linux/drivers/misc/cxl/api.c

673 lines
15 KiB
C
Raw Normal View History

/*
* Copyright 2014 IBM Corp.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/file.h>
#include <misc/cxl.h>
cxl: Add support for interrupts on the Mellanox CX4 The Mellanox CX4 in cxl mode uses a hybrid interrupt model, where interrupts are routed from the networking hardware to the XSL using the MSIX table, and from there will be transformed back into an MSIX interrupt using the cxl style interrupts (i.e. using IVTE entries and ranges to map a PE and AFU interrupt number to an MSIX address). We want to hide the implementation details of cxl interrupts as much as possible. To this end, we use a special version of the MSI setup & teardown routines in the PHB while in cxl mode to allocate the cxl interrupts and configure the IVTE entries in the process element. This function does not configure the MSIX table - the CX4 card uses a custom format in that table and it would not be appropriate to fill that out in generic code. The rest of the functionality is similar to the "Full MSI-X mode" described in the CAIA, and this could be easily extended to support other adapters that use that mode in the future. The interrupts will be associated with the default context. If the maximum number of interrupts per context has been limited (e.g. by the mlx5 driver), it will automatically allocate additional kernel contexts to associate extra interrupts as required. These contexts will be started using the same WED that was used to start the default context. Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-07-14 05:17:10 +08:00
#include <linux/msi.h>
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
#include <linux/module.h>
#include <linux/mount.h>
#include <linux/sched/mm.h>
#include "cxl.h"
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
/*
* Since we want to track memory mappings to be able to force-unmap
* when the AFU is no longer reachable, we need an inode. For devices
* opened through the cxl user API, this is not a problem, but a
* userland process can also get a cxl fd through the cxl_get_fd()
* API, which is used by the cxlflash driver.
*
* Therefore we implement our own simple pseudo-filesystem and inode
* allocator. We don't use the anonymous inode, as we need the
* meta-data associated with it (address_space) and it is shared by
* other drivers/processes, so it could lead to cxl unmapping VMAs
* from random processes.
*/
#define CXL_PSEUDO_FS_MAGIC 0x1697697f
static int cxl_fs_cnt;
static struct vfsmount *cxl_vfs_mount;
static const struct dentry_operations cxl_fs_dops = {
.d_dname = simple_dname,
};
static struct dentry *cxl_fs_mount(struct file_system_type *fs_type, int flags,
const char *dev_name, void *data)
{
return mount_pseudo(fs_type, "cxl:", NULL, &cxl_fs_dops,
CXL_PSEUDO_FS_MAGIC);
}
static struct file_system_type cxl_fs_type = {
.name = "cxl",
.owner = THIS_MODULE,
.mount = cxl_fs_mount,
.kill_sb = kill_anon_super,
};
void cxl_release_mapping(struct cxl_context *ctx)
{
if (ctx->kernelapi && ctx->mapping)
simple_release_fs(&cxl_vfs_mount, &cxl_fs_cnt);
}
static struct file *cxl_getfile(const char *name,
const struct file_operations *fops,
void *priv, int flags)
{
struct qstr this;
struct path path;
struct file *file;
struct inode *inode = NULL;
int rc;
/* strongly inspired by anon_inode_getfile() */
if (fops->owner && !try_module_get(fops->owner))
return ERR_PTR(-ENOENT);
rc = simple_pin_fs(&cxl_fs_type, &cxl_vfs_mount, &cxl_fs_cnt);
if (rc < 0) {
pr_err("Cannot mount cxl pseudo filesystem: %d\n", rc);
file = ERR_PTR(rc);
goto err_module;
}
inode = alloc_anon_inode(cxl_vfs_mount->mnt_sb);
if (IS_ERR(inode)) {
file = ERR_CAST(inode);
goto err_fs;
}
file = ERR_PTR(-ENOMEM);
this.name = name;
this.len = strlen(name);
this.hash = 0;
path.dentry = d_alloc_pseudo(cxl_vfs_mount->mnt_sb, &this);
if (!path.dentry)
goto err_inode;
path.mnt = mntget(cxl_vfs_mount);
d_instantiate(path.dentry, inode);
file = alloc_file(&path, OPEN_FMODE(flags), fops);
if (IS_ERR(file))
goto err_dput;
file->f_flags = flags & (O_ACCMODE | O_NONBLOCK);
file->private_data = priv;
return file;
err_dput:
path_put(&path);
err_inode:
iput(inode);
err_fs:
simple_release_fs(&cxl_vfs_mount, &cxl_fs_cnt);
err_module:
module_put(fops->owner);
return file;
}
struct cxl_context *cxl_dev_context_init(struct pci_dev *dev)
{
struct cxl_afu *afu;
struct cxl_context *ctx;
int rc;
afu = cxl_pci_to_afu(dev);
if (IS_ERR(afu))
return ERR_CAST(afu);
ctx = cxl_context_alloc();
if (!ctx)
return ERR_PTR(-ENOMEM);
ctx->kernelapi = true;
/* Make it a slave context. We can promote it later? */
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
rc = cxl_context_init(ctx, afu, false);
if (rc)
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
goto err_ctx;
return ctx;
err_ctx:
kfree(ctx);
return ERR_PTR(rc);
}
EXPORT_SYMBOL_GPL(cxl_dev_context_init);
struct cxl_context *cxl_get_context(struct pci_dev *dev)
{
return dev->dev.archdata.cxl_ctx;
}
EXPORT_SYMBOL_GPL(cxl_get_context);
int cxl_release_context(struct cxl_context *ctx)
{
if (ctx->status >= STARTED)
return -EBUSY;
cxl_context_free(ctx);
return 0;
}
EXPORT_SYMBOL_GPL(cxl_release_context);
static irq_hw_number_t cxl_find_afu_irq(struct cxl_context *ctx, int num)
{
__u16 range;
int r;
for (r = 0; r < CXL_IRQ_RANGES; r++) {
range = ctx->irqs.range[r];
if (num < range) {
return ctx->irqs.offset[r] + num;
}
num -= range;
}
return 0;
}
int _cxl_next_msi_hwirq(struct pci_dev *pdev, struct cxl_context **ctx, int *afu_irq)
{
if (*ctx == NULL || *afu_irq == 0) {
*afu_irq = 1;
*ctx = cxl_get_context(pdev);
} else {
(*afu_irq)++;
if (*afu_irq > cxl_get_max_irqs_per_process(pdev)) {
*ctx = list_next_entry(*ctx, extra_irq_contexts);
*afu_irq = 1;
}
}
return cxl_find_afu_irq(*ctx, *afu_irq);
}
/* Exported via cxl_base */
int cxl_set_priv(struct cxl_context *ctx, void *priv)
{
if (!ctx)
return -EINVAL;
ctx->priv = priv;
return 0;
}
EXPORT_SYMBOL_GPL(cxl_set_priv);
void *cxl_get_priv(struct cxl_context *ctx)
{
if (!ctx)
return ERR_PTR(-EINVAL);
return ctx->priv;
}
EXPORT_SYMBOL_GPL(cxl_get_priv);
int cxl_allocate_afu_irqs(struct cxl_context *ctx, int num)
{
int res;
irq_hw_number_t hwirq;
if (num == 0)
num = ctx->afu->pp_irqs;
res = afu_allocate_irqs(ctx, num);
if (res)
return res;
if (!cpu_has_feature(CPU_FTR_HVMODE)) {
/* In a guest, the PSL interrupt is not multiplexed. It was
* allocated above, and we need to set its handler
*/
hwirq = cxl_find_afu_irq(ctx, 0);
if (hwirq)
cxl_map_irq(ctx->afu->adapter, hwirq, cxl_ops->psl_interrupt, ctx, "psl");
}
if (ctx->status == STARTED) {
if (cxl_ops->update_ivtes)
cxl_ops->update_ivtes(ctx);
else WARN(1, "BUG: cxl_allocate_afu_irqs must be called prior to starting the context on this platform\n");
}
return res;
}
EXPORT_SYMBOL_GPL(cxl_allocate_afu_irqs);
void cxl_free_afu_irqs(struct cxl_context *ctx)
{
irq_hw_number_t hwirq;
unsigned int virq;
if (!cpu_has_feature(CPU_FTR_HVMODE)) {
hwirq = cxl_find_afu_irq(ctx, 0);
if (hwirq) {
virq = irq_find_mapping(NULL, hwirq);
if (virq)
cxl_unmap_irq(virq, ctx);
}
}
afu_irq_name_free(ctx);
cxl_ops->release_irq_ranges(&ctx->irqs, ctx->afu->adapter);
}
EXPORT_SYMBOL_GPL(cxl_free_afu_irqs);
int cxl_map_afu_irq(struct cxl_context *ctx, int num,
irq_handler_t handler, void *cookie, char *name)
{
irq_hw_number_t hwirq;
/*
* Find interrupt we are to register.
*/
hwirq = cxl_find_afu_irq(ctx, num);
if (!hwirq)
return -ENOENT;
return cxl_map_irq(ctx->afu->adapter, hwirq, handler, cookie, name);
}
EXPORT_SYMBOL_GPL(cxl_map_afu_irq);
void cxl_unmap_afu_irq(struct cxl_context *ctx, int num, void *cookie)
{
irq_hw_number_t hwirq;
unsigned int virq;
hwirq = cxl_find_afu_irq(ctx, num);
if (!hwirq)
return;
virq = irq_find_mapping(NULL, hwirq);
if (virq)
cxl_unmap_irq(virq, cookie);
}
EXPORT_SYMBOL_GPL(cxl_unmap_afu_irq);
/*
* Start a context
* Code here similar to afu_ioctl_start_work().
*/
int cxl_start_context(struct cxl_context *ctx, u64 wed,
struct task_struct *task)
{
int rc = 0;
bool kernel = true;
pr_devel("%s: pe: %i\n", __func__, ctx->pe);
mutex_lock(&ctx->status_mutex);
if (ctx->status == STARTED)
goto out; /* already started */
cxl: Prevent adapter reset if an active context exists This patch prevents resetting the cxl adapter via sysfs in presence of one or more active cxl_context on it. This protects against an unrecoverable error caused by PSL owning a dirty cache line even after reset and host tries to touch the same cache line. In case a force reset of the card is required irrespective of any active contexts, the int value -1 can be stored in the 'reset' sysfs attribute of the card. The patch introduces a new atomic_t member named contexts_num inside struct cxl that holds the number of active context attached to the card , which is checked against '0' before proceeding with the reset. To prevent against a race condition where a context is activated just after reset check is performed, the contexts_num is atomically set to '-1' after reset-check to indicate that no more contexts can be activated on the card anymore. Before activating a context we atomically test if contexts_num is non-negative and if so, increment its value by one. In case the value of contexts_num is negative then it indicates that the card is about to be reset and context activation is error-ed out at that point. Fixes: 62fa19d4b4fd ("cxl: Add ability to reset the card") Cc: stable@vger.kernel.org # v4.0+ Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-10-14 17:38:36 +08:00
/*
* Increment the mapped context count for adapter. This also checks
* if adapter_context_lock is taken.
*/
rc = cxl_adapter_context_get(ctx->afu->adapter);
if (rc)
goto out;
if (task) {
ctx->pid = get_task_pid(task, PIDTYPE_PID);
kernel = false;
ctx->real_mode = false;
/* acquire a reference to the task's mm */
ctx->mm = get_task_mm(current);
/* ensure this mm_struct can't be freed */
cxl_context_mm_count_get(ctx);
/* decrement the use count */
if (ctx->mm)
mmput(ctx->mm);
}
cxl_ctx_get();
if ((rc = cxl_ops->attach_process(ctx, kernel, wed, 0))) {
put_pid(ctx->pid);
ctx->pid = NULL;
cxl: Prevent adapter reset if an active context exists This patch prevents resetting the cxl adapter via sysfs in presence of one or more active cxl_context on it. This protects against an unrecoverable error caused by PSL owning a dirty cache line even after reset and host tries to touch the same cache line. In case a force reset of the card is required irrespective of any active contexts, the int value -1 can be stored in the 'reset' sysfs attribute of the card. The patch introduces a new atomic_t member named contexts_num inside struct cxl that holds the number of active context attached to the card , which is checked against '0' before proceeding with the reset. To prevent against a race condition where a context is activated just after reset check is performed, the contexts_num is atomically set to '-1' after reset-check to indicate that no more contexts can be activated on the card anymore. Before activating a context we atomically test if contexts_num is non-negative and if so, increment its value by one. In case the value of contexts_num is negative then it indicates that the card is about to be reset and context activation is error-ed out at that point. Fixes: 62fa19d4b4fd ("cxl: Add ability to reset the card") Cc: stable@vger.kernel.org # v4.0+ Acked-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by: Vaibhav Jain <vaibhav@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-10-14 17:38:36 +08:00
cxl_adapter_context_put(ctx->afu->adapter);
cxl_ctx_put();
if (task)
cxl_context_mm_count_put(ctx);
goto out;
}
ctx->status = STARTED;
out:
mutex_unlock(&ctx->status_mutex);
return rc;
}
EXPORT_SYMBOL_GPL(cxl_start_context);
int cxl_process_element(struct cxl_context *ctx)
{
return ctx->external_pe;
}
EXPORT_SYMBOL_GPL(cxl_process_element);
/* Stop a context. Returns 0 on success, otherwise -Errno */
int cxl_stop_context(struct cxl_context *ctx)
{
return __detach_context(ctx);
}
EXPORT_SYMBOL_GPL(cxl_stop_context);
void cxl_set_master(struct cxl_context *ctx)
{
ctx->master = true;
}
EXPORT_SYMBOL_GPL(cxl_set_master);
int cxl_set_translation_mode(struct cxl_context *ctx, bool real_mode)
{
if (ctx->status == STARTED) {
/*
* We could potentially update the PE and issue an update LLCMD
* to support this, but it doesn't seem to have a good use case
* since it's trivial to just create a second kernel context
* with different translation modes, so until someone convinces
* me otherwise:
*/
return -EBUSY;
}
ctx->real_mode = real_mode;
return 0;
}
EXPORT_SYMBOL_GPL(cxl_set_translation_mode);
/* wrappers around afu_* file ops which are EXPORTED */
int cxl_fd_open(struct inode *inode, struct file *file)
{
return afu_open(inode, file);
}
EXPORT_SYMBOL_GPL(cxl_fd_open);
int cxl_fd_release(struct inode *inode, struct file *file)
{
return afu_release(inode, file);
}
EXPORT_SYMBOL_GPL(cxl_fd_release);
long cxl_fd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
return afu_ioctl(file, cmd, arg);
}
EXPORT_SYMBOL_GPL(cxl_fd_ioctl);
int cxl_fd_mmap(struct file *file, struct vm_area_struct *vm)
{
return afu_mmap(file, vm);
}
EXPORT_SYMBOL_GPL(cxl_fd_mmap);
unsigned int cxl_fd_poll(struct file *file, struct poll_table_struct *poll)
{
return afu_poll(file, poll);
}
EXPORT_SYMBOL_GPL(cxl_fd_poll);
ssize_t cxl_fd_read(struct file *file, char __user *buf, size_t count,
loff_t *off)
{
return afu_read(file, buf, count, off);
}
EXPORT_SYMBOL_GPL(cxl_fd_read);
#define PATCH_FOPS(NAME) if (!fops->NAME) fops->NAME = afu_fops.NAME
/* Get a struct file and fd for a context and attach the ops */
struct file *cxl_get_fd(struct cxl_context *ctx, struct file_operations *fops,
int *fd)
{
struct file *file;
int rc, flags, fdtmp;
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
char *name = NULL;
/* only allow one per context */
if (ctx->mapping)
return ERR_PTR(-EEXIST);
flags = O_RDWR | O_CLOEXEC;
/* This code is similar to anon_inode_getfd() */
rc = get_unused_fd_flags(flags);
if (rc < 0)
return ERR_PTR(rc);
fdtmp = rc;
/*
* Patch the file ops. Needs to be careful that this is rentrant safe.
*/
if (fops) {
PATCH_FOPS(open);
PATCH_FOPS(poll);
PATCH_FOPS(read);
PATCH_FOPS(release);
PATCH_FOPS(unlocked_ioctl);
PATCH_FOPS(compat_ioctl);
PATCH_FOPS(mmap);
} else /* use default ops */
fops = (struct file_operations *)&afu_fops;
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
name = kasprintf(GFP_KERNEL, "cxl:%d", ctx->pe);
file = cxl_getfile(name, fops, ctx, flags);
kfree(name);
if (IS_ERR(file))
goto err_fd;
cxl: Fix coredump generation when cxl_get_fd() is used If a process dumps core while owning a cxl file descriptor obtained from an AFU driver (e.g. cxlflash) through the cxl_get_fd() API, the following error occurs: [ 868.027591] Unable to handle kernel paging request for data at address ... [ 868.027778] Faulting instruction address: 0xc00000000035edb0 cpu 0x8c: Vector: 300 (Data Access) at [c000003c688275e0] pc: c00000000035edb0: elf_core_dump+0xd60/0x1300 lr: c00000000035ed80: elf_core_dump+0xd30/0x1300 sp: c000003c68827860 msr: 9000000100009033 dar: c dsisr: 40000000 current = 0xc000003c68780000 paca = 0xc000000001b73200 softe: 0 irq_happened: 0x01 pid = 46725, comm = hxesurelock enter ? for help [c000003c68827a60] c00000000036948c do_coredump+0xcec/0x11e0 [c000003c68827c20] c0000000000ce9e0 get_signal+0x540/0x7b0 [c000003c68827d10] c000000000017354 do_signal+0x54/0x2b0 [c000003c68827e00] c00000000001777c do_notify_resume+0xbc/0xd0 [c000003c68827e30] c000000000009838 ret_from_except_lite+0x64/0x68 --- Exception: 300 (Data Access) at 00003fff98ad2918 The root cause is that the address_space structure for the file doesn't define a 'host' member. When cxl allocates a file descriptor, it's using the anonymous inode to back the file, but allocates a private address_space for each context. The private address_space allows to track memory allocation for each context. cxl doesn't define the 'host' member of the address space, i.e. the inode. We don't want to define it as the anonymous inode, since there's no longer a 1-to-1 relation between address_space and inode. To fix it, instead of using the anonymous inode, we introduce a simple pseudo filesystem so that cxl can allocate its own inodes. So we now have one inode for each file and address_space. The pseudo filesystem is only mounted on the first allocation of a file descriptor by cxl_get_fd(). Tested with cxlflash. Signed-off-by: Frederic Barrat <fbarrat@linux.vnet.ibm.com> Reviewed-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-11-18 20:00:31 +08:00
cxl_context_set_mapping(ctx, file->f_mapping);
*fd = fdtmp;
return file;
err_fd:
put_unused_fd(fdtmp);
return NULL;
}
EXPORT_SYMBOL_GPL(cxl_get_fd);
struct cxl_context *cxl_fops_get_context(struct file *file)
{
return file->private_data;
}
EXPORT_SYMBOL_GPL(cxl_fops_get_context);
cxl: Add mechanism for delivering AFU driver specific events This adds an afu_driver_ops structure with fetch_event() and event_delivered() callbacks. An AFU driver such as cxlflash can fill this out and associate it with a context to enable passing custom AFU specific events to userspace. This also adds a new kernel API function cxl_context_pending_events(), that the AFU driver can use to notify the cxl driver that new specific events are ready to be delivered, and wake up anyone waiting on the context wait queue. The current count of AFU driver specific events is stored in the field afu_driver_events of the context structure. The cxl driver checks the afu_driver_events count during poll, select, read, etc. calls to check if an AFU driver specific event is pending, and calls fetch_event() to obtain and deliver that event. This way, the cxl driver takes care of all the usual locking semantics around these calls and handles all the generic cxl events, so that the AFU driver only needs to worry about it's own events. fetch_event() return a struct cxl_event_afu_driver_reserved, allocated by the AFU driver, and filled in with the specific event information and size. Total event size (header + data) should not be greater than CXL_READ_MIN_SIZE (4K). Th cxl driver prepends an appropriate cxl event header, copies the event to userspace, and finally calls event_delivered() to return the status of the operation to the AFU driver. The event is identified by the context and cxl_event_afu_driver_reserved pointers. Since AFU drivers provide their own means for userspace to obtain the AFU file descriptor (i.e. cxlflash uses an ioctl on their scsi file descriptor to obtain the AFU file descriptor) and the generic cxl driver will never use this event, the ABI of the event is up to each individual AFU driver. Signed-off-by: Philippe Bergheaud <felix@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-06-23 21:03:53 +08:00
void cxl_set_driver_ops(struct cxl_context *ctx,
struct cxl_afu_driver_ops *ops)
{
WARN_ON(!ops->fetch_event || !ops->event_delivered);
atomic_set(&ctx->afu_driver_events, 0);
ctx->afu_driver_ops = ops;
}
EXPORT_SYMBOL_GPL(cxl_set_driver_ops);
void cxl_context_events_pending(struct cxl_context *ctx,
unsigned int new_events)
{
atomic_add(new_events, &ctx->afu_driver_events);
wake_up_all(&ctx->wq);
}
EXPORT_SYMBOL_GPL(cxl_context_events_pending);
int cxl_start_work(struct cxl_context *ctx,
struct cxl_ioctl_start_work *work)
{
int rc;
/* code taken from afu_ioctl_start_work */
if (!(work->flags & CXL_START_WORK_NUM_IRQS))
work->num_interrupts = ctx->afu->pp_irqs;
else if ((work->num_interrupts < ctx->afu->pp_irqs) ||
(work->num_interrupts > ctx->afu->irqs_max)) {
return -EINVAL;
}
rc = afu_register_irqs(ctx, work->num_interrupts);
if (rc)
return rc;
rc = cxl_start_context(ctx, work->work_element_descriptor, current);
if (rc < 0) {
afu_release_irqs(ctx, ctx);
return rc;
}
return 0;
}
EXPORT_SYMBOL_GPL(cxl_start_work);
void __iomem *cxl_psa_map(struct cxl_context *ctx)
{
if (ctx->status != STARTED)
return NULL;
pr_devel("%s: psn_phys%llx size:%llx\n",
__func__, ctx->psn_phys, ctx->psn_size);
return ioremap(ctx->psn_phys, ctx->psn_size);
}
EXPORT_SYMBOL_GPL(cxl_psa_map);
void cxl_psa_unmap(void __iomem *addr)
{
iounmap(addr);
}
EXPORT_SYMBOL_GPL(cxl_psa_unmap);
int cxl_afu_reset(struct cxl_context *ctx)
{
struct cxl_afu *afu = ctx->afu;
int rc;
rc = cxl_ops->afu_reset(afu);
if (rc)
return rc;
return cxl_ops->afu_check_and_enable(afu);
}
EXPORT_SYMBOL_GPL(cxl_afu_reset);
void cxl_perst_reloads_same_image(struct cxl_afu *afu,
bool perst_reloads_same_image)
{
afu->adapter->perst_same_image = perst_reloads_same_image;
}
EXPORT_SYMBOL_GPL(cxl_perst_reloads_same_image);
ssize_t cxl_read_adapter_vpd(struct pci_dev *dev, void *buf, size_t count)
{
struct cxl_afu *afu = cxl_pci_to_afu(dev);
if (IS_ERR(afu))
return -ENODEV;
return cxl_ops->read_adapter_vpd(afu->adapter, buf, count);
}
EXPORT_SYMBOL_GPL(cxl_read_adapter_vpd);
int cxl_set_max_irqs_per_process(struct pci_dev *dev, int irqs)
{
struct cxl_afu *afu = cxl_pci_to_afu(dev);
if (IS_ERR(afu))
return -ENODEV;
if (irqs > afu->adapter->user_irqs)
return -EINVAL;
/* Limit user_irqs to prevent the user increasing this via sysfs */
afu->adapter->user_irqs = irqs;
afu->irqs_max = irqs;
return 0;
}
EXPORT_SYMBOL_GPL(cxl_set_max_irqs_per_process);
int cxl_get_max_irqs_per_process(struct pci_dev *dev)
{
struct cxl_afu *afu = cxl_pci_to_afu(dev);
if (IS_ERR(afu))
return -ENODEV;
return afu->irqs_max;
}
EXPORT_SYMBOL_GPL(cxl_get_max_irqs_per_process);
cxl: Add support for interrupts on the Mellanox CX4 The Mellanox CX4 in cxl mode uses a hybrid interrupt model, where interrupts are routed from the networking hardware to the XSL using the MSIX table, and from there will be transformed back into an MSIX interrupt using the cxl style interrupts (i.e. using IVTE entries and ranges to map a PE and AFU interrupt number to an MSIX address). We want to hide the implementation details of cxl interrupts as much as possible. To this end, we use a special version of the MSI setup & teardown routines in the PHB while in cxl mode to allocate the cxl interrupts and configure the IVTE entries in the process element. This function does not configure the MSIX table - the CX4 card uses a custom format in that table and it would not be appropriate to fill that out in generic code. The rest of the functionality is similar to the "Full MSI-X mode" described in the CAIA, and this could be easily extended to support other adapters that use that mode in the future. The interrupts will be associated with the default context. If the maximum number of interrupts per context has been limited (e.g. by the mlx5 driver), it will automatically allocate additional kernel contexts to associate extra interrupts as required. These contexts will be started using the same WED that was used to start the default context. Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-07-14 05:17:10 +08:00
/*
* This is a special interrupt allocation routine called from the PHB's MSI
* setup function. When capi interrupts are allocated in this manner they must
* still be associated with a running context, but since the MSI APIs have no
* way to specify this we use the default context associated with the device.
*
* The Mellanox CX4 has a hardware limitation that restricts the maximum AFU
* interrupt number, so in order to overcome this their driver informs us of
* the restriction by setting the maximum interrupts per context, and we
* allocate additional contexts as necessary so that we can keep the AFU
* interrupt number within the supported range.
*/
int _cxl_cx4_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
{
struct cxl_context *ctx, *new_ctx, *default_ctx;
int remaining;
int rc;
ctx = default_ctx = cxl_get_context(pdev);
if (WARN_ON(!default_ctx))
return -ENODEV;
remaining = nvec;
while (remaining > 0) {
rc = cxl_allocate_afu_irqs(ctx, min(remaining, ctx->afu->irqs_max));
if (rc) {
pr_warn("%s: Failed to find enough free MSIs\n", pci_name(pdev));
return rc;
}
remaining -= ctx->afu->irqs_max;
if (ctx != default_ctx && default_ctx->status == STARTED) {
WARN_ON(cxl_start_context(ctx,
be64_to_cpu(default_ctx->elem->common.wed),
NULL));
}
if (remaining > 0) {
new_ctx = cxl_dev_context_init(pdev);
if (IS_ERR(new_ctx)) {
cxl: Add support for interrupts on the Mellanox CX4 The Mellanox CX4 in cxl mode uses a hybrid interrupt model, where interrupts are routed from the networking hardware to the XSL using the MSIX table, and from there will be transformed back into an MSIX interrupt using the cxl style interrupts (i.e. using IVTE entries and ranges to map a PE and AFU interrupt number to an MSIX address). We want to hide the implementation details of cxl interrupts as much as possible. To this end, we use a special version of the MSI setup & teardown routines in the PHB while in cxl mode to allocate the cxl interrupts and configure the IVTE entries in the process element. This function does not configure the MSIX table - the CX4 card uses a custom format in that table and it would not be appropriate to fill that out in generic code. The rest of the functionality is similar to the "Full MSI-X mode" described in the CAIA, and this could be easily extended to support other adapters that use that mode in the future. The interrupts will be associated with the default context. If the maximum number of interrupts per context has been limited (e.g. by the mlx5 driver), it will automatically allocate additional kernel contexts to associate extra interrupts as required. These contexts will be started using the same WED that was used to start the default context. Signed-off-by: Ian Munsie <imunsie@au1.ibm.com> Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-07-14 05:17:10 +08:00
pr_warn("%s: Failed to allocate enough contexts for MSIs\n", pci_name(pdev));
return -ENOSPC;
}
list_add(&new_ctx->extra_irq_contexts, &ctx->extra_irq_contexts);
ctx = new_ctx;
}
}
return 0;
}
/* Exported via cxl_base */
void _cxl_cx4_teardown_msi_irqs(struct pci_dev *pdev)
{
struct cxl_context *ctx, *pos, *tmp;
ctx = cxl_get_context(pdev);
if (WARN_ON(!ctx))
return;
cxl_free_afu_irqs(ctx);
list_for_each_entry_safe(pos, tmp, &ctx->extra_irq_contexts, extra_irq_contexts) {
cxl_stop_context(pos);
cxl_free_afu_irqs(pos);
list_del(&pos->extra_irq_contexts);
cxl_release_context(pos);
}
}
/* Exported via cxl_base */