5.10.79-rc1 review

From: Paolo Bonzini <[email protected]>

commit 3d5e7a28b1ea2d603dea478e58e37ce75b9597ab upstream.

This is a new warning in clang top-of-tree (will be clang 14):

In file included from arch/x86/kvm/mmu/mmu.c:27:
arch/x86/kvm/mmu/spte.h:318:9: error: use of bitwise '|' with boolean operands [-Werror,-Wbitwise-instead-of-logical]

        return __is_bad_mt_xwr(rsvd_check, spte) |
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                                                 ||

arch/x86/kvm/mmu/spte.h:318:9: note: cast one or both operands to int to silence this warning

The code is fine, but change it anyway to shut up this clever clogs of a compiler.

---
 arch/x86/kvm/mmu/mmu.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

@@ -3545,7 +3545,7 @@ static bool get_mmio_spte(struct kvm_vcp
 		 * reserved bit and EPT's invalid memtype/XWR checks to avoid
 		 * adding a Jcc in the loop.
 		 */
-		reserved |= __is_bad_mt_xwr(rsvd_check, sptes[level - 1]) |
+		reserved |= __is_bad_mt_xwr(rsvd_check, sptes[level - 1]) ||
 			    __is_rsvd_bits_set(rsvd_check, sptes[level - 1],
 					       level);
 	}

From: Petr Mladek <[email protected]>

commit 3cffa06aeef7ece30f6b5ac0ea51f264e8fea4d0 upstream.

The commit 48021f98130880dd74 ("printk: handle blank console arguments passed in.") prevented crash caused by empty console= parameter value.

Unfortunately, this value is widely used on Chromebooks to disable the console output. The above commit caused performance regression because the messages were pushed on slow console even though nobody was watching it.

Use ttynull driver explicitly for console="" and console=null parameters. It has been created for exactly this purpose.

It causes that preferred_console is set. As a result, ttySX and ttyX are not used as a fallback. And only ttynull console gets registered by default.

It still allows to register other consoles either by additional console= parameters or SPCR. It prevents regression because it worked this way even before. Also it is a sane semantic. Preventing output on all consoles should be done another way, for example, by introducing mute_console parameter.

---
 kernel/printk/printk.c |    9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

@@ -2213,8 +2213,15 @@ static int __init console_setup(char *st
 	char *s, *options, *brl_options = NULL;
 	int idx;
 
-	if (str[0] == 0)
+	/*
+	 * console="" or console=null have been suggested as a way to
+	 * disable console output. Use ttynull that has been created
+	 * for exacly this purpose.
+	 */
+	if (str[0] == 0 || strcmp(str, "null") == 0) {
+		__add_preferred_console("ttynull", 0, NULL, NULL, true);
 		return 1;
+	}
 
 	if (_braille_console_setup(&str, &brl_options))
 		return 1;

From: Pavel Skripkin <[email protected]>

commit c052cc1a069c3e575619cf64ec427eb41176ca70 upstream.

Syzbot reported use-after-free in rtl8712_dl_fw(). The problem was in race condition between r871xu_dev_remove() ->ndo_open() callback.

It's easy to see from crash log, that driver accesses released firmware in ->ndo_open() callback. It may happen, since driver was releasing firmware _before_ unregistering netdev. Fix it by moving unregister_netdev() before cleaning up resources.

Call Trace:
...
 rtl871x_open_fw drivers/staging/rtl8712/hal_init.c:83 [inline]
 rtl8712_dl_fw+0xd95/0xe10 drivers/staging/rtl8712/hal_init.c:170
 rtl8712_hal_init drivers/staging/rtl8712/hal_init.c:330 [inline]
 rtl871x_hal_init+0xae/0x180 drivers/staging/rtl8712/hal_init.c:394
 netdev_open+0xe6/0x6c0 drivers/staging/rtl8712/os_intfs.c:380
 __dev_open+0x2bc/0x4d0 net/core/dev.c:1484

Freed by task 1306:
...
 release_firmware+0x1b/0x30 drivers/base/firmware_loader/main.c:1053
 r871xu_dev_remove+0xcc/0x2c0 drivers/staging/rtl8712/usb_intf.c:599
 usb_unbind_interface+0x1d8/0x8d0 drivers/usb/core/driver.c:458

---
 drivers/staging/rtl8712/usb_intf.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

@@ -598,12 +598,12 @@ static void r871xu_dev_remove(struct usb
 
 	/* never exit with a firmware callback pending */
 	wait_for_completion(&padapter->rtl8712_fw_ready);
+	if (pnetdev->reg_state != NETREG_UNINITIALIZED)
+		unregister_netdev(pnetdev); /* will call netdev_close() */
 	usb_set_intfdata(pusb_intf, NULL);
 	release_firmware(padapter->fw);
 	if (drvpriv.drv_registered)
 		padapter->surprise_removed = true;
-	if (pnetdev->reg_state != NETREG_UNINITIALIZED)
-		unregister_netdev(pnetdev); /* will call netdev_close() */
 	r8712_flush_rwctrl_works(padapter);
 	r8712_flush_led_works(padapter);
 	udelay(1);

From: Jan Kara <[email protected]>

commit e96a1866b40570b5950cda8602c2819189c62a48 upstream.

When isofs image is suitably corrupted isofs_read_inode() can read data beyond the end of buffer. Sanity-check the directory entry length before using it.

---
 fs/isofs/inode.c |    2 ++
 1 file changed, 2 insertions(+)

@@ -1321,6 +1321,8 @@ static int isofs_read_inode(struct inode
 
 	de = (struct iso_directory_record *) (bh->b_data + offset);
 	de_len = *(unsigned char *) de;
+	if (de_len < sizeof(struct iso_directory_record))
+		goto fail;
 
 	if (offset + de_len > bufsize) {
 		int frag1 = bufsize - offset;

From: Johan Hovold <[email protected]>

commit 536de747bc48262225889a533db6650731ab25d3 upstream.

USB transfer buffers are typically mapped for DMA and must not be allocated on the stack or transfers will fail.

Allocate proper transfer buffers in the various command helpers and return an error on short transfers instead of acting on random stack data.

Note that this also fixes a stack info leak on systems where DMA is not used as 32 bytes are always sent to the device regardless of how short the command is.

---
 drivers/staging/comedi/drivers/dt9812.c |  115 +++++++++++++++++++++++---------
 1 file changed, 86 insertions(+), 29 deletions(-)

@@ -32,6 +32,7 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/errno.h>
+#include <linux/slab.h>
 #include <linux/uaccess.h>
 
 #include "../comedi_usb.h"

@@ -237,22 +238,42 @@ static int dt9812_read_info(struct comed
 {
 	struct usb_device *usb = comedi_to_usb_dev(dev);
 	struct dt9812_private *devpriv = dev->private;
-	struct dt9812_usb_cmd cmd;
+	struct dt9812_usb_cmd *cmd;
+	size_t tbuf_size;
 	int count, ret;
+	void *tbuf;
 
-	cmd.cmd = cpu_to_le32(DT9812_R_FLASH_DATA);
-	cmd.u.flash_data_info.address =
+	tbuf_size = max(sizeof(*cmd), buf_size);
+
+	tbuf = kzalloc(tbuf_size, GFP_KERNEL);
+	if (!tbuf)
+		return -ENOMEM;
+
+	cmd = tbuf;
+
+	cmd->cmd = cpu_to_le32(DT9812_R_FLASH_DATA);
+	cmd->u.flash_data_info.address =
 	    cpu_to_le16(DT9812_DIAGS_BOARD_INFO_ADDR + offset);
-	cmd.u.flash_data_info.numbytes = cpu_to_le16(buf_size);
+	cmd->u.flash_data_info.numbytes = cpu_to_le16(buf_size);
 
 	/* DT9812 only responds to 32 byte writes!! */
 	ret = usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
-			   &cmd, 32, &count, DT9812_USB_TIMEOUT);
+			   cmd, sizeof(*cmd), &count, DT9812_USB_TIMEOUT);
 	if (ret)
-		return ret;
+		goto out;
+
+	ret = usb_bulk_msg(usb, usb_rcvbulkpipe(usb, devpriv->cmd_rd.addr),
+			   tbuf, buf_size, &count, DT9812_USB_TIMEOUT);
+	if (!ret) {
+		if (count == buf_size)
+			memcpy(buf, tbuf, buf_size);
+		else
+			ret = -EREMOTEIO;
+	}
+out:
+	kfree(tbuf);
 
-	return usb_bulk_msg(usb, usb_rcvbulkpipe(usb, devpriv->cmd_rd.addr),
-			    buf, buf_size, &count, DT9812_USB_TIMEOUT);
+	return ret;
 }
 
 static int dt9812_read_multiple_registers(struct comedi_device *dev,

@@ -261,22 +282,42 @@ static int dt9812_read_multiple_register
 {
 	struct usb_device *usb = comedi_to_usb_dev(dev);
 	struct dt9812_private *devpriv = dev->private;
-	struct dt9812_usb_cmd cmd;
+	struct dt9812_usb_cmd *cmd;
 	int i, count, ret;
+	size_t buf_size;
+	void *buf;
+
+	buf_size = max_t(size_t, sizeof(*cmd), reg_count);
+
+	buf = kzalloc(buf_size, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	cmd = buf;
 
-	cmd.cmd = cpu_to_le32(DT9812_R_MULTI_BYTE_REG);
-	cmd.u.read_multi_info.count = reg_count;
+	cmd->cmd = cpu_to_le32(DT9812_R_MULTI_BYTE_REG);
+	cmd->u.read_multi_info.count = reg_count;
 	for (i = 0; i < reg_count; i++)
-		cmd.u.read_multi_info.address[i] = address[i];
+		cmd->u.read_multi_info.address[i] = address[i];
 
 	/* DT9812 only responds to 32 byte writes!! */
 	ret = usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
-			   &cmd, 32, &count, DT9812_USB_TIMEOUT);
+			   cmd, sizeof(*cmd), &count, DT9812_USB_TIMEOUT);
 	if (ret)
-		return ret;
+		goto out;
 
-	return usb_bulk_msg(usb, usb_rcvbulkpipe(usb, devpriv->cmd_rd.addr),
-			    value, reg_count, &count, DT9812_USB_TIMEOUT);
+	ret = usb_bulk_msg(usb, usb_rcvbulkpipe(usb, devpriv->cmd_rd.addr),
+			   buf, reg_count, &count, DT9812_USB_TIMEOUT);
+	if (!ret) {
+		if (count == reg_count)
+			memcpy(value, buf, reg_count);
+		else
+			ret = -EREMOTEIO;
+	}
+out:
+	kfree(buf);
+
+	return ret;
 }
 
 static int dt9812_write_multiple_registers(struct comedi_device *dev,

@@ -285,19 +326,27 @@ static int dt9812_write_multiple_registe
 {
 	struct usb_device *usb = comedi_to_usb_dev(dev);
 	struct dt9812_private *devpriv = dev->private;
-	struct dt9812_usb_cmd cmd;
+	struct dt9812_usb_cmd *cmd;
 	int i, count;
+	int ret;
 
-	cmd.cmd = cpu_to_le32(DT9812_W_MULTI_BYTE_REG);
-	cmd.u.read_multi_info.count = reg_count;
+	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->cmd = cpu_to_le32(DT9812_W_MULTI_BYTE_REG);
+	cmd->u.read_multi_info.count = reg_count;
 	for (i = 0; i < reg_count; i++) {
-		cmd.u.write_multi_info.write[i].address = address[i];
-		cmd.u.write_multi_info.write[i].value = value[i];
+		cmd->u.write_multi_info.write[i].address = address[i];
+		cmd->u.write_multi_info.write[i].value = value[i];
 	}
 
 	/* DT9812 only responds to 32 byte writes!! */
-	return usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
-			    &cmd, 32, &count, DT9812_USB_TIMEOUT);
+	ret = usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
+			   cmd, sizeof(*cmd), &count, DT9812_USB_TIMEOUT);
+	kfree(cmd);
+
+	return ret;
 }
 
 static int dt9812_rmw_multiple_registers(struct comedi_device *dev,

@@ -306,17 +355,25 @@ static int dt9812_rmw_multiple_registers
 {
 	struct usb_device *usb = comedi_to_usb_dev(dev);
 	struct dt9812_private *devpriv = dev->private;
-	struct dt9812_usb_cmd cmd;
+	struct dt9812_usb_cmd *cmd;
 	int i, count;
+	int ret;
+
+	cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
 
-	cmd.cmd = cpu_to_le32(DT9812_RMW_MULTI_BYTE_REG);
-	cmd.u.rmw_multi_info.count = reg_count;
+	cmd->cmd = cpu_to_le32(DT9812_RMW_MULTI_BYTE_REG);
+	cmd->u.rmw_multi_info.count = reg_count;
 	for (i = 0; i < reg_count; i++)
-		cmd.u.rmw_multi_info.rmw[i] = rmw[i];
+		cmd->u.rmw_multi_info.rmw[i] = rmw[i];
 
 	/* DT9812 only responds to 32 byte writes!! */
-	return usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
-			    &cmd, 32, &count, DT9812_USB_TIMEOUT);
+	ret = usb_bulk_msg(usb, usb_sndbulkpipe(usb, devpriv->cmd_wr.addr),
+			   cmd, sizeof(*cmd), &count, DT9812_USB_TIMEOUT);
+	kfree(cmd);
+
+	return ret;
 }
 
 static int dt9812_digital_in(struct comedi_device *dev, u8 *bits)

From: Johan Hovold <[email protected]>

commit 907767da8f3a925b060c740e0b5c92ea7dbec440 upstream.

The driver uses endpoint-sized USB transfer buffers but had no sanity checks on the sizes. This can lead to zero-size-pointer dereferences or overflowed transfer buffers in ni6501_port_command() and ni6501_counter_command() if a (malicious) device has smaller max-packet sizes than expected (or when doing descriptor fuzz testing).

Add the missing sanity checks to probe().

---
 drivers/staging/comedi/drivers/ni_usb6501.c |   10 ++++++++++
 1 file changed, 10 insertions(+)

@@ -144,6 +144,10 @@ static const u8 READ_COUNTER_RESPONSE[]
 					   0x00, 0x00, 0x00, 0x02,
 					   0x00, 0x00, 0x00, 0x00};
 
+/* Largest supported packets */
+static const size_t TX_MAX_SIZE	= sizeof(SET_PORT_DIR_REQUEST);
+static const size_t RX_MAX_SIZE	= sizeof(READ_PORT_RESPONSE);
+
 enum commands {
 	READ_PORT,
 	WRITE_PORT,

@@ -501,6 +505,12 @@ static int ni6501_find_endpoints(struct
 	if (!devpriv->ep_rx || !devpriv->ep_tx)
 		return -ENODEV;
 
+	if (usb_endpoint_maxp(devpriv->ep_rx) < RX_MAX_SIZE)
+		return -ENODEV;
+
+	if (usb_endpoint_maxp(devpriv->ep_tx) < TX_MAX_SIZE)
+		return -ENODEV;
+
 	return 0;
 }
 

From: Johan Hovold <[email protected]>

commit a23461c47482fc232ffc9b819539d1f837adf2b1 upstream.

The driver uses endpoint-sized USB transfer buffers but up until recently had no sanity checks on the sizes.

Commit e1f13c879a7c ("staging: comedi: check validity of wMaxPacketSize of usb endpoints found") inadvertently fixed NULL-pointer dereferences when accessing the transfer buffers in case a malicious device has a zero wMaxPacketSize.

Make sure to allocate buffers large enough to handle also the other accesses that are done without a size check (e.g. byte 18 in vmk80xx_cnt_insn_read() for the VMK8061_MODEL) to avoid writing beyond the buffers, for example, when doing descriptor fuzzing.

The original driver was for a low-speed device with 8-byte buffers. Support was later added for a device that uses bulk transfers and is presumably a full-speed device with a maximum 64-byte wMaxPacketSize.

---
 drivers/staging/comedi/drivers/vmk80xx.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

@@ -90,6 +90,8 @@ enum {
 #define IC3_VERSION		BIT(0)
 #define IC6_VERSION		BIT(1)
 
+#define MIN_BUF_SIZE		64
+
 enum vmk80xx_model {
 	VMK8055_MODEL,
 	VMK8061_MODEL

@@ -678,12 +680,12 @@ static int vmk80xx_alloc_usb_buffers(str
 	struct vmk80xx_private *devpriv = dev->private;
 	size_t size;
 
-	size = usb_endpoint_maxp(devpriv->ep_rx);
+	size = max(usb_endpoint_maxp(devpriv->ep_rx), MIN_BUF_SIZE);
 	devpriv->usb_rx_buf = kzalloc(size, GFP_KERNEL);
 	if (!devpriv->usb_rx_buf)
 		return -ENOMEM;
 
-	size = usb_endpoint_maxp(devpriv->ep_tx);
+	size = max(usb_endpoint_maxp(devpriv->ep_rx), MIN_BUF_SIZE);
 	devpriv->usb_tx_buf = kzalloc(size, GFP_KERNEL);
 	if (!devpriv->usb_tx_buf)
 		return -ENOMEM;

From: Johan Hovold <[email protected]>

commit 78cdfd62bd54af615fba9e3ca1ba35de39d3871d upstream.

The driver is using endpoint-sized buffers but must not assume that the tx and rx buffers are of equal size or a malicious device could overflow the slab-allocated receive buffer when doing bulk transfers.

---
 drivers/staging/comedi/drivers/vmk80xx.c |   16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

@@ -159,22 +159,20 @@ static void vmk80xx_do_bulk_msg(struct c
 	__u8 rx_addr;
 	unsigned int tx_pipe;
 	unsigned int rx_pipe;
-	size_t size;
+	size_t tx_size;
+	size_t rx_size;
 
 	tx_addr = devpriv->ep_tx->bEndpointAddress;
 	rx_addr = devpriv->ep_rx->bEndpointAddress;
 	tx_pipe = usb_sndbulkpipe(usb, tx_addr);
 	rx_pipe = usb_rcvbulkpipe(usb, rx_addr);
-
-	/*
-	 * The max packet size attributes of the K8061
-	 * input/output endpoints are identical
-	 */
-	size = usb_endpoint_maxp(devpriv->ep_tx);
+	tx_size = usb_endpoint_maxp(devpriv->ep_tx);
+	rx_size = usb_endpoint_maxp(devpriv->ep_rx);
 
 	usb_bulk_msg(usb, tx_pipe, devpriv->usb_tx_buf,
-		     size, NULL, devpriv->ep_tx->bInterval);
-	usb_bulk_msg(usb, rx_pipe, devpriv->usb_rx_buf, size, NULL, HZ * 10);
+		     tx_size, NULL, devpriv->ep_tx->bInterval);
+
+	usb_bulk_msg(usb, rx_pipe, devpriv->usb_rx_buf, rx_size, NULL, HZ * 10);
 }
 
 static int vmk80xx_read_packet(struct comedi_device *dev)

From: Johan Hovold <[email protected]>

commit a56d3e40bda460edf3f8d6aac00ec0b322b4ab83 upstream.

USB bulk and interrupt message timeouts are specified in milliseconds and should specifically not vary with CONFIG_HZ.

Note that the bulk-out transfer timeout was set to the endpoint bInterval value, which should be ignored for bulk endpoints and is typically set to zero. This meant that a failing bulk-out transfer would never time out.

Assume that the 10 second timeout used for all other transfers is more than enough also for the bulk-out endpoint.

---
 drivers/staging/comedi/drivers/vmk80xx.c |   12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

@@ -91,6 +91,7 @@ enum {
 #define IC6_VERSION		BIT(1)
 
 #define MIN_BUF_SIZE		64
+#define PACKET_TIMEOUT		10000	/* ms */
 
 enum vmk80xx_model {
 	VMK8055_MODEL,

@@ -169,10 +170,11 @@ static void vmk80xx_do_bulk_msg(struct c
 	tx_size = usb_endpoint_maxp(devpriv->ep_tx);
 	rx_size = usb_endpoint_maxp(devpriv->ep_rx);
 
-	usb_bulk_msg(usb, tx_pipe, devpriv->usb_tx_buf,
-		     tx_size, NULL, devpriv->ep_tx->bInterval);
+	usb_bulk_msg(usb, tx_pipe, devpriv->usb_tx_buf, tx_size, NULL,
+		     PACKET_TIMEOUT);
 
-	usb_bulk_msg(usb, rx_pipe, devpriv->usb_rx_buf, rx_size, NULL, HZ * 10);
+	usb_bulk_msg(usb, rx_pipe, devpriv->usb_rx_buf, rx_size, NULL,
+		     PACKET_TIMEOUT);
 }
 
 static int vmk80xx_read_packet(struct comedi_device *dev)

@@ -191,7 +193,7 @@ static int vmk80xx_read_packet(struct co
 	pipe = usb_rcvintpipe(usb, ep->bEndpointAddress);
 	return usb_interrupt_msg(usb, pipe, devpriv->usb_rx_buf,
 				 usb_endpoint_maxp(ep), NULL,
-				 HZ * 10);
+				 PACKET_TIMEOUT);
 }
 
 static int vmk80xx_write_packet(struct comedi_device *dev, int cmd)

@@ -212,7 +214,7 @@ static int vmk80xx_write_packet(struct c
 	pipe = usb_sndintpipe(usb, ep->bEndpointAddress);
 	return usb_interrupt_msg(usb, pipe, devpriv->usb_tx_buf,
 				 usb_endpoint_maxp(ep), NULL,
-				 HZ * 10);
+				 PACKET_TIMEOUT);
 }
 
 static int vmk80xx_reset_device(struct comedi_device *dev)

From: Johan Hovold <[email protected]>

commit ce4940525f36ffdcf4fa623bcedab9c2a6db893a upstream.

USB control-message timeouts are specified in milliseconds and should specifically not vary with CONFIG_HZ.

---
 drivers/staging/rtl8712/usb_ops_linux.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

@@ -494,7 +494,7 @@ int r8712_usbctrl_vendorreq(struct intf_
 		memcpy(pIo_buf, pdata, len);
 	}
 	status = usb_control_msg(udev, pipe, request, reqtype, value, index,
-				 pIo_buf, len, HZ / 2);
+				 pIo_buf, len, 500);
 	if (status > 0) {  /* Success this control transfer. */
 		if (requesttype == 0x01) {
 			/* For Control read transfer, we have to copy the read

From: Juergen Gross <[email protected]>

commit 1e254d0d86a0f2efd4190a89d5204b37c18c6381 upstream.

This reverts commit 76b4f357d0e7d8f6f0013c733e6cba1773c266d3.

The commit has the wrong reasoning, as KVM_MAX_VCPU_ID is not defining the maximum allowed vcpu-id as its name suggests, but the number of vcpu-ids. So revert this patch again.

Message-Id: <[email protected]>

---
 arch/x86/kvm/ioapic.c |    2 +-
 arch/x86/kvm/ioapic.h |    4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

@@ -96,7 +96,7 @@ static unsigned long ioapic_read_indirec
 static void rtc_irq_eoi_tracking_reset(struct kvm_ioapic *ioapic)
 {
 	ioapic->rtc_status.pending_eoi = 0;
-	bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID + 1);
+	bitmap_zero(ioapic->rtc_status.dest_map.map, KVM_MAX_VCPU_ID);
 }
 
 static void kvm_rtc_eoi_tracking_restore_all(struct kvm_ioapic *ioapic);

@@ -43,13 +43,13 @@ struct kvm_vcpu;
 
 struct dest_map {
 	/* vcpu bitmap where IRQ has been sent */
-	DECLARE_BITMAP(map, KVM_MAX_VCPU_ID + 1);
+	DECLARE_BITMAP(map, KVM_MAX_VCPU_ID);
 
 	/*
 	 * Vector sent to a given vcpu, only valid when
 	 * the vcpu's bit in map is set
 	 */
-	u8 vectors[KVM_MAX_VCPU_ID + 1];
+	u8 vectors[KVM_MAX_VCPU_ID];
 };
 
 

From: Neal Liu <[email protected]>

commit 7f2d73788d9067fd4f677ac5f60ffd25945af7af upstream.

For Aspeed, HCHalted status depends on not only Run/Stop but also ASS/PSS status.
Handshake CMD_RUN on startup instead.

---
 drivers/usb/host/ehci-hcd.c      |   11 ++++++++++-
 drivers/usb/host/ehci-platform.c |    6 ++++++
 drivers/usb/host/ehci.h          |    1 +
 3 files changed, 17 insertions(+), 1 deletion(-)

@@ -634,7 +634,16 @@ static int ehci_run (struct usb_hcd *hcd
 	/* Wait until HC become operational */
 	ehci_readl(ehci, &ehci->regs->command);	/* unblock posted writes */
 	msleep(5);
-	rc = ehci_handshake(ehci, &ehci->regs->status, STS_HALT, 0, 100 * 1000);
+
+	/* For Aspeed, STS_HALT also depends on ASS/PSS status.
+	 * Check CMD_RUN instead.
+	 */
+	if (ehci->is_aspeed)
+		rc = ehci_handshake(ehci, &ehci->regs->command, CMD_RUN,
+				    1, 100 * 1000);
+	else
+		rc = ehci_handshake(ehci, &ehci->regs->status, STS_HALT,
+				    0, 100 * 1000);
 
 	up_write(&ehci_cf_port_reset_rwsem);
 

@@ -294,6 +294,12 @@ static int ehci_platform_probe(struct pl
 					  "has-transaction-translator"))
 			hcd->has_tt = 1;
 
+		if (of_device_is_compatible(dev->dev.of_node,
+					    "aspeed,ast2500-ehci") ||
+		    of_device_is_compatible(dev->dev.of_node,
+					    "aspeed,ast2600-ehci"))
+			ehci->is_aspeed = 1;
+
 		if (soc_device_match(quirk_poll_match))
 			priv->quirk_poll = true;
 

@@ -218,6 +218,7 @@ struct ehci_hcd {			/* one per controlle
 	unsigned		frame_index_bug:1; /* MosChip (AKA NetMos) */
 	unsigned		need_oc_pp_cycle:1; /* MPC834X port power */
 	unsigned		imx28_write_fix:1; /* For Freescale i.MX28 */
+	unsigned		is_aspeed:1;
 
 	/* required for usb32 quirk */
 	#define OHCI_CTRL_HCFS          (3 << 6)

From: Johan Hovold <[email protected]>

commit 4cfa36d312d6789448b59a7aae770ac8425017a3 upstream.

USB control-message timeouts are specified in milliseconds and should specifically not vary with CONFIG_HZ.

---
 drivers/staging/rtl8192u/r8192U_core.c |   18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

@@ -229,7 +229,7 @@ int write_nic_byte_E(struct net_device *
 
 	status = usb_control_msg(udev, usb_sndctrlpipe(udev, 0),
 				 RTL8187_REQ_SET_REGS, RTL8187_REQT_WRITE,
-				 indx | 0xfe00, 0, usbdata, 1, HZ / 2);
+				 indx | 0xfe00, 0, usbdata, 1, 500);
 	kfree(usbdata);
 
 	if (status < 0) {

@@ -251,7 +251,7 @@ int read_nic_byte_E(struct net_device *d
 
 	status = usb_control_msg(udev, usb_rcvctrlpipe(udev, 0),
 				 RTL8187_REQ_GET_REGS, RTL8187_REQT_READ,
-				 indx | 0xfe00, 0, usbdata, 1, HZ / 2);
+				 indx | 0xfe00, 0, usbdata, 1, 500);
 	*data = *usbdata;
 	kfree(usbdata);
 

@@ -279,7 +279,7 @@ int write_nic_byte(struct net_device *de
 	status = usb_control_msg(udev, usb_sndctrlpipe(udev, 0),
 				 RTL8187_REQ_SET_REGS, RTL8187_REQT_WRITE,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 1, HZ / 2);
+				 usbdata, 1, 500);
 	kfree(usbdata);
 
 	if (status < 0) {

@@ -305,7 +305,7 @@ int write_nic_word(struct net_device *de
 	status = usb_control_msg(udev, usb_sndctrlpipe(udev, 0),
 				 RTL8187_REQ_SET_REGS, RTL8187_REQT_WRITE,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 2, HZ / 2);
+				 usbdata, 2, 500);
 	kfree(usbdata);
 
 	if (status < 0) {

@@ -331,7 +331,7 @@ int write_nic_dword(struct net_device *d
 	status = usb_control_msg(udev, usb_sndctrlpipe(udev, 0),
 				 RTL8187_REQ_SET_REGS, RTL8187_REQT_WRITE,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 4, HZ / 2);
+				 usbdata, 4, 500);
 	kfree(usbdata);
 
 	if (status < 0) {

@@ -355,7 +355,7 @@ int read_nic_byte(struct net_device *dev
 	status = usb_control_msg(udev, usb_rcvctrlpipe(udev, 0),
 				 RTL8187_REQ_GET_REGS, RTL8187_REQT_READ,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 1, HZ / 2);
+				 usbdata, 1, 500);
 	*data = *usbdata;
 	kfree(usbdata);
 

@@ -380,7 +380,7 @@ int read_nic_word(struct net_device *dev
 	status = usb_control_msg(udev, usb_rcvctrlpipe(udev, 0),
 				 RTL8187_REQ_GET_REGS, RTL8187_REQT_READ,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 2, HZ / 2);
+				 usbdata, 2, 500);
 	*data = *usbdata;
 	kfree(usbdata);
 

@@ -404,7 +404,7 @@ static int read_nic_word_E(struct net_de
 
 	status = usb_control_msg(udev, usb_rcvctrlpipe(udev, 0),
 				 RTL8187_REQ_GET_REGS, RTL8187_REQT_READ,
-				 indx | 0xfe00, 0, usbdata, 2, HZ / 2);
+				 indx | 0xfe00, 0, usbdata, 2, 500);
 	*data = *usbdata;
 	kfree(usbdata);
 

@@ -430,7 +430,7 @@ int read_nic_dword(struct net_device *de
 	status = usb_control_msg(udev, usb_rcvctrlpipe(udev, 0),
 				 RTL8187_REQ_GET_REGS, RTL8187_REQT_READ,
 				 (indx & 0xff) | 0xff00, (indx >> 8) & 0x0f,
-				 usbdata, 4, HZ / 2);
+				 usbdata, 4, 500);
 	*data = *usbdata;
 	kfree(usbdata);
 

From: Gustavo A. R. Silva <[email protected]>

commit a44f9d6f9dc1fb314a3f1ed2dcd4fbbcc3d9f892 upstream.

There is a wrong comparison of the total size of the loaded firmware css->fw->size with the size of a pointer to struct imgu_fw_header.

Turn binary_header into a flexible-array member[1][2], use the struct_size() helper and fix the wrong size comparison. Notice that the loaded firmware needs to contain at least one 'struct imgu_fw_info' item in the binary_header[] array.

It's also worth mentioning that

	"css->fw->size < struct_size(css->fwp, binary_header, 1)"

with binary_header declared as a flexible-array member is equivalent to

	"css->fw->size < sizeof(struct imgu_fw_header)"

with binary_header declared as a one-element array (as in the original code).

The replacement of the one-element array with a flexible-array member also helps with the ongoing efforts to globally enable -Warray-bounds and get us closer to being able to tighten the FORTIFY_SOURCE routines on memcpy().

[1] https://en.wikipedia.org/wiki/Flexible_array_member
[2] https://www.kernel.org/doc/html/v5.10/process/deprecated.html#zero-length-and-one-element-arrays

---
 drivers/staging/media/ipu3/ipu3-css-fw.c |    7 +++----
 drivers/staging/media/ipu3/ipu3-css-fw.h |    2 +-
 2 files changed, 4 insertions(+), 5 deletions(-)

@@ -124,12 +124,11 @@ int imgu_css_fw_init(struct imgu_css *cs
 	/* Check and display fw header info */
 
 	css->fwp = (struct imgu_fw_header *)css->fw->data;
-	if (css->fw->size < sizeof(struct imgu_fw_header *) ||
+	if (css->fw->size < struct_size(css->fwp, binary_header, 1) ||
 	    css->fwp->file_header.h_size != sizeof(struct imgu_fw_bi_file_h))
 		goto bad_fw;
-	if (sizeof(struct imgu_fw_bi_file_h) +
-	    css->fwp->file_header.binary_nr * sizeof(struct imgu_fw_info) >
-	    css->fw->size)
+	if (struct_size(css->fwp, binary_header,
+			css->fwp->file_header.binary_nr) > css->fw->size)
 		goto bad_fw;
 
 	dev_info(dev, "loaded firmware version %.64s, %u binaries, %zu bytes\n",

@@ -170,7 +170,7 @@ struct imgu_fw_bi_file_h {
 
 struct imgu_fw_header {
 	struct imgu_fw_bi_file_h file_header;
-	struct imgu_fw_info binary_header[1];	/* binary_nr items */
+	struct imgu_fw_info binary_header[];	/* binary_nr items */
 };
 
 /******************* Firmware functions *******************/

From: Geert Uytterhoeven <[email protected]>

commit a0548b26901f082684ad1fb3ba397d2de3a1406a upstream.

On 64-bit:

drivers/usb/gadget/udc/fsl_qe_udc.c: In function ‘qe_ep0_rx’:
drivers/usb/gadget/udc/fsl_qe_udc.c:842:13: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
842 | vaddr = (u32)phys_to_virt(in_be32(&bd->buf));
| ^
In file included from drivers/usb/gadget/udc/fsl_qe_udc.c:41:
drivers/usb/gadget/udc/fsl_qe_udc.c:843:28: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
843 | frame_set_data(pframe, (u8 *)vaddr);
| ^

The driver assumes physical and virtual addresses are 32-bit, hence it cannot work on 64-bit platforms.

---
 drivers/usb/gadget/udc/Kconfig | 1 +
 1 file changed, 1 insertion(+)

index 8c614bb86c66..69394dc1cdfb 100644
@@ -330,6 +330,7 @@ config USB_AMD5536UDC
 config USB_FSL_QE
 	tristate "Freescale QE/CPM USB Device Controller"
 	depends on FSL_SOC && (QUICC_ENGINE || CPM)
+	depends on !64BIT || BROKEN
 	help
 	   Some of Freescale PowerPC processors have a Full Speed
 	   QE/CPM2 USB controller, which support device mode with 4

2.33.1

From: Viraj Shah <[email protected]>

commit 21b5fcdccb32ff09b6b63d4a83c037150665a83f upstream.

musb_gadget_queue() adds the passed request to musb_ep::req_list. If the endpoint is idle and it is the first request then it invokes musb_queue_resume_work(). If the function returns an error then the error is passed to the caller without any clean-up and the request remains enqueued on the list. If the caller enqueues the request again then the list corrupts.

Remove the request from the list on error.

---
 drivers/usb/musb/musb_gadget.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

@@ -1247,9 +1247,11 @@ static int musb_gadget_queue(struct usb_
 		status = musb_queue_resume_work(musb,
 						musb_ep_restart_resume_work,
 						request);
-		if (status < 0)
+		if (status < 0) {
 			dev_err(musb->controller, "%s resume work: %i\n",
 				__func__, status);
+			list_del(&request->list);
+		}
 	}
 
 unlock:

From: James Buren <[email protected]>

commit 05c8f1b67e67dcd786ae3fe44492bbc617b4bd12 upstream.

These drive enclosures have firmware bugs that make it impossible to mount a new virtual ISO image after Linux ejects the old one if the device is locked by Linux. Windows bypasses this problem by the fact that they do not lock the device. Add a quirk to disable device locking for these drive enclosures.

---
 drivers/usb/storage/unusual_devs.h |   10 ++++++++++
 1 file changed, 10 insertions(+)

@@ -407,6 +407,16 @@ UNUSUAL_DEV(  0x04b8, 0x0602, 0x0110, 0x
 		USB_SC_SCSI, USB_PR_BULK, NULL, US_FL_SINGLE_LUN),
 
 /*
+ * Reported by James Buren <[email protected]>
+ * Virtual ISOs cannot be remounted if ejected while the device is locked
+ * Disable locking to mimic Windows behavior that bypasses the issue
+ */
+UNUSUAL_DEV(  0x04c5, 0x2028, 0x0001, 0x0001,
+		"iODD",
+		"2531/2541",
+		USB_SC_DEVICE, USB_PR_DEVICE, NULL, US_FL_NOT_LOCKABLE),
+
+/*
  * Not sure who reported this originally but
  * Pavel Machek <[email protected]> reported that the extra US_FL_SINGLE_LUN
  * flag be added */

From: Todd Kjos <[email protected]>

commit 32e9f56a96d8d0f23cb2aeb2a3cd18d40393e787 upstream.

When freeing txn buffers, binder_transaction_buffer_release() attempts to detect whether the current context is the target by comparing current->group_leader to proc->tsk. This is an unreliable test. Instead explicitly pass an 'is_failure' boolean.

Detecting the sender was being used as a way to tell if the transaction failed to be sent. When cleaning up after failing to send a transaction, there is no need to close the fds associated with a BINDER_TYPE_FDA object. Now 'is_failure' can be used to accurately detect this case.

---
 drivers/android/binder.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

@@ -2254,7 +2254,7 @@ static void binder_transaction_buffer_re
 		binder_dec_node(buffer->target_node, 1, 0);
 
 	off_start_offset = ALIGN(buffer->data_size, sizeof(void *));
-	off_end_offset = is_failure ? failed_at :
+	off_end_offset = is_failure && failed_at ? failed_at :
 				off_start_offset + buffer->offsets_size;
 	for (buffer_offset = off_start_offset; buffer_offset < off_end_offset;
 	     buffer_offset += sizeof(binder_size_t)) {

@@ -2340,9 +2340,8 @@ static void binder_transaction_buffer_re
 			binder_size_t fd_buf_size;
 			binder_size_t num_valid;
 
-			if (proc->tsk != current->group_leader) {
+			if (is_failure) {
 				/*
-				 * Nothing to do if running in sender context
 				 * The fd fixups have not been applied so no
 				 * fds need to be closed.
 				 */

@@ -3544,6 +3543,7 @@ err_invalid_target_handle:
  * binder_free_buf() - free the specified buffer
  * @proc:	binder proc that owns buffer
  * @buffer:	buffer to be freed
+ * @is_failure:	failed to send transaction
  *
  * If buffer for an async transaction, enqueue the next async
  * transaction from the node.

@@ -3553,7 +3553,7 @@ err_invalid_target_handle:
 static void
 binder_free_buf(struct binder_proc *proc,
 		struct binder_thread *thread,
-		struct binder_buffer *buffer)
+		struct binder_buffer *buffer, bool is_failure)
 {
 	binder_inner_proc_lock(proc);
 	if (buffer->transaction) {

@@ -3581,7 +3581,7 @@ binder_free_buf(struct binder_proc *proc
 		binder_node_inner_unlock(buf_node);
 	}
 	trace_binder_transaction_buffer_release(buffer);
-	binder_transaction_buffer_release(proc, thread, buffer, 0, false);
+	binder_transaction_buffer_release(proc, thread, buffer, 0, is_failure);
 	binder_alloc_free_buf(&proc->alloc, buffer);
 }
 

@@ -3782,7 +3782,7 @@ static int binder_thread_write(struct bi
 				     proc->pid, thread->pid, (u64)data_ptr,
 				     buffer->debug_id,
 				     buffer->transaction ? "active" : "finished");
-			binder_free_buf(proc, thread, buffer);
+			binder_free_buf(proc, thread, buffer, false);
 			break;
 		}
 

@@ -4470,7 +4470,7 @@ retry:
 			buffer->transaction = NULL;
 			binder_cleanup_transaction(t, "fd fixups failed",
 						   BR_FAILED_REPLY);
-			binder_free_buf(proc, thread, buffer);
+			binder_free_buf(proc, thread, buffer, true);
 			binder_debug(BINDER_DEBUG_FAILED_TRANSACTION,
 				     "%d:%d %stransaction %d fd fixups failed %d/%d, line %d\n",
 				     proc->pid, thread->pid,

This is the start of the stable review cycle for the 5.10.79 release. There are 21 patches in this series, all will be posted as a response to this one. If anyone has any issues with these being applied, please let me know.

Responses should be made by Fri, 12 Nov 2021 18:19:54 +0000. Anything received after that time might be too late.

The whole patch series can be found in one patch at:

	https://www.kernel.org/pub/linux/kernel/v5.x/stable-review/patch-5.10.79-rc1.gz

or in the git tree and branch at:
git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git linux-5.10.y
and the diffstat can be found below.

thanks,

greg k-h

-------------
Pseudo-Shortlog of commits:

Greg Kroah-Hartman <[email protected]>

    Linux 5.10.79-rc1

Johan Hovold <[email protected]>
rsi: fix control-message timeout

Gustavo A. R. Silva <[email protected]>
media: staging/intel-ipu3: css: Fix wrong size comparison imgu_css_fw_init

Johan Hovold <[email protected]>
staging: rtl8192u: fix control-message timeouts

Johan Hovold <[email protected]>
staging: r8712u: fix control-message timeout

Johan Hovold <[email protected]>

    comedi: vmk80xx: fix bulk and interrupt message timeouts

Johan Hovold <[email protected]>

    comedi: vmk80xx: fix bulk-buffer overflow

Johan Hovold <[email protected]>

    comedi: vmk80xx: fix transfer-buffer overflows

Johan Hovold <[email protected]>
comedi: ni_usb6501: fix NULL-deref in command paths

Johan Hovold <[email protected]>
comedi: dt9812: fix DMA buffers on stack

Jan Kara <[email protected]>
isofs: Fix out of bound access for corrupted isofs image

Pavel Skripkin <[email protected]>

    staging: rtl8712: fix use-after-free in rtl8712_dl_fw

Petr Mladek <[email protected]>
printk/console: Allow to disable console output by using console="" or console=null

Todd Kjos <[email protected]>
binder: don't detect sender/target during buffer cleanup

James Buren <[email protected]>
usb-storage: Add compatibility quirk flags for iODD 2531/2541

Viraj Shah <[email protected]>
usb: musb: Balance list entry in musb_gadget_queue

Geert Uytterhoeven <[email protected]>
usb: gadget: Mark USB_FSL_QE broken on 64-bit

Yang Shi <[email protected]>
mm: filemap: check if THP has hwpoisoned subpage for PMD page fault

Yang Shi <[email protected]>
mm: hwpoison: remove the unnecessary THP check

Neal Liu <[email protected]>
usb: ehci: handshake CMD_RUN instead of STS_HALT

Juergen Gross <[email protected]>
Revert "x86/kvm: fix vcpu-id indexed array sizes"

Paolo Bonzini <[email protected]>
KVM: x86: avoid warning with -Wbitwise-instead-of-logical

-------------

Diffstat:

 Makefile                                    |   4 +-
 arch/x86/kvm/ioapic.c                       |   2 +-
 arch/x86/kvm/ioapic.h                       |   4 +-
 arch/x86/kvm/mmu/mmu.c                      |   2 +-
 drivers/android/binder.c                    |  14 ++--
 drivers/net/wireless/rsi/rsi_91x_usb.c      |   2 +-
 drivers/staging/comedi/drivers/dt9812.c     | 115 +++++++++++++++++++++-------
 drivers/staging/comedi/drivers/ni_usb6501.c |  10 +++
 drivers/staging/comedi/drivers/vmk80xx.c    |  28 +++----
 drivers/staging/media/ipu3/ipu3-css-fw.c    |   7 +-
 drivers/staging/media/ipu3/ipu3-css-fw.h    |   2 +-
 drivers/staging/rtl8192u/r8192U_core.c      |  18 ++---
 drivers/staging/rtl8712/usb_intf.c          |   4 +-
 drivers/staging/rtl8712/usb_ops_linux.c     |   2 +-
 drivers/usb/gadget/udc/Kconfig              |   1 +
 drivers/usb/host/ehci-hcd.c                 |  11 ++-
 drivers/usb/host/ehci-platform.c            |   6 ++
 drivers/usb/host/ehci.h                     |   1 +
 drivers/usb/musb/musb_gadget.c              |   4 +-
 drivers/usb/storage/unusual_devs.h          |  10 +++
 fs/isofs/inode.c                            |   2 +
 include/linux/page-flags.h                  |  23 ++++++
 kernel/printk/printk.c                      |   9 ++-
 mm/huge_memory.c                            |   2 +
 mm/memory-failure.c                         |  28 +++----
 mm/memory.c                                 |   9 +++
 mm/page_alloc.c                             |   4 +-
 27 files changed, 233 insertions(+), 91 deletions(-)

From: Johan Hovold <[email protected]>

commit 541fd20c3ce5b0bc39f0c6a52414b6b92416831c upstream.

USB control-message timeouts are specified in milliseconds and should specifically not vary with CONFIG_HZ.

Use the common control-message timeout define for the five-second timeout.

---
 drivers/net/wireless/rsi/rsi_91x_usb.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

@@ -61,7 +61,7 @@ static int rsi_usb_card_write(struct rsi
 			      (void *)seg,
 			      (int)len,
 			      &transfer,
-			      HZ * 5);
+			      USB_CTRL_SET_TIMEOUT);
 
 	if (status < 0) {
 		rsi_dbg(ERR_ZONE,

From: Yang Shi <[email protected]>

commit eac96c3efdb593df1a57bb5b95dbe037bfa9a522 upstream.

When handling shmem page fault the THP with corrupted subpage could be PMD mapped if certain conditions are satisfied. But kernel is supposed to send SIGBUS when trying to map hwpoisoned page.

There are two paths which may do PMD map: fault around and regular fault.

Before commit f9ce0be71d1f ("mm: Cleanup faultaround and finish_fault() codepaths") the thing was even worse in fault around path. The THP could be PMD mapped as long as the VMA fits regardless what subpage is accessed and corrupted. After this commit as long as head page is not corrupted the THP could be PMD mapped.

In the regular fault path the THP could be PMD mapped as long as the corrupted page is not accessed and the VMA fits.

This loophole could be fixed by iterating every subpage to check if any of them is hwpoisoned or not, but it is somewhat costly in page fault path.

So introduce a new page flag called HasHWPoisoned on the first tail page. It indicates the THP has hwpoisoned subpage(s). It is set if any subpage of THP is found hwpoisoned by memory failure and after the refcount is bumped successfully, then cleared when the THP is freed or split.

The soft offline path doesn't need this since soft offline handler just marks a subpage hwpoisoned when the subpage is migrated successfully. But shmem THP didn't get split then migrated at all.

---
 include/linux/page-flags.h |   23 +++++++++++++++++++++++
 mm/huge_memory.c           |    2 ++
 mm/memory-failure.c        |   14 ++++++++++++++
 mm/memory.c                |    9 +++++++++
 mm/page_alloc.c            |    4 +++-
 5 files changed, 51 insertions(+), 1 deletion(-)

@@ -169,6 +169,15 @@ enum pageflags {
 	/* Compound pages. Stored in first tail page's flags */
 	PG_double_map = PG_workingset,
 
+#ifdef CONFIG_MEMORY_FAILURE
+	/*
+	 * Compound pages. Stored in first tail page's flags.
+	 * Indicates that at least one subpage is hwpoisoned in the
+	 * THP.
+	 */
+	PG_has_hwpoisoned = PG_mappedtodisk,
+#endif
+
 	/* non-lru isolated movable page */
 	PG_isolated = PG_reclaim,
 

@@ -701,6 +710,20 @@ PAGEFLAG_FALSE(DoubleMap)
 	TESTSCFLAG_FALSE(DoubleMap)
 #endif
 
+#if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
+/*
+ * PageHasHWPoisoned indicates that at least one subpage is hwpoisoned in the
+ * compound page.
+ *
+ * This flag is set by hwpoison handler.  Cleared by THP split or free page.
+ */
+PAGEFLAG(HasHWPoisoned, has_hwpoisoned, PF_SECOND)
+	TESTSCFLAG(HasHWPoisoned, has_hwpoisoned, PF_SECOND)
+#else
+PAGEFLAG_FALSE(HasHWPoisoned)
+	TESTSCFLAG_FALSE(HasHWPoisoned)
+#endif
+
 /*
  * For pages that are never mapped to userspace (and aren't PageSlab),
  * page_type may be used.  Because it is initialised to -1, we invert the

@@ -2464,6 +2464,8 @@ static void __split_huge_page(struct pag
 		xa_lock(&swap_cache->i_pages);
 	}
 
+	ClearPageHasHWPoisoned(head);
+
 	for (i = nr - 1; i >= 1; i--) {
 		__split_huge_page_tail(head, i, lruvec, list);
 		/* Some pages can be beyond i_size: drop them from page cache */

@@ -1367,6 +1367,20 @@ int memory_failure(unsigned long pfn, in
 	}
 
 	if (PageTransHuge(hpage)) {
+		/*
+		 * The flag must be set after the refcount is bumped
+		 * otherwise it may race with THP split.
+		 * And the flag can't be set in get_hwpoison_page() since
+		 * it is called by soft offline too and it is just called
+		 * for !MF_COUNT_INCREASE.  So here seems to be the best
+		 * place.
+		 *
+		 * Don't need care about the above error handling paths for
+		 * get_hwpoison_page() since they handle either free page
+		 * or unhandlable page.  The refcount is bumped iff the
+		 * page is a valid handlable page.
+		 */
+		SetPageHasHWPoisoned(hpage);
 		if (try_to_split_thp_page(p, "Memory Failure") < 0) {
 			action_result(pfn, MF_MSG_UNSPLIT_THP, MF_IGNORED);
 			return -EBUSY;

@@ -3921,6 +3921,15 @@ vm_fault_t finish_fault(struct vm_fault
 		page = vmf->page;
 
 	/*
+	 * Just backoff if any subpage of a THP is corrupted otherwise
+	 * the corrupted page may mapped by PMD silently to escape the
+	 * check.  This kind of THP just can be PTE mapped.  Access to
+	 * the corrupted subpage should trigger SIGBUS as expected.
+	 */
+	if (unlikely(PageHasHWPoisoned(page)))
+		return ret;
+
+	/*
 	 * check even for read faults because we might have lost our CoWed
 	 * page
 	 */

@@ -1232,8 +1232,10 @@ static __always_inline bool free_pages_p
 
 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
 
-		if (compound)
+		if (compound) {
 			ClearPageDoubleMap(page);
+			ClearPageHasHWPoisoned(page);
+		}
 		for (i = 1; i < (1 << order); i++) {
 			if (compound)
 				bad += free_tail_pages_check(page, page + i);

From: Yang Shi <[email protected]>

commit c7cb42e94473aafe553c0f2a3d8ca904599399ed upstream.

When handling THP hwpoison checked if the THP is in allocation or free stage since hwpoison may mistreat it as hugetlb page. After commit 415c64c1453a ("mm/memory-failure: split thp earlier in memory error handling") the problem has been fixed, so this check is no longer needed. Remove it. The side effect of the removal is hwpoison may report unsplit THP instead of unknown error for shmem THP. It seems not like a big deal.

The following patch "mm: filemap: check if THP has hwpoisoned subpage for PMD page fault" depends on this, which fixes shmem THP with hwpoisoned subpage(s) are mapped PMD wrongly. So this patch needs to be backported to -stable as well.

---
 mm/memory-failure.c |   14 --------------
 1 file changed, 14 deletions(-)

@@ -956,20 +956,6 @@ static int get_hwpoison_page(struct page
 {
 	struct page *head = compound_head(page);
 
-	if (!PageHuge(head) && PageTransHuge(head)) {
-		/*
-		 * Non anonymous thp exists only in allocation/free time. We
-		 * can't handle such a case correctly, so let's give it up.
-		 * This should be better than triggering BUG_ON when kernel
-		 * tries to touch the "partially handled" page.
-		 */
-		if (!PageAnon(head)) {
-			pr_err("Memory failure: %#lx: non anonymous thp\n",
-				page_to_pfn(page));
-			return 0;
-		}
-	}
-
 	if (get_page_unless_zero(head)) {
 		if (head == compound_head(page))
 			return 1;

On ARCH_BRCMSTB, using 32-bit and 64-bit ARM kernels:

Florian

5.10.79-rc1 Successfully Compiled and booted on my Raspberry PI 4b (8g) (bcm2711)

Hi Greg,

systemd-journal-flush.service failed due to a timeout resulting in a very very slow boot on my test laptop. qemu test on openqa failed due to the same problem.

https://openqa.qa.codethink.co.uk/tests/365

A bisect showed the problem to be 8615ff6dd1ac ("mm: filemap: check if THP has hwpoisoned subpage for PMD page fault"). Reverting it on top of 5.10.79-rc1 fixed the problem.
Incidentally, I was having similar problem with Linus's tree
for last few days and was failing since 20211106 (did not get the time to check). I will test mainline again with this commit reverted.

--
Regards
Sudip

I have also noticed this problem and Anders bisected and found this first bad commit.

Failed test log link,
A start job is running for Journal Service (5s / 1min 27s)
https://lkft.validation.linaro.org/scheduler/job/3901980#L2234

Bisect log:

# bad: [b85617a6291f710807d0cd078c230626dee60b16] Linux 5.10.79-rc1 # good: [5040520482a594e92d4f69141229a6dd26173511] Linux 5.10.78 git bisect start 'b85617a6291f710807d0cd078c230626dee60b16' '5040520482a594e92d4f69141229a6dd26173511'
# bad: [7ceeda856035991a6c9804916987a03759745fb0] staging: rtl8712:
fix use-after-free in rtl8712_dl_fw
git bisect bad 7ceeda856035991a6c9804916987a03759745fb0
# bad: [8615ff6dd1ac9e01b6fcf0fc0652353f79f524ed] mm: filemap: check if THP has hwpoisoned subpage for PMD page fault
git bisect bad 8615ff6dd1ac9e01b6fcf0fc0652353f79f524ed
# good: [e9cb6ce4690749d42013f1d56874c624d7241740] Revert "x86/kvm:
fix vcpu-id indexed array sizes"
git bisect good e9cb6ce4690749d42013f1d56874c624d7241740 # good: [dc385dfc126d51d7a93db694f8e151afe60eb06a] mm: hwpoison:
remove the unnecessary THP check
git bisect good dc385dfc126d51d7a93db694f8e151afe60eb06a

# first bad commit: [8615ff6dd1ac9e01b6fcf0fc0652353f79f524ed] mm:
filemap: check if THP has hwpoisoned subpage for PMD page fault
commit 8615ff6dd1ac9e01b6fcf0fc0652353f79f524ed
Author: Yang Shi <[email protected]>
Date:   Thu Oct 28 14:36:11 2021 -0700

mm: filemap: check if THP has hwpoisoned subpage for PMD page fault

commit eac96c3efdb593df1a57bb5b95dbe037bfa9a522 upstream.

When handling shmem page fault the THP with corrupted subpage could be PMD mapped if certain conditions are satisfied. But kernel is supposed to send SIGBUS when trying to map hwpoisoned page.

There are two paths which may do PMD map: fault around and regular fault.

Before commit f9ce0be71d1f ("mm: Cleanup faultaround and finish_fault() codepaths") the thing was even worse in fault around path. The THP could be PMD mapped as long as the VMA fits regardless what subpage is accessed and corrupted. After this commit as long as head page is not corrupted the THP could be PMD mapped.

In the regular fault path the THP could be PMD mapped as long as the corrupted page is not accessed and the VMA fits.

This loophole could be fixed by iterating every subpage to check if any of them is hwpoisoned or not, but it is somewhat costly in page fault path.

So introduce a new page flag called HasHWPoisoned on the first tail page. It indicates the THP has hwpoisoned subpage(s). It is set if any subpage of THP is found hwpoisoned by memory failure and after the refcount is bumped successfully, then cleared when the THP is freed or split.

The soft offline path doesn't need this since soft offline handler just marks a subpage hwpoisoned when the subpage is migrated successfully. But shmem THP didn't get split then migrated at all.

 include/linux/page-flags.h | 23 +++++++++++++++++++++++
 mm/huge_memory.c           |  2 ++
 mm/memory-failure.c        | 14 ++++++++++++++
 mm/memory.c                |  9 +++++++++
 mm/page_alloc.c            |  4 +++-
 5 files changed, 51 insertions(+), 1 deletion(-)

--
Linaro LKFT
https://lkft.linaro.org

Compiled and booted on my test system.

dmesg regressions. It took a very long time in trying to start Journal services and finally timed out. Pervious boot was on 5.14.18-rc1 both boot and shutdown were clean.

thanks,
-- Shuah

Hi!

CIP testing did not find any problems here:

https://gitlab.com/cip-project/cip-testing/linux-stable-rc-ci/-/tree/linux-5.10.y

Best regards,
Pavel

DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany

Build test:
mips (gcc version 11.2.1 20211104): 63 configs -> no new failure arm (gcc version 11.2.1 20211104): 105 configs -> no new failure arm64 (gcc version 11.2.1 20211104): 3 configs -> no failure x86_64 (gcc version 11.2.1 20211104): 4 configs -> no failure

Boot test:
x86_64: Regression mail sent earlier. Caused by 8615ff6dd1ac ("mm:
filemap: check if THP has
hwpoisoned subpage for PMD page fault").

arm64: Booted on rpi4b (4GB model). No regression. [1]

[1]. https://openqa.qa.codethink.co.uk/tests/362

--
Regards
Sudip

Reverting mm: filemap: check if THP has hwpoisoned subpage for PMD page fault" worked for me on my test system.

With this commit boots are long and shutdown was at the 20+ minute m ark when I powered it down. This commit isn't in any of the other release candidates.

thanks,
-- Shuah

Build results:

	total: 159 pass: 159 fail: 0
Qemu test results:
	total: 474 pass: 469 fail: 5
Failed tests:
	ppc64:powernv:powernv_defconfig:smp2:nvme:net,i82559a:rootfs
	ppc64:powernv:powernv_defconfig:usb-xhci:net,i82562:rootfs
	ppc64:powernv:powernv_defconfig:scsi[MEGASAS]:net,i82557a:rootfs
	ppc64:powernv:powernv_defconfig:smp2:sdhci:mmc:net,i82801:rootfs
	ppc64:powernv:powernv_defconfig:mtd32:net,rtl8139:rootfs

Reverting commit 8615ff6dd1ac ("mm: filemap: check if THP has hwpoisoned subpage for PMD page fault") fixes the problem.

Guenter

Ugh, ok, I'm going to drop this patch (and the one before it) again.

thanks for the testing.

greg k-h

Thanks, will go drop this commit.

greg k-h

Will go drop the offending patch, thanks.

greg k-h

Thanks, I'm going to go drop this patch again.

This has been the second time we have tried to add it. Yang, are you _SURE_ it needs to be in the 5.10.y tree? So far it's been nothing but build and boot failures :(

thanks,

greg k-h

All tests passing for Tegra ...

Test results for stable-v5.10:

    10 builds:	10 pass, 0 fail
    28 boots:	28 pass, 0 fail
    75 tests:	75 pass, 0 fail

Linux version: 5.10.79-rc1-gb85617a6291f
Boards tested: tegra124-jetson-tk1, tegra186-p2771-0000,

                tegra194-p2972-0000, tegra194-p3509-0000+p3668-0000,
                tegra20-ventana, tegra210-p2371-2180,
                tegra210-p3450-0000, tegra30-cardhu-a04

Jon