL);
return xpc_map_bte_errors(bte_res);
......
......@@ -60,23 +60,37 @@
* the bte_copy() once in the hope that the failure was due to a temporary
* aberration (i.e., the link going down temporarily).
*
* See bte_copy for definition of the input parameters.
* src - physical address of the source of the transfer.
* vdst - virtual address of the destination of the transfer.
* len - number of bytes to transfer from source to destination.
* mode - see bte_copy() for definition.
* notification - see bte_copy() for definition.
*
* Note: xp_bte_copy() should never be called while holding a spinlock.
*/
static inline bte_result_t
xp_bte_copy(u64 src, u64 dest, u64 len, u64 mode, void *notification)
xp_bte_copy(u64 src, u64 vdst, u64 len, u64 mode, void *notification)
{
bte_result_t ret;
u64 pdst = ia64_tpa(vdst);
ret = bte_copy(src, dest, len, mode, notification);
/*
* Ensure that the physically mapped memory is contiguous.
*
* We do this by ensuring that the memory is from region 7 only.
* If the need should arise to use memory from one of the other
* regions, then modify the BUG_ON() statement to ensure that the
* memory from that region is always physically contiguous.
*/
BUG_ON(REGION_NUMBER(vdst) != RGN_KERNEL);
ret = bte_copy(src, pdst, len, mode, notification);
if (ret != BTE_SUCCESS) {
if (!in_interrupt()) {
cond_resched();
}
ret = bte_copy(src, dest, len, mode, notification);
ret = bte_copy(src, pdst, len, mode, notification);
}
return ret;
......
......@@ -683,7 +683,9 @@ extern struct xpc_vars *xpc_vars;
extern struct xpc_rsvd_page *xpc_rsvd_page;
extern struct xpc_vars_part *xpc_vars_part;
extern struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
extern char xpc_remote_copy_buffer[];
extern char *xpc_remote_copy_buffer;
extern void *xpc_remote_copy_buffer_base;
extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);
extern struct xpc_rsvd_page *xpc_rsvd_page_init(void);
extern void xpc_allow_IPI_ops(void);
extern void xpc_restrict_IPI_ops(void);
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册